DETERRENCE, PUNISHMENT SEVERITY AND DRINK-DRIVING SUZANNE BRISCOE

Transcription

1 DETERRENCE, PUNISHMENT SEVERITY AND DRINK-DRIVING SUZANNE BRISCOE A thesis submitted in fulfilment Of the requirements for the degree of Doctor of Philosophy School of Social Science and Policy University of New South Wales August, 2005

2 ABSTRACT This thesis tests one of the major propositions of deterrence theory: that increases in the severity of criminal punishment can reduce the likelihood of offending. To this end, a case study in which the statutory penalties were doubled for almost all drinkdriving offences in New South Wales, Australia, is examined. This case study was considered a good test of deterrence hypotheses regarding the efficacy of punishment severity because it satisfied several preconditions necessary for optimising deterrent effects. Two quasi-experimental studies were undertaken to assess the impact of these legislative changes on drink-driving offending levels: an interrupted time-series analysis of road crash rates (Study 1) and an analysis of drink-driving reoffending rates before and after the penalty changes were implemented (Study 2). Study 1 found a significant increase in monthly single-vehicle night-time crash rates (a surrogate measure of alcohol-related road crashes) after the tougher drink-driving penalties were introduced in New South Wales. The analyses also revealed a significant increase, around the same time, in non alcohol-related crashes (as measured by multiple-vehicle day-time crashes). Further exploratory analyses indicated that the rate of increase observed after the penalty changes, was greater for multiple-vehicle day-time crashes than for single-vehicle night-time crashes. This suggests that the significant increase in the surrogate measure of alcohol-related crashes may have been driven primarily by a rise in non alcohol-related crashes that coincided with the policy s implementation. Two possible conclusions regarding the deterrent effect of the tougher penalties are drawn from these findings: (1) that there was a reduction in alcohol-related road crashes but this effect was overwhelmed by the rise in non alcohol-related crashes occurring around the same time or (2) that there was no reduction in crash rates after the tougher penalties were introduced. To test deterrence hypotheses further, the reoffending rates of two drink-driver cohorts were compared: one consisting of all drink-drivers convicted in NSW Local Courts in the year before the penalty changes and the other consisting of all drink- ii

3 drivers convicted in the year after. Logistic regression and survival analyses demonstrated that those drink-drivers who were convicted under the new penalty regime were less likely, and took longer, to reappear before the courts for a new drink-driving offence than drink-drivers convicted before the introduction of the new penalties. This reduction in reoffending was, however, only apparent for drinkdrivers residing in country and regional areas and was small in magnitude. While these findings are consistent with the possibility that the penalty changes resulted in a reduction in alcohol-related road crashes which could not be detected in Study 1, they suggest that any decrease is likely to have been relatively small. An examination of the context in which the drink-driving sentencing policy was implemented identified two potential explanations for the small deterrent effect observed in the quasi-experimental research. Firstly, NSW Police appeared to have altered their enforcement practices around the time that the legislative amendments were introduced. These changes may have reduced the perceived risk of detection amongst some drink-drivers and therefore undermined the deterrent effect of the increase in punishment severity. Secondly, the tougher drink-driving penalties received limited publicity, thus many drivers may have been unaware of the changes to drink-driving punishment. To further explore the deterrent effect of increases in punishment severity a third study using a scenario-based survey methodology was undertaken. This work builds on Study 1 and Study 2 by examining the relationship between legal sanctions and willingness to drink-drive, in an experimental setting where it was possible to control for other contextual influences. The results of this study showed that participants who made more mistakes in a test of knowledge about drink-driving penalties in New South Wales stated that they were more likely to offend in the drink-driving scenario than were participants who made fewer mistakes, controlling for other factors. This would suggest that were the changes to drink-driving penalties (examined in the case study) communicated more effectively to the target audience then there may have been a greater reduction in drink-driving offending levels in New South Wales. However, these conclusions are tentative given that there was only weak evidence for a relationship between the number of mistakes on the iii

4 knowledge test and participants ratings of penalty severity and, further, that no significant interaction was found between perceptions of punishment severity and perceptions of punishment certainty. The sentencing policy examined in this thesis presented a good opportunity to test the deterrent efficacy of tougher penalties in a real world situation where several necessary conditions were in place for optimising deterrent effects. However, not all relevant contextual factors could be controlled for, and therefore definitive conclusions about the deterrent effect of the penalty changes could not be drawn from its results. Despite these limitations, the small effect of the sentencing changes on offending levels observed in this research, in conjunction with the weak evidence for punishment severity effects found in the scenario-based survey, is consistent with much of the available evidence on deterrence. The implications of these findings for deterrence theory, criminal justice policy and future research in this area are discussed. iv

5 Acknowledgements I would like to acknowledge my thesis supervisors, Professor Janet Chan and Dr Don Weatherburn, for their continual support and guidance throughout this project. In particular, I would like to thank Don for encouraging me to embark on this journey, for allowing me the opportunity to undertake this research in a supportive and resourceful work environment and for always making time. I would also like to thank several staff from the NSW Bureau of Crime Statistics and Research (both past and present) who have helped me at various stages throughout this research. In particular, thanks go to Neil Donnelly, Marilyn Chilvers, Shuling Chen and Bronwyn Lind for their invaluable advice on the statistical methods used in these analyses and also Jiuzhao Hua, whose ingenious work on the Bureau s Reoffending Database made the recidivism analysis reported in this thesis possible. The NSW Roads and Traffic Authority should also be acknowledged for providing the road crash data for the interrupted time-series analyses and for supplying offence data to validate the accuracy of the prior offending variable included in the Bureau s Reoffending Database. Finally, I would like to thank my family and friends who endured much of this journey with me and who were extremely supportive throughout. In particular, I would like to thank my parents who proofread this thesis in record time and especially my husband-to-be for his support and continual encouragement, for listening to my ideas and entertaining my theories, and for all the lost weekends. v

6 Previous publications Some of the analyses appearing in this thesis have previously been published. A paper describing the results from the interrupted time-series analysis of road accident data (Study 1) was published in the international, peer-reviewed journal Accident, Analysis and Prevention in 2004 (Briscoe 2004a). The recidivism analysis, comparing the reoffending rates of drink-drivers convicted in 1997 with those convicted in 1999 (Study 2) was also published in 2004 as an Alcohol Studies Bulletin for the NSW Bureau of Crime Statistics and Research (Briscoe 2004b). All the work reported in this thesis was completed during my PhD candidature at the University of New South Wales and received approval from the Human Research Ethics Committee at this institution. vi

7 TABLE OF CONTENTS Page Number ABSTRACT... II Acknowledgements... v Previous publications...vi CHAPTER 1. INTRODUCTION TO THESIS AND OVERVIEW... 1 History of deterrence theory... 2 General tests of deterrence theory... 3 Deterrence and the certainty of formal punishment... 4 Deterrence and the severity of formal punishment... 7 Limitations of aggregate-level deterrence research Perceptual research in deterrence The current study Thesis structure CHAPTER 2. DETERRENCE THEORY Definitions of deterrence Modern deterrence theory Expanding the Subjective Expected Utility Model Critiques of the Rational-Choice Perspective Deterrence and specific offences The role of informal social control Further interaction effects in deterrence Individual characteristics and sanction perceptions Perceived risk of apprehension and perceptions of punishment severity Drink-driving as a case study CHAPTER 3. DETERRENCE RESEARCH ON DRINK-DRIVING Aggregate-level research Punishment certainty Punishment severity Punishment celerity vii

8 Individual-level research Reoffending studies Perceptual studies Significance of the current study Methodological issues in policy impact studies Knowledge gaps in the deterrence literature CHAPTER 4. INCREASED DRINK-DRIVING PENALTIES IN NEW SOUTH WALES: A CASE STUDY Drink-driving legislation in New South Wales and the 1998 penalty changes The case study Measuring policy effectiveness - The quasi-experiment Alcohol-related road crashes - Study Drink-driving recidivism rates - Study Perceptual research - Study Hypotheses Participants and methodology CHAPTER 5. IMPACT OF THE 1998 PENALTY CHANGES ON OFFENDING RATES IN NEW SOUTH WALES Study 1 - Road crash analysis All road crashes resulting in injury Alcohol-related road crash rates Control series Summary Alcohol-related crashes versus control series Preliminary conclusions Study 2 - Analysis of recidivism rates Implementation of statutory amendments Characteristics of offender cohorts Probability of reappearing for a drink-driving offence Time to first drink-driving offence Summary and discussion of recidivism analysis Conclusions from the quasi-experimental studies CHAPTER 6. CONTEXT OF POLICY IMPLEMENTATION Enforcement of drink-driving offences Police breath tests and public perceptions of enforcement activity Police targeting of high-risk times and locations Media coverage of statutory penalty increase viii

9 Newspaper coverage of penalty changes RTA advertising campaigns and other publicity What went wrong? CHAPTER 7. KNOWLEDGE OF PENALTIES, PERCEPTIONS OF SANCTION SEVERITY AND DRINK-DRIVING Formation of risk perceptions Legal knowledge as a factor in the deterrence process The scenario-based study - Study Hypotheses Results of Study Sample demographics Independent variables Predicting stated likelihood of offending Linear regression models including knowledge and certainty variables Testing the assumptions of linear regression Summary and discussion of survey findings Limitations of Study CHAPTER 8. SUMMARY AND DISCUSSION The deterrent effect of increased drink-driving penalties in New South Wales Changes in drink-driving behaviour after the penalty increase Implementation of the 1998 sentencing policy Knowledge of drink-driving penalties and perceptions of sanction severity Implications for theory and practice Translating policy into a credible threat The marginal deterrent efficacy of punishment severity Future directions for deterrence research and drink-driving policy Future research Policy recommendations REFERENCES APPENDICES Appendix I Summary of drink-driving studies reviewed in Chapter Appendix II Participant information sheet and drink-driving questionnaire ix

10 Appendix III Supplementary analyses x

11 CHAPTER 1. INTRODUCTION TO THESIS AND OVERVIEW Deterrence is one of the major aims of punishments prescribed by legislatures and is a prominent feature of many criminal justice interventions designed to curtail criminal behaviour. It assumes that crime is a rational decision-making process in which potential offenders weigh the costs and benefits associated with an illegal act. If the anticipated costs are greater than the anticipated rewards, individuals will choose not to engage in the prohibited behaviour. It is an appealing penal ideology because it makes sense that people will want to avoid unpleasant consequences and therefore continues to inform public opinion on crime control and, at the same time, play an influential role in public policy. Just how effective deterrence is, however, as a strategy to reduce the incidence of crime still remains somewhat unclear. The question Can punishment deter crime? is one that has been the focus of much criminological research conducted over the last three or four decades, but is not one that can be answered categorically. The term crime covers a wide range of behaviours. The nature of these criminal behaviours, as well as the context in which they are committed, varies substantially. The threat of formal punishment is unlikely to be equally effective in deterring all forms of crime, under all conditions, for all offenders. This differential effectiveness of legal sanctions does not necessarily imply that deterrence theory is irrelevant to offending behaviour, nor does it suggest that deterrence is an ineffective policy tool. It does mean, however, that empirical research on deterrence must consider variations in circumstance that could account for differences in the effectiveness of legal threats. This thesis is primarily concerned with the deterrent effect of the severity of legal punishment. It draws upon theory and research on the effectiveness of formal sanctions, to identify the conditions under which deterrent effects are most likely to be observed. A case study, in which several of these conditions exist, is then presented and the deterrent efficacy of an increase in the severity of statutory penalties is examined. By examining changes in sanction severity within this setting, 1

12 the research provides a valid test of deterrence hypotheses and addresses important limitations in previous research conducted in this area. As a result, this work will lead to more definitive conclusions regarding deterrence and punishment severity, which will have relevance to criminal justice policy and important implications for further theory development. Before exploring these issues further, this chapter will briefly describe the evolution of deterrence theory, present an overview of the present state of knowledge in the deterrence field and discuss the current study in light of previous research findings in this area. History of deterrence theory The notion of deterrence originally appeared in the writings of the 18 th Century Utilitarian philosophers, Cesare Beccaria and Jeremy Bentham. These philosophers viewed human nature as essentially hedonistic and believed that crime can be deterred if punishment is sufficiently unpleasant. Gibbs asserts that many of the ideas of these early deterrence theorists can be reduced to the following statement: the rate for a particular type of crime varies inversely with the celerity, certainty, and severity of punishments for that type of crime (Gibbs 1975, p. 5). However, Gibbs (1975) also notes that these early ideas on deterrence were little more than a series of vague statements or what he terms a deterrence doctrine. To develop a more systematic theory of deterrence, the specific sanctions and relevant circumstances necessary to maximise deterrent effects need to be stipulated in order to determine how punishment will affect offender behaviour in any given situation. The deterrence doctrine proposed by Beccaria and Bentham has, for a long time, held a central position in the philosophy of criminal law, in penal policy and in the administration of criminal sanctions. However it wasn t until the 1960s and 70s that criminologists, sociologists and economists began to take an interest in deterrence and attempted to construct a more systematic theory of deterrence. Their interest arose out of disillusionment with other penal ideologies, in particular rehabilitation. Rehabilitation was found to be a difficult concept on which to base a criminal justice system because there was little evidence that there were rehabilitative programs that 2

13 could substantially affect recidivism (Andenaes 1971). This search for a stronger foundation on which to base criminal justice policy, as well as a growing emphasis on deterrence-based penal policies in the United States of America (USA) and the United Kingdom (UK), led to a large body of literature on deterrence and related issues (Andenaes 1971). A seminal paper, by Becker (1968), emerged during this period which became influential in the development of a more systematic theory of deterrence. It proposed that the decision to offend is a rational one and thus could be treated like any other decision made under conditions of uncertainty. Drawing upon economic theory, Becker asserted that potential offenders calculate the expected costs and benefits of an array of illegal and legal opportunities and commit an offence only if the expected benefits outweigh the expected costs. These calculations will be discussed in greater detail in Chapter 2. It is important to note here, however, that they include an evaluation by the offender of (1) the anticipated gains from committing the crime, (2) the subjective probability of obtaining those gains, (3) the anticipated pain that will arise from punishment for the crime (severity of punishment) and (4) the subjective probability that this punishment will be incurred (certainty of punishment). This view, that criminal behaviour is the outcome of a set of calculated decisions or choices, deviated from more deterministic models of criminal conduct of the time, which saw criminal conduct as the result of psychologically, socially or biologically determined dispositions to offend. In these deterministic models the offender is portrayed as a passive figure, prey to internal or external forces outside personal control (Clarke & Cornish 1985, p. 148). In contrast, the rational choice model depicts the offender as an active decision-maker who, when presented with an offending opportunity, will process and evaluate relevant information from their environment (Homel 1986). General tests of deterrence theory Since Becker s work in the late 1960s, a vast number of studies, of varying quality, have been undertaken to test deterrence hypotheses. These studies fall into one of three categories: ecological studies, quasi-experimental studies and perceptual 3

14 studies (Nagin 1998). Ecological and quasi-experimental studies generally use aggregate data to examine the effect of various deterrence-based policies on rates of particular crimes. Ecological studies look at natural variations in sanction levels and crime rates across time or space, while quasi-experimental examine the effect of specific interventions on measures of offending behaviour. The third type of study used in deterrence research is concerned with individuals perceptions of risk. These studies rely on survey data in order to investigate the relationship between perceptions of sanction risk and self-reported prior offending or intentions to offend. In a recent review of this deterrence research, Nagin concluded, the collective actions of the criminal justice system exert a very substantial deterrent effect (Nagin 1998, p. 3). This concurs with Cook s assessment of the deterrence evidence almost 20 years earlier, from which he affirmed that the criminal justice system, ineffective as it may seem in many areas, has an overall crime deterrent effect of great magnitude (Cook 1980, p. 213). However, as Nagin (1998) rightly observes, these conclusions are of little value in formulating policy. He states that the issue for policy makers is not whether the criminal justice system as an entity deters potential offenders but whether a specific policy can add to its preventative effect. For this reason a large body of deterrence research has focused on the deterrent effect of changes in the certainty and severity of formal punishment. 1 Deterrence and the certainty of formal punishment Many of the early aggregate-level studies that adopted an ecological approach for studying deterrence found consistent evidence for a negative association between crime rates and the probability of formal punishment being imposed by the criminal justice system (Blumstein, Cohen & Nagin 1978). Perhaps the most infamous of these was the work undertaken by Isaac Ehrlich in the 1970s. Building on Becker s economic theory of crime, Ehrlich (1973) developed a rational model of offending which aimed to predict the annual aggregate murder rate in the USA from estimates of the probability of apprehension, conviction and execution for this type of crime. The dependent variable in this analysis was the number of murders/manslaughters 1 Note that very little work has been conducted thus far on the role of celeritous punishment in controlling crime and therefore deterrence research investigating this aspect of punishment is not reviewed here. The efficacy of swift punishment is however discussed in later Chapters. 4

15 per 1,000 civilian population recorded by police each year from 1933 through Estimates of punishment risk over this period came from data on the percentage of all murders cleared by the arrest of a suspect and the fraction of persons charged with murder who were subsequently convicted. Ehrlich s analysis also included measures of other potentially confounding variables, including unemployment, per capita income and the fraction of young people in the population. Ehrlich found a significant independent relationship between the annual murder rate and both the likelihood of arrest and the likelihood of conviction. Ehrlich (1975) estimated that a one per cent change in the risk of arrest for homicide resulted in a per cent change in the aggregate murder rate, while a one per cent change in the risk of conviction was associated with a per cent reduction in the murder rate. Including the risk of execution once convicted, 2 Ehrlich estimated that an additional execution each year could save between seven and eight lives. A number of scholars have since made trenchant criticisms of Ehrlich s study and of ecological research more generally (Cook 1980; Nagin 1998; Spelman 2000; Weatherburn et al. 2000; Chan & Oxley 2004). Most notable of these is the simultaneity problem the reciprocal relationship that potentially exists between sanction risks and crime rates. Tougher crime control policies can affect the incidence of crime but, so can higher crime rates affect sanction risk by influencing the severity of penal policy and/or the amount of resources devoted to law enforcement. This violates the basic assumption of regression analysis that the dependent variable is caused by the independent variables included in the model and thus constitutes a major problem for researchers attempting to infer deterrent effects from ecological research designs. Support for the deterrent effect of increases in the certainty of punishment is, however, also found in quasi-experiments, which overcome many of the methodological limitations that hinder ecological research (Cook 1980; Nagin 1998). Law enforcement agencies often attempt to exert an effect on the incidence of crime 2 Ehrlich (1975) estimated that a one per cent change in the risk of execution reduced the murder rate by between and 0.068%. This measure of punishment risk is correctly subsumed under the severity of punishment. 5

16 by launching crackdowns or blitzes on particular types of crimes or on crime in particular geographical areas. The effect of these crackdowns or blitzes on the actual probability of detection and apprehension for a crime is probably quite limited. However, their effect on the perceived risk of apprehension can be substantial, especially if police increase the ratio of police officers to offenders in specific places or situations, use coercive police powers to stop, search or arrest offenders, or widely publicise police crackdowns in the mass media (Sherman 1990). Sherman (1990) undertook a comprehensive review of empirical research on police interventions in localities where these sorts of initiatives were employed. Of the 18 cases Sherman reviewed, he found that the majority of the crackdowns (15 from 18) were associated with an initial reduction in levels of the target crime, a result that Sherman took as evidence for an initial deterrent effect. Furthermore, the most successful of these police interventions appeared to be those that were implemented over a short-term time frame (i.e. less than six months). These interventions were associated not only with an initial deterrent effect but also with a residual deterrent effect once police efforts were scaled back. In contrast, the initial deterrent effect created by many of the long-term crackdowns showed evidence of decay as the intervention progressed. Sherman proposes that the enhanced effect of the shorter interventions may be due to the fact that police crackdowns increase the uncertainty surrounding sanction risks, thereby causing people to overestimate the probability of detection and apprehension. As the intervention progresses over a longer period, however, people come to learn that this initial reaction was an overestimation and readjust their risk estimates accordingly. Interestingly, however, Sherman (1990) found differential effects for police crackdowns targeting different types of crime. In particular, two case studies of drug crackdowns found evidence of an increase in crime corresponding with the augmented police activity, and three drug crackdowns were associated with displacement of criminal activity to other areas. Sherman suggests that, crackdowns on market-driven criminal activity may have more complex outcomes than those on middle-class offences such as drink-driving and parking violations (Sherman 1990, p. 36). 6

17 More recent work by Braga and his colleagues (Braga et al. 2001) has supported Sherman s conclusions regarding deterrence-based policing interventions but has also shown that the way deterrence initiatives are implemented can be critical to their success. Braga et al. (2001) evaluated the impact of a problem-oriented approach (known as Operation Ceasefire or the Pulling Levers program) to the policing of youth gang violence, particularly gun violence, in the Boston area. A time-series analysis of youth homicide rates, calls for service from shots being fired and gun assaults showed that Operation Ceasefire was associated with a significant reduction in all three of these incident types a reduction that was greater, and more specific to youth gun violence, than that found in other major USA cities over the same period. The authors of this evaluation indicated that the key to the success of this program was the fact that the intervention directly targeted high-risk offenders and adequately communicated the increased threat to this group: the pulling-levers deterrence strategy controlled violence by focusing on particular groups that were behaving violently, subjecting them to a range of discretionary criminal justice system action, and directly communicating cause and effect to a very specific audience (Braga et al. 2001, p. 219). Deterrence and the severity of formal punishment The deterrent efficacy of increases in punishment severity is much less clear. Earlier work on deterrence and the severity of formal punishment dealt predominantly with the impact of capital punishment on homicide rates. State-sanctioned termination of an offender s life is a highly emotive and political topic and thus has attracted a great deal of academic attention. This academic interest was boosted by the fact that many western countries (with the exception of the USA) had abolished the death penalty by the 1970s, which meant that deterrence hypotheses could be tested using natural or quasi-experimental research designs. An example of this approach is the work undertaken by Barber and Wilson (1968) in Australia. These researchers examined trends in homicide rates in Queensland before and after the abolition of the death penalty in 1922, and compared these trends with those occurring in two other Australian States, where capital punishment was still 7

18 available as a sentencing option (i.e. New South Wales and South Australia). Contrary to deterrence hypotheses, this analysis showed a general downward trend in homicide rates in Queensland over the 40-year period examined, which did not alter to any great extent after the death penalty was removed from State legislation. Meanwhile, the homicide rate in New South Wales (NSW) and South Australia generally remained at or around the same level, fluctuating around a mean of 10 murders per million population. Furthermore, Queensland had, historically, recorded a greater number of executions each year, on average, than other Australian States, but at the same time experienced a much higher murder rate than either of the two comparison States. While only a descriptive analysis, and therefore limited in its ability to draw strong inferences, Barber and Wilson conclude from these results that capital punishment is no more effective in deterring potential offenders than the alternative of life imprisonment. Most other empirical studies investigating the deterrent efficacy of capital punishment have been conducted in the USA. This focus on the USA primarily stems from the fact that (1) there is variability across USA States in the uptake and application of the death penalty which allows for cross-sectional comparisons across jurisdictions and (2) a moratorium on executions existed in the USA from 1968 through to 1977, after a USA Supreme Court decision ruled capital punishment unconstitutional, which permits longitudinal, time-series analyses of the effectiveness of different punishment regimes. The majority of studies employing these types of analyses have found little evidence that the use of capital punishment can have a positive impact on murder and/or manslaughter rates (for a review see Chan & Oxley 2004). For example, Peterson and Bailey (1988) examined the extent to which the availability of the death penalty in contiguous USA States was able to predict Statelevel murder rates and changes in these rates over the period 1973 through They conducted a multivariate analysis in which yearly homicide rates were regressed against the number of months where capital punishment was available in each State and measures on seven State-level sociodemographic variables (e.g. % metropolitan population; % poverty; % unemployment in each State). This analysis 8

19 found no relationship between the availability of capital punishment in any given year and the average yearly homicide rate in that State, controlling for other potentially confounding factors. In fact, a further comparative analysis undertaken by these authors showed that in some cases, the relationship between the availability of the death penalty and yearly homicide rates across States was in the opposite direction to that predicted. That is, several States without statutory provisions for capital punishment recorded homicide rates that were, on average, lower than their neighbouring State that permitted capital punishment. Given these findings on the efficacy of capital punishment, many criminologists, sociologists and behavioural scientists have been quick to dismiss the severity of punishment as unimportant in deterring offenders. This is illustrated by a position paper on punishment and behaviour change issued by the Australian Psychological Society in the 1990s. In this paper, Sanson and his colleagues concluded, given that capital punishment represents an extreme form of punishment, its failure to act as a general deterrent suggests that milder forms of punishment are also unlikely to have a strong general deterrence effect (Sanson et al. 1996, p. 160). However, the dismissal of a whole field of inquiry on the basis of the results from capital punishment research alone would seem somewhat premature. Death penalty research only comments on whether capital punishment is any more effective in reducing homicides than the usual alternative of life imprisonment. It says nothing about the deterrent effect of milder forms of punishment or the deterrent efficacy of other types of legal sanctions. Evidence that the abolition of capital punishment has little impact on aggregate-level murder rates therefore cannot be interpreted as proof that reducing the penalty for homicide from life imprisonment to a shorter gaol term will similarly result in no change in offender behaviour. Moreover, capital punishment is, in most cases, reserved only for the most serious crimes, such as homicide, which violate the moral code of society. Most people refrain from committing these types of crimes, not because the costs associated with the formal sanctioning process are too great, but because they believe the act to be immoral or inherently wrong and one that will be harshly judged by other members of society (Andenaes 1966; Gibbs 1975). There are other criminal behaviours, however, that do 9

20 not invoke the same moral approbation and it is possible that, for these offences, changes in the magnitude of criminal penalties can have a significant deterrent effect on rates of offending. Much of the research available on the deterrent effect of non-capital sanctions consists of studies investigating the relationship between imprisonment, either the average length of prison sentences or the likelihood of being sentenced to custody, and aggregate-level crime rates. A review of this research by von Hirsch and his colleagues (1999) found only weak, negative correlations between these measures of punishment severity and crime levels in different jurisdictions. One of the earlier studies included in this review, which attracted considerable attention in the deterrence debate (von Hirsch et al. 1999), was the work by Farrington and Langan (1992). 3 This research examined crime trends in the USA between 1981 and 1986 and in England between 1981 and 1987, and compared criminal justice practices in both these countries over the same time period. Unlike many other aggregate-level studies, this study not only relied on police-recorded offence data (which can be affected by public reporting and police recording practices) to measure crime but also examined data from victim surveys, thereby allowing more valid inferences to be drawn from the trend analysis. These data revealed that, during the early to mid 1980s, England experienced a marked increase in property crime but no change in violent crime, while the USA experienced a reduction in both property and violent crime. However, an analysis of the severity of punishment imposed for these offences revealed no consistent pattern to suggest any association between the degree of onerousness of formal sanctions for violent/property offences and rates of these crimes in each country. The three criminal justice variables measuring punishment severity in this study were the probability of custody given conviction, the length of imprisonment and average prison time served. Only the first of these three measures was negatively correlated with crime rates in England and the USA, and, while this negative relationship was 3 This study was the first of three studies conducted by Farrington and his colleagues which examined the flow of offenders through the criminal justice system, from offending to incarceration, in several different countries. All three studies generally found weak, and statistically non-significant, associations between measures of punishment severity and crime levels (see von Hirsch et al for a review of this research). 10

21 apparent for all violence and property offences in the USA, in England it was observed only for the offences of assault and motor vehicle theft. Nevertheless, von Hirsch et al. (1999) are cautious in their conclusions regarding the marginal deterrent effect of imprisonment, given that the majority of studies conducted in this area suffer from several methodological weaknesses. Firstly, comparisons across jurisdictions or across time must control for other possible influences on crime rates, including social and economic factors. Yet often these controls are missing from econometric analyses of the deterrent effect of punishment severity or, where they have been included, are inappropriate measures of the factors they supposedly represent. Secondly, many punishment severity studies have specified prison population as the independent variable in the research but in doing so fail to distinguish between the certainty and severity of formal punishment. A tough jurisdiction, where a large proportion of offenders are sentenced to custody and to long prison terms, could still have a relatively small prison population if the risk of detection, apprehension and conviction is low in that locale. Thirdly, the bulk of the empirical research available on the effectiveness of imprisonment as a deterrent has adopted an ecological approach and, as noted previously, a major problem with this type of research is the reciprocal relationship that potentially exists between sanction levels and crime rates. Spelman (2000), in his review of the literature on imprisonment and crime, identifies four recent studies that successfully deal with many of the problems associated with earlier ecological research on the efficacy of prison. All four of these studies found that higher rates of imprisonment are associated with reduced crime levels and estimated that a one per cent increase in the USA prison population could reduce aggregate-level crime rates from anywhere between 0.16 and 0.31 per cent. One of these studies, undertaken by Levitt (1996), is particularly noteworthy because it applied a unique approach to deal with the simultaneity bias associated with ecological research. In this study, Levitt (1996) examined the impact of prison overcrowding litigation on crime rates in the USA. This litigation resulted in 12 USA States having to reduce the size of their prison population after the USA Supreme 11

22 Court ruled overcrowding in prisons to be unconstitutional. Importantly, Levitt showed that the overcrowding litigation had the intended effect on the size of prison populations but crime rates, in the years prior to the courts decision, were not predictive of litigation being filed in particular States. Thus, changes in prison populations in these 12 USA States could be considered independent of the level of crime in these jurisdictions. Examining annual crime trends for the years 1973 through 1993, Levitt then demonstrates that the change in prison populations resulting from the litigation was strongly and negatively correlated with recorded crime rates, controlling for several other potentially confounding variables (e.g. unemployment, per capita income, per cent population in metropolitan areas). From this data Levitt estimates that reducing the prison population by one offender is associated with an increase of approximately 15 crimes per year, an estimated elasticity that is two to three times higher than previous studies. Prima facie, these results appear to support deterrence hypotheses but one other plausible explanation still remains: incapacitation. States who reduce their prison population may subsequently experience higher crime rates because motivated offenders, who are released early, are no longer physically prevented from committing crime, not because the broader population are less fearful of the sanctioning process. This is a limitation shared by all four studies included in Spelman s (2000) review and one that is yet to be resolved empirically. Even if we are unable to disentangle the effect of incapacitation from deterrence at the aggregate level it is possible to examine the deterrent effect of imprisonment at the individual level by considering the impact of longer sentence lengths on the likelihood of an offender being reconvicted once released. Again, however, the available evidence in this area is far from conclusive. A recent study by Spohn and Holleran (2002) compared the recidivism rates for a group of 301 drug offenders sentenced to prison with a group of 776 drug offenders who were sentenced to probation and found no evidence that imprisonment reduced the likelihood of recidivism. In fact this research showed that drug offenders who were sentenced to prison had higher recidivism rates and reoffended sooner compared to drug offenders sentenced to the non-custodial alternative. In contrast, research by Weinwrath and 12

23 Gartrell (2001) on 514 incarcerated drink-drivers in Alberta, Canada, found that offenders sentenced to longer periods in detention were less likely to be reconvicted for a new drink-driving offence than drink-drivers sentenced to shorter periods of imprisonment, even after controlling for other factors relevant to sentencing and offending (e.g. age, prior offences, treatment for alcohol addiction, marital status, employment status and education). However, the relationship between sentence length and recidivism found in this study was not linear, with the data suggesting the existence of a punishment threshold (of approximately six months gaol) above which no greater impact on reoffending was observed. Limitations of aggregate-level deterrence research Sherman s (1990) research on police crackdowns and blitzes reviewed earlier in this Chapter, suggests that variations in the probability of formal punishment can deter some potential offenders, at least in the short term. However, the evidence presented above on the deterrent efficacy of alterations to the severity of punishment is equivocal. One reason for the inconsistency of research findings on the impact of punishment severity is that much of the deterrence research on severity effects has been undertaken using an ecological approach and, as previously discussed, this type of research has several methodological limitations that preclude confident conclusions regarding deterrent effects. In addition to these methodological issues, there are two other important contextual factors that may go some way to explaining the minimal, and at times conflicting, evidence currently available on the deterrent efficacy of punishment severity. Firstly, legislation that aims to increase the severity of formal punishment is often unsuccessfully translated into sentencing practice. Policies that raise the severity of statutory sanctions, for example, can sometimes result in little, or no, change in the severity of penalties that offenders actually receive if found guilty. This may be because players within the criminal justice system (e.g. police, prosecutors, judges, jurors and defendants), believing the increase excessive, choose to charge fewer offenders or charge offenders with less serious offences, increase plea bargaining, reduce convictions or impose alternative sanctions, and in doing so undermine the intended deterrent effect of the legislative changes. Barber and Wilson (1968) 13

24 highlight this implementation issue in their research on capital punishment in Australia by showing that jurors are less likely to convict an offender in States where the death penalty is more likely to be imposed. In this case, the threat of punishment cannot be considered equal in each of these jurisdictions. Secondly, to date, investigations of punishment severity conducted at the aggregatelevel have only considered punishment changes in a context that is generally not conducive to deterrent outcomes. The deterrence model proposed by Becker (1968), and indeed the deterrence doctrine proffered by the early utilitarian philosophers, anticipated that punishment certainty and severity combine multiplicatively, such that the severity of punishment matters most when punishment is likely to be incurred. However, much deterrence research has been undertaken in jurisdictions, or for offences, where the certainty of apprehension is relatively low (see von Hirsch et al. 1999; Robinson & Darley 2004). In these instances, where offenders perceive it unlikely that they will be caught, the severity of legal sanctions may matter little in their decisions to offend. This highlights the need to move beyond the simple question of whether punishment deters crime. Gibbs (1975) notes that all of the contradictory general observations about deterrence can be reconciled by one empirical assertion: In some situations some individuals are deterred from some crimes by some punishments (Gibbs 1975, p. 11). But he also observes that this statement is unfalsifiable, and therefore of no theoretical value, unless we can identify (1) the individuals who are most deterred by legal sanctions, (2) the sanctions that are most effective in deterring these individuals, (3) the crimes that are most susceptible to deterrence-based interventions and (4) the context in which sanctions work best. Accordingly, one of the major aims of this thesis is to explore the conditions under which formal punishment is thought to be most efficacious. Only when these conditions have been established can we then undertake valid tests of deterrence hypotheses relating to the probability and severity of legal sanctions. 14

25 Perceptual research in deterrence The aggregate-level research summarised in the previous two sections has, by its very nature another important limitation: the studies in question infer deterrent effects; they do not directly measure them. Suppose a study finds no evidence for a significant association between the availability of capital punishment in a State and murder rates in that State. One might conclude that the death penalty is no better in deterring offenders than life imprisonment. But it is also possible that: (1) people perceive an increase in the cost of committing murder but are not deterred (2) people perceive life imprisonment to be a sanction that is as severe as the death penalty and therefore do not change their offending behaviour or (3) there is a difference in perceived severity of the death penalty compared to life imprisonment but the difference is only slight and unable to be detected by aggregate-level measures. A mass of research emerged during the late 1970s and throughout the 1980s, which utilised survey data to investigate the relationship between risk perceptions and behaviour. This research consisted of three types of studies: cross-sectional surveys, which assess current perceptions of sanction risks and past or future intentions to offend; longitudinal surveys, which measure current perceptions and then, at a later point in time, subsequent criminal behaviour; and scenario-based designs, which experimentally manipulate sanction risks within a hypothetical offending scenario and then measure the likelihood of offending given these conditions. The results from these studies vary somewhat depending upon the methodology employed but are generally consistent with aggregate-level research in finding positive effects for perceived certainty of detection, apprehension and conviction (punishment certainty). That is, those persons who believe it likely that formal punishment will be incurred are less likely to offend (e.g. Waldo & Chiricos 1972; Grasmick & Bryjak 1980; Paternoster et al. 1982; Paternoster 1989; Bachman, Paternoster & Ward 1992). Perceptual studies have also highlighted the role of informal social controls in offender decision-making and the interdependency 15

26 between these informal sanctions and the risk of formal punishment (e.g. Paternoster & Iovanni 1986; Williams & Hawkins 1986; Klepper & Nagin 1989a). On the other hand, very few early perceptual studies have investigated the deterrent effect of sentence severity and those that have, found only weak correlations between perceptions of the severity of legal sanctions and past or future intentions to offend (von Hirsch et al. 1999). One study that did examine the relationship between perceived severity of formal punishment and criminal behaviour was a crosssectional study by Grasmick and Bryjak (1980). In this survey of over 400 respondents, people were asked about their prior involvement in various illegal activities (e.g. theft, assault, tax fraud), the estimated chance of being arrested for each of these offences and how much of a problem the resultant punishment would be for them if caught and found guilty. Both the estimated chance of arrest and ratings of punishment severity were inversely related to previous offending episodes. Thus, those who had offended less often were respondents who thought it likely they would be caught and receive a relatively severe punishment. More recent work by Nagin and Pogarsky (2001) has also examined the marginal deterrent effect of the perceived severity of legal sanctions but in this case a scenario-based methodology was employed. This research found a negative association between perceived severity of legal costs and the stated likelihood of offending in the circumstances described but this relationship was weak and much less robust than that found for the perceived certainty of apprehension and conviction. While the perceptual deterrence research conducted thus far has made an important contribution to our knowledge regarding risk perceptions and criminal behaviour, this type of research has generally failed to consider how perceptions of risk are formed. This is a limitation not only of studies examining the perceived severity of formal punishment but throughout the perceptual deterrence literature (Nagin 1998; von Hirsch et al. 1999). The importance of this neglected area of research becomes immediately obvious when assessing the role of criminal justice policy in the deterrence process. As Nagin observes, even if crime decisions are influenced by sanction risk perceptions absence some linkage between policy and perceptions - behavior is immune to policy manipulations (Nagin 1998, p. 18). Research 16

27 investigating the link between experiences with, or knowledge of, the criminal justice system and perceptions of the formal sanctioning process is therefore needed before we can conclude that criminal justice practice has the potential to affect offender behaviour through deterrent-like mechanisms. The current study The purpose of this thesis is to contribute to the debate on punishment severity by assessing the impact of an increase in the severity of prescribed legal sanctions for drink-driving offences in New South Wales, Australia. Specifically, the research question to be addressed is whether the higher penalties introduced by the NSW State Government in 1998 were a marginal deterrent for drink-drivers and, if so, by how much. Positioning this research within a deterrence framework ensures that a small number of hypotheses can be generated from previous research and empirically tested. Deterrence is also generally considered one of the major aims of legal sanctions prescribed for drink-driving offences (Homel 1980; Ross 1984), has been a potent force in shaping driver behaviour in Australia over the last two decades or more (Homel 2004) and is viewed by the Australian public as the most effective strategy to address drink-driving and related issues (Australian Institute of Health and Welfare 2002a). The case study that has been selected for this analysis is also a good test of deterrence because it is set in a context where several of the preconditions for deterrence success have been met. First of all, drink-driving is an offence that does not attract the same moral condemnation as other types of crime, such as homicide or robbery, and is therefore, potentially more susceptible to changes in the probability and severity of formal punishment than are other criminal behaviours. Furthermore, in Australia, very few people are gaoled for a drink-driving offence. Thus, we can be more confident that any reduction in aggregate-level crime rates emanating from a change in sentencing policy is due to deterrence rather than incapacitation. In Australian jurisdictions there also exists a relatively high perceived risk of detection and apprehension for drink-driving, created by the introduction, and the continued publicity, of random breath testing by police. This high probability of apprehension is not true of many other offences (Robinson & Darley 2004) or, as will be seen later 17

28 in this thesis, the offence of drink-driving in other countries. Finally, the measure typically used to monitor trends in drink-driving behaviour (i.e. road crash rates) is less prone to bias as a result of underreporting or enforcement practice. It is therefore a more valid and reliable measure of offending than other available crime indicators, such as police-recorded arrest data. These features of the selected case study render it a more definitive test of the marginal deterrent effect of punishment severity than preceding work that has been undertaken in this area. This work also builds on previous deterrence research by addressing several methodological limitations of punishment severity studies. Firstly, it employs a quasi-experimental research design to examine the impact of the penalty changes on offending rates in New South Wales. This type of study is able to overcome many of the problems associated with ecological research because it incorporates several features of a true experimental design, including a specific well-defined intervention, comparison groups to control for other possible influences on crime rates and, in the case of interrupted time-series analyses, many observations before and after the intervention to distinguish deterrent effects from other fluctuations in crime levels. Focusing on one specific, well-defined intervention also allows the researcher to consider implementation issues associated with criminal justice policies within a realistic setting. Thus, another important component of this thesis concerns how well the 1998 sentencing policy was translated into a credible threat. In particular, we address the question of what impact the penalty changes had on actual sentencing practice and what potential, if any, the policy had to affect perceptions of sanction severity. These aspects of the research design will improve our ability to make confident conclusions regarding deterrent effects. Thesis structure This thesis begins with a more detailed description of deterrence theory and a discussion of the basic assumptions underpinning the theory. Chapter 2 outlines some of the major criticisms of the rational choice model of offender behaviour and explains how these criticisms have been addressed in modern deterrence theory. It also explores the conditions thought to affect the efficacy of formal punishment and 18

29 explains how the offence of drink-driving, which meets several of these conditions, can be used as a case study to examine deterrent effects. Previous deterrence research that has been conducted in the drink-driving field is reviewed in Chapter 3. The studies described in this chapter present evidence regarding the deterrent effect of changes in the certainty, severity and celerity of formal punishment for drink-driving offences. Both aggregate-level and individuallevel research are included in this review, as are several methodological limitations of the drink-driving studies conducted thus far. This section also highlights some knowledge gaps identified in the broader deterrence literature. The focus of the thesis then shifts towards the 1998 sentencing policy that increased statutory penalties for drink-driving offences in New South Wales and its effect on offender behaviour. In Chapter 4, the case study is described in greater detail and the methodology employed to investigate the impact of the penalty increase is discussed. The results of two major studies undertaken to examine any changes in drink-driving rates which coincided with the increase in punishment severity are presented in Chapter 5. The first study involves an interrupted time-series analysis of alcoholrelated road crashes before and after the legislative changes and a comparison of changes in these crashes with those apparent in rates of non alcohol-related crashes over the same time period. The second study involves a comparison of the reoffending rates of two cohorts of drink-drivers sentenced before and after the penalty changes. The second study also verifies the extent to which the 1998 legislative amendments affected actual sentencing practice and considers any countervailing effects of the policy on the prosecution and conviction of drinkdrivers in New South Wales. Chapter 6 provides further information about the context in which the sentencing policy was implemented. Specifically, it examines data bearing on the level of enforcement of drink-driving offences at and around the time the penalties were raised and on the level of publicity the sentencing policy received in the media at the time of implementation. The issues of enforcement practice and public knowledge of the penalty changes are further explored in the third study undertaken for this thesis, 19

30 the results of which are presented in Chapter 7. This study uses a scenario-based survey approach to examine the relationship between knowledge of applicable drinkdriving penalties, perceptions of formal punishment for drink-driving offences and the stated likelihood of offending. This study builds on the previous two studies by examining, at the individual-level, the link between sentencing policy in New South Wales, perceptions of punishment severity and drink-driving behaviour. It also presents an opportunity to examine the deterrent effect of punishment severity whilst controlling for other potentially confounding contextual factors, such as the level of police enforcement targeting drink-drivers. Finally, in Chapter 8, the results from each of the three studies assessing the deterrent effect of the 1998 sentencing policy are summarised and the implications of these findings for deterrence theory, criminal justice policy and future research in this area are discussed. 20

31 CHAPTER 2. DETERRENCE THEORY Deterrence occurs when a person refrains from committing an offence because they fear the negative consequences associated with the act. Typically, it is the legal sanctions prescribed by the criminal justice system which constitute the subject matter for deterrence research. However, as discussed later in this chapter, informal sanctions or extralegal consequences associated with offending are also potential costs that can influence an individual s decision to offend. Without dismissing the importance of these other types of sanctions, the primary focus of the current thesis is the deterrent efficacy of legal sanctions. The findings from this analysis will have direct implications not only for deterrence theory but also for criminal justice policy more generally. A methodological problem for researchers attempting to investigate the deterrent efficacy of legal punishment is that, by definition, deterrence denotes the absence of a behaviour and is therefore, essentially an unobservable phenomenon. This makes statements such as punishment deters crime difficult to falsify unless they are situated in the broader context of a systematic theory (Gibbs 1975). Deterrence theorists have thus sought to identify the necessary conditions under which deterrence occurs, including the types of offences, sanctions, individuals and contexts that are conducive to deterrent outcomes. Explication of these conditions, a priori, allows one to formulate predictions regarding the deterrent efficacy of particular sanctions and modifications to those sanctions (Beyleveld 1979). This chapter endeavours to explore the factors that are thought to be influential in the deterrence process. It begins by presenting a typology of deterrent effects to more clearly define relevant concepts. The expected utility model of offender decisionmaking, upon which modern deterrence theory is based, is then described and the limitations of this rational choice model of criminal behaviour are discussed. Other factors that may influence the deterrent effect of state-imposed punishment are also highlighted, such as the applicability of the deterrence model to different types of 21

32 crimes and the potential for punishment variables to interact with contextual factors and offender characteristics. Definitions of deterrence General and specific (or special) deterrence are key concepts in the deterrence literature. General deterrence refers to punishment that discourages members of the public, who learn of the sanctions, from violating the law. Specific deterrence refers to the effect that punishment has on the criminal behaviour of those who are punished. Some deterrence theorists claim that general and specific deterrence are not separate mechanisms by which legal sanctions act to reduce offending, but involve the same mechanism operating within different populations (Zimring & Hawkins 1973; Beyleveld 1979; Stafford & Warr 1993). Zimring and Hawkins (1973), for example, describe previous experience of punishment as just one of an enormous variety of factors which condition threat responsiveness (Zimring & Hawkins 1973, p. 225). The distinction between specific and general deterrence is, nevertheless, one that is pervasive throughout the literature on deterrence. Gibbs (1975) argues that it is too narrow in scope to classify punishment effects as simply general versus specific deterrence. He proposes, instead, a typology consisting of 16 different situations in which an individual may consider committing a criminal offence. This typology classifies deterrent effects according to an individual s previous experience with punishment for the particular crime type under consideration, their experience with punishment for other types of crime and their offending history, both in terms of the crime in question and other criminal offences. Gibbs contends that a theory could be formulated for each of the 16 different types of deterrence that result from these four levels of legal threat and punishment experience. A simpler typology offered by Gibbs (1975), which is sufficient for the present purposes, distinguishes between three main types of deterrence: absolute, restrictive and specific deterrence. Absolute deterrence occurs in instances where an individual refrains completely from a particular criminal behaviour because of the risk of being sanctioned. Restrictive deterrence occurs when an individual curtails, or reduces the 22

33 frequency of, criminal behaviour during some period in order to lessen the risk of being sanctioned. These two types of deterrence can be equated with the broader concept of general deterrence, described above. The definition of specific deterrence remains the same as that given previously: the omission, or curtailment of criminal behaviour by an individual who has previously been punished and fears being punished again. The distinction between absolute and restrictive deterrence is based on the frequency with which an individual offends, given the existence of a specific legal threat. Zimring and Hawkins (1973), on the other hand, make a distinction between the deterrent effect of introducing sanctions where none had previously existed (absolute deterrence) and the deterrent effect of incremental changes to an already existing threat (marginal deterrence). Gibbs (1975) asserts that this distinction between absolute and marginal deterrence refers to the size of the threat posed by the criminal justice system rather than two distinctive types of deterrence and thus, should be replaced by the terms absolute deterrent efficacy and marginal deterrent efficacy of punishment. The absolute and marginal deterrent efficacy of legal sanctions is a crucial distinction to make when assessing the impact of penal policies on rates of offending. Much deterrence research has been interested in comparing the rate of crime and the levels of penalty severity across different jurisdictions, thus focusing on the marginal deterrent efficacy of punishment. Advocates for decriminalisation (or alternatively prohibition) have sometimes used findings from this type of research to argue that the removal of some sanction will have no impact on criminal behaviour. Such arguments are misguided. In order to assess the absolute deterrent efficacy of criminal penalties it is necessary to evaluate policies that remove a legal threat completely, or introduce a threat where none had previously existed. Because these types of policies are rarely implemented, deterrence research is mainly focused on the marginal deterrent efficacy of legal sanctions. The distinction made by Zimring and Hawkins (1973) between absolute and marginal deterrent efficacy also highlights the need to clearly define the nature of the 23

34 threat being investigated by deterrence research. As discussed above, state-imposed punishment is the primary focus of the current research. An example of this research would be a comparison of the murder rate across States or counties that have capital punishment for murder with those that do not. However, this type of research does not adequately deal with variations in the application of statutory punishment across jurisdictions. In one State, the criminal courts may never impose capital punishment for murder. In another, several convicted murderers may be on death row at any one time. The threat generated by capital punishment legislation in these two jurisdictions cannot be considered equal. Deterrence research needs to consider both the impact of legal punishment appearing in statutes and the translation of this legislation into actual sentencing practice. A further consideration is people s beliefs about legal punishment. It is often said that legal sanctions operate to deter potential offenders by making them fearful of the consequences of involvement in crime. If the costs being examined are those threatened by the criminal justice system, then this statement assumes that the potential offender has some knowledge, however limited, of the actual or prescribed punishment for that offence. Yet deterrence could also occur if a potential offender mistakenly believes that legal sanctions exist for a particular behaviour or overestimates the magnitude of the penalty attached to an offence. This Beyleveld (1979) labels deterrence by merely believed sanctions to contrast it with deterrence by prescribed or actual sanctions. While Beyleveld notes that this distinction is of theoretical significance, he also concedes that only the deterrent effect of prescribed or actual sanctions will be of relevance to deterrence policies. It does, however, illustrate the importance of legal knowledge in maximising the effectiveness of deterrence-based penal policies, an issue that will be discussed in greater detail throughout this thesis. One final problem for research investigating the deterrent effects of legal sanctions is that deterrence is just one of several mechanisms by which legal sanctions can serve to prevent offending. Gibbs (1975) identifies at least nine other preventive consequences or preventive mechanisms of punishment in addition to deterrence. Three of these, retribution (exacting vengeance through legal means to prevent acts 24

35 of revenge), rehabilitation (changing an offender s motivation to commit crime) and incapacitation (physically preventing an offender from committing further crimes), are commonly identified as major aims of criminal punishment. However, there are several other means by which legal sanctions can serve to reduce crime. For example, the introduction or presence of a legal sanction may demonstrate that a particular act is wrong and is condemned by the wider community. If the behaviour is punished regularly and severely, this shared belief will be reinforced and potentially lead to a reduction in offending (i.e. Normative Validation). Alternatively, sanctions could further public knowledge that a certain behaviour is illegal. This knowledge, combined with an unquestioning acceptance of the law, may improve the legal compliance of some citizens (i.e. Enculturation). These preventative mechanisms of punishment resemble Andenaes (1974) concept of general prevention. Unlike general deterrence, which only refers to compliance resulting from fear of punishment, general prevention refers to a broader range of controls that include moral inhibitions and socialisation of preventive habits. Distinguishing deterrent from other long-term preventative effects of punishment in research, is complicated by the fact that the short-term threat of general deterrence could achieve these broader social controls. As an example, compliance could initially be obtained through fear of punishment but eventually obtained because a social norm develops that inhibits the proscribed behaviour and promotes voluntary compliance with the law (Ross 1984). Similarly, the threat of punishment may have a habituative effect, such that adherence to the law is initially secured through fear but repetition of the law-abiding conduct eventually results in the behaviour becoming routine practice (e.g. seat-belt laws; Elliott 2003). This thesis deals with simple or direct deterrence, that is, behaviour changes resulting from fear of threatened punishment. However, the contribution of other preventative mechanisms of punishment, in securing legal compliance, should be kept in mind when considering the implications of the current research. Deterrence is just one of numerous reasons why people comply with the law. Many citizens refrain from offending simply out of a moral obligation to obey the law or a general acceptance of the legitimacy of authority (Tyler 1990). On the other hand, there is a proportion of 25

36 the population for whom these broader social controls are ineffective, but who are responsive to other preventative mechanisms of punishment, including deterrence. Nagin and Paternoster (1994) categorise these persons as marginal offenders, individuals whose self-control is not high enough to totally eliminate the possibility of breaking the law but neither low enough to make them insensitive to punishment. It is this group of potential offenders who would be the most responsive to changes in the threat of punishment and who are thus the target group for deterrence-based interventions. Modern deterrence theory As discussed in Chapter 1, modern deterrence theory is an application of the rationalchoice paradigm of decision-making. It assumes that human beings are essentially rational actors who, in circumstances of uncertainty, evaluate the consequences of alternative choices and estimate the likelihood of their occurrence. The alternative selected is the one that maximises the expected utility, that is the option maximising the benefits and minimising the costs to the individual. An important component of this calculation is that the utility of each alternative relies on a subjective evaluation of the value of each outcome. For this reason individuals may evaluate the profit and loss associated with equivalent alternatives differently and may, therefore, vary in how they respond to similar rewards and punishments. This notion that people make subjective evaluations rather than relying on objective values differentiates expected utility theory from expected value theory, though both theories assume that the probability of an outcome occurring and the value assigned to that outcome combine in precisely the same way. Expected utility maximisation as applied to decision-making in a criminal context predicts that a person will engage in a criminal act if the subjective expected utility of the criminal act E(U) c is greater than the subjective expected utility of other legitimate alternatives E(U) a. The subjective expected utility of a crime is calculated firstly by evaluating the gains associated with the successful completion of the crime and the losses associated with legal sanctioning if caught. The criminal gains include tangible goods (e.g. cash) or avoided inconveniences (e.g. having to catch public transport home in the case of drink-driving). They also include the anticipated 26

37 benefits of engaging in the behaviour itself (e.g. the thrill obtained from speeding in a car). The losses or costs, on the other hand, are those imposed by the criminal justice system (though crime costs have more recently been extended to include informal social sanctions that may arise from arrest or criminal conviction). As stated previously, the value assigned to these gains and losses will depend on numerous factors, such as current wealth status, and thus utility calculations will vary across individuals. To determine the expected utility of a course of action an individual must also calculate the probability of the outcome occurring, that is, the chance that he or she will be rewarded or sanctioned if he or she were to engage in the behaviour. In a criminal context, this estimate of outcome likelihood is generally thought to be subjective in nature (i.e. the perceived probability of being arrested) rather than objective (i.e. the actual risk of being arrested for an offence), but is otherwise thought to adhere to the general rules of probability theory (note that the use of subjective probabilities distinguishes subjective expected utility theory from expected utility theory). These two steps, assessment of expected outcomes and the application of subjective probabilities, can be represented, more formally, by the following equation: E(U) c = P(G) c U(G) c + P(L) c U(L) c (1) where E(U) c is the expected utility of committing the crime, P(G) c the subjective probability of obtaining the gains associated with the crime, U(G) c is the utility of those gains, P(L) c is the subjective probability of receiving legal sanctions if caught and U(L) c is the disutility of legal sanctioning if caught. Deterrence theory focuses attention on the variables affecting the disutility of crime, which is represented in the right-hand side of the above equation. This theory states that to discourage an individual from committing an offence, the perceived severity (U(L) c ) and the perceived certainty (P(L) c ) of the punishment associated with that crime must be sufficiently large to reduce its subjective expected utility below that of the anticipated gains. Thus, an individual will refrain from offending if: P(G) c U(G) c < P(L) c U(L) c. 27

38 Importantly, however, as shown in equation 1, the utility of formal punishment and the subjective probability of it occurring are thought to combine multiplicatively rather than additively. This means that the deterrent effect of a change in sanction severity (U(L) c ) depends upon the perceived likelihood of the sanction (P(L) c ). If the risk of apprehension is zero then no level of sanction severity will deter an individual from committing a crime. Likewise, in the absence of sanctions, the risk of apprehension will not exert a deterrent effect. Expanding the Subjective Expected Utility Model According to classical deterrence theory the efficacy of legal sanctions is a function of the perceived celerity of punishment. That is, swiftly delivered punishment will deter an individual more than punishment occurring some time in the distant future. While a time variable is not included in the traditional subjective expected utility equation, the deterrence model has recently been extended to include the economic concept of a discount rate to account for the hypothesised celerity effect of punishment (Kenkel 1993; Nagin & Pogarsky 2001). The concept of a discount rate being used to devalue future costs is derived from economic reasoning about the time value of money. According to this reasoning, a given sum of money is worth more now to an individual than it is in the future because it can be used to obtain immediate benefits. Given the choice of receiving $1000 now or $1000 a year later, a rational actor will always select the former option. This is because any delay in receiving the immediate benefits that could be derived from the $1000 will reduce the current worth of that money (example adapted from Nagin & Pogarsky 2001). In financial decision-making, the size of an immediate reward that just balances the value of a delayed reward is called the present value of the delayed reward. The more delayed the reward is, the smaller its present value. Similarly, the more delayed a cost is, the smaller its present value. If the process of evaluating future punishment is similar to financial decisionmaking, long delays in the imposition of punishment for an offence would reduce the present value of the perceived costs of committing the crime. The consequent reduction in the costs associated with the criminal behaviour would diminish the deterrent effects of the prescribed sanctions for that offence. In economic decision- 28

39 making the devaluation of future costs is both a function of the delay period and some discount rate, which is typically the current market interest rate. In an analogous way, Nagin and Pogarsky (2001) and Kenkel (1993) model the magnitude of the celerity effect of punishment as a function of both the length of delay in punishment and the extent to which an individual prefers to delay punishment. The relationship is expressed in the following equation: D t = 1/(1+r) t (2) where r is the degree to which an individual devalues future consequences and t is the delay before the penalty is imposed. Nagin and Pogarsky (2001) argue that preference for delay (r in the above equation) can be equated with an individual s impulsivity or present-orientation. A person who is more concerned with what happens to them in the short run will devalue future punishment to a greater extent than someone who is more concerned about long-term consequences. Thus, the disutility of a crime not only decreases as a function of long delays in the imposition of penalties but also as a function of an individual s preference for delay or present-orientedness. In the expanded deterrence model an individual will offend if the expected gains associated with the crime outweigh the discounted expected costs associated with legal sanctioning if caught: EU(G) c > P(L) c U(L) c D t (3) Thus, ceretis paribus, any increase in the severity, certainty or celerity of punishment for an offence would increase the disutility of the crime and therefore reduce the prevalence of the criminal behaviour. Critiques of the Rational-Choice Perspective Several authors have criticised the rational choice perspective of offending outlined above on the basis that people do not always combine information in the manner predicted by the expected utility equation, an assertion that has been supported by experimental research on human decision-making under uncertainty. For example, Tversky and Kahneman (1981) presented people with a public health dilemma in which 600 lives were at risk and asked whether they would prefer to 29

40 definitely save 200 lives or take a gamble that offered a one-third chance that all 600 lives would be saved. They found that the majority of participants were risk averse in this scenario, that is, most favoured the certainty of definitely saving 200 lives over the gamble that potentially saves all 600. However, when participants were presented the same scenario, but with a slight change of wording from lives saved to lives lost, this preference for risk aversion shifted to one of risk seeking. In the second scenario, the majority of participants favoured the gamble offering a one-third chance that no one would die rather than the certain loss of 400 lives. The finding that an option with the same probability and outcome can be accepted in one situation but rejected in another, depending on how the outcomes are described or framed, is at odds with the subjective expected utility model. Carroll (1978) also provides experimental evidence on human decision-making which is inconsistent with subjective expected utility theory. He presented a group of offenders and a group of nonoffenders with a series of criminal opportunities, varying on four of the dimensions featured in deterrence theory: the probability of a successful crime, the amount of money to be gained if successful, the probability of capture, and the penalty if caught. When deciding whether or not to offend given these conditions, Carroll discovered that participants tended to focus on only one dimension of the problem rather than fully integrating information on all four dimensions. The dimension proving to be the most influential in the decision-making process varied considerably across participants and did not seem to be affected by previous criminal experience. In general, however, he found that people tended to place greater weight on monetary incentives associated with the crime rather than any potential criminal penalties that may be incurred if caught. Tittle (1977) also found a tendency for individuals to focus on the utility or benefits of the crime, rather than evaluating all available information, in his cross-sectional household survey aiming to predict future deviance of almost 2,000 USA residents. From these, and other similar studies appearing in the behaviour-decision literature, Johnson and Payne (1986) identify two important stages in the decision process where criminal choices predictably deviate from the subjective expected utility 30

41 model: (1) the way in which an individual constructs a representation of the problem and (2) the way in which an individual evaluates different alternatives. In discussing problem representation, Johnson and Payne draw on principles of prospect theory, a descriptive model of decisions under uncertainty formulated by Tversky and Kahneman (1981). Prospect theory proposes that the subjective value assigned to a particular outcome will vary depending on whether the choice is framed as one between possible gains or possible losses. For gains, the utility or value function is thought to be concave, but for losses, the utility or value function is convex. Thus, in contrast to the expected utility model which proposes that people will always tend to be risk averse, prospect theory suggests that decision makers will tend to be risk averse for gains but risk seeking for losses. Furthermore, in Prospect Theory the value function for losses is assumed to be steeper than the value function for gains, thus the loss of a sum of money will be considered more painful than the pleasure of winning that same amount. Tversky and Kahneman s public health dilemma, presented previously, exemplifies these deviations from expected utility predictions. When the choice presented to participants was phrased in terms of the number of lives that could be saved, people opted for the certainty of saving 200 lives over the one-third gamble of saving all 600 lives (risk averse). On the other hand, when the problem was presented in terms of lives lost, people were much more willing to accept the gamble in order to prevent any loss of life (risk seeking). For criminal decisions, this could mean that an individual, who chooses not to commit a crime for gain, might still commit the offence if it means avoiding a possible loss (e.g. underreporting your taxable income to reduce your tax liability but not to increase your tax refund, Johnson & Payne 1986). Prospect theory also proposes that people do not adhere to the rules of probability theory when constructing a representation of the problem. Instead, individuals will tend to apply a subjective evaluation or weighting function to outcome likelihood estimates just as they apply a subjective value function to gains and losses. For example, this weighting function has individuals discounting altogether near zero probabilities and overweighting moderately low probabilities. Thus, if there is very little chance of being apprehended for an offence, this will be perceived as equivalent 31

42 to zero and enforcement will have no impact on offending behaviour. But if police are able to alter perceptions to such an extent that the probability of apprehension is believed to be moderately low, then people will tend to exaggerate (out of proportion) apprehension risk and the success of the campaign in reducing offending will exceed subjective expected utility predictions. The second theme in Johnson and Payne s (1986) discussion of decision-making concerns the processes that operate on these problem representations. Several factors in everyday life, such as time pressures, can impinge on available cognitive resources, thereby reducing the capacity of an individual to evaluate all available choice alternatives. As a consequence, individuals will tend to apply various strategies to simplify the task of outcome selection. One common heuristic described by Johnson and Payne is the noncompensatory strategy. This involves screening various alternatives on one or two attributes, choosing or rejecting each alternative on the basis of these attributes alone and then evaluating the selected alternatives more carefully. This screening approach to decision-making would be most effective when cognitive demands are high, such as occurs when an individual is presented with a complex problem or is constrained by time. This is consistent with Carroll s (1978) finding that many people are unidimensional in their approach to criminal decisions, focusing on only one salient feature of the problem presented to them. Johnson and Payne s (1986) descriptive model of decision-making is far from an allencompassing, systematic theory of offender decision-making but it does serve to highlight several ways in which criminal decisions may deviate from subjective expected utility predictions. Cognitive psychology has shown that information is processed in a memory that has limited capacity, duration and ability to permanently store relevant information. This results in heuristics or short-cuts being employed, where necessary, to cope with large amounts of complex information in an economical fashion. Recognising these limitations, most deterrence theorists have long since abandoned the idea that the analysis of criminal opportunities involves a complex mathematical process of utility calculations and probability estimations. What they do propose, however, is that within the limits of their inclination and ability to make such calculations (an individual) will act so as to pursue whatever is 32

43 at the moment, in their own mind, the most important (Stover & Brown 1975, p. 369). This feature of modern deterrence theory is referred to as limited or bounded rationality (Simon 1983), in which the incredibly economic man of decision-making theory (is replaced) with a choosing organism of only limited knowledge and ability (Douglas & Wildavsky 1982, p. 77). In this way, although individuals may not calculate the utility of all available alternatives in precisely the same way as economists describe, they can be regarded as active decision-makers who are affected by information that is supplied to them. Bounded rationality recognises a person s limited capacity to process and acquire information. An individual will often try to avoid exhaustive, complex calculations and instead focus on a few simple and concrete aspects of a problem with the aim to satisfy, rather than optimise, their needs or wants (Johnson & Payne 1986). Their failure to seek out all relevant information also means that their knowledge of risks and rewards associated with various alternatives may be incomplete or biased by the channels through which the information is sourced. This results in behaviour decisions that are far from optimal. Carroll (1978) and Cook (1980) maintain, however, that legal punishment can still modify behaviour through deterrent-like mechanisms by influencing the standing decisions, or rules of thumb, that people adopt in relation to offending. A person may decide, for example, that they will not drive above the speed limit under any circumstances because they fear losing their driver s licence if caught. Adopting this standing decision would eliminate the need for the driver to undertake unnecessary, complex utility calculations whenever presented thereafter with an opportunity to offend. An increase in the severity of a legal threat would aim to persuade more people to adopt such a position in relation to their driving behaviour. Even in instances where information-processing abilities are reduced, Cook (1980) suggests that people can still be guided by these standing decisions because they have been derived through prior contemplation, presumably in a more rational state. Cook s (1980) description of standing decisions is indicative of theorists attempts to apply the rational choice perspective, and by extension, the deterrence model not only to criminal events (crimes) but also to criminal involvement (or criminality). 33

44 This distinguishes deterrence from other criminological theories that also place importance on situational determinants of criminal behaviour, such as routine activity theory (Clarke & Felson 1993). Both routine activity and the rational choice perspective of offender behaviour view crime as purposive behaviour in which an offender makes various decisions or choices about the relative attractiveness of criminal acts. The latter, however, is interested not only in decisions surrounding the commission of an offence but also decisions to adopt crime as a way of life. These criminal involvement decisions entail weighing up of risks and rewards over an extended period of time, rather than simply assessing information that relates to the immediate situation, and are likely to be influenced by different factors than those utilised in event decisions, including any anticipated informal social sanctions associated with offending (Clarke & Felson 1993). While it would be wrong to dismiss deterrence theory on the basis that human decision-making often departs from the mathematical utility equations, it would be equally misleading to claim that deterrence theory is a complete explanation of criminal behaviour. Like most other criminological theories, deterrence fails to account for each and every decision to offend. Individual differences exist in willingness to take risks, in objective circumstances and in the assessment of costs and benefits. Rationality in offender decision-making can also sometimes be compromised by limited knowledge and ability to process relevant information. However, not all potential offenders need to act predictably all of the time, for deterrence theory to be useful in guiding policy and research goals. As Cook (1980) argues, deterrence is concerned with aggregate behaviour. Thus all that is necessary for deterrence assumptions to hold true is that a proportion of criminal acts be the result of a rational decision process (or the result of a prior rational decision to offend in certain circumstances) in which potential offenders consider the possible risks and rewards associated with the behaviour. Deterrence and specific offences Crime covers a wide range of behaviours that have in common only the fact that they are prohibited by the law. Any theory that attempts to model offender decisionmaking must therefore take into account differences in the situational context in 34

45 which crime is committed, as well as the purpose or motivation of offenders at the time a crime is committed. Recognising these important differences across various offence types, deterrence scholars have proposed that the threat of punishment will necessarily be more important for some types of criminal behaviour than for others. As Zimring and Hawkins (1973) note the jealous husband bent on homicide is less likely to be deterred by legal threat than the bank clerk who views embezzlement as a means of securing funds to indulge in speculation (Zimring & Hawkins 1973, p. 130). A distinction appearing in criminal law, which reflects this differential responsiveness to formal punishment, is that which classifies criminal acts as either mala per se or mala quia prohibita. The former are actions that are immoral in their own right (e.g. murder, incest) while the latter are actions that are illegal merely because they are prohibited by the law (e.g. tax evasion, marijuana use). Andenaes (1966) contends that since mala per se crimes are supported by the moral code of society, moral feelings and the fear of social judgement, rather than fear of formal punishment, act to prevent an individual from committing the crime. For mala quia prohibita, on the other hand, the law stands alone as a preventative force: so conformity is only secured by effective legal sanctions. Rates of tax evasion, on this argument, should be more sensitive to changes in the certainty and severity of legal punishment than rates of incest. Other scholars have made the distinction between instrumental crime, which is committed in order to attain some other goal, and expressive crime, which is committed because it is pleasurable in and of itself (Chambliss 1967; Zimring & Hawkins 1973). This distinction focuses attention on offender motivation or purpose for engaging in the criminal behaviour, as well as the emotional state of the offender at the time the decision is made. Instrumental crimes, such as parking violations, are thought to be more normally motivated and rational and are, therefore, more likely to be susceptible to the threat of punishment. On the other hand, expressive crimes, such as assault or murder, are often committed on impulse and in contexts of high emotional arousal. In this situation, where there is little contemplation of the consequences of the behaviour, punishment is likely to be a weak deterrent. 35

46 It is easy, however, to imagine crimes which do not fall neatly into the two categories described above. For example, a domestic violence victim may execute a plan to murder his or her partner, taking into consideration the risks associated with different aspects of the crime location and modus operandi. Alternatively, a dependent drug user who steals valuables from a retail store in order to obtain cash may engage in very little contemplation of the potential risks because of the need to alleviate physically painful withdrawal symptoms. Considerations like these prompt Zimring and Hawkins (1973) to argue that, rather than deterrence functioning for only one group of offences and not for others, deterrence is simply more difficult under certain circumstances than others. In situations where offenders are highly motivated, making decisions in a context of high emotional arousal and where there are no other major barriers to carrying out the crime, the deterrent effect of legal punishment is likely to be significantly reduced regardless of the offence. It is because mala per se and expressive crimes often fit this description, that formal sanctions would seldom deter these offence types. The role of informal social control The distinction between mala per se and mala quia prohibita also highlights the importance of informal social control in understanding compliance with the law. Traditionally, deterrence theorists have focused only on the negative consequences threatened by the criminal justice system, in particular the legal punishment prescribed by legislatures and/or imposed by the courts. However, other criminological theories have stressed the role of social influences in controlling delinquent behaviour and the importance of social and familial norms in explaining participation in and desistance from crime (e.g. Akers et al. 1979; Durkheim 1964; Hirschi 1969). Two such extralegal sources of conformity that have been identified in the deterrence literature are moral commitment to social norms and the threat of social stigma (Meier & Johnson 1977; Grasmick & Bursik 1990; Baum 1999; Nagin & Pogarsky 2001). For an individual with a strong belief that illicit conduct is wrong, the psychological dissonance that flows from violating an internalised norm, and the 36

47 associated feelings of shame, could operate to discourage offending. Likewise, the threat of disapproval from peers, family and colleagues may cause an individual to refrain from committing a crime more than any formal sanction. While evidence from survey data has demonstrated that these extralegal factors are a significant deterrent to crime (Grasmick & Bursik 1990; Grasmick, Bursik & Arneklev 1993; Baum 1999), it is difficult to disentangle the independent effects of informal social controls on criminal behaviour from those arising from the threat of formal punishment. For example, Baum (1999) measured the effect of informal sanctions on drink-driving behaviour by asking survey respondents whether they agreed with the statement my friends will think I am really stupid if I drove after drinking. He found that this factor was more important in deterring drink-driving than the perceived risk of apprehension created by a police enforcement program (Random Breath Testing). However, the threat of arrest created by Random Breath Testing and the threat of peer disapproval for drink-driving may not be factors operating independently of one another. Peers could react negatively to the fact that a person drove after they had been drinking because they know that he or she could be arrested for doing so. For this reason, Williams and Hawkins (1986) distinguish between social disapproval which stems from the stigma of arrest and social disapproval which stems from stigma of the act. Both these types of social stigma are informal social controls that can operate to prevent individuals from committing an offence. However, the former has legal sanctions as the source of control and therefore should ultimately be included as part of the general deterrence process (Williams & Hawkins 1986, p. 561). These authors argue that stigma arising from the act is likely to be least relevant to crimes that lack strong moral condemnation by the general community. For these offences, disapproval from significant others is more likely to arise from apprehension and conviction for the offence than from performing the act itself. To illustrate this concept mathematically, Nagin and Pogarsky (2001) include the extralegal costs arising from apprehension for an offence as an independent component of the disutility of the crime: 37

48 EU(G) c > P(L) c.u(legal costs + extra legal costs) c.d t (4) From equation 4 it can be seen that an increase in P(L) c can trigger both legal and extralegal consequences. In addition to social disapproval associated with the potential for arrest, these extralegal sanctions could include such things as reduced future job opportunities resulting from a criminal conviction or the negative impact of a criminal conviction on an individual s professional standing (e.g. registration as a health professional). Thus, even if the magnitude of legal sanctions for an offence is nominal, the potential disapproval from others for being arrested for an offence could nevertheless act to discourage offending. Feelings of shame from violating an internalised norm can result from a transgression that goes undetected, making it necessary to account for this additional nonlegal source of conformance in any proposed model of offender decision-making. Nagin and Paternoster (1994) do this by adding the term U(Moral Regret) c to the costs side of the deterrence equation. The ensuing model predicts that, even where there is no possibility of apprehension or formal punishment, U(Moral Regret) c can still exceed EU(Gains) c, and subsequently deter an individual from committing an offence. This accounts for those individuals who, irrespective of instrumental concerns, will never consider breaking the law. A further non-instrumental determinant of legal compliance, noted by Tyler (1990), is an individual s normative commitment to legal authorities. If a person perceives legal authorities and the laws they dictate or enforce as fair and legitimate, they will be more likely to adhere to their specifications. Alternatively, if a person believes that a particular law is unjust or unfairly enforced, then they are unlikely to modify their behaviour in accordance with the law. Tyler (1990) differentiates this type of normative commitment from the normative commitment to personal values or beliefs described above. He states that normative commitment through personal morality means obeying the law because one feels the law is just; normative commitment through legitimacy means obeying a law because one feels that the authority enforcing the law has the right to dictate behaviour (Tyler 1990, p. 4). In this way, even if an individual is not personally opposed to engaging in a particular behaviour 38

49 they may still comply with the law because they believe the authority s power to prohibit this behaviour is legitimate. Normative commitment would also be subsumed under the U(Moral Regret) c term added to the costs side of the ledger. Sherman and his colleagues (1992) describe the relationship between informal and formal sanctions depicted in equation 4 as the additive hypothesis in general deterrence. That is, both legal and extralegal consequences deter potential offenders and together these two factors will have a cumulative effect. However, these authors also note two other ways, described in the sociological literature, in which legal and informal threats of punishment can combine to deter offending. The first is the conditional hypothesis, which proposes that legal threats interact with informal social controls to create a more powerful deterrent effect than would be expected from either of these factors in isolation (i.e. a multiplicative relationship between informal and formal social controls). A corollary of this is that legal punishment can only be effective if reinforced by informal social controls. The second means by which formal and informal controls could combine is proposed by the replacement hypothesis, which expects legal control to be only effective if informal social controls are absent. To empirically test these three hypotheses, Sherman et al. (1992) undertook a field experiment, examining the differential effect of police arrest on the recidivism of domestic violence perpetrators who had a high stake in conformity (i.e. married and employed offenders), with those who did not. The subsequent analysis found that police arrest was associated with reduced recidivism for the former offenders, that is those for whom informal social controls were present, but was associated with increased recidivism for the latter offenders, thereby offering some support for the conditional rather than the additive or replacement hypotheses in deterrence. In describing the implications of these findings, the authors suggest that legal and informal control should not be thought of as mutually exclusive concepts that can be readily substituted. For individuals who are strongly bonded to society, informal sanctions may augment the deterrent effect of legal sanctions, but for those individuals where these informal controls are absent, formal punishment may prove ineffective in preventing further offending. In fact, the field experiment described by 39

50 Sherman et al. (1992) indicates that formal sanctioning may actually serve to increase the criminal behaviour of some offenders by labelling the individual as deviant or criminal. This labelling process would perhaps be more problematic for an individual who lacks the available social resources to overcome this type of social stigmatisation (i.e. those with a low stake in conformity). Delineating the relative contributions of informal social controls and legal threats to offender decision-making is a complex undertaking and one which is yet to be adequately resolved. Sherman s research does, however, serve to highlight the interdependent relationship that exists between these two factors. Zimring and Hawkins (1973) observe official actions can set off societal reactions that may provide potential offenders with more reason to avoid conviction than the officially imposed unpleasantness of punishment (Zimring & Hawkins 1973, p. 174). These societal reactions include feelings of embarrassment from the stigma assigned to arrest by significant others, as well as other informal sanctions, such as spill-over effects from an offender s friends and family who are adversely affected by the formal sanction and reduced opportunities for an offender to participate in group activities (Heckathorn 1990). An emphasis on legal sanctions can also serve to reinforce social norms that already exist in relation to proscribed behaviours. For example, the recent focus on the criminal nature of child sex offences aims to strengthen societal condemnation of these acts and, in so doing, increase anticipated offender shame associated with committing the crime (e.g. name and shame campaigns). In this situation, legal punishment operates to amplify if not initiate extralegal costs and thus can be considered a contributor to the deterrent effect. This suggests that, at least to some extent, the efficacy of informal social controls in regulating offender behaviour depends upon the existence of the formal sanctioning process (Heckathorn 1990). Further interaction effects in deterrence Sherman s research, in short, raises the possibility that threat responsiveness depends upon a potential offender s stake in conformity or ties to conventional society. Individuals with social responsibilities or strong social networks are more likely to fear public exposure (in the form of arrest for an offence) than are those for whom 40

51 these sources of informal control do not exist. Survey research examining sanction perceptions and offending behaviour, or intentions to offend, have found similarly heightened deterrent effects of legal sanctions for individuals who have ties to conventional society (see Nagin 1998). The presence of informal social controls is just one of several hypothesised conditions under which punishment may exert a stronger deterrent effect. Other factors that are thought to condition threat responsiveness include enduring individual traits, previous experience with legal punishment, and perceptions regarding the likelihood of being punished for committing an offence. In the deterrence literature, the relationship between these factors and legal punishment has been referred to as interaction effects (Homel 1986). Two further interactions relevant to the current research are: (1) the interaction between individual characteristics and sanction perceptions and (2) the interaction between perceived risk of apprehension and perceived severity of punishment. Individual characteristics and sanction perceptions Traditional deterrence theory says nothing about individual differences in levels of responsiveness to legal threats. It maintains that, because all persons are similarly motivated by pleasure, the costs of punishment should act to deter all people equally. Accordingly, variations in offending rates will be due to differences in punishment likelihoods and levels of severity rather than to characteristics of the potential offender population. This view of criminal behaviour conflicts with other criminological theories that emphasise the role of individual characteristics in explaining criminal propensity. Self-control and impulsivity One enduring personality trait thought to be pertinent to the effectiveness of legal sanctions is self-control. Self-control appears most prominently in Gottfredson and Hirschi s (1990) General Theory of Crime. According to this theory, people who commit crime will tend to be impulsive, insensitive, physical (as opposed to mental), risk-taking, short-sighted and nonverbal (Gottfredson & Hirschi 1990, p ), characteristics which are indicative of a person with low self-control. 41

52 These characteristics are developed at an early age, can be affected by parenting style and, once established, will remain relatively stable throughout life. Persons with a low level of self-control tend to engage in behaviours that bring about the immediate gratification of their wants and desires and, because the negative consequences of crime are often delayed, this inclines them to involvement in crime. Wilson and Hernstein (1985) also view criminality, or law abidingness, as the result of enduring individual differences in criminality, but cite a person s ability to plan for the future or to defer gratification (impulsiveness), as the key individual-level factor. This personality trait manifests itself as a tendency to think in terms of short-term rather than long-term consequences and, in combination with other stable individual characteristics, predisposes a person to engage in reckless behaviours, including those of an illegal nature. These criminological theories of enduring individual differences, like rational choice theories, accept that most crimes are motivated by rewards and the gratification of desires. They also recognise the sensitivity of potential offenders to incentives and disincentives associated with different criminal opportunities. However, they anticipate that people who are impulsive, or who have low self-control, will focus on the short-term benefits that could be derived from the offence rather than any longterm negative consequences, such as being caught and subsequently punished. In economic terms, these impulsive individuals would have a high discount rate and would therefore be less responsive to changes in future legal threats (Kenkel 1993; Nagin & Pogarsky 2001). In a similar vein, it could be argued that, because persons who have low self-control are self-centred and insensitive to others, they will have difficulties in establishing long-term relationships or persisting with education/career goals and as a consequence will have less to lose by committing the offence than would others who invest in conventionality (Nagin & Paternoster 1993; Wright et al. 2004). Both these accounts would suggest that those who are more criminally prone, that is those who are more present-oriented and impulsive, would be less influenced by the threat of formal punishment. Wright et al. (2004) offer an alternative hypothesis, suggesting that the criminally prone persons identified in the above theories may in fact be more likely to be 42

53 deterred by punishment than others. Consistent with some of the informal social control literature, these authors suggest that sanction threats will have minimal influence on individuals who have strong moral beliefs about the appropriateness of criminal behaviours. Their decisions in relation to the law are guided by moral rather than instrumental considerations. On the other hand, those who are less inhibited from crime by moral attitudes or commitments to social norms will be more affected by calculations of costs and benefits. This implies that impulsive and presentoriented people, should be the most responsive to changes in punishment variables. The empirical evidence for each of these theoretical perspectives is not clear-cut. There is some support for the notion that impulsive individuals, or, in Gottfredson and Hirschi s parlance, individuals with low self-control, are more likely to engage in criminal behaviours (Grasmick et al. 1993; Keane, Maxim & Teevan 1993; Nagin & Paternoster 1993) and for the notion that these individuals, being more impulsive, discount future costs associated with offending to a greater extent than other persons (Nagin & Pogarsky 2001; Pogarsky & Piquero 2004). However, other research shows that active offenders do consider the risks involved in committing specific crimes and in continuing their involvement in crime (e.g. Corbett & Simon 1992; Decker, Wright & Logie 1993; Piquero & Rengert 1999), and that they may be more sensitive to punishment contingencies than are other persons (Decker, Wright & Logie 1993; Paternoster & Simpson 1996; Wright et al. 2004). Wright and his colleagues (2004), for example, found that individuals who have low self-control and high self perceived criminality are more deterred by perceived costs of crime than those with relatively low scores on these measures. These authors analysed data from the renowned Dunedin (New Zealand) study, a longitudinal study of 1,000 individuals from birth to age 26. This analysis showed that the perceived risk of getting caught was significantly and negatively related to self-reported offending and, further, that this effect was greatest for those with a very low level of selfcontrol. Meanwhile, perceived risk was not predictive of self-reported crimes for those found to have a very high level of self-control. In discussing their results, Wright et al. (2004) emphasise the social-psychological nature of crime. Any theory of criminal behaviour, these authors propose, should 43

54 address both individual traits that predispose a person to crime as well as the social processes or structures impinging upon that individual. Thus, just as psychological and biological theories of crime need to recognise the sensitivity of criminally prone individuals to their social environment, so too does deterrence theory need to consider individual differences in present-orientedness or impulsivity when determining the role that threatened punishment plays in decisions to offend. All individuals discount future consequences to some extent. Impulsive individuals may discount future costs more than do others but this does not mean that they are incapable of foresight (Wright et al. 2004). In fact, given that criminally prone individuals are often people who lack internal moral restraint, instrumental considerations may feature more in their offending decisions than they do in the decisions of individuals with a high level of self-control. Previous offending Past punishment is another factor thought to influence an individual s sensitivity to legal punishment. A rational model of offending would predict that people who have already been formally sanctioned would believe the certainty of punishment to be higher, compared with those who have had no prior experience of punishment. They should therefore be more sensitive to future sanction threats. The heightened deterrent effect created by prior punishment is referred to in the criminological literature as specific or special deterrence, to distinguish it from general deterrence, in which people modify their behaviour as a result of observing or otherwise learning about formal punishment for particular offences. Stafford and Warr (1993) extend this definition of general and specific deterrence to incorporate further information about the nature of prior experience with legal sanctioning. According to these authors, individuals base their judgements about sanction risks not only on their experience with the sanctioning process but also their experience in avoiding formal punishment. Escaping detection for an offence can serve to reduce the perceived certainty that a particular transgression will result in formal punishment and subsequently increase the likelihood of offending. The more often punishment has been avoided, the greater the erosion of certainty perceptions and thus, the greater the chance that an individual will engage in the behaviour. 44

55 Stafford and Warr also note that one s own experience with punishment and punishment avoidance, and the experience of others (vicarious experiences), will both play a role in the formation of these perceptions. The former, they refer to as specific deterrence, while the latter they call general deterrence. Thus, rather than referring to distinct populations (i.e. those who have been previously punished vs. those who have not), specific deterrence is conceptualised as the deterrent effect of direct experience with punishment and punishment avoidance. General deterrence, on the other hand, is the deterrent effect of indirect experience with punishment and punishment avoidance (Stafford & Warr 1993, p. 127). Piquero and Pogarsky (2002) have recently undertaken an empirical test of the Stafford and Warr model using a scenario-based methodology and found some confirmatory evidence for the model s predictions. In particular, they found strong evidence that both personal and vicarious punishment avoidance is positively related to offending and that punishment and avoidance experiences influence behaviour via sanction risk perceptions. They also found strong evidence for the moderating impact of prior offending, that is, the finding that individuals who report having committed prior offences (in this study, previously driven whilst drunk) place more weight on personal experiences in their decisions to offend than on the experiences of others. Contrary to Stafford and Warr s hypotheses, however, this study found some evidence that punishment experience actually encouraged, rather than discouraged, future offending, with both personal and vicarious punishment experience found to be positively related to offending likelihood. Though somewhat counterintuitive, this finding is consistent with previous tests of the Stafford and Warr model (Paternoster & Piquero 1995; Piquero & Paternoster 1998), as well as more recent survey work on intentions to offend (Pogarsky & Piquero 2004). One obvious reason for a counter-productive positive punishment effect is that previous punishment experience simply identifies those individuals who offend more frequently because they perceive the risk of apprehension and conviction to be low (selection effect). However, Piquero and Pogarsky (2002; Pogarsky & Piquero 2004) suggest two decision-making biases which may also offer a potential explanation for these results: (1) Self-serving bias and (2) Gamblers fallacy. The self-serving bias 45

56 refers to the fact that people tend to see themselves in a favourable light when asked to make judgements. Piquero and Pogarsky (2002) note evidence from the decisionmaking literature showing that people generally view themselves as above average when it comes to several different tasks, such as driving a motor vehicle, managing people and being productive at work. In relation to criminal decision-making this could mean that offenders believe they are superior at avoiding detection than others and the knowledge that someone else has been punished for the same or similar offences may simply serve to bolster this misplaced belief in their offending prowess. The gamblers fallacy, on the other hand, refers to the faulty assumption that the probability of an event increases with each period it fails to occur. For a card player, this may result in a person gambling more money after losing three successive hands of poker on the assumption that the chance of losing four hands in a row is somehow reduced after several consecutive losses. Similarly, for an offender, being apprehended for a crime could be viewed as a rare occurrence and one which is unlikely to occur again soon after it has happened. Thus, punishment would result in a reduction in estimated sanction-risk for the next offending occasion and, consequently, increase the likelihood of further offending. Tests of these hypotheses provide support for the gamblers fallacy theory among individuals who are less experienced offenders, but not among high-risk individuals (Pogarsky & Piquero 2003). Indeed research, focusing on experienced offenders has suggested that offenders do update their certainty perceptions based on new information in the way anticipated by a rational model of offender decision-making. One study involving over 1,000 offenders convicted of serious offences found that felons with higher arrest ratios (i.e. self-reported arrests to self-reported crimes) also reported a higher risk of incurring further punishment (Horney & Marshall 1992). This latter study is particularly noteworthy, not only because it examined a sample of convicted felons, but also because, unlike previous research in this area, punishment experience was considered in conjunction with offending frequency, thereby producing a more valid test of the rational-offender model. Research on the formation of sanction risks is scarce and narrow in scope (Nagin 1998) and it is therefore difficult to draw definitive conclusions about the influence 46

57 of punishment experience on sanction-risk estimates. However, even if punishment experience does not translate into higher perceptions of arrest certainty after sanctioning, this does not mean that offenders with a prior conviction would not be responsive to changes in legal threats. In fact Homel (1986) found that offenders with a prior conviction were more likely to modify their offending behaviour after a significant increase in formal punishment than were other people. This finding is compatible with the view, expressed earlier, that individuals at high-risk of offending are less likely to be influenced by informal social controls and therefore perhaps more responsive to legal threats. Perceived risk of apprehension and perceptions of punishment severity An important contextual factor thought to condition the deterrent efficacy of legal sanctions is the level of perceived risk of apprehension and conviction within a certain jurisdiction. As mentioned previously, a central prediction of the subjective expected utility equation, as applied to deterrence theory, is that the certainty and severity of legal punishment combine multiplicatively rather than additively. Past research examining this aspect of deterrence theory consists mostly of experimental studies in which deterrence variables are manipulated and either prior or future intentions to offend are measured or, in the case of scenario-based methodologies, hypothetical likelihoods of offending are reported. Some of this research has found evidence in favour of the certainty/severity interaction while other studies have suggested that these two deterrence variables combine in an additive rather than multiplicative fashion. Howe and Loftus (1996) conducted three separate experiments in which respondents were asked to rate the extent to which particular punishment conditions might cause someone to refrain from committing a fairly serious crime (i.e. the deterrence value of the punishment threat). The first experiment used a repeated measures design in which respondents were presented with varying levels of punishment certainty, severity and celerity for a particular crime. The other two experiments used an independent groups design with different crime scenarios and sanction risk levels randomly assigned to each group. Across all scenarios and all designs, robust main 47

58 effects for punishment certainty and severity were found, as well as significant interaction effects for these two variables. Further research with active burglars supports the conclusions of Howe and Loftus. Decker, Wright and Logie (1993) presented active burglars with a hypothetical offending scenario in which the punishment risk, the reward to be gained from committing the crime and the anticipated penalty if caught were varied across subjects. The offenders were then asked if they would commit a burglary under the circumstances described. Compared to a matched, non-offender control group, these offenders were far more willing to commit the hypothetical offence but, nevertheless, were found to be sensitive to situational contingencies. In particular, the study found that penalty severity did not, on its own, significantly affect the likelihood of these offenders stating that they would commit the offence. It was only when the perceived risks were high and the anticipated penalties severe that stated offending likelihood was significantly reduced. Grasmick and Bryjak (1980) provide further support for an interactive deterrence model. These authors criticised previous perceptual research that had been conducted in this area because it had assumed that a particular punishment (e.g. one year imprisonment) would be equally severe to all people, thereby neglecting the subjective nature of punishment perceptions. To address this limitation, Grasmick and Bryjak asked their respondents to rate how big a problem the punishment, which they anticipate the court would impose for a particular offence, would create for them personally. Using this refined measure of severity, they found an inverse relationship between perceived severity of punishment and involvement in illegal behaviour, which was concentrated amongst those respondents who perceived the chances of being punished to be relatively high. One major problem with these cross-sectional surveys is that the inverse relationship between previous offending and sanction risk perceptions could reflect an experiential rather than a deterrent effect (see Nagin 1998). Research relying on Grasmick and Bryjak s (1980) measure of perceived punishment severity but utilising panel data instead of data on prior criminal involvement (thus predicting 48

59 subsequent rather than past criminal behaviour) was unable to reproduce the reported interaction effect (Paternoster & Iovanni 1986). More recent research using scenariobased methods also failed to find a multiplicative relationship between punishment certainty and severity (Nagin & Pogarsky 2001). Outside of the experimental literature, empirical research exploring the interaction between perceived certainty and severity is limited. This is not surprising given that the actual probability of being apprehended and convicted for most crime types is relatively low and that it is extremely difficult for law enforcement agencies to influence risk perceptions to any great extent. One study which did manage to investigate the certainty/severity interaction in a field setting was that conducted by Homel (1986), who investigated the impact of the introduction of Random Breath Testing (RBT) for drink-driving offences in New South Wales, Australia. RBT had a considerable impact on the perceived risk of being apprehended for a drink-driving offence because it ensured that any motorist could be stopped and tested for alcohol consumption, regardless of whether the attending police officer suspected the motorist had been drinking. The introduction of this new method of enforcement was widely publicised, with surveys showing that the vast majority of drivers knew about the introduction of these new initiatives (Cashmore 1985). Subsequently there was a significant increase in perceptions of apprehension risk, which resulted in drivers modifying their behaviour to avoid formal punishment (Homel 1988). A new offence for drink-drivers apprehended with a Blood Alcohol Concentration (BAC) above 0.15 was also introduced in New South Wales at the same time as RBT. This new offence was accompanied by severe penalties. Ten weeks following the introduction of RBT and the new penalty regime, Homel (1986) asked a sample of motorists whether they knew about the drink-driving penalty changes. He then questioned these same respondents six weeks later about their drink-driving behaviour since the first interview. A logistic regression analysis showed that respondents who believed drink-driving penalties had increased, at the time RBT was introduced, were 3.5 times less likely to have driven whilst drunk in the period spanning the two interviews than those who were not aware of the penalty changes. Although supportive of deterrence hypotheses, these results should be treated with 49

60 some caution, given that only 12 of the 175 respondents reinterviewed had reported driving whilst impaired during the six-week follow-up period. Other evidence, however, supported the regression analyses (e.g. a significant negative correlation between knowledge of the penalty increase and modifications to travel behaviours), leading Homel to conclude, when the perceived chances of arrest are high, perceived severity can have a deterrent impact additional to that of arrest certainty (Homel 1986, p. 127). Drink-driving as a case study This thesis is focused on the deterrent effect of punishment severity in relation to drink-driving offences. In light of the review we have just conducted, drink-driving is an ideal offence to use in a case study of deterrence for five main reasons. 4 Firstly, it is generally considered that drink-driving offences do not elicit the same moral condemnation as more traditional crimes, such as murder or break and enter. Drink-driving is prohibited in most countries, not because it is an act that is inherently wrong or against the moral code of society, but because alcohol consumption significantly increases a driver s risk of incurring an accident and thus constitutes a behaviour that is a threat to public safety. Accordingly, drink-driving tends to be classified, in criminal law terms, as a mala prohibita offence, or, in Zimring and Hawkins (1973) terminology, as an instrumental offence. These crime types have been identified in the deterrence literature as offences less likely to be 4 Given that alcohol can disrupt cognitive processing abilities, the applicability of a rational model of offending to alcohol-related crimes has been questioned by some academics (e.g. Exum 2002). However, even if alcohol does have a deleterious effect on an intoxicated driver s ability to process information about sanction threats, the fact that drink-driving involves some element of planning (unlike many opportunistic or impulsive crimes) means that formal punishment can still be influential in drink-driving decisions. Consistent with Cook s (1980) standing decisions that guide behaviour, formal sanctions may influence potential offenders through the general stance that they adopt in relation to drink-driving rather than acting to tip the cost-benefit ratio when a decision to drink and drive is being made. Homel (1986) draws upon a key concept in prospect theory known as precommitment (Tversky & Kahneman 1981) to further elaborate on this idea. He argues that punishment can encourage an individual to take actions in the present to render inoperative an anticipated future preference (Homel 1986, p. 33). A driver who has consumed large amounts of alcohol will have difficulty in rationally processing the utility of various alternatives. However, to avoid any possibility of experiencing future legal costs the sober driver may decide to employ preventive strategies, such as leaving the car at home or drinking low-alcoholic drinks. In this way, even if an intoxicated driver cannot be considered fully rational, the decision to drive and then drink can be modelled as a rational choice. 50

61 influenced by informal sources of social control and thus more responsive to threats of legal punishment. Secondly, drink-driving is far more common than most other types of criminal behaviour. In Australia, just over 15 per cent of drinkers aged 14 or over report having driven a motor vehicle whilst under the influence of alcohol in the last 12 months. In contrast, less than two per cent of drinkers admit to physically assaulting another person (Australian Institute of Health and Welfare 2002a). Thus, many people who might never contemplate engaging in other illegal behaviours still decide to drive home after a night out drinking. Surveys suggest that one reason for this is the belief amongst offenders that drink-driving is widespread (Corbett & Simon 1992; Baum 2000) and is viewed by many drivers as a violation of social conventions (Elliott 2003). Rather than being viewed as deviant individuals who seek out offending opportunities, many drink-drivers might be better viewed simply as risk-takers drawn from the general driving population. As such, we would expect situational contingencies, including increased threats of formal punishment, to be more influential for this offence than would be true for other more serious crimes. Thirdly, relatively precise and inexpensive handheld breath testers have been developed to assess a person s blood alcohol level. 5 The convenience of these breathalysers means that they can be widely used by police officers out in the field to enhance the detection of drink-driving offences. The accuracy of these tests means that officers can present reliable, objective evidence of a driver s Blood Alcohol Concentration (BAC) to the courts, thus boosting the chances of a drink-driving offence being proven. Many western jurisdictions have further strengthened the certainty of punishment for these offences by introducing per se laws. This type of legislation removes the necessity to demonstrate intoxication, or intent to drive whilst drunk, when attempting to prove the offence of drink-driving. In these instances, all that the courts require is evidence that, at the time of driving a motor vehicle, the defendant s BAC was over the prescribed legal limit. Legislation allowing for the breath testing of motorists, in conjunction with these per se laws, 5 These tests are based on expired air from the lungs. About 1-5% of alcohol absorbed by the body is expired via the lungs (Jones & Lacey 2001). 51

62 have ensured that the risk of apprehension and conviction for drink-driving offences is relatively high in comparison to most other offences. In this context, deterrence theory would anticipate that the efficacy of punishment severity should be enhanced. Fourthly, valid and reliable measures of offending (i.e. motor-vehicle crash rates) are available to evaluate the impact of drink-driving interventions modelled on deterrence principles (Ross 1984). A common problem associated with traditional measures of offending, such as recorded crime data, is that they necessarily include only crimes which come to the attention of officials. Criminologists have long known that only a small fraction of most criminal acts are ever reported to police. Furthermore, officially recorded offences can vary as a consequence of interventions other than those under investigation, which could lead to erroneous conclusions about the underlying reason for observed changes. Almost all road fatalities and crashes resulting in serious injury, on the other hand, are brought to the attention of authorities because of the need to treat victims. The use of road crash data, therefore, allows for more accurate and reliable conclusions to be made about the effect of a specific policy under examination (Ross 1984). Finally, this thesis examines the deterrent efficacy of a substantial increase in the severity of formal punishment within a real world situation and as such, has significant implications for public policy. Drink-driving is an important social problem that results in significant loss of life and which places a substantial burden on health and criminal justice resources. Over 400 people die, and almost 8,000 persons are hospitalised, for a road injury attributable to alcohol each year in Australia (Chikritzhs et al. 2000). This translates to just over 17,000 person-years of life lost from road fatalities and almost 45,000 bed-days required for road-injury hospitalisations. The social and economic cost of these alcohol-related injuries to the Australian community is estimated to be over AUS$3.4 billion each year (Collins & Lapsley 2001). Any strategies that can be demonstrated to reduce the incidence of this behaviour will therefore be highly relevant for decision-makers working in this area of public policy. 52

63 CHAPTER 3. DETERRENCE RESEARCH ON DRINK- DRIVING As mentioned in the introduction to this thesis, hypotheses generated by deterrence theory have been investigated both at the aggregate- and the individual-level of analysis. Aggregate-level deterrence research consists of two main types: ecological studies 6 and quasi-experimental studies (Nagin 1998). Ecological studies look at natural variations in sanction levels and crime rates across time or space, while quasi-experimental studies examine the effect of targeted and specific interventions on measures of offending behaviour. In the context of drink-driving, ecological and quasi-experimental studies have mostly relied on motor vehicle crash rates as a surrogate measure of offending because of the well-established positive relationship between alcohol consumption and crash risk (Zador 1991; Zador, Krawchuk & Voas 2000). In particular, fatal, serious injury and single-vehicle night-time motor vehicle crashes, which have been demonstrated to include a greater proportion of alcoholrelated incidents (Evans 1991), are frequently used as a proxy for drink-driving in aggregate-level studies. Another substantial deterrence literature that has emerged over the last two decades emphasises the perceptual nature of the deterrence process and, consequently, the importance of investigating deterrent effects at the individual-level. This type of research uses survey data in order to examine perceptions of sanction risk and sanction severity, and the relationship between these perceptions and self-reported offending. Most reviewers have considered perceptual studies, in conjunction with the aggregate-level research described above, to be the mainstay of the deterrence literature (Nagin 1998; von Hirsch et al. 1999; Weatherburn et al. 2000). However, in reviewing the deterrent efficacy of punishment severity, a further type of research, which also features the individual as the unit of analysis, is important: reoffending 6 This type of research has been given various names in the literature including association (e.g. von Hirsch et al. 1999), correlational (e.g. Ross & McCleary 1983), criminal opportunity (e.g. Cook 1980) and passive observational (e.g. Cook & Campbell 1979) studies. Generally speaking, they are all studies that examine how crime rates covary with arrest, conviction or sentencing rates over time or across jurisdictions, or both of these. 53

64 studies. Reoffending studies are primarily concerned with the marginal deterrent effect of particular types of judicial penalties and/or the magnitude of judicial penalties, and typically rely on official records to measure deterrent outcomes. Their importance means that they, too, comprise a major component of this literature review. Aggregate-level research Nagin (1998) notes in his recent review of the deterrence literature that the most important issue for policy makers is not whether the criminal justice system in its totality prevents crimes but whether a specific policy, grafted onto the existing structure, will materially add to the preventative effect (Nagin 1998, p. 3). Accordingly, this section will review aggregate-level deterrence research that has investigated the relative importance of each of the three elements of the deterrence equation, namely the certainty, severity and celerity of drink-driving punishment. A summary of these studies, including a description of the methodology they employ and any significant outcomes found in the research, can be found in Appendix I. Punishment certainty Strategies that increase the probability of detection, apprehension and conviction aim to influence the perceived risk of punishment for drink-driving and in doing so operate to deter an individual from committing the offence. Increasing the perceived probability of detection for drink-driving offences can be achieved by boosting the number of police allocated to the task of drink-driving enforcement, or by adopting strategies such as roadblocks, sobriety checkpoints or Random Breath Testing (RBT). The probability of apprehension and/or conviction can also be increased through legislation that allows police to conduct preliminary breath tests of drivers at the roadside and/or allows the Blood Alcohol Concentration (BAC) of motorists to be used as sufficient proof of a drink-driving offences (per se legislation). Ecological studies Ecological research has compared the impact of several of these deterrence-based policies on State-level motor vehicle fatality rates. A feature of the analytical models utilised in these studies is the inclusion of a broad range of control variables that are 54

65 thought to have an effect on road fatality rates and/or alcohol consumption, over and above that created by the deterrence-based policies being examined. Potentially confounding factors that have been controlled for by econometricians include the number of kilometres travelled, the number of vehicles on the road, weather information, traffic density, alcohol sales, the price of alcohol, unemployment rates and disposable income. Several of these earlier studies found that proxy variables for the certainty of punishment were not correlated with vehicle-fatality rates. Wilkinson (1987), for example, measured differences in law enforcement across USA States in terms of the probability of arrest (arrests per vehicle mile) and conviction (per cent of arrestees found guilty or pleading guilty) for drink-driving but found no marginal effect for these variables on the level of motor vehicle fatalities from 1976 through Similarly, Evans, Neville and Graham (1991) report no conclusive evidence that any specific form of punitive drink-driving legislation implemented in the USA between 1975 and 1986 had a measurable effect on motor vehicle fatalities. However, other ecological research suggests that State laws facilitating arrests and/or convictions can deter drink-driving. Saffer and Chaloupka (1989) found that the introduction of preliminary breath test laws in the USA was associated with a significant reduction in vehicle-fatality rates from and Sloan, Reilly and Schenzler (1994) report that police availability (as measured by police/1,000 population and total police expenditure/1,000 population) reduced USA State vehicle-fatality rates during the period 1982 through Benson, Rasmussen and Mast (1999) tried to make sense of these disparate findings by grouping together several proxies for deterrence-based policies in their analysis of USA vehicle-fatality rates. This enabled the authors to investigate whether a combination of policies aimed at deterring the drinking driver has an impact on alcohol-related road fatalities, even when an individual law appears to have no significant marginal effect. In this analysis, the driver involvement rate was used as a proxy for drink-driving offences, calculated as the number of fatally injured drivers with a positive BAC (information that was available for only 15 USA States) divided by the number of drivers in each State. This analysis found that a subset of 55

66 factors influencing the probability of being stopped (open container laws, anticonsumption laws 7 and police per capita) and/or arrested (illegal per se laws, preliminary breath test laws and implied consent laws) 8 had a significant impact on driver involvement rates, controlling for other confounding factors (including vehicle miles travelled per driver, males aged per capita, per capita disposable income, unemployment rate, metropolitan population). Among these variables, those grouped as influencing the probability of being stopped seemed to be the most important factor in reducing road fatalities. It is possible that some ecological studies have failed to find a deterrent effect for drink-driving policies because the outcome measure used in this research, that is vehicle-fatality data, is an insensitive measure of offending. These data include numerous instances of non alcohol-related fatal crashes and exclude many non-fatal alcohol-related crashes and drink-driving episodes that do not result in a crash. Cook (1980) observes that most policy innovations have a small effect at best, and this effect can easily be submerged and lost in the normal fluctuations of an insensitive measure (Cook 1980, p. 252). In the research described above, Benson and his colleagues (1999) attempted to address this issue by examining only fatal accidents in which the driver had a positive BAC. However, this measure still excludes both non-fatal alcohol-related crashes and drink-driving episodes where no crash results. For this reason, other ecological research has relied on self-reported drink-driving information contained in survey data as a measure of drink-driving offence rates. Kenkel (1993) examined data from a USA national health survey to investigate the deterrent effect of State laws allowing preliminary breath tests and sobriety checkpoints, as well as laws prohibiting plea-bargaining in drink-driving cases. The dependent measures examined in this study included both self-reported alcohol consumption and self-reported drink-driving. The subsequent analysis showed that each of the policies considered, with the exception of plea-bargaining prohibitions, was associated with reduced heavy drinking and with reduced drink-driving 7 Open container laws prohibit the possession of any open alcoholic beverage in the passenger area of a motor vehicle and anti-consumption laws prohibit the consumption of any alcoholic beverage in the passenger area of a motor vehicle. 8 Implied consent laws dictate that any person who operates a motor vehicle is presumed to have given his or her consent to a test to determine their Blood Alcohol Concentration. 56

67 episodes. Moreover, when the effect of the law on heavy drinking was held constant, there was no evidence for an independent effect of these policies on self-reported drink-driving. The latter finding suggests that the State laws in question affected drink-driving behaviour through their impact on alcohol consumption, lending some credence to the notion that the joint decision to drink and drive can be modelled as a rational choice (Chaloupka, Saffer & Grossman 1993). Similarly, Chaloupka and Weschler (1996) considered the effect of drink-driving policies on self-reported binge drinking amongst young people. They used an index developed by the USA advocacy group Mothers Against Drunk Driving (MADD) to measure the restrictiveness of each State s drink-driving laws. This index reflected factors that increased the probability of arrest, eased standards for arrest and conviction and raised penalties for a conviction. Utilising a large data set from a survey of USA college students, the authors concluded that strong State-level policies related to drink-driving can significantly reduce binge drinking in young people. In this instance, however, it was the overall package of deterrence-based policies (including both those affecting the certainty and the severity of punishment), which was found to deter drink-driving (through reduced binge drinking), rather than any specific individual law. Reviews of the deterrence literature have highlighted several methodological shortcomings of studies that rely on natural variations in punishment levels and crime rates to test casual relationships between these two variables (e.g. Cook 1980; Nagin 1998; von Hirsch et al. 1999). Nagin (1998) suggests that the most salient of these is the simultaneity problem: the reciprocal relationship that potentially exists between crime and sanction levels. Consider a model in which drink-driving arrests in each State are used to index punishment certainty and State-level alcohol-related road fatality rates are employed the dependent measure. A significant negative association between these two variables could be interpreted as evidence that arrests reduce road fatalities (i.e. Arrests Fatalities). However, if a particular jurisdiction has a relatively high road fatality rate, public demand for the problem to be addressed may cause authorities to increase the level of police activity targeting these offences and, consequently, increase the number of arrests. In this instance, road fatalities 57

68 would have an effect on the number of drink-driving arrests (i.e. Fatalities Arrests). This reciprocal relationship between crime rates and sanction levels is often ignored in ecological studies investigating deterrent effects, resulting in misspecified models of deterrence. A further shortcoming is that ecological studies require the use of an index of the threat level achieved by a State s drink-driving policies yet often the index used in these analyses is of limited value (Cook 1980). For example, binary variables representing the existence of sobriety checkpoints or preliminary breath test laws have been employed in econometric models as indices of the probability of punishment for drink-driving offences. However, the extent to which these laws are enforced could vary considerably across jurisdictions and a binary variable would be unable to capture these differences in degree. Other studies have employed police expenditure or police allocation per capita as measures of probability of arrest but these indices are similarly flawed. If police expenditure is equal in two States this does not necessarily imply that they allocate the same level of resources to the task of detecting drinking drivers. Indeed, Benson, Mast and Rasmussen (2000) show that, as police shift resources into the control of violent and property crimes, the likelihood of a drink-driving arrest decreases significantly. The use of inadequate estimates of punishment probability is one possible reason why ecological research has produced such inconsistent findings regarding the deterrent effect of drinkdriving enforcement policies. Quasi-experimental studies Quasi-experimental analyses of deterrence-based interventions are thought to be a more promising methodology because they incorporate many features of a true experimental design a well-defined treatment regime, measurement before and after an intervention and (in many cases) a control group (Ross & McCleary 1983; Ross 1984; Nagin 1998). 9 This means that quasi-experimental studies overcome several of the threats to inferential validity associated with correlational designs. In particular, the interrupted times-series technique, which measures the target 9 This experimental design is quasi in the sense that there is no control over the behavioural event which constitutes the intervention variable. 58

69 behaviour at several points before and after an intervention, has frequently been used in quasi-experimental research to examine the effect of specific drink-driving enforcement policies on surrogate measures of offending. This technique addresses rival explanations for observed effects, such as regression to the mean (extreme observations that naturally gravitate back to the average over time), maturation (natural growth processes unrelated to but temporally coincident with the intervention which could cause a pre-post difference), and instability (fluctuations in sampling persons or components) (Cook & Campbell 1979). Although the interrupted time-series analysis cannot itself control for history (some specific event, unrelated to but temporally coincident with the intervention which could cause a prepost difference) it can be supplemented with additional data, such as rates of non alcohol-related crashes, to address this additional threat to inferential validity (Ross 1984). A further added advantage of quasi-experimental designs is that by focusing on the implementation of a specific policy it is possible not only to measure changes in the target behaviour (i.e. drink-driving offending) before and after the intervention, but also ascertain the degree to which the new policy has influenced activity levels and decision-making (Cook 1980). Consequently, it can be determined whether negative findings are due to the policy or law failing to have a deterrent effect or, alternatively, failing to affect punishment certainty in the jurisdiction where it has been implemented. This allows for more informed conclusions regarding the efficacy of specific deterrence-based interventions for drink-driving offences. Much of the early quasi-experimental drink-driving research in the deterrence literature focused on the effectiveness of the Scandinavian approach to controlling the drink driver because of evidently low drink-driving rates in these countries. Scandinavian-type laws give police the power to use chemical testing devices to ascertain a driver s BAC and allow breath test results to be used as sufficient evidence of alcohol intoxication. Jurisdictions without this type of legislation typically rely on behavioural evidence of impairment (or sobriety tests) to discern whether a driver is under the influence of alcohol. The Scandinavian strategy is considered to be superior to behaviour-based approaches for apprehending and 59

70 convicting alcohol impaired drivers because it does not rely on the subjective judgement of police officers to prove that a drink-driving offence has been committed. Ross (1984) conducted a major review of quasi-experimental studies examining the effectiveness of these types of drink-driving countermeasures in several different countries. From this review he concluded that the implementation of Scandinaviantype laws has had a significant impact on alcohol-related traffic injuries and fatalities. The best example of the deterrent effect of this type of legislation comes from Ross (1984) evaluation of the British Road Safety Act This piece of legislation set a blood alcohol limit of 0.08g/100ml and allowed police to demand a preliminary breath test from motorists who committed a traffic infringement or were involved in a crash. An interrupted time-series analysis, examining the effects of this intervention, found a 66 per cent decline in fatal and serious injury crashes occurring immediately after the enactment of the legislation. This effect was shown to be particularly strong in drinking hours but not apparent in non-drinking hours. The effect, moreover, was still evident after controlling for other confounding variables, including total miles travelled by drivers and variations in alcohol sales over time. Although the evidence for an initial deterrent effect created by the enactment of this legislation was strong, Ross observed that the effect dissipated within a few years. His review of similar drink-driving laws implemented in other western countries (including New Zealand, France and Canada) generally found that such interventions were similarly successful in generating an initial deterrent effect but that this effect decayed over time. Simply increasing the numbers of police dedicated to drink-driving enforcement has also been shown, by quasi-experimental research, to have an impact on alcoholrelated road crashes. An enforcement program in Stockton, California, introduced special patrols on weekend nights dedicated to detecting and arresting drink-drivers at high-risk times. Although roadblocks, per se legislation and preliminary breath tests were not available to the Stockton police, there was still a significant impact of the patrols on drink-driving in this community. The beneficial impact of the police patrols was evidenced by a 43 per cent reduction in the number of drivers who 60

71 recorded an illegal BAC in roadside breath test surveys and a per cent reduction in night-time road crashes. Comparisons with night-time crashes in four control cities showed that crash trends in Stockton were significantly different from any of these control sites. However, once the extra police were withdrawn from the task, the observed impact on crashes disappeared (Voas & Hause 1987). A specific enforcement program aiming to increase both the probability of apprehension (through preliminary breath testing) and the probability of detection (through greater use of sobriety checkpoints) provides further evidence of the deterrent effect of the certainty of punishment. This program was introduced in three Californian communities and was accompanied by a media campaign which aimed to increase the perceived risk of punishment by publicising the drink-driving enforcement efforts of local police. Voas, Holder and Greunewald s (1997) evaluation of this program found that there was a significant reduction in singlevehicle night-time crashes in the three treatment communities, without any similar change in crash rates in three pre-selected control sites. Additional evidence showed a significant reduction in the number of drivers with positive BACs in the areas where enforcement had been boosted. Furthermore, drivers surveyed in the test sites reported a greater perceived risk of apprehension for drink-driving and a consequent decrease in the number of times they had driven while drunk. Research on Random Breath Testing (RBT) in New South Wales, Australia, has provided perhaps the strongest evidence that enforcement can reduce drink-driving and subsequent alcohol-related road crashes through the process of deterrence. RBT was introduced in New South Wales in December 1982 with the specific aim of increasing the perceived probability of punishment for drink-driving offences by maintaining high levels of systematic and sustained police efforts against drinkdriving. This enforcement technique resembled those previously discussed, in that it made use of preliminary breath test and per se laws. A defining feature of RBT legislation, however, was that it gave police the authority to stop any motorist and administer a preliminary breath test to them. This method differs from other (less stringent) testing procedures, such as roadblocks and sobriety checkpoints: where only drivers who are judged to have been drinking or who are obviously impaired are 61

72 required to undertake a breath test. The NSW Police aimed to carry out one random breath test for every three licensed drivers, each year, in order to establish a relatively high level of probability of detection. The RBT program in New South Wales was also accompanied by a considerable amount of media coverage and widespread publicity campaigns designed to augment police efforts by increasing the perceived risk of detection and apprehension amongst the general public (see Cashmore 1985). Subsequent analyses found that the introduction of RBT was accompanied by an overall 19 per cent reduction in all serious accidents, a 48 per cent reduction in fatalities and a 26 per cent decline in single-vehicle night-time crashes. Comparable effects on road fatalities and injuries were also observed in Queensland, Western Australia and Tasmania after RBT was introduced in these States (Henstridge, Homel & Mackay 1997). Furthermore, the effect of RBT on drink-driving appeared to be the direct result of an increase in the perceived risk of apprehension for that offence. A survey of drivers immediately after the introduction of RBT in New South Wales found that the perceived chances of arrest increased significantly among NSW motorists and that these perceptions predicted the number of ways in which respondents were modifying both their drinking and their driving behaviour (Homel 1986). Similar findings for the effectiveness of police enforcement in reducing alcohol-related fatalities and injuries are also evident from evaluations of less intensive RBT operations in the Netherlands (Zaal 1994). Importantly, unlike other enforcement initiatives, RBT operations in New South Wales appeared to have a sustained effect on drink-driving rates. Henstridge, Homel and Mackay (1997) estimated that, at least for single-vehicle night-time accidents, the impact of RBT took over 10 years to decline to five per cent of its initial magnitude, meaning that, by 1992, the effect of introducing RBT in New South Wales was still discernible. The most striking result, however, was that these longterm effects appeared to depend greatly on enforcement levels (as measured by the number of breath tests administered by police). Initially the effects of introducing RBT did wear off, but over the longer term the reductions were maintained because police increased the number of breath tests they administered to motorists after

73 In fact the ability of the NSW Police to control drink-driving was shown to be a direct function of how much effort they put into it. An increase in 1,000 breath tests per day was found to correspond with a decline of 19 per cent in single-vehicle nighttime crashes. While significant, this relationship between daily testing rates and crash reductions was not linear, since there was evidence for diminishing returns as enforcement levels increased (Henstridge, Homel & Mackay 1997). The research conducted by Henstridge and his colleagues shows that the long-term success of deterrence-related enforcement relies heavily on high levels of visible police activity to maintain perceptions of risk apprehension at a relatively high level. Ross and LaFree (1986) suggest that an individual s perception of risk does not remain constant but is continually updated on the basis of new experiences. Once people become aware of an increase in enforcement activity, they initially overestimate the risk of apprehension but, with repeat successful drink-driving episodes or reduced publicity of police enforcement efforts, they lower these risk estimates. Homel (1986) maintains that it is necessary to take account of these changes in perceptions when attempting to understand the deterrent effect of enforcement activities. He likens the deterrent effect of RBT to a leaky bucket. That is, it is a delicate balance between the forces maintaining (i.e. visible police enforcement) and those tending to erode (e.g. peer pressure, lack of exposure, successful drink-driving episodes) the perceived risk of apprehension. He concludes that the long-term effect of RBT will depend on the relative sizes of the input and the output effects in other words, how full the bucket can be kept through police enforcement (Homel 1986, p ). Punishment severity So far, we have only discussed the deterrent effect of punishment certainty. However, deterrence theory also suggests that more severe punishments should reduce the incidence of offending. Aggregate-level research has investigated this through two methods. The first involves quasi-experiments capitalising on sudden changes in statutory penalties in order to investigate whether raising the overall threat of punishment is an effective deterrent for drink-driving. A limitation of this research is that sentencing policies often fail to influence the actual penalties 63

74 convicted offenders receive. The second involves studies of actual sentencing practice to determine the impact of increased penalties on offending rates. This latter category includes studies that compare variations in judicial sentencing practices across jurisdictions, as well as studies that investigate the impact of introducing mandatory sentencing policies within a jurisdiction. Increases in the severity of statutory sanctions One method commonly employed by criminal justice policy makers as a means to increase the perceived severity of punishment is to raise the overall threat of sanctions. This can be achieved through laws that increase the magnitude of penalties for all offenders, for particular offender groups (e.g. repeat offenders) or for certain offences (e.g. drink-driving offences where the offender records a high BAC). An evaluation of California s 1982 driving-under-the-influence (DUI) legislative reforms provides support for the deterrent effect of increased statutory penalties for drink-driving offences. These laws introduced more punitive sanctions for drinkdriving offences, including stricter licence restrictions and longer gaol sentences for repeat offenders. Rogers and Shoenig (1994) examined the effect of these legal changes on three road crash categories thought most likely to involve alcohol, namely night-time crashes, single-vehicle night-time crashes and crashes designated by the police as involving a driver that had been drinking. An interrupted time-series analysis found significant decreases in these three road crash categories immediately after the implementation of the new countermeasures. The researchers attempted to isolate the effects of the reforms from changes in the social climate surrounding drink-driving produced by the lobby group Mother s Against Drunk Driving (MADD). This was achieved by including the formation date of MADD as a separate intervention variable in the time-series model. This additional analysis demonstrated that, although the effects of the 1982 legislation were embedded in a larger downward trend, they still remained statistically significant even after accounting for this trend. Peck (1991) also examined the impact of the 1982 Californian legislative reforms on drink-driving recidivism rates in order to investigate the specific deterrent effect of 64

75 these new laws. He examined the failure rates, for both drink-driving offences and major traffic offences (such as reckless driving), of two large offender cohorts who received a drink-driving conviction in either 1980 or All drink-driving and major traffic offences committed up until 1989 were included in the analysis and survival and hazard functions were generated. The analysis showed that the 1984 cohort was less likely to reoffend than the 1980 cohort, both in terms of drink-driving offences and other major traffic offences. It should be noted, however, that, in addition to increased penalties, the new laws mandated alcohol-related treatment programs for some offenders. This additional component of the legislative reforms could have contributed to the observed reduction in reoffending rates. An evaluation of a specific legislative amendment to Swedish drink-driving laws in 1994 provides further evidence for the general deterrent effect of more severe statutory penalties. These legal changes provided for lower BAC levels and increased penalties for aggravated drink-driving offences (from one year to two year imprisonment), with the added requirement that imprisonment should be the usual punishment for this offence. Borschos (2000) examined road traffic crashes between 1986 and 1997 to identify any changes occurring after the introduction of this legislation, while controlling for other confounding factors, such as per capita alcohol consumption and traffic density. This analysis revealed a significant reduction in road fatalities during the study period, over half of which could be explained by the introduction of the harsher penalties, independently of other factors included in the model. A limitation of the quasi-experimental analyses presented above is that the increases in statutory penalties evaluated by these studies formed part of a larger package of legal interventions aimed at deterring the drink-driver. California s severe statutory penalties were introduced at the same time as a 0.1 per se law, a law that increases the certainty of conviction by removing the need to prove behavioural impairment. Likewise in Sweden the introduction of more severe penalties coincided with a lower BAC limit, as well as modifications to punishments for other serious driving violations and improved police control measures. The fact that several legal measures were introduced simultaneously makes it difficult to determine which of 65

76 the interventions produced the desired effect. Since there is strong evidence that increasing the certainty of punishment can have a substantial deterrent effect on drink-driving, the findings reported in these evaluations may be due to changes affecting the likelihood of arrest and conviction, rather than the increased threat of severe sanctions. Other research examining increases in the severity of statutory punishment for drinkdriving offences, with no concurrent increase in the certainty of punishment, has failed to confirm the general deterrent effects of increased statutory penalties. Beirness, Simpson and Mayhew (1993) assessed the impact of legislation introduced in Canada in 1985, which established two new offences of impaired driving causing bodily harm and impaired driving causing death, both which carried maximum penalties of 10 to 14 years in prison. The legislation also increased fines and mandated minimum three-month licence suspensions for existing impaired driving offences. 10 An evaluation of these changes showed that, during the 1980s in Canada, there were significant reductions in several measures of drink-driving behaviour and its consequences, including alcohol-related fatalities and crashes, drink-driving arrests, self-reported offending and positive breath tests in roadside surveys. Unfortunately the downward trend was well established prior to the amendments and there is therefore no clear evidence that could specifically link these reductions to the legislative changes introduced in Beirness and his colleagues suggest that the absence of any observable impact of the increase in drink-driving penalties could have been due to low public awareness of the legislative changes. If drivers are not aware of the increased threat in sanctions for drink-driving, they are unlikely to alter their perceptions of the risk associated with this criminal behaviour and therefore would not be expected to be deterred. Furthermore, the researchers observed that relatively few drivers were charged with the new offences and, among those who were, many were convicted of a less serious offence. 10 The offence of impaired driving in Canada is equivalent to drink-driving offences in Australia 66

77 Ross (1984) also found little evidence for a general deterrent effect of a Finnish law doubling the maximum statutory penalty for drink-driving from two to four years imprisonment. An interrupted time-series analysis of road fatality and injury data showed a statistically significant decline in crashes resulting in an injury but no real change in single-vehicle road fatalities or total road fatalities. Since the latter two crash categories more often involve alcohol, Ross concluded that the increase in the statutory severity of sanctions in Finland did not have observable deterrent consequences. He did, however, note that, although the legislation doubled the maximum penalties for drink-driving, the actual penalties meted out by judges did not change. We noted in Chapter 2 that the certainty and severity of legal punishment are predicted by deterrence theory to combine multiplicatively rather than additively. It follows that increases in perceived severity should exert a bigger effect when punishment is highly probable. The legislative changes introduced in Finland occurred well before per se laws or random breath testing had been adopted in that country. In Canada, a system of random stopping of vehicles is permitted but suspicion of alcohol intoxication is required before a breath test can be requested. Furthermore, most drink-driving enforcement in Canada results from an incident, such as a collision, rather than random stops of drivers (National Highway Traffic Safety Administration 2000). Thus, any potential deterrent effect of the increased severity of punishment created by the legislative reforms in Finland and Canada could also have been moderated by the low certainty of punishment in these jurisdictions. The importance of the interaction between perceptions of arrest certainty and perceptions of penalty severity is illustrated by Hingson et al. s (1987) evaluation of changes to statutory penalties in Massachusetts and Maine. In the early 80s both these USA States introduced new and more severe statutory penalties to increase the overall threat of punishment for drink-driving offences. In the case of Maine, the legislation was deemed the toughest drunk driving law in the nation (Hingson et al. 1987, p. 593). However, no sustained benefits of the legislative changes, in terms of reductions in road fatalities or self-reported drink-driving episodes, were observed 67

78 for either State. Survey data suggested that the null findings might have been influenced by the low perceived probability of arrest in these States. Pre- and postlaw surveys indicated that, while there were substantial increases in the public perception that drunk drivers would be arrested, convicted and given severe penalties, few people believed it was very likely that police would actually stop drunk drivers. In instances such as these, where there exists a relatively low chance of being caught, drivers are acting rationally in ignoring the threat of severe punishment (Ross 1984). The studies reviewed above reveal mixed evidence as to whether severe statutory penalties can reduce the incidence of drink-driving through the process of deterrence. The success of such a strategy however is likely to depend on whether the new sanctions are implemented as intended, the public is aware that more severe penalties have been introduced and whether the perceived probability of punishment is high or low. Policies impacting sentencing practice The fact that policies increasing the severity of statutory penalties may not result in actual changes in sentencing practice, presents a problem for studies examining the deterrent effect of changes to statutory penalties. As Nagin (1998) observes actual policy bears little resemblance to intended policy because the exercise of discretion by key actors of the criminal justice system drives a wedge between the reality of policy and its intention as expressed by its formulators (Nagin 1998, p. 34). According to deterrence theory, a sudden increase in statutory penalties should reduce offending by augmenting perceptions of punishment severity if arrested and convicted for an offence. However, if offenders and the general public come to learn that the legislative changes have had little impact on actual sentences imposed by the courts, individuals may readjust their perceptions of punishment severity and any deterrent effect created by the policy may be diminished. Negative findings of sanction-based policies could therefore be due to the failure of the legislative reforms to influence actual sentences rather than the ineffectiveness of more severe punishment in deterring potential offenders. 68

79 Other research has overcome this problem by taking advantage of naturally occurring variations in the sentencing practice of different judges in order to examine the deterrent efficacy of severe punishment. When individual judges differ significantly in the penalties they impose and court cases are assigned to judges in an arbitrary way, a comparison can be made between groups of offenders, who would be expected to be relatively similar in all respects except for the penalty they receive from the sentencing judge. Martin, Annan and Forst (1993) conducted one such natural experiment to examine the specific deterrent effect of mandatory gaol sanctions. Capitalising on the fact that some judges did not comply with a judicial policy in Minnesota mandating a two-day gaol sentence for first time drink-driving offenders, these researchers compared the recidivism rates of persons sentenced by a gaol judge with that of a no gaol judge. They found no significant difference between the two groups of offenders with regard to the number of drink-driving reconvictions recorded during a 23-month follow-up period. Research conducted by Ross and Voas (1990) also failed to find an effect for a more severe judicial sentencing policy that mandated imprisonment. This study compared drink-driving outcomes of two communities: one where the judiciary had adopted a harsh sentencing policy of mandatory 15-day gaol sentences (accompanied by heavy fines and licence restrictions) for first offenders, and another which did not. Since it was a high profile judge who adopted the harsher sentencing policy, a great deal of media attention surrounded the new penalties and consequently the policy was expected to deter drink-driving in the wider community. To assess the general deterrent effect of the policy the authors relied on several outcome measures of drink-driving, including roadside breath tests and driver surveys. This analysis found that the judicial sentencing policy had a strong effect on the punishment expectations of drivers in New Philadelphia, but there was no difference between the two communities in the proportion of people self-reporting driving under the influence of alcohol or those who tested positive to a roadside breath test for alcohol. This finding conflicts with the prediction of deterrence theory that increases in perceptions of punishment severity should deter individuals from committing an offence. Again, however, these researchers warn that the absence of a deterrent effect may have been due to the low perceived probability of apprehension in this jurisdiction. 69

80 Another test of the deterrent effect of harsher sentencing practices comes from evaluations of policies mandating minimum penalties for drink-driving offences. Mandatory minimum penalties remove some judicial discretion from sentencing decisions, thereby increasing the likelihood that harsher sanctions established by the legislature will be implemented as intended. Ecological research examining variations in road crash fatality rates across USA States provides equivocal evidence on the effectiveness of State laws mandating specific penalties for drink-driving offences. Sloan, Reilly and Schenzler (1994), for instance, demonstrated that mandatory gaol terms, mandatory minimum fines and mandatory minimum periods of licence disqualification did not succeed in deterring offenders. On the other hand, Chaloupka, Saffer and Grossman (1993) concluded that mandatory administrative licence suspensions, high minimum fines and minimum licence suspensions all significantly reduced road fatalities, but mandatory gaol sentences did not. As mentioned previously, a shortcoming of these ecological studies is their reliance on a binary variable to index the threat level achieved by each State s drink-driving policies. With respect to mandatory minimum penalties, a dichotomous variable fails to take account of interstate differences in the actual severity level of penalties imposed on drink-drivers. Research conducted by Votey and Shapiro (1983) overcomes this problem by using information on fine amounts, imprisonment lengths and licence withdrawal periods to estimate the overall expected cost associated with different types of drink-driving penalties. Including these measures of punishment severity in an ARIMA time-series model, the researchers examined the impact of court imposed drink-driving sanctions on fatal and serious injury road crashes in Sweden from 1976 through 1979, while controlling for a host of confounding variables. The findings from this multivariate analysis suggested that, of all sanctions, licence withdrawal appeared to have had the greatest effect in alleviating road injury in Sweden over the study period, since it appeared to influence both fatal and serious injury crashes. Evidence for the deterrent effect of gaol sentence and fines was somewhat less clear. Gaol sentences appeared to be effective in reducing serious injury crashes but not in reducing fatal crashes, whereas fines had some impact on fatalities but not on serious injury crashes. It should be noted, however, 70

81 that when licence withdrawal was included in the models for both classes of crashes, the impact of gaol sentences and fines was substantially reduced. The deterrent efficacy of licence sanctions is further evident from the results of a quasi-experimental study of a mandatory licence suspension law. Blomberg, Preusser and Ulmer (1987) examined the deterrent effect of a 1982 Wisconsin law that mandated three- to six-month licence suspensions for all persons convicted of a drink-driving offence in that State. These researchers compared the 12-month driving records of a cohort of first time offenders convicted under the old law with a cohort of first-time offenders convicted under the new law. A significant reduction in both the number of alcohol-related crashes and drink-driving offences was apparent for drink-drivers who were convicted under the new law and subsequently had their licences suspended for 90 days or more. 11 This law was also found to have a general deterrent effect, as evidenced by a significant reduction in alcohol road crashes (single-vehicle injury and fatal crashes involving male drivers which occurred on a weekend night) immediately after the mandatory penalties were introduced. Again, however, the evaluation was complicated by the fact that the Wisconsin drinkdriving law contained numerous other provisions regarding the offence and it s processing (e.g. the adoption of illegal per se laws), implemented at the same time as the mandatory suspensions. Punishment celerity The final major proposition of classical deterrence theory is that swift punishment celerity can reduce the incidence of a crime. Very few studies have considered whether temporal delays impact adversely on drink-driving offending and those that have provide mixed evidence on the effectiveness of swift adjudication. Despite the limited evidence available for the celerity of punishment, it remains an important area for deterrence research. If drink-driving recidivism is partly a function of time to adjudication, reducing court delay or introducing administrative penalties is an appealing strategy for reducing drink-driving offending. 11 Blomberg, Preusser & Ulmer (1987) report that 45% of convicted Wisconsin drinking drivers lost their licence in 1981 but in 1982, after the mandatory penalties were introduced, 100% of convicted drink-drivers lost their licence for at least 90 days. 71

82 Evaluations of Administrative Licence Revocation (ALR) laws, which authorise immediate seizure and suspension of a driver s licence at the point of arrest, provide perhaps the best evidence for a deterrent effect created by swift adjudication. One such evaluation was of an administrative driver s suspension law introduced in Ontario in 1996, which allowed for a 90-day licence suspension at the time a drinkdriving charge was laid. Evidence for a general deterrent effect of the legal intervention was confirmed by time-series data showing a significant reduction in the proportion of fatally injured drivers with positive blood alcohol levels and a logistic regression analysis which showed a significant decrease in self-reported drinkdriving immediately after the new laws were put into effect (Mann et al. 2000; Stoduto et al. 2000). Likewise, Legge and Park (1994) found administrative licence suspension laws to have a positive impact on alcohol road crash rates. They report the findings of a pooled cross-sectional time-series analysis of fatal road crashes in the 48 contiguous USA States, which showed that, of the several deterrence-based variables considered, administrative laws (alongside per se laws) had the greatest impact on single-vehicle night-time road fatalities. Voas and Tippets (1999) also report that the introduction of administrative licence revocation laws contributed significantly to the downward trend in alcohol-related fatal crashes in the USA from 1982 through However, these authors note that the long-term downward trend evident in this analysis is likely to reflect the growing impact of several new laws over time, as well as other variables not tested by their model, such as enforcement activity. The specific deterrent effect of administrative licence suspension laws is less clear. In their review of this literature, McArthur and Kraus (1999) identified just three studies that had adequate comparison groups for assessing the specific deterrent effect of administrative revocation laws. One of these studies found a positive effect of the laws on recidivism, another found only a short-term (less than two years) reduction in reoffending, while the third found that the law worked in some States but not in others. The largest of these studies was the work undertaken by Stewart, Gruenewald and Parker (1992), examining the impact of administrative licence suspension laws introduced in Louisiana, Mississippi and North Dakota. Recidivism 72

83 rates for pre-law and post-law cohorts in each of the three States were compared using survival analysis and comparisons were made with California, which did not have administrative licence suspension laws in place. In North Dakota, there was a significant reduction in drink-driving recidivism rates following the implementation of the new law. The reduction in recidivism observed in this State was about onethird, a reduction that was maintained during the three-year follow-up period. Evidence for a specific deterrent effect of the North Dakota legislative reforms was further supported by the fact that similar reductions in recidivism were not apparent in the comparison State, California, over the same period. However, there were no differences in drink-driving recidivism rates across the pre- and post-law cohorts in either Louisiana or Mississippi. The researchers suggested that differential police enforcement amongst the various States may have been a reason for this disparity in results. The findings from research evaluating administrative licence suspension laws appear to support the deterrent effect of punishment celerity but are limited by the fact that these types of legal reforms influence both the celerity and the certainty of punishment. 12 This characteristic of administrative laws makes it difficult to determine which component of the deterrence equation is contributing to the observed effect. In fact, individual-level research, using time to conviction as a measure of celerity, has failed to confirm the deterrent effect of punishment swiftness found in evaluations of administrative laws (Yu 1994; Howe & Loftus 1996). Furthermore, experimental research that has considered not only the temporal delay in formal punishment but also a potential offender s preference for delay or impulsivity, as per economic discounting models, has similarly found little evidence for punishment celerity affecting offender decision-making (Nagin & Pogarsky 2001). Given what we know about the certainty of punishment, the effects of administrative licence revocation laws may simply be due to their effect on the perceived probability of drink-driving punishment, rather than their ability to reduce time to adjudication. 12 Nichols and Ross (1990) argue that the administrative process is more certain in applying sanctions due to the differences in standards of proof. 73

84 Individual-level research Although aggregate-level effects of deterrence-based interventions are arguably of greater importance to policy-makers (Gibbs 1975), much of the existing research has evaluated policies that are either poorly implemented or which vary in intensity across jurisdictions. Several of the quasi-experimental studies on punishment certainty, cited above, have managed to overcome these methodological limitations by focusing on well-defined interventions that have influenced enforcement practice and risk perceptions as intended. However, several questions regarding the deterrent efficacy of punishment severity remain unanswered. Specifically, it is unclear from aggregate-level studies what form drink-driving punishment should take and whether the magnitude of different penalties has a significant marginal deterrent effect. Furthermore, aggregate-level studies on punishment severity have generally failed to examine perceptions of punishment severity and whether these perceptions affect offending likelihood. Reoffending studies Numerous studies of deterrence have explored the effect of judicial penalties on reoffending rates of drink-drivers. These studies generally compare different degrees or types of sanctions rather than examining alternatives to legal sanctions or the imposition of sanctions where none existed previously. Thus, they present evidence on marginal rather than absolute deterrent effects of punishment. 13 One problem confronting research on the marginal deterrent effect of penalties is that many of the factors influencing the type and severity of the penalty allocated to a particular offender can also affect reoffending. For example, offenders sentenced to imprisonment could have numerous prior convictions of a serious nature and therefore be at a greater risk of reoffending, irrespective of what penalty they receive. To conclude that imprisonment is an ineffective deterrent, because recidivism rates are high amongst this group of offenders, is problematic. In order to conclusively establish a causal relationship between type and severity of penalty and subsequent offender behaviour it is necessary to randomly allocate sanctions to offenders. This ensures that offenders receiving different penalties do 13 A summary of these studies also appears in Appendix I. 74

85 not vary systematically. One such experimental study examining the deterrent effect of drink-driving penalties was conducted by Holden (1983) in the USA. For this study, the court randomly assigned first-time drink-driving offenders to four experimental conditions: (i) no sanction (control group), (ii) education and therapy, (iii) supervised probation, and (iv) supervised probation plus education/therapy. No significant differences in drink-driving rearrest rates between the four groups of offenders were evident from a two-year follow-up of conviction records, suggesting that none of the four sanction types examined were more likely to deter drink-drivers from reoffending. There was, however, some evidence that severe sanctions (in the form of supervised probation) were more effective in reducing rearrest rates but only for offenders identified as problematic drinkers. Although experiments are the most methodologically sound means by which to investigate the differential impact of penalties on drink-driving recidivism, ethical concerns and practical difficulties in randomising punishment prevents the vast majority of researchers using this paradigm in recidivism analyses (Ross & Blumenthal 1975; Farrington 1983). Most studies have therefore tended to employ statistical controls to separate out the effects of penalties from those exerted by offender characteristics. The extent to which valid comparisons between penalties can be made from such analyses will necessarily depend on appropriate offender characteristics being identified and adequately measured (Homel 1980). Ideally, these analyses would include data on a broad range of psychological and sociological characteristics of offenders. However, reoffending studies have, for the most part, been limited to information contained in official records, such as age, gender, prior convictions, BAC at time of arrest and concurrent convictions for other offences. Reoffending studies relying on these types of statistical controls to account for potentially confounding factors have found discouraging results for the specific deterrent effect of gaol sentences. DeYoung (1997) examined the reconviction rates of all Californian drivers convicted of a drink-driving offence during 1990 and Drink-driving reconvictions within 18 months was employed as the dependent variable in the analysis and a number of demographic and prior personal driving history measures were included as covariates. DeYoung s analysis found that first 75

86 offenders receiving a two-day gaol sentence for drink-driving offences recorded twice as many drink-driving reconvictions within the 18-month follow-up period, as offenders allocated other types of punishment. In his three-year follow-up study of 1,000 drink-drivers convicted in New South Wales, Homel (1980) found that, controlling for other confounding variables, imprisonment was no more effective than any other penalty for drink-driving. In fact, there was some evidence that long periods of imprisonment (especially longer than six months) actually increased, rather than decreased, the probability of reconviction for a drink-driving offence. In the natural experiment cited earlier, Martin, Annan and Forst (1993) directly compared the recidivism rates of persons who received a two-day gaol sentence, under the severe judicial sentencing policy in Minnesota, with those who did not. This additional analysis generally supported the findings from their aggregate-level research. Specifically, the results showed that a two-day stay in gaol was no more effective in reducing the reoffending of first-time drinkdrivers than alternative sanctions, even after controlling for BAC at arrest, age, gender, prior convictions, seriousness of offence and plea. Despite the limited deterrent efficacy of imprisonment, there is some evidence that other penalty types, particularly licence disqualification, can reduce the reoffending rates of convicted drink-drivers. Evidence from an Australian study, for example, shows that licence actions are successful, not only in reducing reoffending but also in limiting the number of road crashes involving convicted drink-drivers. Siskind (1996) examined the driving records of over 25,000 drivers convicted of a drinkdriving offence in Queensland, who had received at least one licence restriction in He found evidence for a two-thirds reduction in both crash and drink-driving offence rates for offenders under disqualification from driving compared with offenders who had not been disqualified. Moreover, offenders who were apprehended for a new offence whilst their licence was disqualified were more likely to be apprehended in the early stages of their disqualification period. This latter finding would suggest that improvements in driver behaviour resulting from a licence sanction would be more pronounced for persons disqualified for longer periods of time. 76

87 Yu (1994) examined the effect of various penalty types on the recidivism rates of almost 14,000 drivers with at least one conviction for drink-driving in the State of New York, USA. He found that drivers who received higher monetary fines had significantly reduced drink-driving recidivism rates during the three-year follow-up period. However, there was no evidence that longer periods of licence withdrawal had a stronger impact on drink-driving recidivism than shorter periods. A limitation of both these studies is that they failed to adequately control for prior convictions. The New York study included no information on previous drink-driving convictions in its recidivism analysis. The Queensland study included a variable representing any charge for a drink-driving offence in the preceding five years but did not specify the number of convictions recorded for each offender. Previous convictions is one of the best predictors of drink-driving reoffending (Ryan et al. 1996) and is also likely to affect both the length and type of penalty imposed by the court. Research examining the differential impact of penalties on recidivism must take this fact into account. A study that did adequately control for previous driving record demonstrates that more severe licence actions can reduce reoffending, but that high fines and gaol sentences do not succeed in deterring offenders. Mann et al. (1991) conducted a three-year follow-up study of approximately 4,000 Ontario drivers who were convicted of at least one drink-driving offence in Importantly, this study used data on the number of convictions recorded for each offender in the five years prior to the index offence to examine the deterrent effect of different penalty types among individuals with differing numbers of previous offences. The results of this analysis indicated that longer periods of licence suspension were associated with significantly fewer total and alcohol-related crashes for first and multiple offenders, as well as reduced drink-driving charges for first offenders. On the other hand, higher fines (for first time offenders) and more days in gaol (for all offenders) were associated with an increase in the number of crashes and convictions. 77

88 Reviews of the drink-driving literature on specific deterrence and penalty severity generally concur with the conclusions of Mann and his colleagues that licence disqualification is the most effective deterrent for drink-driving and that gaol has little effect on offender behaviour (Nichols & Ross 1990; Zaal 1994). This is also consistent with aggregate-level research presented earlier in this chapter on mandatory licence suspension laws. With gaol being more severe than licence disqualifications, the superior effect of licence sanctions in reducing offending would seem at odds with deterrence predictions. However, surveys of offenders have suggested that a large proportion of drivers fear losing their driver s licence and the freedoms that come with it more than they do serving a relatively short imprisonment term. In particular, persons who rely heavily on their motor vehicle for employment purposes perceive the threat of licence restrictions as extremely costly (Homel 1986). It is also worth noting that, rather than eliminating drink-driving completely, licence restrictions result in an offender driving less often and more carefully during the period of disqualification than they would otherwise (Ross & Gonzales 1988). Since many drink-drivers have concurrent convictions for other traffic offences (Homel 1980), these modifications to driver behaviour through fear of further punishment are likely to have an ancillary effect on other traffic offending, which could also serve to heighten the overall impact of licence sanctions. Perceptual studies Deterrence is essentially a perceptual theory, in which criminal behaviour is thwarted not by the actual risk of punishment being imposed or the actual severity of punishment, but rather by what people believe these to be. Recognising this, a large body of research has emerged investigating the relationship between individuals perceptions of sanction risks and their self-reported offending behaviour. As discussed in Chapter 1, perceptual studies can be broadly classified into three categories: cross-sectional, longitudinal or panel, and scenario-based (Nagin 1998; von Hirsch et al. 1999). Cross-sectional studies collect information, at one point in time, on current perceptions of sanction risks and prior offending behaviour, or future intentions to offend. Longitudinal or panel studies, on the other hand, measure perceptions and behaviour in the same sample on two or more occasions. These two 78

89 types of perceptual research have provided conflicting evidence on the deterrent effect of formal punishment. Cross-sectional studies have consistently shown negative associations between self-reported offending and perceived certainty of arrest or punishment, though the evidence for an effect of perceived punishment severity is weaker (Grasmick & Bryjak 1980; Grasmick & Bursik 1990; Paternoster & Simpson 1996). In contrast, panel and longitudinal studies have found little or no deterrent effect of varying levels of perceived certainty or severity on offender behaviour (Paternoster et al. 1983; Paternoster & Iovanni 1986; Lanza-Kaduce 1988). The validity of both these research designs in investigating hypotheses generated by deterrence theory has, however, been questioned (e.g. Williams & Hawkins 1986; Nagin 1998; von Hirsch et al. 1999). Cross-sectional designs, in which current perceptions are used to predict past behaviour, suffer from the problem of causal ordering. A significant negative relationship between perceptions of apprehension risk and prior drink-driving episodes could reflect a deterrent effect: people who perceive the risks as lower will offend more often. Alternatively, it could reflect an experiential effect: people who offend more often, having escaped detection on numerous occasions, will perceive the risks as lower. Rational models of offending would, on the other hand, predict that perceptions are continually updated on the basis of new information. Thus longitudinal research, which assesses current perceptions and then observes behaviour at some future point, may be subject to measurement error, particularly if there is a long delay between the two testing occasions. Ideally, deterrence research would measure perceptions of formal punishment at the point the decision to offend is made (Williams & Hawkins 1986). The difficulties of achieving this in practice are self-evident. Researchers have therefore attempted to deal with the temporal lag between measures of perceptions and behaviour through the use of scenario-based studies. This research design presents participants with vignettes, describing realistic offending scenarios, and then requires them to indicate the likelihood that they would offend in the given circumstances. Perceptions of sanction risks can be measured at the same time as willingness to offend. Although 79

90 this methodology does not question participants on actual offending behaviour 14 it does succeed in modelling the decision-making process so that perceptions can be measured concurrently with stated offending likelihoods. Furthermore, as noted by Nagin (1998), research on situational crime prevention has demonstrated the importance of the context in which crimes are committed to perceptions about sanction risks. In drink-driving research, for example, these situational factors could include such things as distance from destination, amount of alcohol consumed and inconveniences avoided. Scenario-based research provides detailed information about the circumstances of the crime and, in doing so, these situational variables are held constant across the varied conditions or groups. Previous research using scenario-based designs has provided consistent evidence for deterrent-like effects of formal sanctions on offending behaviour (e.g. Klepper & Nagin 1989b; Bachman, Paternoster & Ward 1992; Nagin & Paternoster 1993; Nagin & Paternoster 1994; Tibbetts 1997). Furthermore, these studies have shown that the threat of formal punishment has a deterrent effect which is independent of time-stable individual differences, such as poor self-control and shame-proneness. Nagin and Paternoster (1993), for example, presented college students with three scenarios describing the commission of different offences: drink-driving, theft and sexual assault. Respondents were asked to estimate their chances of arrest or public exposure (without arrest), the probability of being formally sanctioned by the criminal justice system (either by way of jail or licence disqualification) or informally sanctioned by others (e.g. peers, family, the university) if caught, and the likelihood they would offend in the given scenario. Feelings of shame arising from the act itself and levels of self-control were also measured in the survey. Subsequent analyses showed that, even after individual differences in self-control and shame were taken into account, a significant inverse relationship between intentions to offend and perceived formal costs was still apparent. 14 Most scenario-based studies do however measure prior offending and have shown a strong, positive relationship between previous offending episodes and stated likelihood of offending in the described scenario (e.g. Nagin & Pogarsky 2001). This attests to the veracity of respondents answers on the dependent variable. 80

91 Using a similar paradigm, Tibbetts (1997) demonstrated that a measure of the overall threat of formal sanctions (which combined information on both the risk of apprehension and severity of sanctions) had a significant negative effect on an individual s intention to drink-drive, controlling for both low self-control and shame proneness. Moreover, these external sanctions appeared to be more influential in drink-driving decisions than in decisions relating to other types of offending behaviour. Shoplifting, for example, was found to be inhibited primarily by internal sanctions related to anticipated shame, rather than the risk of formal punishment. For the most part, however, perceptual research has restricted its focus to the impact of informal social sanctions on offending. Where survey-based research has considered formal sanctions it mostly concentrates on the relationship between offending and the perceived certainty of punishment or the overall threat of punishment, rather than the independent effect of severity (e.g. Paternoster 1989; Grasmick & Bursik 1990; Baum 1999). An exception to this general observation is the recent work conducted by Nagin and his colleagues in the USA (see Nagin & Pogarsky 2001; Pogarsky 2002; Nagin & Pogarsky 2003). Using scenario-based designs, these authors experimentally manipulated the severity of formal sanctions for drink-driving offences by randomly assigning respondents three, six and nine month licence suspensions. Respondents were then asked to estimate the likelihood of apprehension if they drove home in these circumstances and the likelihood they would offend given these conditions. The researchers also collected information on a host of other variables thought to condition deterrence effectiveness, including impulsivity and anticipated extralegal consequences. Although subsequent analyses found some evidence that variations in sanction severity predict stated offending likelihood, in general these data confirmed the conclusion from ecological and quasiexperimental studies that the certainty of punishment is a far more robust deterrent than the severity of punishment. A shortcoming of Nagin and Pogarsky s approach to evaluating punishment severity effects is it assumes that people know the magnitude of the penalty (e.g. the length of licence disqualification) they would receive if convicted of a drink-driving offence. Some evidence, however, suggests that public knowledge of criminal penalties is in 81

92 fact quite poor (Williams, Gibbs & Erickson 1980; Hough & Roberts 1998; Kenkel & Koch 2001). This failure to consider the link between perceptions of sanction severity and actual sentencing policy is another notable limitation of the perceptual deterrence research conducted thus far. As Nagin (1998) points out: While great effort (in the deterrence literature) has been committed to analysing the links between sanction risk perceptions and behaviour, comparatively little attention has been given to examining the origins of risk perceptions and their connection to actual sanction policy (Nagin 1998, p. 5). Studies like those above provide information on the likelihood of offending, were a particular sanction risk perceived by respondents. However, they tell us nothing about the extent to which changes in criminal justice policy will alter perceptions of punishment severity. This knowledge gap needs to be addressed before a complete test of the deterrence model can be realised (Gibbs 1975; Cook 1980). Miller and Iovanni (1994) attempted to address this problem by examining the determinants of sanction risk perceptions in relation to intimate partner violence. They conducted a two-way panel survey of university students, some of whom selfreported previous experiences with violent conflict in an intimate relationship, while others reported no prior experience with partner violence. To disentangle experiential from deterrent effects, the perceived certainty of arrest for partner violence and the perceived severity of the consequences arising from an arrest were measured at two time periods conducted three-months apart. Lagged measures of perceived risk and perceived severity at Time 1 were then included as independent variables in the regression models predicting risk perceptions at Time 2. Time 1 measures of prior use of partner violence, as well as measures on constructs derived from social learning and social control theories, were also independent variables in the study. These latter variables included measures of the attitudes of significant others toward violence and prior exposure to models of violence, as well as parental attachment, peer attachment and commitment to conventional goals. 82

93 The analysis of these data found variations across gender in terms of the factors influencing perceived risk. For certainty of arrest, prior use of violence in an intimate relationship was associated with reduced perceived risk of apprehension, but this effect was evident only for males. Furthermore, males who believed that partner violence was illegal (a measure of legal knowledge), perceived arrest for these events to be more likely. On the other hand, perceptions regarding the severity of consequences arising from an arrest for partner violence, or the consequences arising from police being called to an incident, appeared to be more influenced by informal social controls, such as attachment to parents and peers, but again this effect was significant only for males. The study also found some ambiguous and contradictory findings. For example, peer violence was found to be significantly related to arrest risk for females but in the opposite direction to that hypothesised. That is, females who reported having more violent peers perceived the risk of arrest to be higher. Also, contrary to predictions, beliefs regarding the illegality of partner violence were negatively related to the perceived severity of consequences arising from an arrest, for the male group. That is, males who believed partner violence was against the law were less likely to view arrest as severe. The authors suggest that these unexpected results could be due to the fact that non offenders may have had difficulty in imagining themselves in a violent relationship. The nature of partner violence and the low risk of formal intervention for these type of offences would also mean that informal social controls, such as moral beliefs toward violence, may be a stronger predictor of perceptions than would be the case for other offence types. Further investigations of this type with a more common offence, like drink-driving, would improve our knowledge of relevant factors influencing risk perception formation. Significance of the current study This thesis explores one of the major hypotheses of deterrence theory that more severe punishment can have a marginal deterrent effect on offending rates. Despite a wealth of studies examining this issue, the literature reviewed above reveals mixed evidence for deterrent-like effects from more severe punishment. Evaluations of sentencing policies that increase the severity of statutory penalties are an important 83

94 source of evidence on the deterrent efficacy of punishment severity. However, the policy impact studies that have been conducted in the drink-driving domain so far contain several methodological shortcomings that impede our ability to make confident assessments regarding the effectiveness of this strategy. By identifying and addressing these weaknesses, this thesis builds on prior deterrence research and, in so doing, enhances our understanding of the relative importance of punishment severity in controlling crime. It also explores two areas related to sanctioning policy where few deterrence researchers have yet ventured: (1) the link between actual and intended policy and (2) the formation of risk perceptions. Methodological issues in policy impact studies Drink-driving policies that stipulate more severe statutory penalties are often introduced in response to public concern about alcohol-related road crashes or as a result of extensive lobbying by victim and community groups. To address these societal concerns, legislatures sometimes implement policies that affect more than one component of the deterrence equation. The modifications to Californian drinkdriving laws in 1982 exemplify this process. These legal reforms arose partly out of the publicity generated by the advocacy group MADD, and were aimed at strengthening the perception amongst drink-drivers that punishment would be unavoidable and severe (Rogers & Schoenig 1994). To achieve this, penalties for drink-driving offences were raised substantially. However, 0.1 per se laws and treatment services for repeat offenders were also introduced at the same time. In these circumstances it is difficult to disentangle the effect of increased penalty severity from other changes made to drink-driving laws. While it is reassuring to know that increasing the overall threat of formal punishment exerts a substantial deterrent effect, this research is unable to provide specific information for policy makers and legislators as to which aspect of the legislative package is most effective. Many evaluations of drink-driving interventions have also been carried out in jurisdictions where the certainty of punishment is relatively low. The certainty and severity of legal punishment are predicted by deterrence theory to combine multiplicatively rather than additively, thus severity is most likely to matter when 84

95 punishment is highly probable. In North America, where much drink-driving research has been conducted, drink-driving enforcement takes the form of sobriety checkpoints or roadblocks. This is a limited version of RBT that requires reasonable suspicion of drinking before an officer can obtain a breath sample from a driver. While checkpoints have been shown to reduce drink-driving and associated alcoholrelated road crashes under certain conditions (Shults et al. 2001), the reduced risk of being tested for alcohol impairment (Homel 1988) and the infrequent use of checkpoints (DeJong & Hingson 1998; Shults et al. 2001) suggest that the perceived risk of apprehension in these jurisdictions is unlikely to be sustained at a high level. However, in Australian States, such as New South Wales (where the case study for this thesis is situated), enforcement of drink-driving through RBT has continued at a relatively high level since its introduction in the early 1980s. As a consequence, significant reductions in alcohol-related road crashes (Homel 1994; Span 1995; Span & Stanislaw 1995; Henstridge, Homel & Mackay 1997) and self-reported drinkdriving episodes (Homel 1986; Homel, Carseldine & Kearns 1988), as well as increases in perceptions of apprehension risk (Homel 1986), have been observed in these jurisdictions. An investigation of increased penalties in this context, where there exists a relatively high level of visible enforcement of drink-driving offences, would permit more informed conclusions regarding the effectiveness of severe punishment. Poor implementation of sentencing policies is a further reason why several drinkdriving studies have failed to find deterrent effects for severe sanctions (Ross 1984; Mayhew, Beirness & Simpson 1995; Nagin 1998). Ross and Voas (1990), for example, found that appeals quadrupled when a New Philadelphian judge adopted a harsh sentencing policy providing for jail terms and the elimination of pleabargaining in all drink-driving cases within his jurisdiction. Similarly, Mayhew, Beirness and Simpson (1995) found that between 20 and 40 per cent of offenders charged with a newly introduced drink-driving offence (i.e. impaired driving causing harm or death), which incurred severe penalties upon conviction were eventually convicted of a less serious offence that had lower prescribed penalties. Voas (2001) also reports that many drink-drivers escape licence suspension in the USA because the courts want to encourage offenders to plead guilty and/or to avoid calls for jury 85

96 trials. Increasing penalty severity may not have a deterrent effect if actual sentencing policy does not change. If offenders get off more lightly than they had originally anticipated they may adjust their estimates of sanction risk downward, thus reducing any initial specific deterrent effect created by the sanction policy. This may also have implications for the general deterrent effect of more severe penalties by decreasing the perceived credibility of drink-driving sanctions as other people become aware of this leniency in sentencing practice. Only once a policy prescribing increased statutory penalties has been shown to affect sentencing practice, can its impact on subsequent criminal behaviour be considered. Finally, since increasing the statutory penalties for an offence is posited to prevent future legal violations by influencing perceptions of sanction severity, there must be some minimum level of general awareness before a sanctioning policy can be expected to have a deterrent-like effect. Several studies, particularly those evaluating police interventions (Ross 1984; Cashmore 1985; Voas, Holder & Gruenewald 1997), have demonstrated that broadcasting the implementation of new drink-driving initiatives can greatly enhance deterrence outcomes. However, amendments to statutory penalties are generally not supported by similar enthusiastic publicity campaigns and therefore often are not widely known in the general community. Under these circumstances, perceptions of sanction severity would vary little as a result of changes in punishment levels and, as a consequence, offending rates would remain relatively unaffected. Knowledge gaps in the deterrence literature Nagin (1998) identifies four important research areas that have thus far received little attention in the deterrence literature but which need to be addressed before confident conclusions regarding the deterrent efficacy of various policy options can be drawn. These priority areas are: (1) long term effects (2) the implementation of policies across different population units (3) the gap between intended and actual policy which means laws are generally not administered as intended (4) the relationship between risk perceptions and actual sanctioning policy. While addressing all four of these areas is beyond the scope of the current research, this thesis will bring evidence to bear on the latter two areas highlighted as priorities for deterrence research. 86

97 CHAPTER 4. INCREASED DRINK-DRIVING PENALTIES IN NEW SOUTH WALES: A Case Study The purpose of the Traffic Amendment (Penalties and Disqualification) Bill is to make our roads safer by implementing new and increased penalties for certain serious traffic offences. Adoption of the legislation will give effect to new penalties that aim to reflect the safety implications of the offences, decreases in real value of fines since the current provisions were put in place and increased community concern about road safety issues, particularly when driver behaviour is regarded as highly irresponsible. The increased penalties greatly enhance the deterrent effect of our road penalties and will help to improve road safety (Mr Scully; Minister for Transport, and Minister for Roads, Second Reading of the Traffic Amendment (Penalties and Disqualifications) Bill, 21/05/98). On September , the NSW Traffic Amendment (Penalties and Disqualifications) Act was introduced by the NSW State Government to enhance the deterrent effect of drink-driving penalties and improve road safety. This legislation essentially doubled the maximum penalties for almost all drink-driving offences. 15 Various members of the NSW Parliament voiced their opposition to these legislative amendments during the Second Reading of the NSW Traffic Amendment (Penalties and Disqualifications) Bill. Concerns highlighted by these parliamentarians focused mainly on the financially excessive nature of the penalty increases and the potential benefits that the State Government would reap from any subsequent increase in revenue. However, the opposition also claimed that the imposition of severe punishment is unlikely to influence offending behaviour to any great extent. This point was echoed by other non-government organisations, including The Law Society of New South Wales who, in a written response to the Government s proposal, stated the available evidence shows that the most effective deterrent is the 15 A small number of drink-driving offences, such as occupying a seat next to a learner driver while intoxicated were not subject to the penalty changes. 87

98 fear of being caught rather than the severity of penalties for speeding, drinking and drug driving (Mr Souris, Deputy Leader of the National Party; Second reading of Traffic Amendment (Penalties and Disqualifications) Bill to the Legislative Assembly, 03/06/98). Despite this opposition, the imposition of severe penalties for drink-driving is an extremely popular policy amongst NSW residents. In a survey, measuring public support for policies designed to address alcohol-related harm, almost nine in 10 persons aged 14 or over supported more severe penalties for drink-driving offences (Australian Institute of Health and Welfare 2002b). Thus, the strengthening of legal sanctions to counter this criminal behaviour was at least reflective of public opinion at the time. This thesis examines whether these changes in the severity of drink-driving penalties in New South Wales had the intended deterrent effect on drink-driving rates in this State. As will be described in this Chapter, this case study represents a good opportunity to test deterrence theory predictions regarding changes to penalty severity because several of the preconditions for deterrence success have been met. If more severe penalties are a deterrent for drink-drivers then we would expect the effect of harsher penalties to be maximised under these conditions. This chapter also describes the two quasi-experimental studies that were undertaken to investigate the impact of the penalty changes: an interrupted time-series analysis of road crash rates and a recidivism analysis comparing drink-drivers convicted pre-1998 with those convicted post Before describing the methodology used in these studies, it is important to gain an understanding of the context in which this case study is situated and the sentencing policy that forms the independent variable in this research. The following section therefore discusses drink-driving legislation in New South Wales and the amendments that were made to statutory penalties for drink-driving offences in

99 Drink-driving legislation in New South Wales and the 1998 penalty changes In New South Wales, drink-driving encompasses four distinct offences based on the alleged offender s BAC at the time of the infringement. 16 These four offences include a special-range Prescribed Concentration of Alcohol (PCA) offence 17 with a BAC limit of 0.02g/100ml, 18 a low-range PCA offence with a BAC limit of 0.05g/100ml, a mid-range PCA offence with a BAC limit of 0.08g/100ml and a highrange PCA offence with a BAC limit of 0.15g/100ml. 19 For each of these offences the culpable act is defined as operating a motor vehicle with a BAC in excess of the specified limit. Thus there is no need to prove behavioural impairment resulting from alcohol intoxication. Drink-driving laws such as these, where evidence of an illegal BAC is sufficient to prove the offence, are known in the drink-driving literature as per se laws. They are also strict liability offences, since there is no mental element (mens rea) required for the offence to occur. The term per se refers to the fact that neither mens rea nor behavioural impairment due to intoxication needs to be proven to successfully prosecute this offence. In 1998, the penalties prescribed for these offences were defined under section 4E of the NSW Traffic Act This legislation prescribed maximum monetary fines for each of the four PCA offence categories and maximum gaol sentences for mid- and high-range PCA offences. The magnitude of these penalties varied as a function of the PCA category within which the current conviction fell and whether the current conviction was deemed to be a repeat offence 20 (though the maximum gaol term did 16 These offences were originally defined under section 4E of the NSW Traffic Act 1909 but can now be found under section 9 of the NSW Road Transport (Safety and Traffic Management) Act PCA is the offence category for drink-driving in NSW. 18 Special range offences apply to special category drivers and supervisors. This primarily includes holders, and supervisors, of a learner licence or a first-year provisional licence, under 25 drivers who have had a licence for less than 3 years, persons not authorised to drive in NSW, drivers of heavy vehicles and drivers of public vehicles or vehicles that carry dangerous loads. 19 More recently, an offence for the Novice PCA range was introduced in NSW (see NSW Road Transport (Safety and Traffic Management) Amendment (Alcohol) Act 2004). This offence applies to learner or provisional drivers and prohibits these drivers from driving with any alcohol present in their blood. Since this legislation was not implemented until 2004, after the completion of the current analysis it won t be considered in the remainder of the thesis. 20 Penalties for a second or subsequent offence are applicable if a person has been convicted of a drink-driving or other major offence in the previous 5 years. Major offences primarily consist of driving offences resulting in death or bodily harm to another person and any of the four PCA offences (see Part 1 of NSW Road Transport 89

100 not vary by number of offences for mid-range offences). All PCA offences also attracted some form of licence disqualification under this Act. Maximum licence disqualification periods were specified for each PCA offence category, as were automatic disqualification periods to be applied in the absence of a specific court order. Minimum licence disqualification periods were specified for all PCA offences 21 with the exception of low- and special-range offences in cases where it was an offender s first offence. Furthermore, any person charged with a high-range PCA offence could have their licence suspended by a police officer within 48 hours of being charged and the suspension remained in effect until a court determined the charge. With the implementation of the NSW Traffic Amendment (Penalties and Disqualification) Act 1998, these statutory penalties underwent several changes. Table 4.1 shows the penalties applicable to drink-driving offences before these legislative amendments came into effect and the penalties applicable after these changes. As seen here, the legislative amendments significantly increased the magnitude of almost all penalties for drink-driving offences. These harsher penalties included doubling the maximum gaol terms applicable to mid-range and high-range PCA offences and doubling the maximum monetary fines for all PCA offences. Mandatory three-month minimum licence disqualification periods were also introduced for first offenders charged with a special-range or low-range PCA offence (offences which previously had no such mandate) and minimum licence disqualification periods were doubled for all other PCA offences. Furthermore, the legislative amendments allowed for the licences of mid-range offenders to be immediately suspended by police until the offender s appearance in court, a regulation that had previously only been applicable to high-range PCA offences, and introduced a new licence disqualification penalty of up to five years for offenders who accumulated three or more convictions within a five-year period (these (General) Act 1999). 21 Although there were other legislative provisions that allowed a court to avoid disqualification, notably section 556A of the NSW Crimes Act 1900 which is now equivalent to the section 10 of the NSW Crimes (Sentencing Procedure) Act

101 offenders would have to be declared an habitual offender by the court to be eligible for this new penalty). 22 Table 4.1: Statutory penalties for drink-driving offences in New South Wales before and after the 1998 statutory amendments (penalty changes are in bold text) PCA offence Penalty type First offence Second and subsequent offences pre-1998 post-1998 pre-1998 post-1998 Special range PCA (BAC.02 to <.05) Max. fine $500 $1,100 $1,000 $2,200 Low range PCA (BAC.05 to <.08) Mid range PCA (BAC.08 to <.15) High range PCA (BAC.15 & above) Licence Disqualification - min. - max. - auto* none 3 months none 3 months 6 months 6 months 3 months Unlimited 12 months 6 months Unlimited 12 months Max. fine $500 $1,100 $1,000 $2,200 Licence Disqualification - min. - max. - auto* none 6 months none 3 months 6 months 6 months 3 months Unlimited 12 months 6 months Unlimited 12 months Max. fine $1,000 $2,200 $1,000 $3,300 Max. gaol term 6 months 9 months 6 months 12 months Licence Disqualification - min. - max. - auto* 3 months Unlimited 12 months 6 months Unlimited 12 months 6 months Unlimited 3 years 12 months Unlimited 3 years Immediate No Yes No Yes suspension** Max. fine $1,500 $3,300 $2,000 $5,500 Max. gaol term 9 months 18 months 12 2 years months Licence Disqualification - min. 6 months 12 months 1 year 2 years - max. Unlimited Unlimited Unlimited Unlimited - auto* 3 years 3 years 5 years 5 years Immediate suspension** Yes Yes Yes Yes * Automatic is the disqualification period that applies in the absence of a specific court order ** Suspension of licence within 48 hours of being charged for this offence and until the charge is determined by the court 22 The NSW Traffic Act 1909 was repealed on 1 st December Maximum monetary penalties and gaol terms applicable to drink-driving offences in NSW can now be found in the NSW Road Transport (Safety and Traffic Management) Act Relevant licence sanctions can be found in the NSW Road Transport (General) Act

102 The legislative amendments did not, however, affect the power of police to conduct Random Breath Testing (RBT) to ascertain a driver s BAC level. In New South Wales, in accordance with section 4E of the NSW Traffic Act 1909 (now Division 3 of the NSW Road Transport (Safety and Traffic Management) Act 1999), police officers can request a preliminary breath test from any person driving a vehicle, attempting to start a motor vehicle or supervising a learner licence while the learner is driving. When requesting a breath sample from a driver, the police officer does not have to have any suspicion that the person has been drinking. This section of the Act also permits police to request or signal a driver to stop his or her vehicle for the purpose of a preliminary breath test and if a driver is found to be over the legal limit or refuses to undergo a breath test, the police officer has the authority to immediately arrest the person and take them into custody. Once in police custody an evidential breath analysis is conducted by an authorised police officer specially trained in breath analysis techniques. So long as the breath analysis is conducted within two hours of the original incident, the reading at the time of the evidential test is deemed to be the BAC recorded at the time of the incident. It is an offence for a driver to refuse to undergo a preliminary breath test 23 or to refuse to stop if requested to do so by police. Failure to submit to an evidential breath analysis and refusing to stop are offences attracting the same penalties as a high-range PCA offence. The case study The 1998 changes to drink-driving penalties in New South Wales, which are described above, represent a critical case in testing propositions of deterrence theory because (1) the legislation substantially raised the severity of formal punishment without any concurrent changes to punishment likelihood and (2) the offence and the jurisdiction to which the penalties relate meet several of the conditions necessary for optimising deterrent effects. A key assumption of deterrence theory is that more severe punishment can prevent future offending. The NSW legislative changes to statutory penalties for drinkdriving represented a substantial increase in the threat of formal punishment for this 23 Unless the person is unable to do so on medical grounds. 92

103 offence. The sentencing policy essentially doubled the prescribed maxima and minima for drink-driving offences, and these were crimes that already attracted relatively harsh sanctions. Also the penalty changes, unlike others previously evaluated in the deterrence literature, were not accompanied by any legislative measure designed to increase the perceived risk of apprehension. The policy, thus, affected only one element of the deterrence equation, meaning that any subsequent effect on offending can more confidently be attributed to increases in the level of punishment severity. There are also thought to be several different circumstances within which deterrence predictions are thought to hold true. As highlighted in Chapter 2, the presence (or lack) of informal social controls is one important factor thought to condition threat responsiveness. For crimes that lack strong moral condemnation by the general community, social disapproval from apprehension and conviction for an offence is thought to be more influential in offending decisions than social stigma arising from performing the act itself. Given that drink-driving offences generally do not elicit the same level of moral condemnation as more traditional crimes, we would anticipate deterrence-based interventions aimed at drink-drivers to be more efficacious. South (1998) identifies a further six conditions necessary for deterrence to succeed in influencing offender behaviour: (1) The proscribed behaviour must be adequately defined (2) Potential offenders must believe they cannot avoid apprehension if caught (3) Sufficiently severe penalties must be attached to the offence (4) The targeted behaviour must be the result of a deliberate decision made in a reasonably rational state (5) Potential offenders must perceive a high risk of being detected if they commit the offence (6) The punishment prescribed for the offence must be known and be unavoidable The offence of drink-driving meets the first four of these six criteria. As discussed in Chapter 2, per se laws have removed any doubt about what constitutes an offence of this type and technological advancements in breath testing devices have significantly 93

104 improved the chances of apprehension and conviction for a drink-driving offence. In most cases, drink-driving also involves an element of planning, given that the drink driver must choose to both drive to the drinking establishment and to consume alcohol. This behaviour can therefore be considered the result of a deliberate decision, rather than an impulsive or an automatic reaction. Furthermore, over the last two or more decades legislatures have tended to adopt a tough stance on drinkdriving and related behaviours. Consequently, it is an offence that, in many States and Provinces, attracts quite severe sanctions relative to other traffic violations. The final two conditions for deterrence success articulated by South are also present in the particular jurisdiction that has been selected for this case study: Potential offenders must perceive a high risk of being detected if they commit the offence As discussed in previous chapters, RBT is an essential component of drinkdriving enforcement in Australia and has been successful in creating a public perception that the likelihood of detection for these offences is high. This goal has not been achieved to the same extent in many other western countries. Evidence from an international survey shows that eight out of 10 Australians report having ever been checked by the police for the presence of alcohol but only 63 per cent of Canadians, 29 per cent of Americans and 16 per cent of respondents from the United Kingdom report personal experience with police checks for drink-driving (Williams, Fergusson & Camissa 2000). The introduction of RBT in the Australian State of New South Wales was particularly successful in raising perceptions of punishment certainty and was demonstrated to have a long-term impact on alcohol-related road crashes (Henstridge, Homel & Mackay 1997). Evidence suggests that this high level of drink-driving enforcement is still present today. A 2001 survey of NSW drivers found that 71 per cent had seen an RBT unit in operation during the previous six months and almost one in five recalled being personally breath tested during this six-month period (Mitchell-Taverner 2002a). If deterrence theory is right, increasing punishment severity should have a suppression effect on drink-driving in this context, ceretis paribus. 94

105 The punishment prescribed for the offence must be known and be unavoidable The severity-based policy being examined in this research, introduced mandatory penalties for drink-driving offences by stipulating minimum licence disqualification periods for all drink-driving offences. To ensure that the message was conveyed to drivers that severe punishment for this offence is unavoidable, the statutory amendments made no exceptions for any particular offences or types of offenders. 24 Whether this threat was successfully translated into actual sentencing practice and adequately conveyed to the target audience are questions that will be explored in greater detail throughout this thesis. The intention of the policy, however, was to create a perception that severe punishment will be imposed on all persons detected for a drink-driving offence. The 1998 sentencing policy introduced in New South Wales, Australia, thus provides a good test of one of the main propositions of deterrence theory: that an increase in the severity of punishment can reduce the likelihood of offending. 25 Measuring policy effectiveness - The quasi-experiment A major step in designing and conducting a case study in order to test deterrence hypotheses is defining the units of analysis. In this case study, the unit of analysis could be the entire State, or each individual within the State. The former is appropriate when we seek to answer questions regarding the policy s effect on offending rates at the aggregate level. The latter is appropriate when we seek information about a policy s influence on individual risk perceptions and behaviour. An aggregate-level analysis is essential in determining the social policy implications of alterations to punishment severity (Gibbs 1975) and therefore comprises a major 24 Note that in some Australian jurisdictions (e.g. Qld) drivers can be given a licence restriction as opposed to a licence disqualification which allows certain offenders to use their vehicle at particular times (e.g. to go to work). This type of licence sanction would reduce perceived punishment severity and raise doubts about the certainty of punishment being imposed. Licence restrictions are not currently available as a sentencing option in NSW. 25 Being a single-case study, one question that may arise in relation to this work is how generalisable these findings are to other areas of offending or other types of offenders. This is not, however, the intended aim in case study research. As Yin (1994) notes case studies, like experiments, are generalisable to theoretical propositions and not to populations or universes (Yin 1994, p. 10). 95

106 component of this research. This takes the form of a quasi-experiment examining the impact of the 1998 legislative amendments on alcohol-related road crash rates (Study 1) and drink-driving recidivism rates (Study 2) in New South Wales. If the policy did achieve its deterrence objectives, then we would expect to see evidence from these two studies for: 1. A significant reduction in alcohol-related road crashes immediately after the introduction of more severe statutory penalties in A significant reduction in the likelihood of reoffending and the time taken to reoffend for drink-drivers convicted after The next sections describe the purpose of these two studies and the methodologies employed in this research. Alcohol-related road crashes - Study 1 The first study relies on statistics of alcohol-related road fatalities and serious crashes to measure changes in drink-driving offending rates after the penalty changes were introduced in New South Wales in Road crash data are frequently used in drink-driving studies to index offending levels because of the significant positive relationship that exists between crash risk and alcohol consumption. Case-control studies have consistently shown that the risk of incurring a road crash increases as a function of a driver s Blood Alcohol Concentration (BAC). Birrell (1974), for example, estimates that, at a BAC of 0.05g/100ml, the risk of crashing is about double what would be expected at a zero BAC. At a BAC of 0.10g/100ml it is about seven times greater and at 0.15g/100ml BAC it is about 25 times greater than the risk of crashing at zero BAC. The degree of crash severity is also significantly related to levels of alcohol intoxication, with more severe crashes recording a greater proportion of intoxicated drivers (Evans 1991; Zador 1991; Zador, Krawchuk & Voas 2000). Crash data is also considered a more reliable and well-documented source of information about drink-driving offending than other readily available measures. A common problem associated with traditional measures of offending, such as arrest statistics, is that they necessarily include only crimes which come to the attention of 96

107 officials. However, criminologists have long known that only a small fraction of most criminal acts are ever reported to police. Moreover, officially recorded offences can vary as a consequence of interventions other than those under investigation, which could lead to erroneous conclusions about the underlying reason for observed changes. For example, variations in recorded drink-driving arrests or positive breath tests may simply reflect changes in enforcement practice rather than actual variations in drink-driving behaviour. Almost all road fatalities and crashes resulting in serious injury, on the other hand, are reported to authorities because of the need to treat victims (Ross 1984). The use of road crash data, therefore, allows for more accurate and reliable conclusions to be made about the impact of a specific policy on offending rates in that jurisdiction. Although a strong association exists between alcohol consumption and injury crashes, a large proportion of injury crashes are not alcohol-related and therefore would not be expected to vary as a function of drink-driving legal innovations. For this reason, more refined measures of alcohol-related crashes need to be employed to investigate trends in drink-driving. Ideally, trends in road crashes where the driver records a BAC over the legal limit would be used for this purpose. However, the BAC of drivers is not always reliably recorded for non-fatal road crashes. This is particularly true for the current case study. In New South Wales, despite legislative provisions requiring that all drivers involved in a crash resulting in injury are tested for the presence of alcohol, 28 per cent of all injury crashes occurring between 1994 and 2001 had no information available on the BACs of the drivers involved. This figure increases in the latter years of this study period, peaking at 39 per cent in These missing data would undoubtedly affect the usefulness of recorded BACs as a measure of drink-driving trends. Two surrogate measures of alcohol-related road crashes, all fatal crashes and singlevehicle night-time crashes (i.e. crashes occurring between 8pm and 4am which involve only one vehicle; Voas, Holder & Gruenewald 1997), are therefore employed in this study. Both of these categories include a greater proportion of alcohol-related 26 These figures are derived from the data set obtained from the NSW RTA for the interrupted timeseries analysis. 97

108 incidents, with 22 per cent of fatal crashes and 40 per cent of single-vehicle nighttime crashes occurring between 1994 and 2001, where alcohol involvement was known, having at least one controller with a BAC over the legal limit (cf. nine per cent for all injury crashes and 29 per cent for all night-time crashes). These two surrogate measures of alcohol-related crashes also have frequently been used in drink-driving studies to investigate the effect of deterrence-based initiatives (Ross 1984; Hingson et al. 1987; Henstridge, Homel & Mackay 1997; Voas, Holder & Gruenewald 1997). The interrupted time-series analysis An interrupted time-series analysis is used in Study 1 to examine changes in crash rates after the 1998 legislative reforms took effect. Interrupted time-series designs infer the effects of a treatment by comparing measures of the outcome variable many times before the treatment with measures taken many times after the treatment has been implemented. A model can be constructed from the pre-observational data as to what the trends in the outcome measure are likely to be in the absence of any intervention. The actual trends in the outcome measure can then be compared with those in the predicted model to see whether they differ significantly. The extent to which they do differ may then be attributable to the intervention (Cook & Campbell 1979). To undertake this analysis, unit record crash data for a period of eight years, from January through December , were extracted from the Traffic Accident Database System (TADS). This database, which is maintained by the NSW Roads and Traffic Authority (RTA; the motor vehicle licensing authority in New South Wales), includes information for all crashes in New South Wales reported to police in which a person was killed or injured, or at least one motor vehicle was towed away. 27 No identifying information, such as names or vehicle registration numbers, was obtained. Consistent with previous Australian research (Henstridge, Homel & Mackay 1997; Chikritzhs et al. 2000), only crashes resulting in the death or injury of at least one person involved in the collision are considered. These crashes, defined 27 Two additional criteria for the inclusion of crashes into the RTA data collection are (a) that the crash occurred on a road open to the public and (b) that the crash involved at least one moving road vehicle. 98

109 here as injury crashes, are thought to be both more reliably reported and more likely to be alcohol-related. 28 Since injury crash numbers are likely to vary as a function of the number of vehicles on the road, monthly crash rates per 100,000 motor vehicles registered in New South Wales were examined in order to control for road usage. 29 Vehicle registration data is not, however, reported on a monthly basis in New South Wales. Instead, the RTA reports two six-monthly snapshots of the number of vehicles registered in New South Wales as at the June 30 and December 31. Monthly vehicle registrations were extrapolated for the current analysis from these six-monthly data. 30 Using interrupted time-series analysis, the monthly rate of injury crashes 31 per 100,000 vehicles was thus examined to determine whether the legal intervention was associated with a reduction in crash rates. The primary aim of using an interrupted time-series analysis in this instance is to isolate the effects of the legal changes from other trends or variability in the crash data. Additional interrupted time-series analyses are carried out for the two alcoholrelated road crash categories, that is single-vehicle night-time and fatal crashes. Again, crash rates per 100,000 vehicles registered are considered in order to control for changes in road usage. In an attempt to control for potentially confounding variables, such as weather conditions, interrupted time-series analyses are also carried out for a low alcohol-involvement crash category (or control series), namely vehicle-to-vehicle daytime crashes (i.e. multiple-vehicle crashes occurring between 28 Unlike previous road crash research conducted in NSW (Henstridge, Homel & Mackay 1997; Chikritzhs et al. 2000), serious injury crashes could not be used for the current study. This is due to the fact that, since 1997, NSW police no longer make the distinction between serious and other injury when recording details about a road crash. This information, therefore, was not available in TADS for all crashes recorded during the study period. 29 Registrations were considered to be a more accurate measure of road usage than licences on issue since many people have driver s licences for identification purposes, no longer drive or do not have access to a vehicle. 30 Excluded from the vehicle registration data were registrations of trailers and plant equipment. It should also be noted that the RTA changed their registration data collection from December 1994 and onwards by excluding outstanding registrations. These are registrations that have been transferred to another person but are yet to be confirmed with the RTA by the new registrant. This accounts for anywhere between 30,000 and 50,000 registrations each year. Since this only represents 1% of all registrations in the State it was decided that this discrepancy was not sufficient to bias the crash rates and therefore, 1994 registration statistics could be used for the analysis. 31 Fatal crashes are included in the injury crash category. 99

110 9am and 3pm; Henstridge, Homel & Mackay 1997). For this crash category, only one per cent of crashes occurring between 1994 and 2001, where alcohol involvement was known, had at least one controller who recorded an illegal BAC. Trends in high alcohol-involvement crash rates are then compared with trends in crash rates for the low alcohol-involvement category. Since the legal intervention would be expected to affect only drink-driving offending, any apparent impact of the policy should be less apparent for crashes less likely to involve alcohol. Ninety-six monthly observations were used as data points in the series (57 preintervention observations and 39 post-intervention observations), which is considered an adequate number of time-points for an interrupted time-series analysis (Cook & Campbell 1979; Box, Jenkins & Reinsel 1994). The intervention point for the change in penalties was October 1998, the first full month following the enactment of the Traffic Amendment (Penalties and Disqualifications) Bill on 30 September A Dickey-Fuller test for the presence of a unit root was initially conducted for each crash series to ascertain whether a regression procedure could be utilised in the interrupted time-series analysis (Koop 2000). Estimates of the Augmented Dickey Fuller test statistic (ADF) were obtained using E-Views4. For each series the test statistic was significant at the 0.05 level (injury rate => ADF = , fatal rate => ADF = , single-vehicle night-time rate => ADF = , multiple-vehicle day-time rate => ADF = ), indicating that the null hypotheses, in regard to the presence of a unit root, could be rejected. 33 Therefore it was concluded that all four data series were trend stationary and that an Ordinary Least Squares regression approach was an appropriate procedure to adopt for the interrupted time-series analyses. 32 It could be argued that July should be used as the intervention point since the legislative changes were announced by the government in late June and assented to on the July However further interrupted time-series analyses using July 1998 as the intervention point revealed no change in the significance of the intervention variables and only slight changes in the size of the estimated coefficients in the model. 33 For the injury and multiple-vehicle day-time series the Augmented Dickey-Fuller (ADF) test was based on the last 56 observations only. The reason for not using the first part of the series concerned the change in the level of the series after the intervention. A visual inspection of the time series plot suggested that the change in the level of the series in the latter half of the study period might be large enough to overwhelm the identification statistics, which could lead to incorrect specification of the model (Enders 1995). 100

111 Examination of the autocorrelation function plots of each data series indicated that all four series had seasonal autocorrelation. To control for this autocorrelation, 11 dummy variables corresponding to months of the year were included as independent variables in the models. September was used as the referent category in the model since it was thought to be the month that would be least likely to be affected by either school/public holidays or changes in rainfall. The regression model also contained variables for: 1. The underlying trend 2. The level of the series from October The underlying trend from October 1998 The two intervention terms included in the model to quantify the impact of the new policy on crash rates were the variables for the level of the series and the underlying trend of the series from October The first of these variables modelled an intervention effect where the level of crash rates changed immediately after the more severe penalties were introduced. This term was a binary variable corresponding to the implementation of the new penalties in October The second of these variables modelled an intervention effect where there was a change in the underlying trend of the data series after the legal intervention. This variable was coded 0 for the first 58 months prior to the new penalties being introduced and 1-39 thereafter. It is the significance and magnitude of the estimated parameters for these two terms, which are of most interest in this analysis. If the more severe penalties introduced in New South Wales did deter drinking drivers we would expect to see either one or both of these terms statistically significant and have negative coefficients in the regression model. Once the parameters of the model were estimated, diagnostic checks of the residuals were conducted to test for the presence of autocorrelation. This was done by examining the autocorrelation function (ACF) and partial autocorrelation function (PACF) plots for the estimated models, as well as Ljung-Box Q-statistics and their p- values The Q-statistic at lag k is a test statistic for the null hypothesis that there is no autocorrelation up to order k. 101

112 Drink-driving recidivism rates - Study 2 Study 2 compares the reoffending patterns of two cohorts: one group consisting of offenders convicted of a drink-driving offence in New South Wales in 1997 (pre-law cohort), the other consisting of offenders convicted of a drink-driving offence in New South Wales in 1999 (post-law cohort). The overall aim of this study is to examine whether drink-driving offenders sentenced after the new more severe penalties were introduced, differ significantly in their rates of recidivism from those sentenced under the old penalty system. This study builds on Study 1 in two important ways. Firstly, it provides an additional source of data to measure the effectiveness of the legislative reforms on drinkdriving offending. One concern with relying on motor-vehicle crash rates as a measure of drink-driving offending is that small changes in offending rates may not be evident in these aggregate-level data. This is because many drink-driving episodes do not result in road crashes and subsequently would not be captured in monthly crash data. Null findings from the interrupted time-series analysis may, therefore, be due to the policy having no discernable impact on drink-driving offending or, alternatively, due to the measure of offending not being sensitive enough to detect any effect. Study 2 attempts to rule out the latter interpretation. Secondly, if the results of the interrupted time-series analysis do show that the introduction of the new law had a significant effect on crash rates, Study 2 can investigate the mechanism by which the law has impacted offending. For example, did the legislative changes reduce the probability of offending, decrease the time to first offence or reduce offending amongst certain groups of offenders (e.g. highrange PCA offenders)? This study also examines the specific deterrent effect of the policy that is, whether offenders, with a previous conviction for a drink-driving offence, are responsive to changes in punishment variables. A threshold question when examining the effect of changes to sentencing law on recidivism is whether the legal changes resulted in any increase in the severity of penalties actually imposed by the courts. If they did not affect sentencing practice, 102

113 any change in recidivism cannot be attributed to the specific deterrent effect of the new laws. In addition to sentencing practice it is also important to consider whether the legislation had any negative impact on the prosecution of drink-driving offences. As discussed in Chapter 3, circumvention of the more severe penalties could reduce the perceived certainty of punishment and subsequently diminish any impact the sentencing policy may hope to have in deterring offenders. Given this, before assessing changes in reoffending rates, an analysis of drink-driving prosecutions and average drink-driving penalties imposed by magistrates in the local courts, before and after the legislative amendments, is undertaken. Measures of recidivism In Study 2, the number of convictions for a new offence is used to index recidivism. Reconviction rates represent only one of several measures that could be used to assess reoffending. A number of other drink-driving studies have relied on crash statistics as an indicator of reoffending because of its close relationship with drinkdriving and its public health significance. However, as mentioned previously, missing data on driver BAC, particularly in the latter years of the study period, means that these data would not be a reliable indicator of recidivism. Although more refined surrogate measures like single-vehicle night-time crashes could be used, the total number of these crashes occurring in New South Wales each year is relatively small (approximately 500 annually). The number of convicted drinkdrivers incurring this type of crash would be even smaller, making it difficult to detect any significant change in reoffending rates after the penalty increase. An alternative measure of offending is participant self-report. This measure of recidivism is advantageous in that it allows for a relatively accurate estimate of the true rate of reoffending. Court or crime data, on the other hand, only reflect those offending episodes that come to the attention of law enforcement agencies. The retrospective nature of the current study did not permit the use of self-report but an estimate of the true number of new offences committed by each offender is not necessary for this study. The question of interest in this analysis is whether the prelaw cohort differs significantly from the post-law cohort in their pattern of 103

114 offending and all that is needed to answer this is an unbiased indicator of reoffending (Luke & Lind 2002). The issue then becomes whether offenders from one group are more likely to be arrested and convicted for a new offence than offenders in the other. It would be reasonable to assume that there would be little difference between the two cohorts in terms of the probability of conviction for a new offence since the legislative reforms were intended to affect only the severity of punishment. However, the probability of arrest remains a potential source of bias if drink-driving enforcement levels changed substantially over the study period. It should be noted here that the actual probability of arrest for drink-driving is extremely low: estimated to be somewhere between one in 82 trips, based on data from self-report surveys (Liu et al. 1997), and one in 300 trips based on roadside breath-test surveys (Voas & Hause 1987). Furthermore, RBT, which is the primary means of enforcement of drink-driving offences in New South Wales, aims not to catch more offenders but to increase the perceived likelihood of apprehension for drink-driving amongst the general motorist population through highly visible and unpredictable road-side testing (Homel 1988). Thus, even if the number of breath tests were substantially different during the two follow-up periods this would not be expected to affect the actual probability of arrest substantially enough to render conviction rates unreliable. Increased breath testing by NSW Police during the study period could, however, have an additional deterrent effect on drink-driving offending (whether measured by road crash or conviction statistics) and therefore will need to be taken into account when interpreting the findings from Study 1 and Study 2. The recidivism analysis Study 2 employs a non-equivalent control group design to examine changes in reconviction rates. In these types of studies, the responses of a treatment group and comparison group are measured before and after an intervention. The treatment group in Study 2 consists of drink-drivers convicted after the increase in statutory 104

115 penalties (post-law cohort) and the comparison group consists of drink-drivers convicted before the penalty changes (pre-law cohort). The validity of inferences made from this design relies on the treatment and comparison groups being as near identical as possible, but since there is no random allocation to groups there still exists the possibility that one group differs systematically from the other. If the initial differences between the two groups are sufficiently large, any observable effect of the intervention may be due to selection bias. Therefore, attempts are made to select groups that would be relatively similar in all respects other than the variable of interest. Where this cannot be achieved, statistical controls can be incorporated into the analysis in order to account for any pre-existing differences between the groups. Data for the two offender groups being compared were drawn from NSW Bureau of Crime Statistics and Research Reoffending Database for this analysis. This database has assigned a unique identifier, based on several decision criteria, for every individual convicted of an offence in the NSW Local Courts since This identifier allows court appearances to be aggregated across years for each offender, giving an overall total of the number and type of offences for which any individual has been convicted (see Weatherburn, Lind & Hua 2003 for more discussion on matching criteria employed in the construction of this database). The pre-law cohort consisted of individuals convicted of a Prescribed Concentration of Alcohol (PCA) offence during the period January through December , while the post-law cohort was individuals convicted of a PCA offence during the period January through December Offenders in the pre-law cohort, who reappeared for a drink-driving offence in 1999, were excluded from the 1999 cohort to avoid duplication. The follow-up period in which reoffending is measured commences from the date of the first PCA conviction and continues for at least 36 months post-conviction. For the pre-law cohort, the cut-off date was December and for the post-law cohort, December For offenders convicted earlier in the year of interest the actual follow-up period exceeds 36 months, which will be taken into account in the survival analysis. A three-year 105

116 follow-up period has previously been estimated to capture approximately 60 per cent of offenders eventually reconvicted for a drink-driving offence (Homel 1980). Two measures of reoffending are used in this study to examine the effectiveness of the new penalties: (1) the proportion of offenders who were convicted of a new drink-driving offence during the follow-up period and (2) the number of days to first reconviction for a drink-driving offence. These measures of offending have been used successfully in previous offender studies to establish differences in recidivism (Peck 1991; Stewart, Gruenewald & Parker 1992; Voas, Holder & Gruenewald 1997; Lind et al. 2002; Luke & Lind 2002). Frequency of offending is also a useful measure of recidivism. However, in order to minimise the effect of other contextual variables on offending rates the two groups selected for the analyses were close in temporal proximity. Some of the follow-up period for the pre-law cohort would therefore have overlapped with the implementation of the new penalties, rendering the frequency comparison between groups problematic. A reconviction for drink-driving is defined in this analysis as any finalised court appearance 35 for a PCA offence during the follow-up period. Date of finalisation is used to identify the time at which the new offence occurred. It was not possible to use arrest date or offence date in this analysis since many of these dates are missing from the NSW Bureau of Crime Statistics and Research s Local Courts database. Although there would be a long delay between the offence date and the finalisation date, this would bias the results only if drink-driving offences were finalised sooner for one of the two groups. The average time from offence date (where these dates are available) to court finalisation date for drink-driving offences is therefore examined for each cohort. Using these court recorded offence data, two comparisons are made between the offender cohorts. Firstly, the proportion of offenders in each group who have been reconvicted for a drink-driving offence during a three-year follow-up period is 35 A finalised appearance is a group of one or more offences, against a single individual or a company, which are finalised by the Local Court on a single day. 106

117 compared. 36 This analysis provides a descriptive measure of the presence or absence of a reduced percentage of recidivists after the introduction of the more severe statutory penalties. A shortcoming of this analysis is that it assumes the two offender cohorts do not differ in any systematic way. Systematic group differences can be addressed, however, by conducting a logistic regression analysis. In this analysis the offender is the unit of analysis and the outcome is a binary variable, coded for whether or not the offender reappeared before the court during the follow-up period. A dummy variable is included in the model to designate whether an offender was sentenced prior to or after the legislative reforms took effect. Controlling for other confounding influences, such as age, gender and area of residence, differences between the offender groups can then be examined. Secondly, a survival analysis is conducted to compare the two groups on their time to first offence. Some individuals in the pre- and post-law cohorts will not be convicted of a second offence during the follow-up period, giving rise to censored data, and the follow-up periods for offenders in each group will also differ in length. These characteristics of the data point to survival analysis as being the most appropriate technique for the analysis. This approach provides an estimate of the likelihood of reoffending at any given time during the follow-up period. For this study, if the increased statutory penalties have the effect of reducing recidivism, then it should take longer on average for the post-law cohort to be convicted of a new PCA offence than the pre-law cohort. In the survival analysis, Kaplan-Meier survival functions are generated for the two offender groups. This procedure produces a survival curve showing the proportion of each group surviving beyond a given duration of time, that is, the proportion of each group that has not reappeared for a new offence. The two survival curves can then be directly compared, using nonparametric tests (e.g. log-rank tests), in order to determine whether the two groups differ significantly in terms of the time taken, on average, to reappear before the court. Alternatively, the two samples can be compared using Cox Proportional Hazard models. This method examines the 36 Note that in the logistic regression analysis the outcome measures is any reconviction within three years of the index offence. However in the survival analysis the outcome is any reconviction before the cut-off date specified for the relevant cohort. 107

118 proportion of each group reoffending at various points or intervals of time following the originating offence. More importantly, this analysis permits the use of covariates to adjust for a priori differences between groups. Therefore, it will be possible to test for significant differences in recidivism rates between the pre- and post-law groups while controlling for other factors which may affect reoffending (e.g. age, gender and area of residence). Two issues that need to be considered for the survival analysis are: the effect of imprisonment on time available to reoffend during the follow-up period and the assumptions underlying the proportional hazards model. While offender free time during the follow-up period may be somewhat reduced by a custodial sentence, very few drink-drivers in New South Wales receive an imprisonment term. This is particularly true for offenders convicted of low-range and special-range PCA offences, since there is no prescribed gaol term for these offences. However, imprisonment is a possibility for offenders convicted of mid- and high-range PCA offences. Therefore, survival analyses, excluding offenders who were sentenced to imprisonment, are also conducted to confirm the results obtained from the larger dataset. In any analysis it is also important to consider the assumptions underpinning the estimated models. A major assumption of a Cox Proportional Hazards Model analysis is that the hazard (i.e. risk of failure) for any one individual is proportional to the hazard for any other individual, where the proportionality constant is independent of time (Kleinbaum 1996, p. 133). This assumption will be tested in this study by graphing the log-log survival plots for the estimated model to ensure that the plots do not cross for one or more categories of predictors (Kleinbaum 1996). Perceptual research - Study 3 An essential factor to consider in any investigation of deterrent effects is people s perceptions of formal punishment. Ideally, this research would therefore have measured knowledge about the increases in statutory penalties for drink-driving offences contemporaneously with the implementation of the legislative changes, to investigate whether the legislative reforms had the intended effect on severity perceptions. The retrospective nature of this project prevented this analysis from 108

119 being undertaken. What could be measured, however, was the extent to which knowledge of the sanctions currently applicable to drink-driving offences in New South Wales predict the likelihood of committing this offence. For this reason a scenario-based study (Study 3) was designed to examine whether people who have more accurate knowledge of the current sanctions for drink-driving offences in New South Wales are less likely to state that they would drink and drive under certain conditions. Study 3 also contributes to the broader deterrence literature by examining the determinants of sanction perceptions, specifically, the influence of legal knowledge on perceptions of sanction severity. As noted previously in Chapters 1 and 3, perceptual studies examining deterrent effects have focused, almost exclusively, on the relationship between risk perceptions and intentions to offend or risk perceptions and actual offender behaviour. By restricting their focus in this way, these investigations neglect to consider an assumption of the deterrence doctrine which is essential for policy, that being the link between information known to an individual and their judgment of sanction risks (Piquero & Pogarsky 2002). This gap in the literature was highlighted by Nagin (1998) in his recent review of the deterrence research, but has been noted by scholars over the last two decades or more as a neglected area of research (e.g. Cook 1980). Miller and Iovanni (1994) also echo this concern when they note one must move beyond the focus on the effects of perceived risk and must investigate the determinants of perceived risk themselves (Miller & Iovanni, p. 282). By examining the factors influencing sanction risk perceptions, particularly the influence of legal knowledge on severity perceptions, this scenariobased research aims to bridge the gap between policy and behaviour. Hypotheses As discussed in Chapter 3, the validity of cross-sectional and longitudinal surveys in investigating hypotheses generated by deterrence theory has been questioned (Nagin 1998). For this reason a scenario-based design was employed to examine the relationship between legal knowledge and offending likelihood. In this survey, participants are presented with a vignette, describing a realistic drink-driving 109

120 scenario, and are then asked to indicate the likelihood they would offend under the circumstances described. The scenario-based survey undertaken for this thesis was designed to test four specific hypotheses generated by deterrence theory: 1. Increased perceived risk of arrest reduces offending likelihood One of the most robust findings from the deterrence literature is that the perceived certainty of apprehension can affect the likelihood of offending (Hingson et al. 1987; Homel 1988; Henstridge, Homel & Mackay 1997; Voas, Holder & Gruenewald 1997; Nagin & Pogarsky 2001). Several scenario-based studies have tested punishment certainty by having respondents estimate, on a continuum, the likelihood of apprehension and conviction for an offence. However, from a policy perspective it is important to establish whether police practice can successfully manipulate perceptions and subsequent behaviour. That is, can previous experience of RBT patrols in an area influence a person s current decision to drink and drive via increased perceptions of apprehension risk. For this reason, the level of RBT activity was varied across scenarios and its subsequent effect on stated offending likelihood examined. Deterrence theory would anticipate that those allocated the scenario with a high-level of police activity would be less likely to state that they would offend. 2. Enhanced knowledge of drink-driving penalties reduces offending likelihood Given that we were unable to investigate whether people knew about the 1998 drinkdriving policy, one of the key objectives of the scenario-based study was to examine knowledge of the current drink-driving penalties in New South Wales. One point to note here is that penalties applicable to drink-driving can take two forms: statutory penalties prescribed by the legislature and actual penalties imposed by the courts. This distinction is important in a deterrence context because perceptions of punishment severity are thought to depend not only on the magnitude of the prescribed penalty but also the likelihood of being sanctioned if convicted (von Hirsch et al. 1999). In this way, a person may know that severe statutory maximums are in place for an offence yet still offend because they believe that they will receive a much lighter penalty if caught. Thus, the survey tested knowledge of both statutory 110

121 and actual penalties for drink-driving offences. Utilising natural variation in levels of knowledge we then explored whether those individuals who were better informed about drink-driving penalties were less likely to report they would offend. 3. Increased perceptions of sanction severity reduces offending likelihood Within a deterrence framework we would anticipate that knowledge of legal penalties influences behaviour by raising perceptions of sanction severity and subsequently, the anticipated costs associated with the act. Unless knowledge of penalties alters perceptions of penalty severity then any relationship found between knowledge and behaviour cannot be attributed to deterrence. This study therefore not only tested respondents knowledge of the current penalties applicable to drinkdriving offences but also had them rate the severity of these penalties. Using a measure employed in other perceptual research (Homel 1986; Nagin & Paternoster 1993), respondents were asked how much of a problem the penalties would create for them personally. This question was designed to capture the subjective nature of severity perceptions, as they relate specifically to formal punishment prescribed by the law (statutory penalties) and imposed by the criminal justice system (actual penalties). 4. Perceptions of sanction severity matter more when punishment is perceived to be certain According to the deterrence model, a multiplicative relationship exists between the certainty and the severity of punishment. Thus punishment severity matters more in situations where offenders perceive it as likely they will be apprehended for the offence. While making intuitive sense, empirical research exploring this aspect of deterrence theory provides mixed evidence on the existence of such a relationship (Homel 1986; Howe & Brandau 1988; Howe & Loftus 1996; Grosvenor, Toomey & Wagenaar 1999). The survey formally tests the interaction between perceptions of punishment certainty and severity by including an interaction term in the regression models predicting stated offending likelihood. 111

122 Proposed model of the deterrence process Figure 4.1 presents a proposed model of the deterrence process that provides a framework for analysing the survey data. This model is loosely based on Homel s (1986) deterrence model describing the effect of RBT on future drink-driving behaviour (see Homel 1986, p. 41). Additional pathways have been included in this revised version to describe the relationship between knowledge of penalties for drink-driving offences (statutory and actual penalties) and decisions to offend. Underpinning this model is the basic assumption from subjective expected utility theory that formal punishment acts to reduce an offending behaviour by increasing the disutility (or costs) of the act. Accordingly, an individual will be deterred by the extent to which the anticipated costs associated with the behaviour outweigh the benefits. Expanding upon these basic concepts, the model also incorporates the influence of anticipated informal sanctions, that arise from a drink-driving arrest, on the decision to offend. Thus, in addition to the costs potentially imposed by the criminal justice system, decisions to drink and drive will be affected by informal sanctions from peers, family and the wider community when the offence is publicly exposed. Not all pathways described in the model are being investigated in Study 3 (the boxes in bold denote the variables being measured in the survey) and some interaction effects are implied. Despite these limitations, Figure 4.1 is a useful graphical depiction of the deterrence hypotheses described above that are being investigated in this study. Firstly, it proposes that individuals must have knowledge of the applicable penalties for drink-driving offences before they can be deterred by the formal punishment prescribed for this offence. While seemingly commonsensical, this relationship between knowledge and behaviour is one that is generally assumed rather than empirically tested in deterrence investigations. Secondly, for knowledge of drink-driving penalties to influence behaviour it must increase an individual s perceptions of sanction severity. Unless a significant relationship is found between knowledge and perceptions any relationship evident between knowledge and behaviour may be spurious. Thirdly, exposure to police activity will influence an individual s perceptions of apprehension certainty, thereby lowering the likelihood of drink-driving. 112

123 There are several other factors that could impact on both knowledge of drink-driving penalties and perceptions of arrest certainty, which will also be considered in this scenario-based research. Most importantly, consistent with previous research in the deterrence domain, the model proposes that prior offending, alcohol consumption and prior involvement in alcohol-related crashes could be determinants of the level of legal knowledge. Marginal offenders, that is people who have previously driven while drunk, who are at high-risk of acute alcohol-related harm or who have previously experienced an alcohol-related road crash, will be more motivated to seek out and retain relevant information about the formal costs associated with drinkdriving behaviours and therefore be more knowledgeable of drink-driving laws (Kenkel & Koch 2001). People who refrain from offending on moral grounds, on the other hand, are less likely to change their beliefs on the basis of this information and will potentially be less knowledgeable in this area. For these individuals, the model assumes that extralegal constraints, such as moral adherence to the law or feelings of guilt, reduce the benefits gained from committing the act and subsequently reduce the likelihood of offending. 113

124 Figure 4.1: Proposed model of the deterrence process Prior offending/ alcohol consumption/ prior accident involvement Knowledge of statutory penalties Prior convictions/ Media publicity/ friends, family, acquaintances Informal costs arising from apprehension Perceptions of penalty severity Perceived applicability of statutory penalties Total anticipated formal costs associated with DD (disutility) Decision to drink & drive Total benefits associated with DD (utility) Perceived risk of apprehension Media publicity/ friends, family, acquaintances Moral inhibitions associated with drink-driving Enforcement activity Prior successful drinkdriving episodes 114

125 Participants and methodology Students attending the University of New South Wales participated in the survey, which had received approval from the University of New South Wales Human Research Ethics Committee. These students were recruited from several criminal law, 37 engineering and general studies 38 classes to ensure that a broad cross-section of students studying various disciplines was included in the study. Students were considered a relevant target group since young people are at greater risk of injury from alcohol-related road crashes (Chikritzhs et al. 2000) and of drinking alcohol at acute-risk levels for alcohol-related harm (Chikritzhs et al. 2003). Furthermore, these students are also likely to be more knowledgeable of driving laws applicable in New South Wales, given that many would have obtained their full licence in recent years. The survey was self-completed by the students after the researcher gave a brief introduction and explained the questionnaire format (the participant information sheet and drink-driving questionnaire can be found in Appendix II). In the survey, knowledge of the drink-driving penalties currently in force in New South Wales was tested through a series of multiple-choice questions asking about the applicability of penalties and the statutory maximums and minimums that are prescribed for drink-driving offences. Given that the vast proportion of drink-drivers appearing before the Local Courts in New South Wales are persons convicted of a low- or mid-range drink-driving offence (NSW Bureau of Crime Statistics and Research 2002b), the questions asked in the survey pertained only to these two offences. As an example, the question for a low-range drink-driving offence was as follows: 37 Students in the criminal law classes were in their first year of study and therefore would not be expected to be any more knowledgeable on drink-driving penalties than other types of students. Later analyses confirmed that law students were, on average, no more knowledgeable about the penalties than other participants. 38 All University of New South Wales students are required to complete four general studies courses to fulfil the requirements of their degree. Most faculties offer these courses and students must attend a course offered by a school other than their own. For this research, students from a general studies course offered by the law and medical faculties were approached to participate in the survey. 115

126 Here are some questions about what the law can and cannot do when a person is convicted of drink-driving in NSW. Please answer all questions in the order in which they appear by ticking the appropriate box. If a person was caught driving with a blood alcohol level just over the legal limit (i.e. just over 0.05g/100ml) and it was their first drink-driving conviction. Participants were then asked to indicate whether a person could be fined, have their driver s licence disqualified or be goaled and, if these penalties were applicable, they were asked to indicate the magnitude of each sanction. Participants who correctly indicated the exact penalty applicable to the drink-driving offence described (i.e. either low- or mid-range drink-driving offences) received no points. For those correctly nominating that a particular penalty could be applied to the offence but did not know the magnitude of that penalty, a number of points based on their underestimation/overestimation were allocated. Those who incorrectly thought that a penalty was not applicable to the offence received the maximum points. Points for each response were totalled and participants were given an overall score reflecting the number of mistakes they made in answering these questions. 39 In a similar way, participants were asked a series of multiple-choice questions on the actual penalties they would receive were they caught and convicted for a low- or mid-range drink-driving offence in New South Wales. By way of example, the question pertaining to actual penalties for a low-range drink-driving offence was as follows: The previous questions asked about the minimum and maximum penalties, allowed by NSW law, for drink-driving offences. 39 This scoring process replicated a methodology previously used by Kenkel and Koch (2001) in their research on deterrence and knowledge of the law. 116

127 Suppose you were caught with a blood alcohol limit just over the legal limit (i.e. just over 0.05g/100ml). Give your best estimate as to what the actual penalty (or penalties) would be if you were convicted of your first drink-driving offence. Again, participants indicated whether they would be fined, have their licence disqualified or be gaoled and the magnitude of each of these penalties. In order to assess the accuracy of these responses, the multiple-choice answers were compared to the average penalties offenders received upon conviction for a low- or mid-range drink-driving offence in NSW Local Courts, between 1999 and A composite score was then assigned to reflect participant knowledge of the actual penalties imposed for drink-driving offences in New South Wales. Participants were then randomly allocated one of three scenarios, each varying in the level of RBT activity described, but following the general format shown below: Suppose you drove by yourself one evening to meet some friends at a pub. You didn t intend to drink much alcohol but by the end of the evening, you ve had enough drinks so you re pretty sure your blood alcohol level is above the legal limit. You haven t seen police conducting Random Breath Tests (RBT) in this area for some time. You live about 10kms away and you have to be at work early the next morning. You can either drive home or find some other way home, but if you leave the car at the pub, you will have to return early the next morning to pick it up. 41 The sentence italicised in the above scenario is the information that was varied across the three certainty conditions. You haven t seen police conducting Random Breath Tests (RBT) in this area for some time was the description given to participants allocated the low-risk of apprehension scenario. Participants allocated the mediumrisk scenario were told You have often seen police conducting Random Breath Tests (RBT) in this area and participants allocated the high-risk scenario were told You 40 These data were obtained from the Local Courts Database maintained by the NSW Bureau of Crime Statistics and Research. 41 Nagin and Pogarsky (2001) previously used a similar scenario to examine the relationship between punishment costs and intentions to offend amongst a sample of university students in the USA. 117

128 have often seen police conducting Random Breath Tests (RBT) in this area and since it is a Friday night you know that there will be more police on the roads. After reading the scenario, participants were asked to estimate the likelihood, on a scale of 0 (not at all) to 100 (definitely), that they would drive home under the circumstances provided. To confirm that the three different scenarios did affect perceptions of apprehension certainty, participants were also asked to estimate the likelihood, on a scale of 0 (not at all) to 100 (definitely), of being caught and convicted if they did drive home. Finally, to obtain subjective ratings of penalty severity, participants were asked to indicate how big a problem, on a scale of 0 (no problem) to 100 (a very big problem), the drink-driving penalties would create for them. This latter measure is one that has been used in previous research to gauge subjective estimates of punishment severity (Grasmick & Bryjak 1980; Homel 1986; Nagin & Paternoster 1993). Prior deterrence research has suggested an important role for extralegal factors, such as moral inhibitions, moral attachments, conscience and shame, in offending decisions. To account for these extralegal costs when modelling deterrence, some researchers have included a moral component to the costs side of the expected utility equation (e.g. Nagin & Paternoster 1994), such that the decision rule becomes offend if: p.u(benefits) > p.u(sanctions) + U(Moral Regret) However, Nagin and Pogarsky (2001) account for these same moral costs through the utility function: that is, as moral inhibitions associated with offending increase, U(Benefits) decrease. Thus, for persons who are so opposed to breaking the law that they will not offend under any circumstances it can be assumed that the crime is of no benefit to that individual (i.e. U(Benefits) = 0). Nagin and Pogarsky (2001) have identified these individuals who are morally opposed to offending by asking survey respondents in scenario-based research to report the likelihood of offending if there was no possibility of being sanctioned in the given circumstances. Participants who answer 0 to this question as well as 0 to the question on offending likelihood can be assumed to be insensitive to changes in formal sanctions and can therefore be excluded from the analysis. This methodology was also used in the current survey to 118

129 remove from the sample, those individuals for whom drink-driving would derive no benefit. Demographic information thought to influence offending decisions and/or knowledge of the law was also collected in the survey. This information included age, gender, university degree currently enrolled in, number of times respondent drove a motor vehicle after drinking too much alcohol, number of convictions for a drink-driving offence, individual and relatives involvement in alcohol-related crashes and frequency of high-risk drinking in previous 12 months. These data were collected to verify that groups receiving the three different scenarios, which contain varying levels of apprehension risk, did not differ in any systematic way. It was also necessary to include these measures as control variables in the regression analysis, given that legal knowledge and severity perceptions were not experimentally manipulated in the scenarios. Before commencing data collection, the scenario-based survey was piloted with a sample of third-year criminology students from the University of New South Wales (n=26). This pilot was undertaken to ensure that the drink-driving scenario presented in the survey was realistic and that it elicited a number of non-zero offending probabilities. Perceptions of apprehension risk were also compared across the three scenario conditions to confirm that the different levels of police activity described, affected perceptions of punishment certainty as intended. As a result of this pilot several minor changes were made to the design of the scenario. Most importantly, the wording contained in the medium- and high-risk scenarios was changed from you remember seeing police conducting RBTs of drivers in this area recently and you remember seeing police conducting RBTs of drivers in this area recently and since it is a Friday night you know that there will be more police on the roads, respectively, to you have often seen police conducting Random Breath Tests (RBT) in this area and you have often seen police conducting Random Breath Tests (RBT) in this area and since it is a Friday night you know that there will be more police on the roads. These alterations were made to the scenario design after observing little difference in risk perceptions between the low- and medium-risk scenarios. 119

130 CHAPTER 5. IMPACT OF THE 1998 PENALTY CHANGES ON OFFENDING RATES IN NEW SOUTH WALES As discussed in Chapter 4, two aggregate-level studies were undertaken to examine the impact of the penalty changes made to drink-driving offences in New South Wales, in Study 1 consists of interrupted time-series analyses of four road crash series: all injury road crashes, fatal crashes, single-vehicle night-time crashes and multiplevehicle day-time crashes occurring in New South Wales between 1994 and Fatal and single-vehicle night-time crashes are the two surrogate measures of alcohol-related road crashes used in this study and multiple-vehicle day-time crashes is the control series against which the alcohol-related series are compared. The second study involves an analysis of the recidivism rates of two drink-driver cohorts: one convicted before the penalty changes in 1998 and the other convicted after the penalty changes. Reoffending is measured as any new court appearance for a drinkdriving offence recorded during the follow-up period and both the probability of offending and the time to first new offence are compared across the two cohorts. If more severe penalties have an effect on offending we would expect to see a significant reduction in alcohol-related road crashes and drink-driving recidivism rates after the implementation of the new penalty regime in This Chapter presents the results of Study 1 and Study 2 and discusses their implications in light of current knowledge about deterrence and the context in which the penalties were implemented. Study 1 - Road crash analysis Overall, 164,573 motor vehicle crashes, in which a person was either injured or killed, occurred on NSW roads between 1994 and 2001, with the vast majority (

131 per cent) of these being non-fatal crashes. These 164,573 crashes resulted in a total of 4,684 fatalities and 210,084 casualties. All road crashes resulting in injury The first crash series examined in this study consists of all road crashes recorded in New South Wales from 1994 to 2001, that resulted in the death or injury of at least one person. These crashes are presented in Figure 5.1 by the year in which they occurred. As shown in this figure, injury crashes remained fairly stable in New South Wales from 1994 to 1998, with the exception of a notable decrease in However, from 1999 through 2001 there was an increase in the total number of injury crashes recorded each year. Figure 5.1: Total injury crashes in NSW, ,000 20,000 19,488 19,306 19,366 18,378 19,388 19,886 21,852 22,704 Total number 15,000 10,000 5, To examine trends in injury crashes, controlling for road usage, the rate of monthly injury crashes per 100,000 vehicles registered in New South Wales was calculated as the dependent variable for the interrupted time-series analysis. 42 The final regression 42 Another technique to control for the number of vehicle on the roads would be to use the amount of petrol sold in NSW each month as the denominator for monthly crash rates. Automotive petrol sales data were therefore sought from the Commonwealth Department of Industry, Sciences and Resources for the period 1994 to Interrupted time-series analyses were then carried out for each of the four crash series previously examined using monthly crashes per kilolitres of petrol sold as the dependent 121

132 model estimated for these data is shown in Table 5.1. Included in this model is a term for the underlying trend in injury crashes, a term for the change in the underlying trend from October 1998 and dummy variables to control for seasonal autocorrelation in the data. 43 The term modelling a change in the level of the injury crash series was not significant (t=0.011, p=0.991) and was therefore excluded from the final model. The estimated model accounted for 71 per cent of the total variance in injury crash data but an examination of the autocorrelation and partial autocorrelation function plots revealed significant autocorrelation in the residuals at lag 3. Re-estimating the regression model with the inclusion of a third order autocorrelation term (AR3), resulted in trend terms that had near identical magnitudes and statistical significance to those shown in Table 5.1. The Ordinary Least Squares estimates shown in this table were therefore retained. Table 5.1: Linear Regression Model for Injury Crashes per 100,000 vehicles registered, New South Wales, Variable Coefficient Standard t p value error Intercept < Underlying trend < Change in underlying < trend from Oct. 98 Seasonal terms Jan Feb Mar < Apr May < Jun Jul Aug Oct Nov Dec Notes: R 2 = & Durbin-Watson = 1.633, 44 significant autocorrelation at lag 3 p = variable. The results of these supplementary analyses were similar to those found using vehicle registrations as the denominator. 43 Note that September is the referent month for the seasonal variables. 44 Durbin-Watson is a test statistic for the presence of first order serial correlation. If the value of the test statistic is close to 2 then the model is considered a good fit to the data. 122

133 As seen in Table 5.1, the coefficient for the underlying trend is negative and statistically significant (t=-6.57, p<0.001). This indicates that there was a general decreasing trend in injury crash rates over the study period of 0.09 crashes, per 100,000 vehicles registered, per month. The term for the change in trend, post intervention, was also significant, indicating that there was a significant change in this downward trend in injury crashes from October 1998 onwards (t=8.36, p<0.001). The magnitude of this change can be calculated by adding together the coefficients for the initial decreasing trend and the increasing trend post intervention. Thus, after the legal policy had been implemented, it was estimated that injury crashes per 100,000 vehicles registered increased in New South Wales at a rate of 0.15 per month. The significant increase in injury crashes after the intervention apparent from the results of the interrupted time-series analysis, is displayed graphically in Figure 5.2. This figure presents the injury crash rate per month from 1994 through 2001, with a regression line showing how the underlying trend in crash rates changed after the new drink-driving penalties were introduced in September To construct this graph, the seasonal variables were removed from the regression model and the underlying trend and change in the underlying trend were estimated. By plotting the observed crash rates against the estimated values from this model, a clearer picture of the trend in injury crashes over the study period emerges. 123

134 Figure 5.2: Monthly injury crashes per 100,000 vehicles registered incorporating a fitted trend line, NSW, Rate per 100,000 vehicles Jul-01 Jan-01 Jul-00 Jan-00 Jul-99 Jan-99 Jul-98 Jan-98 Jul-97 Jan-97 Jul-96 Jan-96 Jul-95 Jan-95 Jul-94 Jan-94 Penalties increase Injury rate Trend line Alcohol-related road crash rates As noted previously, two surrogate measures of alcohol-related crashes, all fatal crashes and single-vehicle night-time crashes, were employed in Study 1 to better estimate the impact of the penalty increase on drink-driving offending. Figure 5.3 presents the raw number of each of these crash types recorded in New South Wales each year, from 1994 through to As seen here, the yearly number of fatal crashes remained relatively stable over this eight-year period, ranging from a high of 563 fatal crashes in 1995 to a low of 486 fatal crashes in At the same time, the number of single-vehicle night-time crashes appeared to be steadily declining from 1994 through However, after a low of 1,427 crashes in 1997, single-vehicle night-time crashes began to increase in the latter half of the study period, with crashes peaking at 1,643 in

135 Figure 5.3: Monthly fatal and monthly single-vehicle night-time crashes, NSW, Total number 2,000 1,800 1,600 1,400 1,200 1, SVNT Fatal Table 5.2 and Table 5.3 display the estimated regression models for fatal crashes per 100,000 vehicles registered and single-vehicle night-time crashes per 100,000 vehicles registered, respectively. Again, the only intervention variable included in the final models is a term for the change in the underlying trend from October 1998 onwards, given that there was no significant change in the level of these series post intervention (for fatal crashes; t=-0.12, p=0.91 and for single-vehicle night-time crashes; t=-0.15, p=0.88). Both the regression model for fatal crash rates and the regression model for single-vehicle night-time crash rates contained no autocorrelation (as assessed by the Durbin-Watson statistic and the autocorrelation and partial autocorrelation plots) and these models accounted for 37 per cent and 49 per cent of the total variance in crash rates, respectively. Taken together, this suggests that the estimated regression models for both alcohol-related series could be considered a good fit to the data. 125

136 Table 5.2: Linear Regression Model for Fatal Crashes per 100,000 vehicles registered, New South Wales, Variable Coefficient Standard t p value error Intercept < Underlying trend Change in underlying trend from Oct. 98 Seasonal terms Jan Feb Mar Apr May Jun Jul Aug Oct Nov Dec Notes: R 2 = & Durbin-Watson = 2.319; no autocorrelation or partial autocorrelation in residuals Table 5.3: Linear Regression Model for Single-Vehicle Night-Time Crashes per 100,000 vehicles registered, New South Wales, Variable Coefficient Standard t p value error Intercept < Underlying trend < Change in underlying trend from Oct. 98 Seasonal terms Jan Feb Mar Apr May Jun Jul Aug Oct Nov Dec < Notes: R 2 = & Durbin-Watson = 2.034; no autocorrelation or partial autocorrelation in residuals The model for fatal crash rates shown in Table 5.2 reveals that there was a downward trend in this crash series in New South Wales over the period 1994 through The magnitude of this decreasing trend was in the order of fatal crashes, per 126

137 100,000 vehicles registered, each month (t=-3.09, p=0.003). This significant downward trend was also observed for all injury crashes over the same time period. However, for fatal crash rates, the term modelling the change in this underlying trend, post intervention, was not statistically significant (t=0.63, p=0.53). This suggests that the implementation of the new laws in September 1998 had no impact on the general downward trend in fatal crash rates in New South Wales between 1994 and The data series representing single-vehicle night-time crash rates also showed a significant downward trend, prior to the intervention, of 0.01 crashes, per 100,000 vehicles registered, (t=-3.95, p<0.001) per month. However, unlike fatal crash rates, there was evidence from the regression model that there was a significant change in this underlying trend from October 1998 onwards (t=2.41, p<0.05). The positive coefficient for the term modelling the change in the underlying trend indicates that, after the penalties for drink-driving offences were increased in September 1998, single-vehicle night-time crashes began to rise. The magnitude of this change was in the order of additional crashes, per 100,000 vehicles registered, per month. Given that single-vehicle night-time crashes contain a greater proportion of alcoholrelated incidents, this significant increase in crash rates raises some doubts about the hypothesised deterrent effect of the 1998 sanctioning policy. Figure 5.4 and Figure 5.5 display the monthly rates for the fatal and single-vehicle night-time crash series, as well as regression lines showing the underlying trend and the change in trend from October Again, these graphs were constructed by plotting the actual monthly crash rates against the predicted monthly crash rates, where the predicted values are estimated from the regression model with seasonal variables removed. These figures illustrate the decline in both alcohol-related crash series over the initial part of the study period and the significant increase in singlevehicle night-time crash rates after the implementation of the new penalties, an increase which is not apparent in the fatal crash series. 127

138 Figure 5.4: Monthly fatal crashes per 100,000 vehicles registered incorporating a fitted trend line, NSW, Rate per 100,000 vehicles Jul-99 Jan-99 Jul-98 Jan-98 Jul-97 Jan-97 Jul-96 Jan-96 Jul-95 Jan-95 Jul-94 Jan-94 Penalties increase Jul-01 Jan-01 Jul-00 Jan-00 Fatal rate Trend line Figure 5.5: Monthly single-vehicle night-time accidents per 100,000 vehicles incorporating a fitted trend line, NSW, Rate per 100,000 vehicles Jul-01 Jan-01 Jul-00 Jan-00 Jul-99 Jan-99 Jul-98 Jan-98 Jul-97 Jan-97 Jul-96 Jan-96 Jul-95 Jan-95 Jul-94 Jan-94 Penalties increase SVNT rate Trend line 128

139 Control series The control series in Study 1 was the number of multiple-vehicle crashes occurring between 9am and 3pm (of which approximately one per cent are alcohol-related). Figure 5.6 presents the raw number of these types of crashes occurring in New South Wales each year between 1994 and Following a similar pattern to all injury crashes, the yearly number of multiple-vehicle day-time crashes appeared to be relatively stable in New South Wales between 1994 and However, in the latter half of the study period these crashes began to increase in number, particularly in the final years of the eight-year period (i.e and 2001). Figure 5.6 Multiple-vehicle day-time crashes in NSW, ,000 6,000 5,000 5,004 5,030 5,098 4,813 5,051 5,186 5,731 6,077 Total number 4,000 3,000 2,000 1, The estimated regression model for the rate of multiple-vehicle day-time crashes per 100,000 vehicles registered is shown in Table 5.4. Included in this model are terms for the underlying trend in this crash series, the change in this trend after the penalties were raised in September 1998 and 11 dummy variables to control for seasonal autocorrelation in the data. Again, the regression model revealed that the intervention had no statistically significantly impact on the level of the series (t=- 0.03, p=0.98) and, therefore, this variable was excluded from the final model. The model shown in Table 5.4 accounts for 58 per cent of the variance in the control series and from an examination of the autocorrelation and partial autocorrelation 129

140 plots, as well as the Durbin-Watson test statistic, the model contained no significant autocorrelation. The significant, negative coefficient for the term modelling the underlying trend in these data indicates that, over the eight-year period, there was a decreasing trend in multiple-vehicle day-time crashes of 0.02, per 100,000 vehicles registered, per month (t=-4.14, p<0.001). From October 1998 onwards, however, there was a statistically significant change in this downward trend, with multiple-vehicle day-time crashes beginning to increase at a rate of 0.05 crashes, per 100,000 vehicles registered, per month (or a 4.5 per cent change from October 1998). Table 5.4: Linear Regression Model for Multiple-Vehicle Day-Time Crashes per 100,000 vehicles registered, New South Wales, Variable Coefficient Standard t p value error Intercept < Underlying trend < Change in underlying < trend from Oct. 98 Seasonal terms Jan Feb Mar Apr May Jun Jul Aug Oct Nov Dec Notes: R 2 = & Durbin-Watson = 1.775; no autocorrelation or partial autocorrelation in residuals The significant increase in the control series post intervention is displayed graphically in Figure 5.7. This figure presents the monthly rates of multiple-vehicle day-time crashes and a fitted trend line derived from an estimated regression model with the seasonal terms removed. As with the trends observed for the two other nonfatal crash series, rates of multiple-vehicle day-time crashes were decreasing up until the implementation of the intervention, at which point these crash rates reversed direction and began to rise. 130

141 Rate per 100,000 vehicles Figure 5.7: Monthly multiple-vehicle day-time crash rates per 100,000 vehicles registered incorporating a fitted trend line, NSW, Jul-01 Jan-01 Jul-00 Jan-00 Jul-99 Jan-99 Jul-98 Jan-98 Jul-97 Jan-97 Jul-96 Jan-96 Jul-95 Jan-95 Jul-94 Jan-94 Penalties increase MVDT rate Trend line Summary To summarise so far, the interrupted time-series analyses found statistically significant increases in injury crash rates, single-vehicle-night-time crash rates and multiple-vehicle day-time crash rates after the more severe penalties for drinkdriving offences were introduced in New South Wales in The only crash series in which there was no significant increase during the latter half of the study period was monthly fatal crashes. What is of most interest for our test of deterrence is whether there was any change in alcohol-related crashes after the legislative amendments were enacted in New South Wales. To determine this, two surrogate measures of alcohol-related crashes, single- 45 The magnitude of these effects should warrant some caution when interpreting the public health significance of these findings. For single-vehicle night-time crashes, the observed increase in crashes after the policy was implemented was crashes each month for every 100,000 vehicles registered in New South Wales. With at least three and a half million motor vehicles registered in New South Wales during this time, this increase equates to five additional single-vehicle night-time crashes occurring each year. In a State of six and a half million persons this rise in crash rates, although statistically significant, may seem relatively small. However, it is arguable that any increase in the risk of injury or death can be considered an important finding. 131

142 vehicle night-time crashes and fatal crashes, were examined. The differing trends in these two series during the latter half of the study period were therefore unexpected. Here it should be noted however that fatal crashes in New South Wales have been decreasing at a fairly constant rate over the last two decades (Australian Transport Safety Bureau 2001). One of the reasons for this reduction is certainly the introduction of RBT (Henstridge, Homel & Mackay 1997) but this phenomena of falling road mortality rates is not exclusive to New South Wales, nor indeed to Australia, but has been observed in many Western countries during the second half of this century, despite a concurrent increase in the number of vehicles on the road (Australian Transport Safety Bureau 2001). It is possible that this general downward trend in road fatalities is not due to fewer people being involved in serious, and potentially fatal, road crashes but due to other factors, such as enhancements in automotive design, road engineering and trauma care (Nathens et al. 2000), improving the chances of a crash victim surviving. If the lethality of road crashes has reduced in more recent times, then single-vehicle night-time road crashes resulting in an injury might be the best indicator of changes in alcohol-related road crashes, particularly when measuring changes over the short term. It is also the crash category that contains a greater proportion of drivers with an illegal BAC and therefore should be more sensitive to changes in drink-driving levels. The increased drink-driving penalties introduced in 1998 were therefore expected to have the biggest impact (in the form of a significant reduction) on single-vehicle night-time crashes. The control series, multiple-vehicle day-time crashes, was expected to be the least affected because fewer of these crashes involve alcohol. The findings from the interrupted time-series analyses described above, which suggest that there was a significant increase in both single-vehicle night-time and multiplevehicle day-time crash rates after the 1998 penalty changes, are therefore, prima facie, inconsistent with deterrence theory. When interpreting these results however it needs to be recognised that the significance test in an interrupted time-series analysis is not a test of cause (Cook & Campbell 1979). While the above analyses did find a statistically significant change 132

143 in these two non-fatal crash series, at the specified time point, the analysis does not offer any explanations for the change. This means that any other number of factors, apart from the sentencing policy, may have led to the observed increase in crashes post intervention. In fact sensitivity analyses, in which the intervention point was placed six months prior to the penalty increase being implemented, resulted in the same conclusions. Thus, rather than the sentencing policy actually increasing road crashes, it is more likely that the intervention occurred at, or around, the time that crash rates were already on the rise in New South Wales. Alcohol-related crashes versus control series Further support for the conclusion that the drink-driving penalty changes did not contribute to the rise in road crashes is apparent when we compare the postintervention increase in single-vehicle night-time crashes with the increase in multiple-vehicle day-time crashes. If the penalty changes had a negative effect on alcohol-related crashes then we would expect to see a greater increase in our surrogate measure of alcohol-related road crashes after the changes were enacted than in the control series. However, Figures 6.1 and 6.2 suggest that multiple-vehicle day-time road crashes (the control series) may have been increasing at a greater rate after the intervention point than single-vehicle night-time road crashes. To more formally test this possibility, the following regression model was estimated: Y = b0 + b1group + b2trend + b3trendch + b4trend2 + b5trendch2 where Y is the crash rates for both multiple-vehicle day-time crashes and singlevehicle night-time crashes; GROUP is a binary variable representing the two crash categories (coded 0 for multiple-vehicle day-time rates and 1 for single-vehicle night-time crash rates); TREND is a linear variable to model the underlying trend in the series, TRENDCH is a binary variable to model the change in the underlying trend after October 1998 and TREND2 and TRENDCH2 are interaction terms for GROUP*TREND and GROUP*TRENDCH. Dummy variables corresponding to the months of the year were also included in the final model to deseasonalise the data. 133

144 The variable of interest in this analysis is the last term in the regression equation, that is the interaction between group membership and the change in the underlying trend post intervention (GROUP*TRENDCH). This term essentially compares the rate of increase after the intervention across the two crash categories (single-vehicle nighttime crash rates and multiple-vehicle day-time crash rates), taking into account other trends or variability in the data series. The regression approach adopted here was considered appropriate because there was no serial autocorrelation in either the single-vehicle night-time or the multiple-vehicle day-time crash series and it would be reasonable to assume that these types of crashes can be considered two separate observations at a single point in time. 46 The final regression model, which is shown in Table 5.5, reveals that the interaction term for group by trend change (GROUP*TRENDCH) is statistically significant (t=- 4.76, p<0.001). This suggests that the change in the underlying trend from October 1998 (i.e. the post-slope of the series) was significantly different for the two crash categories. The estimated post-slope for multiple-vehicle day-time crashes from the regression equation is and the post-slope for single-vehicle night-time crash rates is This significant finding provides some evidence that alcohol-related crashes were increasing at a slower rate after the 1998 sentencing policy was introduced than were non alcohol-related crashes. 46 While collinearity diagnostics suggested that there was some multicollinearity in the model, both the condition index and variance proportion estimates were within an acceptable range (Kleinbaum, Kupper & Muller 1988). Comparisons of the predicted rates for each crash series with the observed rates also confirmed the validity of the estimated model. 134

145 Table 5.5: Linear Regression Model for Multiple-Vehicle Day-Time and Single- Vehicle Night-Time Crash rates, NSW, Variable Coefficient Standard t p value error Intercept < GROUP < Underlying trend (TREND) < Change in underlying trend < from Oct. 98 (TRENDCH) Group*Underlying trend (TREND2) Group*Change in < underlying trend from Oct. 98 (TRENDCH2) Seasonal terms Jan Feb Mar Apr May Jun Jul Aug Oct Nov Dec Preliminary conclusions To assess changes in drink-driving levels after the sentencing policy was implemented in New South Wales, single-vehicle crashes occurring at night were selected as one of the surrogate measures for alcohol-related road crashes. This series was chosen because a greater proportion of these crashes involve drivers with illegal BACs. However, it needs to be remembered that over half of the crashes included in this category are unrelated to alcohol. Thus, if some other unmeasured factor was driving an increase in non alcohol-related crashes then it would be reasonable to expect some residual effect to also be evident in this surrogate measure. The comparison between the control and alcohol-related crash series described above provides some support for this assertion. If the observed rise in single-vehicle night-time crashes was driven mostly by the increase in non alcohol-related crashes, then two possible conclusions regarding the policy s effect on alcohol-related crashes can be drawn from the findings presented 135

146 above: (1) that the increase in statutory penalties had a beneficial impact on alcoholrelated road crashes but the effect was swamped by the secular rise in non alcoholrelated crashes occurring at the same time or (2) that the increase in statutory penalties for drink-driving had no beneficial impact on alcohol-related road crashes. Given that single-vehicle night-time crashes were found to be increasing at a slower rate post-intervention than multiple-vehicle day-time crashes, it is possible that there was a significant reduction in alcohol-related road crashes after the intervention point but that this decrease was overwhelmed by the rise in non alcohol-related crashes occurring at the same time. This would suggest that the sentencing policy introduced in New South Wales in 1998 had a deterrent effect on drink-driving but that our surrogate measure of alcohol-related crashes was not sensitive enough to detect the change. However, there may be other factors that could potentially account for the differential rise in these two crash series. One possibility is that there was an increase in traffic density in the latter half of the study period, which resulted in more road crashes. Because traffic is heaviest during the day, this change would have a greater effect on multiple-vehicle day-time crashes than on single-vehicle night-time crashes. Thus, it also remains possible that the sentencing policy failed to have any impact on alcohol-related road crashes and that the differential rise in alcohol and non-alcohol crash rates found in this study was simply due to other factors. Other potentially confounding factors such as traffic density that could account for the differential rate of increase in the two crash series are difficult to measure at the aggregate level. Nevertheless, the second quasi-experimental study undertaken for this thesis will be able to shed further light on the findings from these time-series analyses. Recalling from Chapter 4, Study 2 compares the recidivism rates of drinkdrivers convicted before the penalty changes with the recidivism rates of drinkdrivers convicted after the rise in penalties. If a significant difference in reoffending is found in this second study, and it is in the direction that deterrence theory would predict, then it remains possible that alcohol-related road crashes decreased after the penalty changes but that this reduction was swamped by the rise in non alcoholrelated crashes. If, on the other hand, there are no significant differences found between these two offender groups, or if the differences are not in the expected 136

147 direction, then we can rule this out as a potential explanation for the results from Study 1. An examination of the impact of the penalty increase on drink-driving reoffending rates is therefore necessary before we can make confident conclusions regarding the deterrent effect of the 1998 sentencing policy. Study 2 - Analysis of recidivism rates As discussed in Chapter 4, before changes in recidivism can be considered it is important to establish that the 1998 statutory amendments were successfully translated into sentencing practice. If these legal changes failed to affect the severity of penalties actually imposed by the courts, any reduction in recidivism cannot be attributed to the specific deterrent effect of the new laws. In addition, it is also important to consider whether the legislation had any negative impact on the prosecution of drink-driving offences. Circumvention of the more severe penalties by players within the judicial system could reduce the perceived certainty of punishment and subsequently diminish any impact the sentencing policy may hope to have in deterring offenders. The following section examines both these issues by comparing all drink-driving offences prosecuted in the NSW Local Courts in 1997 with those prosecuted in These data were extracted from the Local Courts Database maintained by the NSW Bureau of Crime Statistics and Research (BOCSAR). Implementation of statutory amendments There are essentially five ways in which the intended deterrent effects of the new sentencing laws for drink-drivers could have been undermined: (1) a reduction in drink-driving charges brought before the courts, (2) an increase in court delay, (3) a reduction in guilty pleas, (4) a reduction in proven offences or (5) no subsequent change to the severity of drink-driving penalties imposed by the courts (MacCoun 1993; Nagin 1998). As shown in Table 5.6, the number of drink-driving prosecutions and the proportion of these that were proven in court increased after the penalties were raised. On the other hand, there was also no significant change in the average time from the offence being committed to the matter being finalised in the Local Court. There was a significant decrease from 1997 to 1999 in the proportion of offenders who pleaded 137

148 guilty for a drink-driving offence, however this decrease was mostly due to more offenders not turning up before the court to submit a plea rather than an increase in the proportion of offenders pleading not guilty. 47 Taken together, these data provide no evidence to suggest that the new sentencing laws were undermined by factors associated with the prosecution of drink-driving offences in NSW courts. Table 5.6: Drink-driving offences prosecuted in NSW Local Courts, 1997 & % change Total prosecutions (n) 19,237 23, Average time to finalisation (days) Offences with a guilty plea (%) Offences proven (%) Importantly, the new drink-driving laws did increase the severity of penalties that drink-driving offenders received upon conviction. As seen from Table 5.7, average gaol terms, fines and licence disqualification periods imposed by the courts for drinkdriving offences increased after the implementation of the new legislation in Statistical tests showed these increases in the magnitude of penalties to be significant. 48 Table 5.7: Penalties imposed for drink-driving offences in NSW Local Courts, 1997 & 1999 Penalty type % change Gaol % gaoled Average (mths) Fine % fined Average ($) Licence disqualification % disqualified Average (mths) However, the magnitude of this rise in penalties was not uniform across the State (see Table 5.8). In Sydney metropolitan courts the average gaol term rose by 28 per cent, the average fine by 51 per cent and the average licence disqualification period by nine per cent. For non-sydney courts, the average gaol term did not change significantly from 1997 to 1999 but the average fine increased by 40 per cent and the 47 Offences proven (x 2 =5.57, df=2, p=0.02); time to finalisation (t=0.72, df=42,519, p=0.47); offences with a guilty plea (x 2 =123.74, df=2, p=0.001). 48 Gaol (t=-3.43, df=737, p<0.001); Fine (t=-57.04, df=30,279, p<0.001); Licence Disqualification (t=-13.36, df=33,736, p<0.001). 138

149 average licence disqualification period by 24 per cent. 49 This differential rise in the severity of drink-driving penalties across the State is an important outcome to consider in the recidivism analysis. The harsher penalties imposed on non-sydney offenders could affect their perceptions of the costs associated with drink-driving and therefore have a greater impact on their offending behaviour. Furthermore, with less access to alternative forms of transport, people living outside of major cities are heavily reliant on motor vehicles and thus may be more likely to perceive licence disqualification as a relatively severe sanction. Given this, a greater reduction in reoffending rates for offenders residing in country and regional areas would be anticipated. Table 5.8: Penalties imposed for drink-driving offences in NSW Local Courts broken down by court location, 1997 and 1999 Sydney courts non-sydney courts Penalty type % change % change Gaol % gaoled Average (mths) Fine % fined Average ($) Licence % disqualification disqualified Average (mths) The only trend toward greater leniency in sentencing was a statistically significant but small decrease in the proportion of offenders who received a licence disqualification and in the proportion of offenders who were fined in 1999 (see Table 5.7). 50 This decrease in the proportion of drink-drivers disqualified was due to more drink-drivers having their conviction dismissed under Section 10 of the NSW Crimes (Sentencing Procedures) Act 51 in 1999 than in This increase in Section Variations across localities in the average licence disqualification were still apparent when differences in the type of drink-driving offences prosecuted in these courts (i.e. high-, mid-, low- or special-range PCA) were taken into account. 50 x 2 =8.00, df=1, p= Section 10 of the NSW Crimes (Sentencing Procedures) Act 1999 permits a court, in certain circumstances, to direct that a relevant charge be dismissed for a guilty offender or that the guilty offender be conditionally discharged. In the case of drink-driving offences this means that the licence 139

150 dismissals was most apparent for offences that had minimum licence disqualification periods introduced in 1998, that is low- and special-range drink-driving offences (see Figure 5.8). In contrast there was little change between 1997 and 1999 in the imposition of Section 10s for those convicted of a mid- or high-range offence, offences that already had prescribed minimum licence disqualifications prior to the current legislative amendments. 52 These data lend some support to the claim that the removal of judicial discretion through the introduction of statutory minima resulted in magistrates circumventing the severe penalties through the imposition of alternative sanctions. Figure 5.8: Percentage of proven drink-driving offences dismissed without a conviction (s10), by offence type and year percentage Special Low Mid High Drink-driving offence type In summary, the 1998 legislation resulted in a significant increase in the average penalties imposed for drink-driving offences without having any negative impact on the prosecution of drink-driving offences. The only setback in terms of the deterrence aims of the legislation was a statistically significant, though small, decrease in the proportion of offenders being disqualified from driving. However, four out of every five drink-driving offenders convicted in 1999 did receive a licence disqualification, disqualification provisions under the NSW Road and Transport (General) Act also are not applied. This section of the NSW Crimes (Sentencing Procedures) Act 1999 is equivalent to Section 556A of the NSW Crimes Act 1900 which was repealed in Special-range PCA (x 2 =8.36, df=1, p=0.004); Low-range PCA (x 2 =13.65, df=1, p<0.001); Midrange PCA (x 2 =3.45, df=1, p=0.063); High-range PCA (x 2 =1.04, df=1, p=0.307). 140

151 which, on average, was longer in duration than licence disqualifications given to their 1997 counterparts. Since evidence suggests that longer periods of licence disqualification can have a specific deterrent effect on drink-driving offending (Blomberg, Preusser & Ulmer 1987; Mann et al. 1991; Siskind 1996), the 1998 legislative amendments still had the potential to reduce recidivism rates in New South Wales. Characteristics of offender cohorts To compare recidivism rates before and after the penalty changes, data for two offender cohorts were extracted from BOCSAR s Reoffending Database: one consisting of individuals convicted of drink-driving in 1997 (pre-law cohort) and the other consisting individuals convicted of drink-driving in 1999 (post-law cohort). A total of 18,017 persons were included in the pre-law cohort and 20,759 persons were included in the post-law cohort. All these offenders had been convicted of at least one drink-driving offence in 1997 or 1999, respectively. Another 697 offenders were excluded from the post-law cohort since they had previously been convicted of a drink-driving offence in Table 5.9 presents descriptive data on the characteristics of offenders included in each cohort. These data show that the 1997 offender group, that is those convicted under the old penalty system, were more likely to be male, Aboriginal, 53 reside outside of the Sydney metropolitan area, be convicted of a high-range PCA offence, have a previous conviction for a drink-driving offence, as well as previous convictions for other offences, and have concurrent offences finalised at the time of conviction, in comparison to their 1999 counterparts. These differences in offender characteristics across the two cohorts underline the need to take into account preexisting differences between the groups when examining reoffending rates. 53 Care should be taken when comparing Aboriginality across cohorts since a greater proportion of offenders convicted in 1997 had missing information on this variable. 141

152 Table 5.9: Comparison of offender characteristics for drink-driving offenders included in the 1997 and the 1999 cohorts Offender characteristics p-value 54 n % n % Gender Male 15, , * Female 2, , Aboriginality Non-Aboriginal 13, , <0.001* Aboriginal 1, , Unknown 3, , Age <25 5, , , , , , , , , , , , Unknown Area of residence Sydney metropolitan area 8, , <0.001* Rest of NSW 8, , Other Offence type High PCA 5, , <0.001* Other 12, , PCA convictions in previous 2 yrs 0 17, , <0.001* Convictions for any other offence in previous 2 yrs 0 15, , <0.001* 1 1, , Concurrent offences 0 13, , <0.001* 1 2, , , , TOTAL 18, , *Significant at the 0.05 level 54 Gender (x 2 =8.95, df=1); Aboriginality (x 2 =45.25, df=1); Age (x 2 =4.10, df=5); Area of residence (x 2 =117.85, df=2); Offence type (x 2 =95.26, df=1); PCA convictions in previous 2yrs (x 2 =118.78, df=2); Convictions for any other offence in previous 2yrs (x 2 =23.70, df=3); Concurrent offences (x 2 =56.107, df=2). 55 This category includes offenders who, at the time of conviction, were living interstate (92.5%), in a State institution (2.7%), overseas (1.3%) or who had no fixed abode (1.3%). It also included a small number of offenders for whom there was no information available on the area of residence (2.1%). 142

153 Probability of reappearing for a drink-driving offence The first measure of offending considered in this analysis is the probability of reappearing in the Local Court for a drink-driving offence within three years of the initial drink-driving conviction. The percentage of drink-drivers appearing before the courts for a new drink-driving offence by area of residence 56 and year of conviction is shown in Table The most notable feature of this table is that the vast majority of drink drivers, regardless of where they lived or when they were convicted, did not reappear for a drink-driving offence during the follow-up period. In fact, about 90 per cent of persons in all four groups had no new appearances for drink-driving within three years of the reference offence. Table 5.10: Number and percentage of drink-driving offenders who reappeared for a drink-driving offence during the 3-year follow-up by offender group and area of residence Sydney Non-Sydney New PCA n % n % n % n % appearance No 7, , , , Yes Total 8, , , , From Table 5.10 it can also be seen that there is little difference between Sydney offenders convicted in 1997 and Sydney offenders convicted in 1999, in terms of the percentage who reappeared for a drink-driving offence within three years of conviction. The difference between 1997 and 1999 in the proportion reappearing was slightly greater for offenders residing elsewhere in New South Wales but statistical tests showed this difference to be non-significant. 57 However, as described previously, the differences in offender characteristics apparent between the two offender cohorts mean that a direct comparison of these percentages could be biased. Thus logistic regression techniques were applied to compare adjusted differences in reoffending rates across groups. 56 Persons residing interstate, in State institutions, overseas or who had no fixed abode were excluded from the analyses. 57 Sydney offenders (x 2 =0.079, df=1, p=0.783); non-sydney offenders (x 2 =0.922, df=1, p=0.345). 143

154 The main effects model from the logistic regression analysis is shown in Table As seen in this table, offender group, area of residence, gender, age, Aboriginality, prior convictions for non drink-driving offences 58 and concurrent offences were predictive of reappearing for a drink-driving offence within the three-year follow-up period. The type of drink-driving offence (high-range v. other) and prior convictions for a drink-driving offence were not independently predictive of reappearing and therefore these variables were excluded from the final model. Table 5.11: Odds ratios for risk of appearing for a new drink-driving offence Main effects model Variable 95% Confidence p-value Odds ratio Interval s lower upper Offender group 1999 v < Area of residence Non-Sydney residence v. Sydney < Gender Female v. male < Aboriginality Aboriginal v. non-aboriginal < Age v. < < v. < < v. < < v. < < v. < < Convictions for any other offence in previous 2 yrs 1 (other) prior v. none < (other) priors v. none < (other) priors v. none < Concurrent offences 1 v. none v. none < As described previously, the marginal increase in the length of licence disqualification periods after the implementation of the sentencing policy was found to be greater for offenders residing outside of the Sydney metropolitan area than was the case for their city counterparts. In order to examine whether the deterrent effect 58 This variable excluded prior convictions for a PCA offence to avoid multicollinearity problems. 144

155 of the penalty changes was greater for drink-drivers residing from non-sydney areas an interaction term for year of conviction by area of residence was included in the logistic regression model. 59 This analysis found evidence for a significant interaction between these two variables (p=0.037) after controlling for other confounding factors. Table 5.12 shows the odds ratios for this interaction effect adjusted for the covariates presented in Table For offenders residing in the Sydney metropolitan area there was no significant difference in the odds of reappearing for a new drinkdriving offence when comparing drink-drivers convicted in 1997 with those convicted in However, the odds of reappearing did decrease for the non- Sydney drink-drivers who received the more severe penalties. The odds for non- Sydney drink-drivers convicted in 1997 were 23 per cent higher than Sydney drinkdrivers convicted in the same year, but by 1999, Non-Sydney drink-drivers had the same odds of reappearing as the Sydney drink-drivers convicted in Table 5.12: Adjusted odds ratios for risk of appearing for a new drink-driving offence Interaction model Variable Odds ratio 95% Confidence Interval s lower upper Sydney 1999 v. Sydney Non-Sydney 1997 v. Sydney Non-Sydney 1999 v. Sydney These differences across offender cohorts can be more clearly understood by comparing the probability of reappearing for a new drink-driving offence. In order to calculate probabilities 60 from the logistic regression output a base model for comparison needs to be specified. In this instance a male, non-aboriginal offender, who is aged less than 25 years and has no prior convictions or any concurrent offences, was chosen. Figure 5.9 displays the probability of reappearing before the court for a new drink-driving offence, given this set of offender characteristics, by 59 The Hosmer and Lemeshow Goodness-of-fit test statistic for this interaction model was not significant at the 0.05 level (chi-square =11.747; p=0.163). This means we retain the null hypothesis that there is no difference between the observed and predicted values of the dependent variable and, thus, can conclude that the model s estimates fit the data at an acceptable level. 60 Probabilities were estimated from the regression model using the following equation; p ln = α 1 p + β 1 X 1 + β 2 X 2 + β 3 X β k X k, where p is the probability of reappearing for a new drinkdriving offence and X k are the predictor variables included in the regression model. 145

156 area of residence and year of conviction. As seen from this figure the probability of a young, male, non-aboriginal, first offender from outside of the Sydney metropolitan area reappearing for a new drink-driving offence was three percentage points less after the penalties were raised than before (14 per cent in 1997 v. 11 per cent in 1999). This equates to a reduction of 21 per cent in the probability of reappearing for a drink-driving offence after the statutory penalties were increased. In contrast, there was no change in the probability of non-sydney drink-driver reoffending given this set of offender characteristics. Figure 5.9: Probability of reappearing for a new drinkdriving offence by offender residence and year of conviction; male, non-aboriginal, aged <25, with no prior or concurrent offences Probability Sydney Non-Sydney Time to first drink-driving offence Time to first reappearance for a drink-driving offence was the second measure of reoffending used in the recidivism analysis. Survival time for non-recidivists was calculated as the number of days from the finalisation date of the reference drinkdriving offence until the cut-off date specified for the relevant offender groups (i.e. December for the 1997 cohort and December for the 1999 cohort). In the case of recidivists, survival time was the number of days from the finalisation date of the reference drink-driving offence until court finalisation of the new drink- 146

157 driving offence. Calculating survival time in this way gives rise to the following mean times 61 to first offence for Sydney and non-sydney drink-drivers: Sydney drink-drivers convicted in 1997: 1,370 days Sydney drink-drivers convicted in 1999: 1,370 days Non-Sydney drink-drivers convicted in 1997: 1,357 days Non-Sydney drink-drivers convicted in 1999: 1,361 days Log-rank tests of equality for the survival functions of the two offender cohorts (i.e and 1999) showed no significant differences in the time to first drink-driving reappearance for both drink-drivers residing in the Sydney metropolitan area and offenders residing elsewhere in New South Wales. 62 At this point, other explanatory variables related to offending have not been controlled for. In order to take account of pre-existing differences between offender groups, a Cox proportional hazards model was used to fit data on time to first drinkdriving reappearance. The main effects model, as shown in Table 5.13, found significant effects for offender group, area of residence, gender, Aboriginality, age, prior convictions for a non drink-driving offence, 63 prior convictions for a drinkdriving offence and concurrent offences. Current conviction for a high-range PCA offence was not significant and was therefore excluded from the final model. 61 Median survival times could not be calculated because more then 50% of offenders in each group had not reappeared for a drink-driving offence by the end of the follow-up period. 62 Sydney drink-drivers (log-rank: p=0.744); non-sydney drink-drivers (log-rank: p=0.477). 63 Graphing the hazard function for the previous convictions category indicated that the hazard function for the 2 and 3+ categories crossed, suggesting that the proportional hazards assumption is not met for this predictor variable. The 2 and 3+ categories were therefore combined for the Cox regression analyses. 147

158 Table 5.13: Hazard ratios for time to new appearance for a drink-driving offence Main effects model Variable 95% Confidence p value Hazard ratio Interval s lower upper Offender group 1999 v Area of residence Non-Sydney residence v. Sydney < Gender Female v. male < Aboriginality Aboriginal v. non-aboriginal < Age v. < < v. < < v. < < v. < < v. < < PCA convictions in previous 2 yrs Prior PCA v. none Convictions for any other offence in previous 2 yrs 1 (other) prior v. none < (other) priors v. none < Concurrent offences 1 v. none v. none < Once again an interaction term was included in the model to examine the differential effect of the sentencing policy across different areas of New South Wales. This analysis found a significant interaction effect between offender group and area of residence (p=0.011). Table 5.14 shows the hazard ratios for this interaction effect after adjusting for the covariates presented in Table These hazard ratios indicate that, controlling for other confounding variables, non-sydney drink-drivers had a decreased risk of reappearing for a new drink-driving offence after the penalties were raised but the reappearance rate for Sydney drink-drivers remained the same. Before the penalties were raised, non-sydney offenders had a reappearance rate that was 21 per cent higher than Sydney drink-drivers but after the penalty changes there was no significant difference between non-sydney and Sydney offenders. 148

159 Table 5.14: Adjusted Hazard ratios for time to new appearance for a drink-driving offence Interaction model Variable Hazard ratio 95% Confidence Interval s lower upper Sydney 1999 v. Sydney Non-Sydney 1997 v. Sydney Non-Sydney 1999 v. Sydney Figures 5.10 and 5.11 display the survival curves, adjusted for the explanatory variables included in the Cox regression model, for Sydney and non-sydney offenders, respectively. 64 These adjusted survival curves show the proportion of each group who had not yet reappeared before court for a drink-driving offence plotted against the number of days in the follow-up period. Again base model needs to be specified to calculate the survival functions for each group. Our base model is a male, non-aboriginal drink-driver aged less than 25 years, who had no prior convictions or concurrent offences. As seen from these figures, offenders who reside outside of the Sydney metropolitan area and who were convicted of drink-driving in 1999 generally remained offence-free for longer than non-sydney drink-drivers convicted of an offence in In contrast, there was almost no difference in the time to first new drink-driving offence for Sydney drink-drivers convicted after the penalties were raised compared to those convicted prior to the legislative amendments. 64 The adjusted survival curves were estimated separately for Sydney and non-sydney offenders. However, the estimates derived from these new models, in comparison to the full interaction model reported in this Chapter, were almost equal. 149

160 Figure 5.10: Time to first offence, Sydney drink-drivers Proportion Days to first reappearance for drink-driving Figure 5.11: Time to first offence, non-sydney drinkdrivers Proportion Days to first reappearance for drink-driving A major assumption of the Cox proportional hazards model employed in this analysis is that the hazard ratio for one category is proportional to the hazard ratio for any other category of an explanatory variable. One method for checking whether this 150

161 assumption is satisfied is to plot the log-log survival curves 65 for the different categories of explanatory variables. Parallel survival curves for each category indicates that the proportional hazards assumption is satisfied and the estimated model can be considered valid (Kleinbaum 1996). This procedure was undertaken for each of the predictor variables appearing in the Cox regression models shown above. These curves were parallel for all predictor variables indicating the proportional hazards assumption is met for the predictors included in the final model. Summary and discussion of recidivism analysis The results from Study 2 provide some evidence for a beneficial effect of the 1998 sentencing policy on drink-driving recidivism. At least for offenders residing outside of the Sydney metropolitan area, drink-drivers sentenced before the statutory penalties were increased, had higher odds of reappearing for a new offence and reoffended sooner than drink-drivers sentenced after the penalties were raised. This effect was not apparent for Sydney drink-drivers. Furthermore, examination of the implementation of the sentencing policy, indicates that increases in court-imposed penalties for drink-driving offences coincided with the legislative changes and that negative effects on prosecution rates, conviction rates, guilty pleas and court delay were not apparent. Prima facie these results would suggest that harsher penalties can decrease rates of drink-driving reoffending, at least for one specific group of offenders. However, three other explanations for this effect need to be discounted before such a conclusion can be reached. The first of these is changes to drink-driving enforcement practice during the followup period. If the perceived risk of apprehension was greater after the penalties were increased then this could account for the observed reduction in recidivism. To examine this possibility, data on the number of breath tests administered to motorists over the period 1994 through 2001 were obtained from the NSW Police. These data showed that the number of breath tests conducted by police did increase in the later 65 The log-log survival curve is a transformation of an estimated survival curve that results from taking the natural log of an estimated survival probability twice. 151

162 years of the study period (i.e and 2002). However, the overall number of breath tests in the first year of follow-up for the 1999 cohort was still at a slightly lower level in comparison to the first year of follow-up for the 1997 cohort (1,961,448 v. 2,077,034 tests). Restricting the outcome used in the logistic regression analysis to any reappearance within the first 12 months after conviction (instead of three years) resulted in the same conclusion that non-sydney drink-drivers from the 1999 cohort had reduced odds of reoffending (note that this issue of enforcement of drink-driving offences is discussed in greater detail in the next Chapter). A second plausible explanation for the findings from Study 2 is that the reduction in reoffending observed for drink-driving offences simply reflects a secular downward trend in all offending rather than a specific deterrent effect created by the penalty increase. Additional logistic regression and survival analyses, comparing recidivism rates before and after 1998, were therefore conducted for two groups of non drinkdrivers: offenders convicted of a property crime and offenders convicted of a violent crime. 66 Property and violence offences were chosen as comparisons because these crime types were unaffected by the penalty changes implemented in 1998 and are less likely to be influenced by changes in police practice than other offences (e.g. drug offences). These analyses showed a significant reduction for violent offences in the odds of reappearing for a new offence within three years of conviction and also a significant increase, between 1997 and 1999, in the time it took this group to reoffend. However, unlike drink-driving offences, this reduction in reoffending did not vary by an offender s area of residence. No differences in reoffending were apparent for persons convicted of a property offence in 1997 compared to those convicted of a property offence in It could be argued that the observed increase in time to reoffend for the 1999 offender cohort stemmed not from a deterrent effect but rather an incapacitation effect. That is, it could be argued that drink-drivers convicted in 1999 took longer to reappear before the court for a new drink-driving offence than their Comparisons of reoffending rates for driving offenders could not be undertaken since statutory penalties for driving offences were increased at the same time as drink-driving penalties. For practical reasons, offenders who appeared before the Local Courts on every third day, and were convicted of a property or violence offence, were considered here. 152

163 counterparts because they were sentenced to longer custodial terms. This argument can be rejected, however, because the proportion of drink-drivers that were imprisoned for the reference offence was only one per cent in both cohorts and the average custodial sentence received was only around four and a half months. It is true that the actual free time that this small group of offenders had to appear before the courts for a new offence during the follow-up period was reduced after However, using proportional hazard models to fit data on time to first offence, excluding offenders who were imprisoned, confirmed the significant interaction effect between year of conviction and area of residence, controlling for other explanatory variables. The data from Study 2 therefore indicate that the changes to drink-driving penalties in New South Wales in 1998 were associated with a significant reduction in the rate of reoffending for offenders from country and regional areas. When considering the importance of these findings, however, it needs to be noted that the overall effect of the increased penalties on recidivism rates was relatively small. The probability of a drink-driver reoffending was reduced by just three percentage points in non-sydney locations. Given such a small effect size from what was essentially a doubling of the statutory penalties for all drink-driving offences, and keeping in mind the associated costs with administering the new penalty regime, the value of the tougher penalties strategy in controlling drink-driving remains questionable. It is possible, however, that Study 2 underestimated the overall impact of the 1998 sentencing policy on reoffending. In order to minimise the potential effect of other contextual variables on reoffending rates (e.g. changes in police activity), the two cohorts selected for comparison were relatively close in temporal proximity. This meant that some of the follow-up period for offenders included in the 1997 cohort extended beyond the introduction of the new policy. A general deterrent effect created by the penalty increase therefore had potential to influence the rate at which this group reoffended. Any bias relating to this aspect of the research design would have been towards the null hypothesis. Thus, the overall conclusions from the analysis remain valid, but it is still possible that the true magnitude of the group difference in reoffending was greater than the current research suggests. It is difficult 153

164 to ascertain the extent to which the announcement of the drink-driving penalty changes may have affected the reoffending rate of drink-drivers included in the 1997 cohort. However, as will be seen in the next chapter, the 1998 increase in drinkdriving penalties received very little publicity, therefore any underestimate of the policy s impact on recidivism in Study 2 is likely to be minimal. It is also possible that the impact of the 1998 legislation would have been greater if licence disqualifications were more systematically applied for drink-driving offences. While the legislative amendments had the effect of increasing the average licence disqualification for drink-driving offences across the State, 20 per cent of guilty offenders still escaped licence disqualification on being found guilty of a drinkdriving offence (via a Section 10 dismissal) despite the existence of these mandatory minima (for more details on trends in Section 10 dismissals in New South Wales see Moffatt, Fitzgerald & Weatherburn 2004). Ensuring that almost all offenders are recipients of a licence disqualification once found guilty for a drink-driving offence could have amplified the effects found here. Conclusions from the quasi-experimental studies To summarise so far, Study 1 found a significant increase in one of the surrogate measures used to index alcohol-related road crashes (single-vehicle night-time crashes) after the harsher penalties were introduced in New South Wales in September Further exploratory analyses suggested that this increase was mostly driven by a rise in non alcohol-related road crashes that occurred at or around the same time as the penalty changes. From these findings two possible conclusions about the sentencing policy could be drawn: (1) that there was a reduction in alcoholrelated crash rates after the penalty changes but that this reduction was overwhelmed by the rise in non alcohol-related crashes and therefore could not be detected in our surrogate measure or (2) there was no change in alcohol-related road crashes after the sentencing policy was implemented. Study 2 showed that average gaol terms, fines and licence disqualification periods for drink-driving offences significantly increased after the statutory penalties were raised in September 1998, with some variations in the magnitude of this rise across the 154

165 State. There was also little evidence for a negative effect of the sentencing changes on the prosecution of drink-driving offences. There was a small reduction in the proportion of drink-drivers who received a licence penalty after the legislative amendments were enacted. However the vast proportion of drink-drivers appearing before the courts in 1999 did have their licence disqualified and for longer than their 1997 counterparts. The results of the logistic regression and survival analyses found a significant reduction in the probability of drink-drivers reoffending and the time to first new drink-driving offence after these tougher penalties were enacted. This reduction in reoffending was small and specific to offenders residing in country and regional areas. In light of the findings from Study 2 it remains possible that there was a significant reduction in alcohol-related crashes after the penalty changes but, because of the overwhelming rise in non alcohol-related crashes around the same time, this decrease could not be detected in our analysis of single-vehicle night-time crash rates (Study 1). However, the results of Study 2 also suggest that if there were a reduction in alcohol-related crashes immediately after the policy was implemented, the magnitude of this effect would have been relatively small. These findings indicate that the doubling of statutory penalties for drink-driving offences in New South Wales in 1998 and the subsequent increase in average court penalties for these offences resulted in only a small reduction in offending. The question then becomes why didn t the policy have a greater impact on drink-driving rates in New South Wales? Does punishment severity matter little in offender decision-making or are people who have previously been sanctioned for an offence more sensitive to changes in punishment variables? Before the latter questions can be addressed, we need to investigate other factors that may have affected the deterrent efficacy of the penalty changes within this real-life setting. Therefore, in the next Chapter we will examine the context in which the 1998 sentencing policy was implemented. 155

166 CHAPTER 6. CONTEXT OF POLICY IMPLEMENTATION Chapter 3 identified several reasons why a sentencing policy that raises punishment severity may have minimal influence on offending rates. These include: (1) the policy failing to alter actual sentencing practice (2) the penalty increase having countervailing effects on the prosecution and conviction of the targeted offences (3) the perceived risk of apprehension being relatively low in the jurisdiction where the new law is implemented and (4) the target audience being unaware that formal punishment for the offence has changed. The data provided in Chapter 5 demonstrated that the magnitude of drink-driving penalties increased significantly after the statutory amendments were enacted in On the whole, there was also no indication that the penalty changes had negative, countervailing effects on the prosecution and conviction of drink-driving offences. There was some evidence that a proportion of drivers escaped the mandatory minimum licence disqualifications prescribed by the new laws, which could have contributed somewhat to the diminished effect of the policy on reoffending. However, the vast majority of convicted drink-drivers were recipients of more severe penalties after the legislation was implemented and yet there was still only a small observed effect of these changes on offending levels. Two other reasons for the failure of the 1998 sentencing policy to exert a greater deterrent effect on drink-driving remain. Firstly, the perceived risk of apprehension may have altered over the course of the study period. Henstridge and his colleagues (1997) have shown a direct relationship between the number of breath tests that police undertake and subsequent reductions in alcohol-related road crashes. If the NSW Police altered their enforcement activity at or around the time the new penalties came into operation then this could have undermined the intended deterrent effect of the penalty changes. Secondly, it is possible that many drink-drivers failed to change their offending behaviour because they were simply unaware that drinkdriving penalties had increased. These two accounts, for the relatively small effect 156

167 observed in the quasi-experiment, are explored in greater detail in the following sections. Enforcement of drink-driving offences As described in Chapter 2, deterrence theory anticipates that the severity of punishment matters more when punishment is highly probable. Given the relatively high level of drink-driving enforcement in New South Wales, through RBT, it was thought that raising the severity of drink-driving penalties should have a greater effect in this State than would be the case for many other jurisdictions. It is therefore essential to consider any variations in the level of RBT at or around the time the penalties were increased in New South Wales in 1998 before any definitive conclusions can be reached about the effectiveness of punishment severity. To investigate this issue further the number of breath tests administered by NSW Police and the number of police-recorded drink-driving incidents (or charges), were examined over the eight-year period covered by the quasi-experimental study. 67 Data for 2002 are also included in this analysis, given that the follow-up period in which reoffending was measured for the post-law cohort fell within this year. Drink-driving charges are included here as an indicator of enforcement activity, rather than as a measure of offending because, as discussed in Chapter 4, the actual probability of arrest for drink-driving is extremely low. Figure 6.1 displays both the number of breath tests administered by NSW Police each month from April 1994 through December 2002 and the number of drinkdriving charges recorded from January 1995 through December As seen in this figure, these two indicators appear to follow a similar pattern during the earlier years of the study period but from 1998 the series begin to diverge, with the number of breath tests dropping as drink-driving charges begin to rise. Interrupted time-series analyses (using a regression approach) were conducted for each of these data series, 67 Monthly breath tests administered by NSW Police between April 1994 and December 2002 were examined. However, the COPS (Computerised Operational Police System) database, used by NSW police to record drink-driving incidents, was only introduced in April 1994 and there is some concern about data reliability for the first few months following the statewide rollout. Therefore, only drinkdriving incidents recorded during the period 1995 through 2002 are considered here. The NSW Police supplied breath test data and police charge data were obtained from the NSW Bureau of Crime Statistics and Research (see NSW Recorded Crime Statistics 1998 to 2002). 157

168 with the intervention point corresponding to the implementation of the new penalties. 68 For breath tests, this analysis showed a significant decreasing trend prior to the penalty change, followed by a drop in breath tests that coincided with the intervention point and then a significant change in the underlying trend, as the number of breath tests began to increase from October In contrast, there was a small, but significant, upward trend in the number of drink-driving charges before the penalty change, then a significant rise in charges at the time the penalties were raised but no change in the underlying upward trend thereafter. In summary, at the time drink-driving penalties were increased in New South Wales, there were fewer overall breath tests being conducted by police but there were more people being caught for drink-driving offences. 69 Figure 6.1: Monthly police breath tests and recorded drink-driving incidents, NSW, Apr1994-Dec2002 Number of incidents Oct-02 Apr-02 Oct-01 Apr-01 Oct-00 Apr-00 Oct-99 Apr-99 Oct-98 Apr-98 Oct-97 Apr-97 Oct-96 Apr-96 Oct-95 Apr-95 Oct-94 Apr-94 Drink-driving incidents Breath tests Number of breath tests 68 A regression approach was adopted for the interrupted time-series analyses, after a Dickey-Fuller test confirmed that both series were trend stationary. The intervention point was October 1998, the first full month following the enactment of the Traffic Amendment (Penalties and Disqualifications) Bill. Eleven dummy variables corresponding to the months were also included in the models to control for seasonal autocorrelation in the data. Diagnostic checks of the residuals suggested that the models contained no significant autocorrelation. The final models are included in Appendix III. 69 Note that multivariate modelling of the relationship between breath testing rates and alcohol-related crashes was not possible due to the variability in the breath testing data and the lack of data points in both series to enable the use of quarterly data. 158

169 At first glance these results would seem somewhat incongruous, given both breath tests and drink-driving charges represent levels of police enforcement. However, what appears to have happened over this time period (and this was verbally confirmed by Commander Ron Sorrenson, head of Traffic Services at the NSW Police when data on breath tests were sought for this study) is that the police switched their enforcement strategy from one of general deterrence, with stationary RBTs in highly visible locations, to the targeting of high-risk times and locations (e.g. weekend nights and major roads) for drink-driving. Thus, while they were not conducting as many breath tests post September 1998, the police were in the right places, at the right times, and were therefore catching more offenders. The probable impetus for this switch in tactics (and the subsequent increase in drinkdriving charges) was the launch of a NSW Government initiative, known as the Enhanced Enforcement Campaign. This was a statewide program, managed by the RTA, which funded police Local Area Commands (LACs) that wished to enhance the enforcement operations in their local area (including those targeting drinkdrivers). The program commenced operation in NSW in March 1998, and by the middle of 1998 many applications for funding had been received and approved by the RTA (Barry Wirrick, Enhanced Enforcement Program Manager, RTA; personal communication 2003). LACs that were successfully granted funds used these monies to set up random breath testing sites late at night/early in the morning and during holiday periods when a greater proportion of drink-driving occurs. Because raising the perceived certainty of apprehension amongst drink-drivers is associated with reductions in offending (Voas & Hause 1987; Henstridge, Homel & Mackay 1997; Voas, Holder & Gruenewald 1997), the increase in drink-driving charges and the decrease in breath tests had the potential to affect offending behaviour, albeit in opposing directions. It is possible that the reduced level of police breath testing from 1997 onwards had a deleterious effect on the perceptions of arrest risk held by the general driving population and that this weakened the deterrent effect of the 1998 sanctioning policy. It is also possible, however, that the targeting of highrisk times and locations increased the perceived risk of apprehension, particularly 159

170 amongst frequent offenders, and thus contributed somewhat to the decrease in reoffending apparent in Study 2. Police breath tests and public perceptions of enforcement activity There is good reason to believe that the drop in total breath tests administered by NSW Police compromised the deterrent effect of the 1998 penalty changes. Previous research conducted in New South Wales has demonstrated a direct relationship between the overall number of breath tests conducted by police and subsequent reductions in alcohol-related road crashes (Henstridge, Homel & Mackay 1997). Thus, even modest reductions in breath tests during the study period could have reduced perceptions of apprehension risk. Due to the retrospective nature of the quasi-experimental design, perceptions of apprehension risk before and after the drink-driving penalties were increased could not be assessed. However, data bearing upon this issue was available from an annual, national survey conducted by the Australian Transport Safety Bureau (ATSB). The ATSB Community Attitudes Survey is a telephone survey, of Australian residents aged 15 years and over, which measures respondents attitudes toward a range of road safety issues (see Mitchell-Taverner 1994; 1995; 1996; 1997; 1998; 1999; 2000; 2002a; 2002b). Importantly, for this thesis, the survey asks people in all Australian States about their exposure to random breath tests and their perceptions of RBT activity. Figures 6.2 and 6.3 present NSW data, from this survey, for each year between 1994 and Shown in Figure 6.2 is the proportion of NSW respondents that thought RBT activity had increased, and the proportion that thought RBT activity had decreased, in the previous two years. In Figure 6.3, is the proportion of the NSW sample that reported seeing RBT in operation in the last six months and the proportion reporting having been personally breath tested during this same period. In both figures the number of breath tests administered by NSW Police each year during the study period is also plotted for comparative purposes. 160

171 Percentage Figure 6.2: Yearly breath tests and perceptions of enforcement activity, NSW, Increased Decreased Total yearly breath tests No. breath tests Percentage Figure 6.3: Yearly breath tests by reported experience with RBT, NSW, Seen in operation Tested Total yearly breath tests no. breath tests The NSW sample included in this survey is relatively small (n equals approximately 250 for the years shown here) and, therefore, these data are quite variable across the study period. Despite this variability, there does appear to be some congruency 161

172 between perceptions of drink-driving enforcement reported in the survey each year and trends in the number of breath tests recorded by police. When comparing these data, it should be remembered that the survey inquired about the level of RBT activity occurring during the previous two years or exposure to RBT in the previous six months. Thus, we would expect to see a lag between breath tests administered by police and perceptions recorded in the survey. Accordingly, from Figure 6.2 it can be seen that the peak in breath tests evident in 1996, corresponds to fewer people reporting a decrease, and more people reporting an increase, in police activity in As the number of breath tests begins to decline after 1996, the proportion perceiving a decrease in RBT activity commences to rise each year, reaching a peak of 21 per cent in The relationship between trends in experiences with RBT activity and trends in police breath tests is a little less clear. Figure 6.3 shows a peak in 1997 (25 per cent) in the proportion reporting having been breath tested, as well as a peak in the same year (79 per cent) in the proportion reporting that they had seen RBT in operation, which closely corresponds to the peak in breath tests administered by NSW Police in Then in the 1998 survey, we see a drop in RBT experience, and RBT exposure, which is consistent with the decreasing trend in the number of breath tests after However, reported experience with RBT testing jumps up in 1999 but drops off in the remaining years of the study period. Meanwhile the proportion exposed to RBT steadily increases from 1998 until Overall, these data clearly show that after the statutory penalties were raised in 1998, more NSW residents perceived that RBT activity had decreased and fewer people had seen RBT in operation. In 1997, 12 per cent of the respondents surveyed thought that RBT activity in New South Wales had decreased, but by 1999 this percentage had risen to 20 per cent, or one in five. Similarly, in 1997, 79 per cent of respondents reported having seen RBT in operation, but by 1999 only 67 per cent reported exposure to RBT operations in New South Wales. Given strong evidence for a relationship between drink-driving enforcement levels and subsequent reductions in alcohol-related crashes, this decline in exposure to police activity raises some concern about the level of perceived certainty of apprehension at the time the 162

173 sentencing legislation was implemented. Reductions in the intensity of drink-driving enforcement may also have contributed somewhat to the increase observed in non alcohol-related crashes after the intervention in Although RBT is specifically designed to increase the perceived risk of apprehension for drink-driving offences, the fact that it involves many police in highly visible locations means that it may also have an ancillary effect in deterring (or not) other traffic offending, particularly drug driving and speeding. Police targeting of high-risk times and locations While the reduction in the number of breath tests may have diminished the general deterrent effect of the statutory changes, it might be thought that the hotspotting techniques adopted by police contributed to the reduction in drink-driving recidivism observed after the penalty changes. Upon closer examination of drink-driving charges across different areas of New South Wales, this explanation for the findings from Study 2 appears unlikely. Despite being a statewide campaign, the enhanced enforcement program was taken up more readily by some LACs than by others (Barry Wirrick, Enhanced Enforcement Program Manager, RTA; personal communication 2003). This disparity across areas is clearly shown in Table 6.1, which presents a breakdown of the number of drink-driving charges by Statistical Division, from 1997 through As shown here, all Statistical Divisions in New South Wales experienced an increase in drink-driving charges over this six-year period. The areas that experienced the greatest increase in charges during this time were located in the Sydney Metropolitan region. Yet the reduction in reoffending, observed in Study 2, was found only for drink-drivers residing in country and regional areas. 163

174 Table 6.1: Number of police recorded drink-driving incidents and percentage change from 1997 to 2002 by Statistical Division where incident occurred, New South Wales. Statistical Division % change Sydney 7,559 9,279 12,099 12,126 12,009 12,795 69% Inner Sydney ,202 1,374 1,459 1, % Eastern Suburbs % St George- Sutherland ,058 1,140 1,313 93% Canterbury % Bankstown Fairfield Liverpool , , % Outer South West % Sydney Inner Western Sydney % Central Western % Sydney Outer Western Sydney ,216 1,416 1, % Blacktown Baulk , % Hills Lower Northern , , % Sydney Hornsby Kuring-gai % Northern Beaches ,089 46% Gosford Wyong ,167 1,121 1,212 56% Hunter 1,797 1,674 1,929 2,303 2,742 3,000 67% Illawarra 891 1,043 1,143 1,150 1,166 1,402 57% Richmond Tweed 967 1,024 1,148 1,270 1,383 1,415 46% Mid-North Coast 1,078 1,171 1,286 1,257 1,309 1,163 8% Northern % North Western % Central West % South Eastern ,204 1,127 1,225 1,201 37% Murrumbidgee % Murray % Far West % NSW TOTAL 16,507 18,633 23,028 23,557 23,933 25,087 52% Source: NSW Bureau of Crime Statistics and Research, NSW Recorded Crime Statistics 1998 to It is true that many country and regional areas also experienced noticeable increases in drink-driving charges after 1998 and is conceivable that police targeting of highrisk times and locations in these areas had a greater marginal deterrent effect on offending than in the Sydney Metropolitan areas. Further data would be needed on perceptions of enforcement activity across different areas of New South Wales and the localities where breath tests were administered to investigate this possibility. This 164

175 information, however, is not available. Nor is there any research on the relative effectiveness of hotspotting in deterring different types of drink-drivers (i.e. city versus country). It is therefore difficult to assess the extent to which the switch in enforcement tactics, by the NSW Police, may have contributed to the reduction in drink-driving recidivism observed in Study 2. All that can be said is that changes in perceived certainty of drink-driving punishment cannot entirely be ruled out as a potential explanation for our findings. Media coverage of statutory penalty increase A central question in this thesis is whether the 1998 increase in statutory penalties for drink-driving offences, which occurred in New South Wales, deterred offending by increasing perceptions of sanction severity. We should therefore not only examine changes in drink-driving behaviour but also assess people s beliefs about the severity of drink-driving punishment before and after the implementation of the new laws. Unfortunately, the current investigation, being a retrospective study of the legislative changes, does not permit this analysis. As an alternative we examine media coverage of drink-driving and related issues, at and around the time of the introduction of the new policy, in order to gain insight into the extent to which the increased threat was adequately communicated to the target audience. Newspaper coverage of penalty changes To examine the extent of print media covering the 1998 rise in drink-driving penalties, an interactive media database, Factiva, was used to gather any relevant newspaper articles published in National or NSW Metropolitan newspapers, between June and June Factiva combines media sources from both the Dow Jones Interactive (Fairfax holdings) and Reuters Business Briefing databases and provides full text access to news from more than 8,000 international sources. The publications of most relevance to the current analysis (and which are included in this database) are The Sydney Morning Herald, The Sun Herald, The Australian, the Australian Financial Review, The Daily Telegraph, the Sunday Telegraph, the Newcastle Herald and the Illawarra Mercury. 165

176 NSW country newspapers could not be accessed because the information gathered from this historical search was limited to the sources contained in the Factiva database. Although this necessarily restricts the analysis, the National newspapers and daily Sydney Metropolitan newspapers that are included here, (in particular The Australian, Sydney Morning Herald and Daily Telegraph) are also widely distributed throughout New South Wales and therefore would be accessed by many living in country and rural areas. Furthermore, while the quantity of drink-driving articles may possibly be greater in country newspapers, there is no reason to suspect that the content would differ to any considerable degree. This database search identified a total of 448 newspaper articles that included the two keywords of drink driving and NSW. Only 139 of these specifically referred to issues associated with the offence of drink-driving in New South Wales. Figure 6.4 displays the monthly distribution of these drink-driving articles across the two-year period examined. Also shown in this figure is the month in which drink-driving penalties were raised in New South Wales. It can be seen from this figure that in 1997 and 1998, drink-driving articles appeared most frequently in June and December, which corresponds to the Christmas/New Year holiday period and the Queen s birthday holiday in June. There was also a peak in newspaper articles in the month immediately following the implementation of the statutory amendments to drink-driving penalties (October 1998), which was not evident for the same month in the previous year. 166

177 Figure 6.4: Number of drink-driving articles appearing in NSW metropolitan papers each month, Jun1997- Jun1999 Number of articles Penalties increase Jun-97 Aug-97 Oct-97 Dec-97 Feb-98 Apr-98 Jun-98 Aug-98 Oct-98 Dec-98 Feb-99 Apr-99 Jun-99 A coding system was developed to extract further relevant information from these 139 articles. This coding system was similar to that used by Cashmore (1985), in her analysis of the impact of RBT in New South Wales, and essentially consisted of three categories of information. The first category concerned the physical characteristics of the article and included data such as the name of the newspaper, page number, size of the article (in number of words) and the headline. The second category referred to sources cited in the article, such as police, government officials/departments and non-government organisations. The third category concerned the type of issues discussed in the article: (1) Description of drink-driving penalties (2) Reference to drink-driving penalty increase in 1998 (3) Description of Random Breath Testing operations (4) Other issues related to drink-driving and road safety. The coding also indicated whether these topics were the main or secondary issue of the article and whether the article contained an overall evaluative element (i.e. whether negative or positive). As can be seen in Table 6.2, the majority of articles discussing drink-driving offences over the period examined appeared in National and Sydney Metropolitan papers, with almost 40 per cent published in either The Daily Telegraph or Sunday 167

178 Telegraph but a substantial minority came from Wollongong and Newcastle newspapers, which service much smaller populations. Table 6.2 also shows that the NSW Police was most often the source for information on drink-driving and related issues discussed in these articles, with the courts and the NSW Government being the next most frequent information source. The size of the drink-driving articles published ranged from 64 to 1,985 words in length, with a median of 346 words. Approximately one-third of the 139 drink-driving articles gathered in the search appeared on the first to the third page of the newspaper (only seven were front-page articles). Table 6.2: Newspaper outlets and sources used in drink-driving articles, June 1997 June 1999 No. % Newspaper The Australian Sydney Morning Herald Daily Telegraph Sun Herald Sunday Telegraph Sunday Age Newcastle Herald Illawarra Mercury Total Source Police Courts Government officials/departments Political parties Staysafe parliamentary committee Non-government associations Other Total The number of references to each of the four content areas into which the articles were collapsed is shown in Table 6.3. Also indicated in this table is whether each content area was discussed as a main or secondary topic of the article. As shown here, only 15 articles describing the penalty increase for drink-driving offences in 1998 were published in NSW Metropolitan newspapers over the two years examined. The penalty changes were the main topic discussed in just seven of these. Most of the articles referring to the new laws were published either at the time the statutory 168

179 amendments came into effect or in the two weeks following its enactment (i.e. September October ) but two coincided with the announcement of the penalty changes by the State Government in June A further 51 articles discussed drink-driving penalties in New South Wales in more general terms, usually with reference to a specific drink-driving case prosecuted in the courts. However, the majority of drink-driving articles concerned RBT operations or road crashes arising from episodes of drink-driving. Table 6.3: Type of reference by content area, June 1997 June 1999 Main Secondary Total Content Area n % n % n % Drink-driving penalties Drink-driving penalty increase in 1998 RBT operations Other issues Table 6.4: Nature of evaluation by content area, June 1997 June 1999 Negative Positive Total Content Area n % n % n % Drink-driving penalties Drink-driving penalty increase in 1998 RBT operations Other issues If the articles identified in the database search contained an evaluative component they were also coded as negative or positive depending on the general view expressed in relation to the content area discussed (see Table 6.4). As seen from Table 6.4, many of the articles that discussed drink-driving penalties during this period expressed a negative view of the legal implications for these offences. Often these newspaper articles implied that drink-drivers once apprehended frequently escape the severe penalties that accompany this offence, conveying the impression that the courts are soft on these types of crimes (e.g. Soft pca sentences slammed by Collins, Newcastle Herald, 27 th June 1998). This was particularly true around the middle of 1998, only a few weeks after the State Government announced their intentions to increase drink-driving penalties. 169

180 During this period, a former NSW Minister for Corrective Services was charged with a high-range drink-driving offence. Instead of receiving the harsh penalties that were prescribed for this offence, the former politician escaped a criminal conviction (via the Section 10 equivalent available to magistrates at that time) and was given a $500, 12-month good behaviour bond (The Sydney Morning Herald, 18 th June 1998; The Sydney Morning Herald, 19 th June 1998). Around the same time, the media also highlighted a similar case, where a police officer charged with high-range drinkdriving escaped conviction (The Daily Telegraph, 24 th June 1998; Newcastle Herald, 25 th June 1998). Yet another occurred in the latter part of 1998, in which a Supreme Court judge escaped conviction after pleading guilty to driving with a BAC that was twice the limit (The Daily Telegraph, 26 th November 1998; Illawarra Mercury, 27 th November 1998; The Daily Telegraph 27 th November 1998). Several articles suggested that these cases illustrated the preferential treatment in sentencing accorded to high-profile and wealthy people found guilty of a drinkdriving offence: Three times the limit, but Yapsley keeps licence (The Sydney Morning Herald, 18 th June 1998), When justice is no longer (The Daily Telegraph, 23 rd June 1998), Celebrity drunks get off lightly / One law for the rich and one for the rest of us (Sunday Telegraph, 28 th June 1998) and Penalties should be mandatory (Sunday Telegraph, 28 th June 1998). This negative view conveyed in relation to drink-driving penalties could have reduced the deterrent effect of the 1998 sentencing reforms. As noted in Chapter 2, an individual s normative commitment to legal authorities is a non-instrumental determinant of legal compliance (Tyler 1990). The impression conveyed by the media during 1998 that drink-driving laws are not equally enforced amongst all members of the driver population could have reduced the perceived legitimacy of authority and, consequently, resulted in fewer drivers being willing to modify their behaviour in accordance with the law. These newspaper articles also highlight a court s ability to dismiss a drink-driving offence without a conviction if the person is considered to be of good character. Knowledge of this possible exemption from criminal penalties could also serve to reduce the perceived certainty of conviction for drink-driving in New South Wales. If this were the case, the negative publicity would have substantially undermined the deterrent effect of the severe penalties for drink-driving. 170

181 There was also a general tendency by the media to view the 1998 penalty changes in a negative light. The implementation of the statutory amendments coincided with the Labour Day holiday weekend in New South Wales and, over this holiday weekend, NSW Police reported an increase in charges for drink-driving and charges for other major traffic offences. Several newspapers took this as evidence that the more severe penalties introduced by the State Government in September had had little impact on driver behaviour: Demerits damned (The Australian, 6 th October 1998), Tougher laws no deterrent (Illawarra Mercury, 6 th October 1998) and Shocking behaviour on roads (Newcastle Herald, 6 th October 1998). A further point to note here is that the few articles that did describe the penalty changes for drink-driving tended to focus on penalty changes for other driving offences, such as speeding. For example, headlines used to describe the 1998 penalty changes included Big jump in fines for traffic offences (The Sydney Morning Herald, 5 th June 1998), Serious speeders to lose licences (The Daily Telegraph, 5 th June 1998) and Speeding points doubled (The Daily Telegraph, 29 th September, 1998) and similarly, the headline used to discuss the impact of these more severe penalties: Double demerits damned (The Australian, 6 th October, 1998). Many drink-driving strategies that have attempted to modify behaviour have stressed the criminal nature of the act of drink-driving and the criminal conviction that could result if apprehended. Yet the media consistently packaged drink-driving along with other non-criminal driving offences. This may lead many people to view drinkdriving as just another traffic offence which results in nothing more than administrative licence sanctions or monetary penalties. RTA advertising campaigns and other publicity The small amount of print media devoted to the drink-driving penalty changes was reflected in the attention the legislative amendments received elsewhere in the media, in particular on radio. Data pertaining to radio coverage of the sentencing policy were obtained from an on-line transcript library maintained by the Australian media company Rehame. The information that could be gleaned from this database was extremely limited. The records that needed to be accessed were over five years old 171

182 and for this reason transcripts were not available from the media agency. However, summaries of announcements were provided, as were details of the time, date and station on which the announcement was made. Given the restricted information that was available, the search was limited to records between January 1998 and March 1999 and drink-driving was used as the keyword. This search returned a total of 49 references, only four of which appeared from the summary to be specifically discussing the 1998 penalty changes in New South Wales. Relevant details for these four radio announcements are shown in Table 6.5. Although little information was available on the content of the announcement, the few records retrieved in the search are indicative of how little publicity the 1998 sentencing policy received in the NSW media. Table 6.5: Summaries of radio references to 1998 penalty increase, January 1998 March 1999 Date Time Station Announcer Summary 05/05/1998 4:00pm Sydney 2GB 05/05/1998 5:13pm Sydney 2UE 06/05/1998 5:44am Sydney 2UE 06/05/1998 7:08am Sydney 2UE Radio news Mike Carlton Alan Jones Alan Jones The opposition has accused the government of revenue raising by lifting the penalties for speeding and drink-driving. Interview with Shadow Minister for Roads, George Souris. Discussion on tough new penalties for driving offences. Interview with NSW Minister for Roads and Transport, Carl Scully. Announcer comments on new penalties for driving offences Announcer comments on new penalties for driving offences Information was also sought from the NSW Roads and Traffic Authority on any media campaigns launched to communicate the changes to drink-driving penalties to the general driver population. The road traffic licensing authorities were aware of only one publicity campaign ( Unintended Consequences ) that was run in conjunction with the penalty increase. Unfortunately, these authorities could only provide limited information on the campaign s details (John Bruton, Corporate Policy & Communications, RTA; personal communication 2002). This campaign was a two-part television advertisement that attempted to convey the serious implications arising from driving with a blood alcohol over the legal limit. 172

183 The first advertisement depicted a male driver who, after drinking with his friends at a local hotel, drove into a young girl crossing the road. The advertisement was extremely emotive, showing the child incurring horrific injuries from the crash and the mother s subsequent reaction to the injuries sustained by her child. The advertisement also showed the informal sanctions (e.g. shame and embarrassment) that can arise from this offence, with the driver having to return home to his wife and young daughter after the crash to explain what had just occurred. The second advertisement utilised the same characters and extended the drink-driving scenario to a courtroom setting. The scene depicted the driver being sentenced by a magistrate for the drink-driving accident in which the young girl was killed. A gaol sentence was imposed and, after the magistrate read out the penalty, court officers immediately escorted the driver out of the courtroom. The scene also showed the distress of the driver s wife and young daughter upon hearing that the driver would be gaoled for the offence. The second advertisement in this two-part series was shown on NSW television in late 1998 and early 1999, after the introduction of the new, more severe penalties. While this publicity would have been relatively successful in conveying the serious consequences that can arise from an episode of drink-driving, it did not specifically describe the 1998 penalty changes or convey to the average drink-driver the penalties attached to these offences. Instead, it highlighted the penalties that would be imposed in what many drivers would see as an extreme scenario. The consequences depicted in the advertisement may therefore not have had enough personal relevance to exert a deterrent effect. What went wrong? Criminal justice policies that raise statutory maximums often fail to reduce offending because they are not successfully translated into sentencing practice. This can happen because statutory maximums are raised but there is no subsequent change in the actual penalties imposed by the court or because players within the criminal justice system circumvent the more severe penalties by imposing alternative sanctions. The 1998 statutory amendments to drink-driving penalties did result in larger fines and longer licence disqualification periods for the majority of drink-drivers convicted 173

184 after the penalty changes came into effect, yet only a small change in offending levels was observed. Over the period in which drink-driving behaviour was measured in the quasiexperimental study, NSW Police appeared to alter their enforcement tactics from one where they predominantly deployed stationary RBTs to highly visible locations, to the more frequent use of hotspotting techniques that target high-risk drink-driving times and locations. The extent to which these changes in enforcement practice affected potential offenders perceptions of apprehension certainty is unclear. However, survey evidence suggests that, after 1998, fewer people had seen RBTs in operation and more people believed that enforcement activity targeting drink-drivers had decreased. While it is possible that hotspotting may have contributed somewhat to the decrease in recidivism found in Study 2, it is also possible that the drop-off in police breath tests reduced the perceived risk of apprehension and thus the deterrent efficacy of the new laws. The extent to which the penalty changes affected people s perceptions of sanction severity is also unclear from the quasi-experiment. As noted earlier, the retrospective nature of this study made it impossible to measure severity perceptions pre- and postintervention. However, the media analysis presented above revealed that very little publicity accompanied the enactment and the announcement of the drink-driving penalty increases. Furthermore, the media attention the statutory amendments did receive was ineffectively communicated. It was packaged with other non-criminal driving offences and was surrounded by negative publicity regarding the certainty of severe penalties being imposed by the courts. This, too, may have undermined the deterrent efficacy of the new penalties. The current chapter illustrates the difficulties that arise when attempting to test the deterrence model in a field setting, particularly where the research is conducted retrospectively and a quasi-experimental methodology must be employed. Attempts were made to introduce statistical controls but, in a quasi-experimental design, it is impossible to control for all extraneous variables. Further work is therefore necessary at the individual level, to strengthen our conclusions regarding the deterrent efficacy 174

185 of punishment severity. An individual-level analysis will allow for more effective measurement of the policy-perception-behaviour link, in an experimental setting where it is possible to control for other contextual influences. 175

186 CHAPTER 7. KNOWLEDGE OF PENALTIES, PERCEPTIONS OF SANCTION SEVERITY AND DRINK-DRIVING The aim of policies that raise the severity of statutory penalties is often simply to communicate the fact that severe punishment accompanies violations of the law (Tonry 1996). Since they endeavour to modify behaviour by influencing perceptions of sanction severity rather than by imposing harsher punishments these policies are consistent with deterrence principles. In order to achieve positive outcomes in relation to offending however, the general public has to be adequately informed of the legislative changes. As demonstrated in the previous chapter, statutory increases are often implemented without sufficient media attention or publicity and thus, are unlikely to affect offender behaviour. To date, little research has been conducted on factors that influence the formation of risk perceptions (Nagin 1998). The research that has been conducted focuses mostly on the role of individual attitudes, previous offending or social influences, rather than on the effect of criminal justice policy. Study 3 aims to fill this knowledge gap by investigating whether knowledge of criminal penalties influences perceptions of sanction risk and, consequently, decisions about whether to offend. Specifically, this study uses a scenario-based methodology to determine (1) whether people who have more accurate knowledge of the legal sanctions currently applicable to drink-driving offences in New South Wales perceive the penalties to be more severe and (2) whether they say they are less likely to offend, compared with those who have little or no knowledge of the law. Study 3 also aims to further elaborate on the findings from the quasi-experimental study reported in Chapter 5. Study 1 and 2 it will be recalled, found only small reductions in measures of offending after the drink-driving penalties were increased in The media analysis discussed in Chapter 6 suggested that one possible reason for this weak effect is that the public may have had insufficient knowledge of 176

187 the penalty changes. If knowledge does influence perceptions of punishment severity and decisions to offend then inadequate threat communication is a plausible explanation for the small deterrent effect observed. Alternatively, if there is no such relationship, the efficacy of raising punishment severity as a method for deterring offenders remains in doubt. Formation of risk perceptions Pogarsky, Piquero and Paternoster (2004) describe deterrence as a two-stage decision-making process. In the first stage external factors affect a potential offender s perceptions regarding sanction risks. In the second, these risk perceptions affect offending behaviour (see Linkage 1 and Linkage 2 in Figure 7.1). The perceptual stage is a dynamic process, whereby perceptions of threats are continually updated on the basis of new information acquired from external sources or from one s own offending experiences. The second stage of the deterrence process (the behavioural linkage) occurs contemporaneously with the offending opportunity. If, when an offending opportunity arises, present perceptions are that the costs associated with the act are too great, the expected utility is reduced and deterrence ensues. Importantly, this two-stage process does not imply that the perceptual/behavioural linkages are sequential. Instead, the behavioural linkage is thought to be nested within a continual process of perception modification and evolution (Pogarsky, Piquero & Paternoster 2004). This means that perceptions have the potential to influence offending behaviour but criminal conduct and its consequences also have the potential to affect threat perceptions. Figure 7.1: Two linkages in the deterrence process (from Pogarsky, Piquero & Paternoster 2004) Objective properties of punishment Committing crimes Perceptions of Offending Consequences of crime punishment behaviour Other events and experiences Linkage 1 Linkage 2 Perceptual linkage Behavioural linkage Pogarsky, Piquero and Paternoster (2004), along with other scholars, note that deterrence research thus far has focused almost exclusively on the behavioural 177

188 linkage (Cook 1980; Miller & Iovanni 1994; Nagin 1998) and has neglected to investigate how perceptions of sanction risks are formed. To address this limitation of prior deterrence research, Pogarsky and his colleagues (2004) tested the model shown in Figure 7.1, using data from a panel-based survey of 1,530 high school students in the USA. This survey consisted of two waves of data collection conducted approximately 12 months apart. At both survey administrations (Time 1 and Time 2) participants were asked to estimate the likelihood of being caught by police for three different offences (shoplifting, vandalism, marijuana use). The number of times participants were arrested between Time 1 and Time 2 was also measured in the survey, as was offending experiences: including both the number of times between Time 1 and Time 2 that a participant committed the offence being contemplated (specific offending) and the number of times a participant committed other similar offences (general offending). In addition, participants were also asked about the offending experience of their peers between Time 1 and Time 2. Pogarsky and his colleagues included this latter measure in the analyses to reflect vicarious punishment avoidance. Analysis of these data found some support for the proposed two-stage process. Both the number of arrests and the extent of peer offending between the first and subsequent interview waves were found to be significantly associated with changes in perceived certainty of arrest over that period. The direction of these relationships was consistent with deterrence predictions. That is, increased arrests between Time 1 and Time 2 were associated with increased perceptions of arrest certainty at Time 2. Having more peers who had offended, on the other hand, reduced the perceived likelihood of arrest at Time 2. The extent to which these experiences influenced risk perceptions also appeared to depend upon the perceived certainty of arrest prior to the new information being received. Arrest had little impact on individuals who already had relatively high estimates of punishment certainty at Time 1. It appeared that punishment avoidance for these offenders was the most influential factor in altering perceptions. On the other hand, the risk perceptions of individuals who had low estimates of arrest certainty at Time 1 were more affected by punishment experiences. Pogarsky, Piquero and Paternoster suggest that this may reflect the disparity between participant s expectations of punishment at Time 1 and 178

189 punishment experience or avoidance in the intervening period. If punishment experiences are contrary to what they expected then participants will update their perceptions of sanction risk accordingly. These findings are generally consistent with other research examining the effect of offending experiences on perceptions of punishment certainty, several of which were described in detail in Chapter 2 (see Paternoster & Iovanni 1986; Piliavin et al. 1986; Horney & Marshall 1992). However, this study only investigated punishment certainty. It remains to be seen whether the formation of severity perceptions follows a similar pattern to that observed for perceived certainty of punishment. Legal knowledge as a factor in the deterrence process This Chapter focuses on the question of whether knowledge of the statutory penalties prescribed by the legislature or knowledge of the actual penalties imposed by the courts influences severity perceptions and subsequent decisions to drink and drive. The model being used to inform this research was introduced in Chapter 4 and is represented here as Figure 7.2. As indicated previously, not all pathways shown in this model are being investigated in the scenario-based study (those appearing in bold text identify the variables that are being measured) and some interactions are implied. Nevertheless, Figure 7.2 is a useful graphical depiction of the deterrence hypotheses being investigated in Study 3 and provides a framework for the analysis and interpretation of data collected from participants. The two linkages of the deterrence process described by Pogarsky and his colleagues (2004), namely the perceptual and behavioural linkages, are an additional feature of Figure 7.2. Because a cross-sectional design is being employed in Study 3, this research will have to examine the link between knowledge and perceptions (Linkage 1) concurrently with the link between perceptions and behaviour (Linkage 2). This means that the dynamic process by which risk perceptions are updated on the basis of new information cannot be measured in this study. Nevertheless, the cross-sectional approach was considered appropriate given that this is one of the first investigations of the relationship between legal knowledge, severity perceptions and behaviour using a scenario-based research design. In the absence of evidence that current legal 179

190 knowledge affects perceptions of sanction severity, then measuring changes in risk perceptions over time becomes unnecessary. Furthermore, Pogarsky and his colleagues (2004) note that perceptions of apprehension risk may be far more dynamic than perceptions regarding the severity of formal punishment because certainty perceptions are, by their very definition, probabilistic. This view is consistent with the work undertaken by Miller and Iovanni (1994) described in Chapter 3. These authors found evidence from their panel-based survey that prior use of violence in an intimate relationship had a negative effect on perceptions of arrest certainty for courtship violence. However, there was no significant relationship between prior violence use and severity ratings of the problems arising from an arrest for this type of offence. This would suggest that a cross-sectional study would be appropriate when investigating factors affecting perceptions of punishment severity. Consistent with Pogarsky, Piquero and Paternoster (2004), the proposed deterrence model depicted in Figure 7.2 is premised on the idea that the objective properties of formal punishment influence behaviour by modifying perceptions of punishment severity. The model however includes other factors that may influence an individual s knowledge of the law, including previous offending experiences, the level of offending-risk (as indicated by alcohol consumption levels and involvement in alcohol-related accidents) and social influences from peers, family and the media. These potential determinants of legal knowledge, as well as previous research examining the relationship between legal knowledge and offending, are discussed in greater detail in the following section. 180

191 Figure 7.2: Proposed model of the deterrence process Prior offending/ alcohol consumption/ prior accident involvement Knowledge of statutory penalties Prior convictions/ Media publicity/ friends, family, acquaintances Linkage 1 Informal costs arising from apprehension Perceptions of penalty severity Linkage 1 Perceived applicability of statutory penalties Linkage 2 Linkage 2 Total anticipated formal costs associated with DD (disutility) Linkage 2 Decision to drink & drive Total benefits associated with DD (utility) Linkage 2 Perceived risk of apprehension Linkage 1 Media publicity/ friends, family, acquaintances Moral inhibitions associated with drink-driving Linkage 1 Linkage 1 Enforcement activity Prior successful drinkdriving episodes 181

192 Previous research Within a rational choice paradigm of decision-making, knowledge of the law logically precedes perceptions of sanction severity. As von Hirsch et al. (1999) observe, a potential offender can not fear consequences of which he is unaware (von Hirsch et al. 1999, p. 46). Previous research examining the level of legal knowledge held by members of the general public suggests that people are often poorly informed when it comes to the laws and criminal penalties applicable in their State, District or County (e.g. Williams, Gibbs & Erikson 1980; Snortum & Berger 1989; Hough & Roberts 1998; Kenkel & Koch 2001). This research seems difficult to reconcile with drink-driving studies that have found an association between the introduction of various drink-driving laws, such as preliminary breath test or mandatory licence disqualification laws, and subsequent reductions in aggregatelevel measures of offending (e.g. Saffer & Chaloupka 1989; Evans, Neville & Graham 1991; Kenkel 1993; Benson, Rasmussen & Mast 1999). If people are generally poorly informed about criminal laws and penalties, why does the implementation of these drink-driving laws reduce offending? Part of the explanation, Kenkel and Koch (2001) propose, lies in the fact that particular groups of people are more knowledgeable about the law than others. From a general population survey conducted in the USA, Kenkel and Koch (2001) found that those at greater risk of drink-driving, that is heavy drinkers and prior offenders, knew more about the applicability of drink-driving laws in their State than did other respondents in the sample. Peer group behaviour was also an important predictor of knowledge. Individuals who have peers that tend to limit their drinks when driving know more about drink-driving laws, than those individuals whose peers abstain from drinking. Thus even if the general public has a poor knowledge of drink-driving laws, those who are being targeted by such legislation demonstrate a better understanding of the legal consequences of drink-driving. Although Kenkel and Koch (2001) identified potential factors influencing legal knowledge, they failed to consider whether those who are more knowledgeable about the laws are less likely to offend. Previous research examining the relationship 182

193 between legal knowledge and offending behaviour provides mixed results. Williams, Gibbs and Erickson (1980) found some evidence for an inverse relationship between perceived statutory maximums and aggregate-level crime rates. These authors surveyed 2,400 adult residents in Tuscon, Arizona, on their perceptions of the maximum and the applicability of five different penalty types (e.g. whether or not a person can be imprisoned or whether or not a person can be fined) for nineteen different types of crime. These perceptions were found to be moderately correlated with actual maximums stipulated by the law. Furthermore, the analysis showed strong negative correlations between perceived statutory maximums and official crime rates in that area, suggesting that enhanced public knowledge of criminal penalties is associated with reduced criminal activity. However, further analyses of these data revealed that, when a measure of public disapproval of these crimes was included in the analytical model, the significant relationship between perceived maximum penalties and crime rates disappeared (Williams & Gibbs 1981). This, the authors propose, demonstrates that the inverse relationship between perceived maximums and actual crime rates is not due to deterrence, but rather the shared normative values of the public and legislators in relation to these types of crime. It should be noted here, however, that the data used in this analysis came from a survey examining a range of criminal offences, most of which were serious (e.g. first-degree murder, first-degree rape and aggravated assault). Serious offences have traditionally been thought of as less susceptible to deterrence-based initiatives because they are already supported by the moral code society (Andenaes 1966; Zimring & Hawkins 1973). Yet no distinction is made in the analysis between these crimes and others that are more instrumentally motivated (e.g. petty theft) and thus more likely to be affected by formal sanctions. Furthermore, as discussed in Chapter 3, ecological designs such as this suffer from several methodological weaknesses, including reverse causality, which preclude confident conclusions regarding deterrent effects. Homel s longitudinal study of drink-drivers in NSW (previously referred to in Chapter 2; Homel 1986) overcomes the limitations of William et al. s ecological research and presents some support for the notion that improved legal knowledge can 183

194 result in modifications to behaviour, particularly for those who are more likely to offend. When RBT was introduced in NSW a high-range drink-driving offence that carried harsh penalties was also introduced. To examine the impact of this offence, Homel (1986) asked drivers whether they knew about the penalty for this offence and also asked about the number of modifications that they had made to their driving behaviour since its introduction. He then reinterviewed these drivers six weeks later to measure any changes in drink-driving behaviour during the follow-up period. His results indicated that those who knew about the penalty increase were not only more likely to modify their drinking and driving behaviour at the first interview but were also less likely to drink and drive during the six-week follow-up period (this was independent of the effects of RBT on drink-driving behaviour). Furthermore, these changes in drink-driving behaviour were more apparent for respondents who had a previous conviction for drink-driving than for those without a conviction. Despite this initial evidence for a deterrent effect, Homel (1986) also found that two other measures of perceived punishment severity used in the survey did not significantly predict changes in offending behaviour. He asked participants to imagine the punishment they would receive if caught for drink-driving and then rate (on a five-point scale) how big a problem this punishment would be for them. He also asked participants to rate their chances (on a five-point scale) of receiving no penalty from the court if arrested for drink-driving. Logistic regression analyses found neither of these indicators to be significant predictors of having driven in the intervening period between the two interviews. This latter finding seems somewhat surprising, given the significant relationship between knowledge and behaviour, however it should be noted that just 12 (6.9 per cent) out of the 175 participants reinterviewed in the second survey reported offending during the follow-up period. This low base rate would reduce the power to detect significant effects in the regression analysis and could explain why the study failed to find a relationship between the measures of severity perceptions and behaviour. A further possibility is that the categorical variable used in the analysis to measure perceived severity may not have sufficiently captured individual variation in severity perceptions. A continuous measure of severity perceptions as is sometimes 184

195 used in deterrence research to measure perceived likelihood of punishments, would improve the power of the analysis to detect significant effects if they exist. The scenario-based study - Study 3 The primary aim of Study 3 is to examine the policy-perception-behaviour link and in doing so, address an important knowledge gap identified in the deterrence literature. Specifically this research investigates the relationship between knowledge of the drink-driving penalties currently applicable in New South Wales, perceptions of punishment severity and willingness to offend in a given set of circumstances. This research will also allow for more confident conclusions regarding the deterrent efficacy of punishment severity to be drawn from the results of the case study. Chapter 6 suggested that a possible reason for the small effect of the 1998 penalty changes was that the sentencing policy was inadequately publicised. Examining the influence of current knowledge of drink-driving penalties on willingness to offend will allow us to determine the extent to which this account for the small observed deterrent effect is valid. This research also builds on the quasi-experimental research in that it allows the level of apprehension risk to be experimentally manipulated across groups. Chapter 6 demonstrated that the level of enforcement activity associated with drink-driving offences in NSW altered over the period in which drink-driving was measured for the quasi-experimental study. This modification to RBT activity could have potentially affected offending behaviour at the same time that the penalties were raised. In this scenario-based research, however, the effect of punishment severity on offending decisions can be examined at various levels of perceived certainty of apprehension. Hypotheses As outlined in Chapter 4, the scenario-based study was designed to test four specific hypotheses generated by deterrence theory: 1. Increased perceived risk of arrest reduces offending likelihood To investigate whether police practice can influence behaviour through altering perceptions of arrest certainty, the level of RBT activity was varied across scenarios. The three conditions of punishment certainty varied in the scenario were: You 185

196 haven t seen police conducting Random Breath Tests (RBT) in this area for some time (low-level of enforcement activity); You have often seen police conducting Random Breath Tests (RBT) in this area (medium-level); and You have often seen police conducting Random Breath Tests (RBT) in this area and since it is a Friday night you know that there will be more police on the roads (high-level). Deterrence theory would anticipate that participants allocated the scenario with a high-level of police activity would be less likely to state that they would offend under the circumstances described. 2. Enhanced knowledge of drink-driving penalties reduces offending likelihood by increasing perceptions of sanction severity Natural variation in levels of legal knowledge was exploited in Study 3 by way of a series of multiple choice questions asking about the penalties for low- and mid-range drink-driving offences. These questions pertained to both the maximum penalties prescribed by the law and the actual penalties the individual would receive from the court if found guilty of drink-driving. We would anticipate that individuals who are more knowledgeable about either actual or statutory drink-driving penalties, would be less likely to state they would offend in the given scenario. Within a deterrence framework, legal knowledge can only affect behaviour by altering perceptions of punishment severity. For this reason, we would expect a significant relationship, not only between knowledge of penalties and drink-driving behaviour, but also between legal knowledge and severity perceptions. 3. Increased perceptions of sanction severity reduces offending likelihood Perceptions of sanction severity were measured by asking respondents to indicate how much of a problem the penalties would create for them personally. This question has been employed in previous perceptual research (Grasmick & Bryjak 1980; Homel 1986; Nagin & Paternoster 1993) in order to capture the subjective nature of severity perceptions. However, unlike previous studies, participants in Study 3 are asked to rate their perceptions regarding penalty severity on a scale of 0 (no problem) to 100 (a very big problem). This continuous variable was employed in an attempt to measure more accurately the range of severity perceptions held by survey participants. Again, consistent with deterrence predictions, we would expect 186

197 that participants who perceive the penalties as more severe to state that they are less likely to offend given the conditions described in the scenario. 4. Perceptions of sanction severity matter more when punishment is perceived to be certain Study 3 also formally tests the interaction between certainty and severity by including an interaction term in the regression models predicting stated offending likelihood. One condition thought necessary to maximise the deterrent efficacy of severe sanctions is a relatively high-level of perceived risk of detection and apprehension. If this is true, then we would anticipate that perceptions of sanction severity matter most in the offending decisions of participants allocated the scenario with a high-level of police activity targeting drink-driving. The model depicted in Figure 7.2 also describes several other factors that may influence the formation of risk perceptions and decisions to offend. In particular, it includes extralegal consequences arising from apprehension for a drink-driving offence as an independent factor affecting the disutility of the crime. This is consistent with Williams and Hawkins (1986) proposal that informal social controls arising from arrest and conviction should be included as part of the general deterrence process (Williams & Hawkins 1986, p. 561). This additional influence on offending decisions is not, however, investigated in Study 3. Much of the perceptual research that has previously been undertaken in the deterrence domain has examined the influence of informal social controls on offending decisions. This research has made an important contribution to the deterrence literature by demonstrating that informal social sanctions, under some conditions, can operate like legal threats in preventing offending (e.g. Grasmick & Bursik 1990; Grasmick, Bursik & Arneklev 1993; Baum 1999). 70 However, perceptual research has said little about the policy-behaviour link upon which deterrence-based criminal justice interventions are modelled. Unless a relationship exists between actual sentencing policy and offending behaviour then the efficacy of raising the threat of formal punishment to prevent future offending remains in doubt. This area of research is one 70 Though it should be noted that moral inhibitions associated with offending are accounted for in the analysis by excluding individuals who would not offend even where there is no possibility of punishment if caught. 187

198 of the gaps in the deterrence literature identified by Nagin (1998) and one that needs further exploration given the results obtained from the quasi-experimental study described in Chapter 5. Results of Study 3 The following sections present the results of Study 3. Sample demographics and measures on the independent variables relevant to our hypotheses are initially described and then bi-variate relationships between these independent variables and the dependent variable (i.e. the stated likelihood from 0 to 100 of driving home under the prescribed conditions) are examined. Following this, the results of a linear regression analysis (estimated using Ordinary Least Squares) in which the independent variables were regressed against the dependent variable are presented and any findings pertinent to the deterrence hypotheses being tested are discussed. Finally, the assumptions underpinning the estimated regression models are tested and further necessary exploratory analyses are undertaken. Sample demographics The sample for Study 3 consisted of 445 participants recruited from various legal, engineering and general studies classes at the University of New South Wales. Just over half (56 per cent) of the participants recruited for this study were female. The average age of participants was 21 years, and 91 per cent were aged between 18 and 25 years. Twenty-six per cent of the sample reported that they had, on at least one occasion, driven after drinking too much alcohol but only three participants reported a previous conviction for a drink-driving offence. One in ten participants had themselves been involved in an alcohol-related crash or knew a relative who had. Over one-quarter of participants reported that they had consumed alcohol at an acuterisk level 71 at least once weekly in the previous 12 months and 10 per cent reported drinking at these levels on more than one occasion every week. 71 Drinking alcohol at or above the Australian low-risk guidelines for acute harm, that is, more than four standard drinks for females and more than six standard drinks for males (National Health and Medical Research Council 2001) 188

199 Independent variables Perceptions of apprehension certainty As described previously, respondents were randomly assigned one of three scenarios varying in the level of RBT activity described; low (n=146), medium (n=148) and high (n=150). Figure 7.3 shows the mean estimates of apprehension certainty across these three groups. There was a significant difference in perceived certainty of apprehension between the low- and high-risk groups (t=-2.140, df=235, p=0.033), as well as between the low and medium groups (t=-3.147, df=241, p=0.002), but no significant difference between the medium and high groups was found (t=0.822, df=240, p=0.412). For this reason the medium and high groups were combined in later analyses. The participants assigned to these two groups showed no significant differences in terms of age, gender, alcohol consumption, prior offending or prior accident involvement. 72 Figure 7.3 Mean estimates of certainty of apprehension and conviction by scenario group Av. perceived certainty of apprehension low medium high Scenario Knowledge of statutory penalties Generally, participants had a good understanding of the types of penalties that are prescribed by legislation for drink-driving offences according to the recorded BAC 72 Age, t=-0.986, df=362, p=0.325; gender, x 2 =0.547, df=1, p=0.459; alcohol consumption, t=0.693, df=363, p=0.489; prior offending, t=-0.641, df=364, p=0.522; accident involvement, t=-0.508, df=364, p=

200 level, but did not have a good understanding of statutory maxima and minima relevant to particular offences. Knowledge of the applicability of fines was high, with over 95 per cent of the sample correctly reporting that a person could be fined for a low- or mid-range drink-driving offence in NSW. However at least half of the respondents underestimated the maximum fine for both these offences. Most participants knew that gaol terms were not applicable to low-range drink-driving offences but 43 per cent incorrectly believed that gaol terms were not applicable to mid-range drink-driving offences. The results from the survey, moreover, suggest that knowledge of mandatory minimum licence disqualification periods for drink-driving offences was quite low. Fifty-five per cent of participants did not know about minimum disqualification periods for low-range drink-driving offences (minima that were introduced in NSW as part of the 1998 legislative changes to driving penalties) and 35 per cent did not know about the minima for mid-range drink-driving offences. Participants were assigned a score based on the number of errors they made in relation to questions on statutory penalties for drink-driving offences (mistake1). A low score represented few errors and thus good knowledge of the applicable statutory penalties, while a high score represented many errors, or poor knowledge. The frequency distribution for these scores is shown in Figure

201 Figure 7.4: Frequency distribution of scores on knowledge of statutory penalties No. of persons mistakes Knowledge of actual penalties As with statutory penalties, a large proportion of participants (57 per cent and 34 per cent for low-range and mid-range respectively) underestimated the actual average fine imposed on people convicted of drink-driving offences. Most, however, knew that they would be fined (98 per cent for both low- and mid-range) and knew that they would not be gaoled (90 per cent low-range; 73 per cent mid-range). On the other hand, a substantial proportion of participants (40 per cent) mistakenly thought that they would not have their licence disqualified for a low-range drink-driving offence. For a mid-range drink-driving offence, nearly all of the participants (98 per cent) correctly thought that their licence would be disqualified, but only one-third reported correctly that the average disqualification period would be nine months or more. Again, based on the answers to these questions, a knowledge score (mistake2) was calculated, the frequency distribution of which is shown in Figure

202 Figure 7.5: Frequency distribution of knowledge of actual penalties No. of persons mistakes Perceptions of penalty severity Figure 7.6 displays the frequency distribution of the ratings of penalty severity for all participants. The distribution of severity ratings shown here suggests that the vast majority of participants perceived the penalties to be very severe, with almost onethird rating the severity of drink-driving penalties at 100. Only three respondents reported that the penalties would be no problem at all for them. 192

203 Figure 7.6: Frequency distribution of severity scores No. of respondents Severity of penalties Predicting stated likelihood of offending Estimates of stated offending likelihood When presented with the drink-driving vignette, almost one-third of participants stated that they would not drive home under the circumstances described but almost one-quarter stated that there was a 50 per cent chance or greater that they would do so. If there were no possibility of being caught and punished for the offence, 18 per cent of participants said that they would still not drive home, but 18 per cent said they would definitely drive home. Respondents who would not drive home even if there were no chance of being caught or convicted can be considered a group who is morally opposed to drink-driving (Nagin & Pogarsky 2001). This group was therefore excluded from further analyses (79 respondents in total). The frequency distribution of the dependent variable (excluding these nondeterrable respondents) is shown in Figure 7.7. The average reported offending likelihood in this reduced sample was 37 per cent, with a median of 30 per cent (SD=32.1) It is worth noting that these estimates of reported offending likelihood are slightly higher than those obtained by Nagin and Pogarsky (2001) in their scenario-based drink-driving study using a student sample from a US university (av. 31%; median 20%). 193

204 Figure 7.7: Frequency distribution of offending likelihood estimates No. of persons Likelihood of offending Zero order correlations Table 7.1 shows descriptive measures and zero order correlations for the variables of interest in the study. As seen from this table, there are a number of significant bivariate relationships between the dependent and the independent variables and among the independent variables themselves. Firstly, there is a moderate, positive correlation between mistakes made with regard to statutory sanctions for drinkdriving offences and mistakes made with regard to the actual penalties imposed by the courts. However, only the latter variable has a significant positive correlation with stated offending likelihood. This indicates that those who made more mistakes when asked about the actual penalties drink-drivers receive upon were more likely to offend in the scenario. There are also significant relationships between stated offending likelihood and the two deterrence variables measured in the survey. The direction of these zero-order correlations is negative, indicating that persons who were assigned to the low certainty of apprehension group and those who perceive the penalties to be less of a problem are more likely to report that they would drink and drive under the circumstances described in the scenario. 194

205 Earlier we hypothesised that knowledge of penalties would influence offending likelihood through its impact on perceptions of penalty severity. Consistent with this prediction, the zero-order correlations between ratings of sanction severity and both knowledge scores are negative. Thus, the more mistakes a respondent made on the questions tapping knowledge of drink-driving penalties in NSW, the lower the rating of punishment severity. Note, however, that these correlations are weak and only the relationship between severity perceptions and knowledge of actual penalties is statistically significant at the 0.05 level. There is a positive correlation between prior drink-driving episodes and three other independent variables; age, number of alcohol-related crashes and frequency of highrisk drinking. Participants who report having frequently driven after drinking too much alcohol were older, had been involved in more alcohol-related crashes and drank alcohol at high-risk levels more frequently. We would expect those who drink and drive more often to be involved in more alcohol-related crashes and to drink alcohol more frequently at high-risk levels. The positive correlation between age and prior offending may seem contrary to the results of many criminological studies, but it should be remembered that the prior offending variable in this study measures the number of occasions ever driven whilst drunk. Older people in this sample would have had more opportunity to offend than younger persons and would therefore be more likely to record numerous drink-driving episodes. Prior offending also has a significant positive correlation with stated likelihood of offending in the hypothetical scenario. This indicates that those who had previously offended were also more likely to state that they would offend in the circumstances described in the scenario. While not surprising, this latter finding does attest to the truthfulness of participant responses on the dependent variable. The number of previous drink-driving episodes is negatively correlated with both of the variables measuring legal knowledge. This suggests that persons who offend more often know more about the statutory penalties applicable to drink-driving offences (mistake1) and about the court penalties imposed on drink-drivers (mistake2). A similar relationship is evident for levels of alcohol consumption and 195

206 legal knowledge, with those who frequently drink at high-risk levels making fewer mistakes about drink-driving penalties. These findings are consistent with the results from Kenkel and Koch s (2001) research, which showed that people who had previously driven whilst drunk knew more about illegal per se laws and also that heavy drinkers make fewer mistakes about laws related to licence suspension. Table 7.1: Zero order correlations between independent and dependent variables (n=366) x 1 x 2 x 3 x 4 x 5 x 6 x 7 x 8 x 9 y Age (x 1 ) 1.0 Gender (x 2 ) Prior.21* offending (x 3 ) Accident * 1.0 involvement (x 4 ) Alcohol *.11* 1.0 Consumption (x 5 ) Mistake1 (x 6 ) * * 1.0 Mistake2 (x 7 ) * *.38* 1.0 Certainty gp (x 8 ) Severity (x 9 ) *.11* * Offending * * -.12* -.11* 1.0 likelihood (y) Mean Standard Dev * Significant at the 0.05 level ** Where ordinal variables are correlated spearman s correlation coefficient is presented Linear regression models including knowledge and certainty variables To predict stated likelihood of drink-driving in the scenario, controlling for the influence of the factors appearing in Table 7.1, multivariate analyses were undertaken using Ordinary Least Squares regression. Three separate models were estimated and the details of these models are shown in Table 7.2. Model 1 includes the composite variable measuring knowledge of statutory penalties, Model 2 includes the composite variable measuring knowledge of actual penalties and Model 3 includes a categorical variable measuring estimates of perceived penalty severity. All 196

207 three models also contain a dichotomous variable measuring perceived certainty of apprehension (low vs. high) and measures on other relevant control variables. Table 7.2: Linear regression models for stated likelihood of offending Model 1 Model 2 Model 3 b p b p b p CERTAINTY High v. low (3.39) KNOWLEDGE OF PENALTIES 0.88 (Statutory penalties) (0.32) KNOWLEDGE OF PENALTIES < (3.37) < (0.53) < (3.42) < 0.01 (Actual penalties) SEVERITY Medium v. Low (3.86) High v. Low (4.06) DRINK-DRIVING FREQUENCY 2.37 < < (no. of times) (0.47) (0.46) (0.47) RISKY ALCOHOL CONSUMPTION Monthly v. Rarely/never (3.89) (3.81) (3.88) Weekly v. Rarely/never (4.24) ALCOHOL-RELATED 0.61 ACCIDENT FREQUENCY (no. (2.80) of times) (4.23) (2.77) (4.23) (2.82) < < 0.01 AGE (0.49) (0.48) (0.49) GENDER (3.20) (3.17) (3.22) Notes: Standard error of estimates appear in parentheses. For all models n=364 (note that two cases had missing data for age) Perceptions of apprehension certainty As seen in Table 7.2, the coefficient for perceived certainty of apprehension is negative and statistically significant (at α=0.01) in all three models. The magnitude of this certainty effect is such that respondents who received the high-risk of apprehension scenario were, on average, 10 per cent less likely to report that they would drive home than were respondents who received the low-risk scenario. 197

208 Knowledge of penalties Table 7.2 also shows significant positive coefficients for both knowledge variables. This indicates that the more mistakes respondents made with regard to the drinkdriving penalties, the more likely they were to say they would offend in the hypothetical scenario, controlling for other variables in the model. In the case of statutory penalties, an increment of five on the knowledge scale increases the stated probability of drink-driving by a little over four per cent. An increment of five on the knowledge scale for actual penalties increases the stated probability of drink-driving by 10 per cent. The difference in regression coefficients for the knowledge variables, as well as the difference in variance accounted for by Models 1 (R 2 =0.106) and 2 (R 2 =0.124), indicates that knowledge of actual penalties is a better predictor of stated offending likelihood than knowledge of statutory penalties. 74 If the knowledge effect found in the above analysis was due to deterrence then we would also expect to see a significant relationship between legal knowledge and perceptions of punishment severity. Table 7.1 shows that there were in fact negative correlations between ratings of penalty severity and both knowledge scores, however only the relationship between severity perceptions and knowledge of actual penalties imposed on drink-drivers was statistically significant. This accords with the regression models shown in Table 7.2, which suggest that knowledge of actual penalties imposed for drink-driving is a better predictor of reported offending likelihood than is knowledge of statutory sanctions. Perceptions of penalty severity Main effects for severity perceptions on stated offending likelihood were also considered. Due to the highly skewed distribution of the severity estimates, it was not possible to include severity perceptions in the linear regression analysis as a continuous variable. Thus, a new categorical variable was created for this analysis which took one of three values: low (0-70), medium (71-90), and high (91-100). Approximately one-third of respondents in the sample fell into each of these three categories. The average reported likelihood of offending for each of these severity 74 Achen (1982) also suggests looking at the b coefficient multiplied by the mean for the corresponding independent variable to determine the level of importance of variables in the regression analysis. This calculation also found knowledge of actual penalties to be a better predictor. 198

209 groups across the two certainty conditions is presented in Figure 7.8. This figure shows that respondents assigned to the low-risk of apprehension scenario were more likely to state that they would offend than respondents assigned to the high-risk condition. Conversely respondents who rated the penalties as a very big problem were less likely to state that they would offend compared to those who considered the penalties to be less of a problem. Inclusion of these subjective ratings of penalty severity in the linear regression models described previously produced a small effect for severity group membership (see Model 3 in Table 7.2). Respondents in both the medium and high severity groups reported that they were less likely to offend in the scenario than those in the low severity group (medium group b=-5.49, p=0.15; high group b=-7.51, p=0.07) but only the difference between the high severity and the low severity groups approached significance at the 0.05 level. Figure 7.8: Likelihood of offending by perceptions of penalty severity and apprehension risk Av. likelihood of offending low risk high risk low medium high perceptions of severity Interaction between certainty and severity The interaction between punishment certainty and severity is also of importance to our hypotheses. Perceptions of punishment severity, in other words, should matter more in offending decisions when punishment is highly probable. Looking back at Figure 7.8 we see that, on average, participants allocated to the low-level of RBT activity scenario stated, that they were more likely to offend in the circumstances described than those allocated to the high-level scenario. Participants in both these groups who perceived the penalties to be less of a problem stated that they were more 199

210 likely to offend. However, if severity were more likely to matter at high levels of punishment certainty than we would expect to see divergence of the two lines shown in this figure with the severity effect being greater for the high-risk of apprehension group. As seen from this graph the two lines are parallel. Moreover when product terms for the severity and certainty groups are included in the regression models there is no evidence for a significant interaction effect. 75 These findings are inconsistent with the assumption that the relationship between perceived risk of apprehension and perceived punishment severity is multiplicative. Testing the assumptions of linear regression The regression models described above show a significant relationship between knowledge of drink-driving penalties (both actual and statutory penalties) and stated offending likelihood, as well as between the perceived risk of detection and stated offending likelihood. While these two findings are supportive of our deterrence hypotheses, they need to be treated with some caution because the residual diagnostic plots for all three models indicate lack of homogeneity of variance. These plots suggest that the models tend to overestimate low offending probabilities and underestimate high probabilities. 76 This violates the assumption of linear regression that the residuals should be dispersed randomly throughout the range of the estimated dependent variable. 77 This assumption violation could mean that (1) there is an interaction effect not included in the model or (2) independent variables predictive of the outcome measure have been omitted from the model. 75 The interaction between certainty and severity was modelled using the following regression equation; Y = α + β 1 X 1 + β 2 X 2 + β 3 X 3 + β 4 X 1 X 2 + β 5 X 1 X β k X k, where Y is the stated likelihood of offending in the scenario, X 1 is certainty group (high v. low), X 2 is a binary variable representing inclusion in the medium severity group (SEVERITY1), X 3 is a binary variable representing inclusion in the high severity group (SEVERITY2) and X 1 X 2 and X 1 X 3 are interaction terms for GROUP*SEVERITY1 and GROUP*SEVERITY2. Other predictor variables were also included in the regression model. The final model found that neither of the two interaction terms in this model were significant (p=0.816; p=0.696, respectively). The incremental change in R 2 was also small, increasing from to Note that similar results were evident even after excluding respondents who reported a zero likelihood of offending in the scenario. 77 Transformation of the dependent variable did not improve the model fit. However, similar results to those from the linear regression analysis were found from a logistic regression analysis where the dependent variable was 50% or greater chance of offending in the given scenario. This type of analysis is not subject to the same assumptions as a linear regression analysis. 200

211 To investigate the first of these possibilities further exploratory analyses were conducted, partitioning out the effect of the punishment variables for different subsamples. Given previous evidence in the literature that people who are at greater risk of offending are more sensitive to changes in punishment variables (e.g. Homel 1986; Decker, Wright & Logie 1993; Wright et al. 2004), high-risk drinkers and those reporting prior offending episodes were examined separately. The mean plot graphs shown below provide some evidence for an interaction effect between risky alcohol consumption and perceived risk of apprehension (Figure 7.9), as well as between prior offending and perceived risk of apprehension (Figure 7.10). The relationship is such that the effect of the perceived certainty of punishment on stated offending likelihood appears to be more pronounced for the high-risk groups (i.e. frequent risky drinking and greater than five previous offending occasions). Figure 7.9: Likelihood of offending by certainty group and drinking status Av. likelihood of offending never/not often (n=201) monthly/1 a wk (n=119) frequently (n=45) 0 low certainty group medium 201

212 Figure 7.10: Likelihood of offending by certainty and prior offending episodes Av. likelihood offending low Certainty group no priors (n=257) 1-5 priors (n=88) >5 priors (n=21) high To test the interaction between drinking status and the deterrence variables measured in the survey, the sample was divided into three groups, based on self-reported frequency of drinking. Separate regressions of the dependent variable (i.e. stated offending likelihood) on the remaining independent variables were then carried out. 78 The results of these analyses are shown in Tables 7.3 and 7.4 (note that nonsignificant control variables have been removed). The most notable feature of both these tables is that Model 3, which includes only participants who report drinking at acute-risk levels on a weekly basis, accounts for substantially more of the variance in the dependent variable than do Models 1 or 2 (i.e. R 2 =0.32 v. R 2 =0.06 & R 2 =0.12, respectively). Furthermore, examination of the residual plots for Model 3 reveals no serious departures from linear regression assumptions, indicating that the model is a relatively good fit for stated likelihood of offending. 78 Including dummy variables in the regression equations to represent this interaction increased R 2 from 0.12 to 0.14, but tests of this incremental change were non-significant (F=1.64 < F c =2.37). However, given that there were only small numbers of persons in the high-risk group (n=45) it is possible that there was insufficient power to detect a significant interaction effect using this approach. 202

213 Table 7.3: Linear regression models for stated likelihood of offending specified for different drinking groups Model 1 (rarely/never n=201) Model 2 (regular n=119) Model 3 (frequent n=45) b p b p b p CERTAINTY High v. low (4.69) (5.77) (9.73) MISTAKE1 (STAT < PENALTIES) (0.41) (0.57) (1.05) DRINK-DRIVING FREQUENCY (no. of times) 1.75 (1.20) (0.66) < (0.68) < 0.01 R-square Notes: Standard Error of estimates appear in parentheses Table 7.4: Linear regression models for stated likelihood of offending specified for different drinker groups Model 1 (rarely/never n=201) Model 2 (regular n=119) Model 3 (frequent n=45) b p b p b p CERTAINTY High v. low (4.66) (5.63) (9.34) MISTAKE2 (ACTUAL < PENALTIES) (0.74) (0.87) (1.44) DRINK-DRIVING FREQUENCY (no. of times) 1.88 (1.19) (0.65) < (0.68) < 0.01 R-square Notes: Standard Error of estimates appear in parentheses In terms of the independent variables, we see that the effect for punishment certainty is much larger for frequent risky drinkers than it was for the full sample. Model 3 shows that frequent drinkers who were assigned to the high-risk of apprehension scenario were, on average, 25 per cent less likely to state that they would offend than frequent drinkers assigned to the low-risk scenario. However, knowledge of penalties is no longer a statistically significant predictor of stated offending likelihood for this subgroup. This could be because there is an interaction between drinking status and knowledge, such that knowledge is more predictive of offending at low-frequency drinking. Alternatively it could be due to the small sample size (n=45) (i.e. there is insufficient power to detect an effect). Since the beta coefficients for the knowledge variables in Model 3 are similar to those reported for the full sample, the latter explanation would seem to be the most accurate interpretation of this result. 203

214 In Models 1 and 2, R 2 is relatively low and the variable representing perceived risk of apprehension is not significant in the case of the low-risk drinkers. Furthermore, the residual plots for both of these models display a similar pattern to that seen for the full sample, that is overestimation of the low offending probabilities and underestimation of high probabilities. This suggests that it may not be fear of punishment that is influencing the offending decisions for these low-risk groups but some other variable that has been omitted from the model. Similar results as those described above were found when the sample was separated on the prior offending variable. That is, the effect of perceived risk of apprehension on stated offending likelihood was much greater for those participants who had driven home on numerous occasions after drinking too much alcohol than for those who rarely offended. The variance accounted for by the regression models was also greater for the frequent offender group and the residual plots showed no serious violations of regression assumptions. To minimise repetition, however, the linear regression models for these supplementary analyses have been placed in Appendix III. Summary and discussion of survey findings Study 3 confirms previous research findings in showing that perceived certainty of apprehension can affect the likelihood of offending. Simply knowing that RBTs have been operating within the immediate area on a frequent basis is enough to reduce the chances that someone will drink and drive. While there was an overall effect of apprehension risk it is also worth noting that there was no difference in perceived certainty of apprehension between the medium- and high-risk scenarios. The highrisk scenario attempted to mimic a police blitz or hotspotting operation that tasked more police to the job of breath testing drivers on weekends. Contrary to expected utility predictions, this increase in police numbers did not significantly raise perceptions of apprehension risk any more than that created by frequent sightings of police in the area. This unexpected finding can perhaps be explained by Sherman s (1990) idea of ambiguity aversion in situations of uncertainty people will tend to play it safe 204

215 and overestimate the risks associated with getting caught. Since the day of the week was not identified in the medium-risk scenario, respondents may have assumed the worst and estimated the risk as if it were a weekend and more police would be patrolling. This suggests that keeping potential offenders guessing about the certainty of being breath-tested on any one day may be just as useful in drink-driving enforcement as increasing the actual risk of being tested at certain high-risk times and locations. A further finding from this scenario-based research is that punishment certainty appears to be more of a deterrent for people who are at greater risk of offending. This result is consistent with previous research conducted in the deterrence domain, including studies that have focused on drink-driving. For example, Grosvenor, Toomey and Wagenaar (1999) conducted a cross-sectional survey of adolescents (12 th grade students) asking them about their drink-driving behaviour in the previous 30 days. These students were also asked the penalty they thought they would receive if caught drink-driving, the chances of being caught by the police for drink-driving and their drinking behaviour in the previous two weeks. Controlling for other individual demographic characteristics, perceived certainty of apprehension was only an effective deterrent for binge drinkers (i.e. those reporting drinking five or more drinks on one occasion in the previous two weeks). Among non-binge drinkers there was no relationship between drink-driving and perceived certainty of apprehension for an offence. Similarly, in our sample, participants allocated the high-risk of apprehension scenario were, on average, 10 per cent less likely to state they would offend. However, for participants who frequently drink alcohol at high-risk levels, the reported offending likelihood was reduced by up to 25 per cent in the scenario where the perceived risk of apprehension was high. Data from this study also provide some support for the notion that increased knowledge of applicable penalties is associated with reduced offending likelihood under scenario-based conditions. The regression analyses described above showed that participants who made more mistakes about the penalties applicable to drinkdriving offences in New South Wales were more likely to state that they would offend under the circumstances described in the scenario than were participant who 205

216 made fewer mistakes about the sanctions. Interestingly, however, these analyses suggest that statutory penalties are not the most influential factor in perceptions of punishment costs. Instead, severity perceptions appear to be more closely related to the penalties people think they would receive if convicted for drink-driving rather than to the maximum penalties prescribed by the legislature. This was demonstrated by the significant bivariate relationship between severity perceptions and knowledge of actual penalties for drink-driving, as well as evidence indicating that knowledge of actual drink-driving penalties is a better predictor in the linear regression models predicting stated offending likelihood. This result is not a particularly surprising finding but it does have some important implications for criminal justice policy. It suggests that simply raising the maximum or minimum penalties will not be enough to prevent future offending. For deterrence to occur, the public needs to know about the penalty changes and believe that this will result in more severe penalties if convicted. This need to increase awareness of drink-driving penalties is underlined by the fact that, overall, knowledge of applicable penalties in NSW was relatively poor amongst survey participants. A substantial proportion of participants was not aware that there are minimum licence disqualification periods prescribed for drink-driving offences in this State and, though many knew that they could be fined for drink-driving, most participants significantly underestimated the monetary penalties that would ensue. While it is acknowledged that the sample recruited for this survey is not representative of the general driving population in NSW, these results mirror those found from a large telephone survey of licensed drivers conducted by the NSW Roads and Traffic Authority in This survey involved Computer Administered Telephone Interviews with 1,143 drivers aged 17 to 69 years who reported having consumed alcohol at least once a month in the previous 12 months. Analysis of the data from these interviews revealed that 35 per cent of the drivers were not aware that they could lose their licence if charged with drink-driving, while 73 per cent were not aware that they could receive a fine of over $1,000. Furthermore, 40 per cent of the drivers surveyed were not aware that a drink-driving charge could result in a criminal record (Bryant, Hawkins & ACNielsen 2004). 206

217 So where does this leave the NSW Government s drink-driving policy which raised the statutory penalties for drink-driving offences in 1998? If the public was better informed would the implementation of the policy have led to a reduction in drinkdriving or alcohol-related road crashes? The significant effect of knowledge found in this study, as well as the generally low level of knowledge about drink-driving penalties, would suggest that greater publicity of the policy may have decreased the likelihood of some people drink-driving. As discussed in Chapter 6, the publicity that surrounded the penalty increase in NSW was sparse and sometimes negative. If a well-coordinated media campaign had been launched by the Government to inform the public of the 1998 legislative amendments, we may have seen a greater reduction in drink-driving offending levels after the penalty changes. However, although the knowledge variables were significant in the regression analyses shown above, only tentative conclusions about the relationship between legal knowledge, perceptions of punishment severity and offending decisions can be reached on the basis of these data. Only one of the legal knowledge variables was significantly correlated with perceptions of penalty severity and, although statistically significant, the relationship in question was weak. Furthermore, when subjective ratings of penalty severity were included in the regression models predicting stated offending likelihood, only the difference between the low and high severity groups was close to significance and no interaction effect between perceived certainty and severity of punishment was apparent. We also cannot ignore the fact that diagnostic checks of the linear regression models suggested the omission of an important variable from the analysis (at least for the full sample). Very little research has been conducted thus far on the role that knowledge of criminal penalties plays in risk perceptions and deterrence. In fact, the current study appears to be the first attempt using the scenario-based methodology to investigate the extent to which legal knowledge can predict offending behaviour. The validity of the measures used in the current research to assess respondents legal knowledge need to be tested further before any definitive conclusions can be made. Given that knowledge of statutory penalties was predictive of stated likelihood of offending but not correlated with perceptions of sanction severity, it is possible that some 207

218 construct, other than a participant s understanding of the criminal penalties prescribed for drink-driving offences was being measured by the multiple-choice questions asked in the survey. The accuracy of responses to these questions could be indicative of levels of general knowledge or intelligence rather than specific knowledge relating to maximum and minimum penalties. The former may also be associated with participants own beliefs about what the penalties should be for drink-driving offences rather than what they actually know the sanctions to be. It is also possible, however, that the weak association between legal knowledge and severity perceptions is due to the fact that the measure used to estimate perceived sanction severity was capturing more than just the disutility of State-imposed punishment. The question assessing perceived penalty severity was intentionally phrased in such a way as to invite participants to include in their estimates only the formal costs associated with a conviction for drink-driving. Many participants, despite these instructions, may have incorporated other informal costs into their estimates of penalty severity, such as the cost of a conviction on future job prospects. If this were the case, then the significant relationship between knowledge of statutory penalties and stated offending likelihood may still be evidence for a deterrent effect, even though no association was found between statutory knowledge and perceptions of penalty severity. Nevertheless, what is clear from the results shown above is that the effect of perceived certainty of apprehension on stated offending likelihood was much more robust than that found for perceptions of penalty severity. This finding - that certainty matters more than severity - is a recurring theme throughout the perceptual research and indeed much of the fieldwork conducted in the deterrence domain. There are two potential reasons why survey research has failed to find a significant effect of the perceived severity of criminal punishment. Firstly, as mentioned previously in this thesis, being apprehended for an offence imposes more costs on an individual than the sanction imposed by the criminal justice system. Survey-based research suggests that there is potential for informal sanctions from peers and family, who might learn of the arrest, to deter offenders through evoking feelings of shame 208

219 and embarrassment (e.g. Miller & Iovanni 1994; Nagin & Pogarsky 2001). Unlike the formal sanctions imposed by the criminal justice system, these informal costs associated with arrest are likely to be incurred almost simultaneously with the offending episode. Given that people tend to discount future costs at a greater rate than they do current costs, it is possible that the informal costs arising from an arrest may be more immediate for many potential offenders than those associated with formal sanctioning (von Hirsch et al. 1999). The second reason is that there may be insufficient variation in perceptions of the severity of criminal punishment to detect a significant effect. As seen from Figure 7.6, most participants in our survey perceived criminal penalties as being very severe. Over two-thirds of the respondents rated the penalties for drink-driving, on a scale of 100, at 70 or above. One-third rated them at 100. In contrast, even for the high-risk scenario, the average perceived risk of apprehension was estimated at less than 50 per cent. Again, it should be recognised that this is not a representative sample of NSW drivers. However, if a similar distribution of perceptions of penalty severity exists in the broader population then this suggests that enhanced RBT activity may have greater scope to influence offending decisions than do policies aiming to increase perceptions of sanction severity. Limitations of Study 3 Two further limitations of Study 3 should be kept in mind when assessing the implications of these findings for deterrence theory and criminal justice policy. Firstly, the survey sample used in this research is not representative of the wider driver population. A student-based sample was recruited for Study 3 on the premise that young people more frequently engage in risky drinking and driving behaviour. However, it is possible that for many of these university students, factors additional to those threatened by the criminal justice system would also be predictive of their likelihood of engaging in illegal behaviours. Indeed, other perceptual deterrence research has shown that people who have strong ties to conventional society are more likely to factor informal social sanctions (i.e. feelings of shame and embarrassment) arising from public exposure of the act itself into their decisions to offend, rather 209

220 than the formal costs associated with legal sanctioning once caught (Nagin & Paternoster 1994; Nagin 1998). Further exploratory analyses in Study 3, which partitioned out the effect of the deterrence variables for high-risk sub samples (i.e. risky drinkers and prolific offenders), improved the amount of variance accounted for by the estimated models. But the power to detect significant effects in these analyses, at least for perceptions of sanction severity, was substantially reduced by the small number of participants meeting the high-risk classification. The current research should therefore be extended to include surveys of active or convicted drink-drivers when trying to establish the importance of legal sanctions in models of offender decision-making. Secondly, in the scenario-based design used in Study 3, participants are asked only to estimate the likelihood that they would offend under the conditions described. Thus we are examining the relationship between legal knowledge and hypothetical offending rather than legal knowledge and actual episodes of drink-driving. A longitudinal study that measures participants knowledge of formal sanctions and then subsequent drink-driving occasions during a certain follow-up period would further expand on the work described here. 210

221 CHAPTER 8. SUMMARY AND DISCUSSION The major aim of this thesis has been to assess the deterrent efficacy of increases in the severity of criminal punishment. A considerable literature already exists on this topic. However, as highlighted in the introductory chapter to this thesis and in Chapter 3, the available evidence is far from conclusive. The current study improved on prior research by examining changes in the threat of formal punishment in a context where deterrent effects created by modifications to the severity of legal sanctions should be optimised. The case study selected for this research was a sentencing policy that substantially raised the severity of statutory penalties for drink-driving offences in New South Wales, Australia, in Drink-driving is an ideal offence to use in a case study of deterrence because it is an offence that (1) is instrumentally motivated and not viewed as inherently wrong or against the moral code of society (2) has a high risk of apprehension, prosecution and conviction (3) nevertheless results in very few offenders being imprisoned (thus minimising incapacitation effects) and (4) has valid and reliable measures of offending available. The specific policy examined in this case study also was thought to provide a more definitive test of deterrence hypotheses than previous investigations of the severity of formal punishment because it was (1) implemented in a jurisdiction where the perceived risk of detection for drink-driving offences is relatively high (2) intended to affect only the severity of formal punishment but not the probability of punishment and (3) designed to make licence sanctions mandatory, and therefore unavoidable, for all persons convicted of drink-driving in New South Wales. Furthermore, the quasi-experimental methods used in this research were able to overcome several limitations of previous ecological investigations of punishment severity because they involved a specific well-defined intervention, included comparison groups to control for other possible influences on crime rates and, in the case of the interrupted time-series analyses, measured crime at many points before and after the intervention (thereby making it possible to distinguish deterrent effects 211

222 from other fluctuations in offending levels). Another strength of this thesis is that data from the quasi-experiment were supplemented by data collected in a scenariobased survey. This survey explored an area of deterrence which, to date, has received little attention in the empirical literature, namely the relationship between sentencing policy, sanction perceptions and criminal behaviour. This chapter summarises the results from the three studies undertaken to examine the marginal deterrent effect of the changes made to drink-driving penalties in New South Wales. The findings from these studies are presented within the broader context of deterrence theory and the implications for criminal justice policy and future research are discussed. The deterrent effect of increased drink-driving penalties in New South Wales Changes in drink-driving behaviour after the penalty increase Two aggregate-level studies were undertaken to determine the impact of the 1998 penalty changes on drink-driving offending rates in New South Wales. Study 1 assessed changes in alcohol-related road crashes that were contemporaneous with the penalty increase. Study 2 examined variations in reoffending rates before and after the legislative changes were implemented. It was expected that, if the sentencing policy had the intended deterrent effect on drink-drivers, then a reduction should be apparent in both alcohol-related road crashes and drink-driver recidivism rates after the penalties were raised. The interrupted time-series analyses of road crash data (Study 1) showed that, prior to the penalty changes being implemented in New South Wales, there was a significant downward trend in two surrogate measures of alcohol-related road crashes: single-vehicle night-time crashes and crashes resulting in a fatality. After the enactment of the new penalty regime in September 1998, there was a significant change in one of these surrogate measures, with the monthly rate of single-vehicle night-time crashes beginning to rise. The magnitude of this increase (though 212

223 relatively small) appeared to be inconsistent with the hypothesised deterrent effect of the new sentencing policy. Further analyses, however, indicated that non alcohol-related road crashes (as measured by multiple-vehicle day-time crashes) were also increasing after the intervention point and, more importantly, appeared to be increasing at a greater rate than single-vehicle night-time crashes. This additional observation offers a possible explanation for the unexpected results found in Study 1. Single-vehicle night-time crashes are more likely to involve alcohol than other types of crashes but a substantial proportion of these road crashes are unrelated to alcohol. As such, if there was an overall rise in non alcohol-related crashes in New South Wales after the penalties were increased in 1998, then it would be reasonable to expect some residual effect to also be evident in our surrogate measure of alcohol-related crashes. Two possible conclusions regarding the effect of the sentencing policy on alcohol-related road crashes could be drawn from these findings: (1) there was a reduction in alcohol-related crash rates after the penalty changes but this reduction was overwhelmed by the rise in non alcohol-related crashes and therefore could not be detected in our surrogate measure or (2) there was no change in alcohol-related road crashes after the sentencing policy was implemented. Study 2 compared the probability of reoffending and the time to new offence for two drink-driver cohorts, one consisting of persons convicted of drink-driving in the year before the penalty changes and the other convicted in the year after. Data from this study showed that, controlling for other potentially confounding variables, offenders convicted under the new penalty regime were less likely and took longer to reappear before the court for a new drink-driving offence than those convicted prior to the legislative amendments. This effect, however, was only apparent for drink-drivers residing in country and regional areas and the size of the effect was small, with the probability of reoffending being reduced by just three percentage points. No differences in reoffending were apparent for drink-drivers residing in the Sydney metropolitan region. 213

224 The significant reduction in reoffending after the drink-driving penalties were increased in 1998, which was found in Study 2, is consistent with the possibility that the tougher penalties resulted in fewer alcohol-related road crashes but the surrogate measure used in Study 1 was unable to detect this effect. However, the fact that drink-driving recidivism was reduced by only a small amount and only for offenders residing outside the Sydney Metropolitan area suggests that any decrease in alcoholrelated road crashes after the intervention is likely to have been relatively small. Thus, the results from these quasi-experimental studies provide only limited support for a marginal deterrent effect created by the more severe sanctions. Implementation of the 1998 sentencing policy Previous studies have identified poor policy implementation as a potential explanation for the failure of tougher sentencing policies to achieve their deterrence aims. Changes to statutory punishments have, in other words, failed to affect the actual penalties convicted offenders receive from the court (e.g. Ross 1984; MacCoun 1993; Nagin 1998) or have produced unintended effects that undermine the certainty of apprehension, prosecution and conviction of the targeted offence (e.g. Barber & Wilson 1968; Ross & Voas 1990; Beirness, Simpson & Mayhew 1993; Mayhew, Beirness & Simpson 1995; Voas 2001). A comparison of drinkdriving prosecutions and court imposed sanctions before and after the penalty changes in New South Wales suggested that the 1998 sentencing policy examined in this thesis was, for the most part, implemented as intended by its formulators. Additional analyses, reported in connection with Study 2, for example, showed that average drink-driving penalties imposed by NSW courts in 1999 were significantly higher than those imposed in Gaol terms were found to have increased by 19 per cent (or almost one month), fines by 47 per cent (or approximately $240) and licence disqualification periods by 16 per cent (or two months). In fact, for drinkdrivers sentenced in non-sydney courts, the increase in average licence disqualification periods was found to be even greater at 24 per cent (or almost three months). Given previous research suggesting that licence disqualification is the most effective sanction for deterring drink-drivers (Nichols & Ross 1990; Zaal 1994), the longer disqualification periods imposed by magistrates from non-sydney courts may 214

225 go some way in explaining the reduced recidivism of offenders residing in country and regional areas. In terms of the prosecution and conviction of drink-driving offences, there was also generally no evidence for the new sentencing laws being undermined by players within the criminal justice system. Comparing prosecutions in 1997 with those in 1999 showed a significant increase in the number of drink-drivers brought before the courts, a slight increase in the proportion of offenders found guilty and no change in court delay or not guilty pleas being submitted to the court. The only setback in terms of the deterrence aims of the legislation was a small but statistically significant increase in the proportion of drink-drivers who escaped a licence disqualification upon being found guilty. This was particularly evident for offences where mandatory minima were introduced as part of the 1998 legislative changes (i.e. special- and lowrange drink-driving offences). If all convicted drink-drivers received a licence disqualification as intended by the sentencing policy then we may have seen a greater impact of the penalty increase on reoffending rates. Nevertheless, four out of five drink-drivers convicted under the new penalty regime did receive a licence disqualification, which was, on average, longer than their 1997 counterparts. Two further possible reasons why the sentencing policy failed to have a larger impact on drink-driving rates in New South Wales were also explored in Chapter 6. The first of these concerns the level of enforcement at and around the time that the legislative amendments were enacted. Modifications to the perceived risk of detection and apprehension have been shown to have a significant effect on offending rates of drink-drivers (e.g. Ross 1984; Voas & Hause 1987; Henstridge, Homel & Mackay 1997; Voas, Holder & Gruenewald 1997). Thus any substantial reduction in drinkdriving enforcement practices over the study period could have compromised the intended deterrent effect of the penalty changes. From an examination of breath testing data, it would seem that the intensity of drink-driving enforcement did drop off in New South Wales around the time that the drink-driving penalties were raised. From 1994 through 1996 the total number of breath tests conducted by NSW Police increased from 2.4 to 2.7 million but from 1997 onwards they began to decline, dropping to a low of 1.9 million tests in 1999 and 2.0 million tests in This 215

226 decline in yearly testing rates was also reflected in people s perceptions of drinkdriving enforcement activity, with fewer residents from New South Wales reporting in the 1999 ATSB survey that they had seen RBTs in operation and more reporting that they thought RBT activity had decreased. For a sanctioning policy to meet its deterrence objectives, perceptions of sanction severity must be increased and this can only occur if the target population is aware of the policy change (McLean, Kloeden & McCaul 1991; Voas, Holder & Gruenewald 1997; Nagin 1998). A media analysis of the publicity surrounding the penalty changes revealed that, from the beginning of June 1997 to the end of June 1999, 139 drink-driving articles appeared in New South Wales Metropolitan newspapers but only 15 of these described the details of the 1998 penalty increases. Furthermore, where the sentencing policy was discussed it appeared to be ineffectively communicated. It was packaged with more minor traffic offences and was accompanied by other drink-driving articles that questioned the certainty of more severe penalties being imposed by the courts. Also, over the same time period, there was very little discussion of the penalty changes on radio. Road traffic licensing authorities were aware of only one publicity campaign, an emotive TV advertisement, that was run in conjunction with the penalty increase. Although public risk perceptions could not be assessed in this study, the scarce and sometimes negative publicity that the policy received, raises the possibility that public awareness of the new penalties was not sufficient to produce any real change in the level of drink-driving amongst the general motoring population. Knowledge of drink-driving penalties and perceptions of sanction severity Aggregate-level analyses of changes in offending rates comprised a major component of this thesis. These types of investigations are essential in assessing the extent to which deterrence theory can be applied to criminal justice policy. As Tittle (1980) notes, in the final analysis, laboratory manipulations are too artificial and removed from everyday life to satisfy serious scholars who try to judge the work of the criminal justice system (Tittle 1980, p. 386). However, our research 216

227 demonstrates the inherent difficulties in measuring deterrent effects in a real world situation where the researcher has little or no control over the behavioural event: The problem is this, of course, that useful comparisons of a rate of crime before and after the changes in punishment policy depend, in each case, on the assumption that any changes noted in the rate of a particular threatened behaviour can be attributed to a specific shift in a facet of punishment policy, and to that change exclusively. The world rarely holds still for the researcher in this way (Morris & Zimring 1969, p. 145). Reduced levels of drink-driving enforcement activity and poor knowledge of the legislative amendments remain potential explanations for the minimal impact of the 1998 sentencing policy on drink-driving offending rates. Yet while both are plausible accounts for the small effect observed in the quasi-experimental research there is no way of knowing whether the marginal deterrent effect of the policy would have been any greater had the penalty changes been more widely known and enforcement intensity did not change during their implementation. The only way to assess this possibility was to examine the relationship between sentencing policy, sanction perceptions and behaviour in a setting where other potentially confounding contextual variables could be controlled. The third study in this research attempted to achieve this aim by seeing whether current knowledge of drink-driving penalties influenced perceptions of the severity of drink-driving punishment and the stated likelihood of offending in a hypothetical scenario. It was anticipated that those who knew more about the current penalty regime in New South Wales would perceive drink-driving punishment as more severe and thus would be less likely to state they would offend. Again, however, there was only minimal evidence for the hypothesised deterrent effect of penalty severity on willingness to drive while intoxicated. Instead, consistent with previous research, the deterrent effect observed for increases in the perceived risk of apprehension was much more robust (Paternoster 1989; Grasmick & Bursik 1990; Nagin & Pogarsky 2001). 217

228 Study 3, for example, found a significant relationship between legal knowledge and offending, as evidenced by the fact that respondents who made more mistakes in a test of knowledge about drink-driving penalties were more likely to state that they would drive home in the scenario described, even after controlling for other confounding variables. Moreover, knowledge of the actual penalties that drinkdrivers receive upon conviction was found to be a better predictor of behavioural intentions than knowledge of the statutory maxima and minima stipulated by drinkdriving legislation. Prima facie, these results are consistent with deterrence hypotheses and would lead one to conclude that, had the 1998 sentencing policy introduced in New South Wales been more widely publicised, the penalty increases may have had a greater impact on levels of offending. This conclusion is supported by the fact that, consistent with data from other surveys conducted in New South Wales (Bryant, Hawkins & ACNielsen 2004), the level of knowledge of drinkdriving penalties found in Study 3 was generally quite low. It should be noted, however, that legal knowledge is thought to deter offending only through its impact on perceptions of sanction severity. Our analyses of the relationship between knowledge scores and ratings of penalty severity in Study 3 revealed weak associations between these variables. Only the correlation between knowledge of actual penalties and perceived severity of drink-driving penalties was significant at the 0.05 level. Furthermore, when severity perceptions were included in the regression models predicting stated likelihood of offending, only the difference between those respondents rating penalty severity as low and those rating it as very high (between 91 and 100 out of 100) was approaching significance. There was no significant interaction effect between perceived risk of apprehension and perceptions of penalty severity. These somewhat contradictory findings from the scenario-based survey may be due in part to measurement error in the knowledge variable. The multiple-choice questions employed in this research were designed to assess current knowledge of the applicable penalties for drink-driving offences. However it is possible that the knowledge scores were measuring factors other than legal knowledge that have a 218

229 negative relationship with offending likelihood but are not related to perceptions of punishment severity. For example, individuals who have a higher level of intelligence may be more skilled at answering these types of questions but also be less likely to commit a criminal offence. An alternative explanation for the study findings is that respondents incorporated other informal costs arising from apprehension and conviction for a drink-driving offence into their ratings of penalty severity. Previous perceptual research has clearly demonstrated that anticipated social sanctions, like embarrassment and shame, can operate to deter some individuals from offending (e.g. Grasmick & Bursik 1990; Grasmick, Bursik & Arneklev 1993; Nagin & Pogarsky 2001). If these additional costs were included in respondents ratings of severity then we would not necessarily expect to find a very strong relationship between legal knowledge and severity perceptions. The significant effect of knowledge found in Study 3 would therefore need to be further tested before confident conclusions regarding the role of legal knowledge in deterrence can be drawn. The only true clear result to emerge from the scenario-based survey is that increases in the perceived risk of apprehension reduce the likelihood of drink-driving, a finding that is consistent with the extant research on deterrence (e.g. Homel 1986; Nagin & Pogarsky 2001; Nagin & Pogarsky 2003). Simply knowing that police have been conducting frequent RBTs in the immediate area reduced stated likelihood of offending, in the described scenario, by as much as 10 per cent. For individuals who had previously driven whilst drunk or who frequently drank alcohol at high-risk levels, willingness to offend was reduced by an even greater amount (up to 25 per cent) when the perceived risk of apprehension was high. This suggests that perceptions regarding the risk of being punished may be more important in the offending decisions of high-risk individuals than they are for other groups of potential offenders. Implications for theory and practice The case study selected for examination in this thesis was considered a good real-life test of deterrence hypotheses because many of the preconditions for deterrence 219

230 success were thought to be in place. However, a number of contextual factors were identified in this research that could have served to undermine the penalty changes and which therefore preclude definitive conclusions regarding the deterrent efficacy of the legislative reforms. Perhaps the most important of these is poor policy implementation: the possibility that the increase in formal punishment had minimal impact on offending because it was either inadequately communicated to the target audience or implemented in such a way that the actual threat was not raised to the intended level. By addressing these barriers to successful implementation early on in policy development it may be possible to enhance the effectiveness of deterrencebased interventions. With this in mind, the next section of this Chapter discusses some strategies that could be employed to increase the credibility of sanction-based policy. Translating policy into a credible threat (T)he success of any deterrence process will be determined by the degree to which this message is successfully transmitted to the population of offenders (Geerken & Gove 1975, p. 499; cited in Miller & Iovanni 1994). Survey data on public awareness of penalties for different types of crimes show that legal knowledge is generally quite poor. People often do not know whether particular types of penalties are applicable to certain offences or they substantially underestimate the magnitude prescribed by legislation (e.g. Williams, Gibbs & Erickson 1980; Snortum & Berger 1989; Kenkel 1993; Hough & Roberts 1998; Kenkel & Koch 2001). It is not surprising, then, that changes to penalties frequently result in little or no change in offender behaviour. Complete and accurate knowledge of criminal penalties is not necessarily a precondition for deterrence to work. However, if sentencing policies are to improve compliance with the law, there must at least be a general perception that the threat of punishment has increased. This can only be achieved if appropriate publicity and education campaigns accompany any changes that are made to the severity of statutory punishments. While seemingly commonsensical, this step is one that is often ignored during the development and implementation of new sentencing regimes. 220

231 An example whereby publicity has improved the effectiveness of modifications to the severity of formal punishment is presented by Beha (1977). This research examined the impact of a sentencing policy dictating mandatory minimum prison terms for defendants found guilty of illegally carrying a firearm in Massachusetts. This case study is particularly noteworthy because the sentencing changes were accompanied by publicity that not only educated citizens about the legal changes but also emphasised the certainty of the increased penalties being imposed, with the overall message conveyed to the public being that, once caught, nobody can get you out (Beha 1977, p. 98). Relevant media attention to the legislative amendments was generated by the office of the sponsor of the legislation (the then-speaker of the House), as well as by an advertising agency. Furthermore implementation of the changes was postponed long enough for this information to be adequately conveyed to the community. Beha s subsequent examination of the impact of the firearm legislation showed a dramatic decrease in the number of arrests and prosecutions for the target offence after the penalty rise, a decrease which did not appear to result from changes in enforcement or prosecutorial practices. There was also a substantial improvement in compliance with ancillary firearms laws, especially those dealing with the licensing and identification of personally owned firearms. Intensive media campaigns publicising sentencing changes can also have unintended beneficial effects on perceptions of the certainty of apprehension and conviction. Blomberg and his colleagues (1987), for example, considered the effect of a publicity campaign accompanying a new law mandating licence disqualifications for first-time drink-drivers on perceptions of the certainty and severity of drink-driving punishment. These authors found that media attention surrounding the new laws significantly improved general awareness of the mandatory licence sanctions, with knowledge of the law amongst survey respondents increasing from 50 per cent to over 80 per cent. However, they also found that survey participants reported an increase in the number of arrests and conviction for drink-driving after the penalty 221

232 changes were publicised, despite the media campaign failing to refer to this aspect of punishment and there being no change in arrest or conviction rates. 79 An alternative to mass media education campaigns is to communicate the increased threat directly to those most at risk of offending. This approach was used to great effect in the Boston gun control intervention, referred to earlier in this thesis, known as Operation Ceasefire (Braga et al. 2001; previously cited in Chapter 1). This intervention aimed to generate a strong message to chronic gang offenders that violence would no longer be tolerated and that formal punishment would immediately follow any such incidents. When violence did occur, several legal strategies were used to formally sanction the gang members involved. Other gangs were then informed that this gang did violence, we responded with the following actions, and here is how to prevent anything similar from happening to you (Braga et al. 2001, p. 200). Communication of the increased threat was achieved by way of formal meetings between members of the working group and gang members, through police, probation and outreach worker contacts with the gang members and through meetings with individuals in juvenile detention facilities. This program proved to be a success in deterring offenders, with evidence for substantial reductions in levels of gun crime in the Boston community and, in particular, gun crime involving young people. 80 Operation Ceasefire and Beha s (1977) study of the changes to Massachusetts firearm laws highlight another important requirement if a sentencing policy is to pose a credible threat: the belief that harsh punishment will be imposed once apprehended for an offence. The scenario-based survey described in Chapter 7 indicates that perceptions of sanction severity are more closely associated with knowledge of the 79 The reasons for these changes in perceived certainty of punishment are unclear. One possibility is that the publicity of the new penalties simply served as a reminder that criminal justice agencies are targeting drink-driving and, for those who choose to offend, formal punishment will ultimately ensue. These spillover effects do, however, present an additional challenge for researchers who are attempting to delineate the cause of any resultant change in behaviour given the overwhelming evidence for the deterrent effect of increases in the perceived certainty of punishment. 80 The extent to which this strategy can be used on a larger scale with more high volume crimes like drink-driving is unclear. It is possible that the credibility of the increased threat would be compromised once the size of the would-be offender population becomes too large and scarce resources are overstretched (Nagin 1998). The Boston intervention does, however, provide some promising evidence for improved communication of the formal consequences associated with offending altering perceptions and subsequent behaviour, even amongst persistent offenders. 222

233 actual penalties offenders receive upon conviction, than with the prescribed maximum statutory penalties. This suggests that, simply informing the public that formal penalties have increased, is insufficient for deterrence to occur. Offenders also need to believe that, once introduced, the court will impose these harsher penalties on all persons apprehended and prosecuted for the target offence. Strategies employed to improve public knowledge of penalty changes must therefore be designed to create this perception that severe punishment will be imposed. The media analysis presented in Chapter 6 suggests that achieving this change in perceptions is not necessarily an easy task. Much of the publicity drink-driving penalties receive in the media comes from cases where the court has failed to impose a penalty or has imposed a penalty that is viewed as particularly light. Focusing on these isolated but high-profile cases undermines the deterrent effectiveness of courtimposed sanctions. As the comments of the Chief Justice of New South Wales below illustrate, this problem appears to be relevant, not only to drink-driving offences, but to criminal offences more broadly, particularly in the jurisdiction where the case study for this thesis was situated: A key objective of criminal sentencing is, of course, deterrence. For deterrence to work, potential offenders must have an understanding of the likely consequences of criminal conduct. If, as I believe is the case, media reporting gives excessive emphasis to light sentences and gives the impression that such sentencing is typical, when it is not, then deterrence will not work (Spegilman 2005, p. 3). These remarks suggest that one way to improve the deterrent effect of penalties is not to increase them but to communicate the current threat of formal punishment for particular offences in a more effective way or more directly to the target audience. This could be by providing offenders, at the time of conviction, with information regarding the actual penalty they will receive if they reappear before the court or, alternatively, by actively publicising the average penalties being imposed in the courts in order to counterbalance media coverage of atypical cases. 223

234 The latter strategies are particularly attractive given that increases in the severity of statutory punishment can sometimes result in little or no change in the actual penalties that convicted offenders receive. The case study described in this thesis provides some evidence highlighting this possibility. The 1998 sentencing policy introduced minimum licence disqualification periods for two drink-driving offences which previously had not attracted such a mandate and doubled the minimums already applicable to other offences. These amendments resulted in an increase the average length of licence disqualification periods but did not appear to affect the proportion of drink-drivers whose licence was disqualified. In fact there was a slight decrease in the percentage of drink-drivers who had their licence disqualified after the penalty changes, an effect that was much more pronounced for offences where mandatory minimums had been introduced. The drink-driving literature offers strong evidence for a specific deterrent effect of licence disqualification, with drink-drivers who are disqualified from driving being involved in fewer alcohol-related crashes and recording fewer offences (e.g. Votey & Shapiro 1983; Blomberg, Preusser & Ulmer 1987; Mann et al. 1991; Siskind 1996). Thus, the failure of 1998 sentencing policy to affect the extent to which licence disqualification is applied to drink-driving offences is an obvious reason for the small marginal deterrent effect observed in the quasi-experimental study. Where there are wide disparities in sentencing (see Moffatt, Fitzgerald & Weatherburn 2004) defendants may come to believe that they can reduce the severity of punishment if they can get their case listed before a lenient judge. Greater consistency in the sentencing of drink-drivers may therefore also enhance the credibility of the threat generated by formal punishment. This could potentially be achieved through the development and application of sentencing guidelines for these types of offences (as has since happened in New South Wales), 81 or, through better judicial education on the deterrent efficacy of licence sanctions. As discussed in Chapter 3, there is good evidence starting to emerge from North America showing that administrative licence suspension laws can have a beneficial impact on alcohol- 81 Application by the Attorney General under Section 37 of the Crimes (Sentencing Procedure) Act for a Guideline Judgment Concerning the Offence of High Range Prescribed Concentration of Alcohol Under Section 9(4) of the Road Transport (Safety and Traffic Management) Act 1999 (No. 3 of 2002) [2004] NSWCCA

235 related road crashes and, to some extent, drink-driving recidivism rates. The deterrence mechanism by which this effect is achieved is not yet clear. Nevertheless, these laws have the potential to increase the perceived certainty of licence sanctions for drink-driving offences without necessarily changing the level of judicial discretion available in the sentencing process. An alternative approach therefore would be to introduce administrative penalties. The marginal deterrent efficacy of punishment severity It is true that the implementation problem limits our ability to draw definitive conclusions regarding the deterrent efficacy of the penalty increase examined in this thesis. With greater consistency in sentencing and greater publicity of the policy, the tougher penalties could have had a larger impact on drink-driving levels in New South Wales. It is also possible that the deterrent effect of the harsher punishment could have been further enhanced if NSW Police maintained a high level of enforcement around the time of the legislative reforms. Despite these limitations however, the results of our quasi-experimental research are consistent with those from many other aggregate-level studies reported in the deterrence literature (see von Hirsch et al for a review), in suggesting that penalty increases have limited marginal deterrent efficacy. The results from our scenario-based study (Study 3), which was able to examine punishment severity effects whilst controlling for other potentially confounding factors, is further support for this conclusion. Here it was found that the deterrent effect of the perceived severity of punishment was much less robust than the deterrent effect of the perceived risk of detection: a finding that is consistent with the extant perceptual research on deterrence. Thus, even if sentencing policies could be translated into a credible threat, the marginal deterrent effect of tougher penalties might still be relatively small. There are at least three possible reasons identified in the research literature why policies that raise the level of punishment severity might have limited efficacy in deterring offenders: (1) the vast majority of potential offenders are deterred by the anticipated informal social sanctions associated with public exposure of the offence 225

236 rather than the formal punishment prescribed by legislatures (2) there is a fixed cost associated with arrest for an offence and the penalties incurred later in the criminal justice process are mostly discounted by potential offenders (3) a severity threshold exists beyond which changes in criminal penalties will have no additional impact on offending likelihood. These three accounts for weak severity effects, and their relevance to the results from the case study, are discussed below. Since the introduction of RBT in New South Wales and Australia there has been a strong emphasis on the serious risk that drink-drivers pose to themselves and to other drivers on the road. For example, a recent national anti-drink-driving advertisement used the slogan if you drink and drive you are a bloody idiot, which was accompanied by graphic images of a road crash victim. It is possible that this portrayal of a drink-driver as a killer drunk has resulted in drink-driving behaviours being more widely condemned in the community and the social stigma attached to the behaviour therefore being more pronounced. With this increased stigmatisation of drink-driving behaviours, anticipated feelings of shame from violating an internalised norm would become more influential in deterring potential drink-drivers and the sanctions attached to formal apprehension and conviction for the offence less efficacious. Thus, during the early stages of RBT in Australia drink-driving levels may have decreased primarily because of fear of formal punishment but over time a social norm has developed that drink-driving is wrong and voluntary compliance with the law has improved (Ross 1984). While social stigma associated with the act is undoubtedly important in the offending decisions of some would-be drink-drivers, the significant effect of punishment certainty found in the scenario-based research described in Chapter 7 suggests that the fear of arrest is still a powerful deterrent for many people. In Study 3, respondents who were told that police had been conducting RBTs in the immediate area, on a frequent basis, were less likely to state that they would offend in the hypothetical scenario than respondents who were told that police had not been conducting RBTs in the area for some time. On the other hand, only weak deterrent effects were found for the perceived severity of formal punishment for drink-driving offences. This suggests that there are additional costs associated with public 226

237 exposure of the offence through conviction or mere apprehension which are serving to discourage many people from offending and which are independent of the formal costs prescribed by the criminal justice system. As discussed in previous chapters, these additional extralegal costs associated with an arrest would incorporate the sense of shame or feelings of embarrassment evoked by the knowledge that significant others will learn of the offence but might also include anticipated nonlegal costs such as reduced future job opportunities because of a criminal conviction or the negative impact of a criminal conviction on an individual s professional standing (e.g. registration as a health professional). For our student-based sample in the scenario-based study, as is perhaps true for the broader driver population, it is likely that these extralegal costs play an important role in their decisions to offend. The severity of legal punishment will then matter only to the extent that it can have an additional effect over and above this fixed cost associated with an arrest. Given also that informal social sanctions would be incurred almost simultaneously with an arrest, unlike criminal penalties which are a future contingent cost that tends to be discounted (von Hirsch et al. 1999), the marginal deterrent efficacy of sanction severity will be limited in many cases. Nevertheless, there are those individuals who frequently engage in risky behaviour (as do their peers) and who are less influenced by informal social sanctions. For these people (as argued in Chapter 2) punishment contingencies are expected to play a greater role in deterring further offending than extralegal costs. This is one possible explanation why the deterrent effect of the 1998 sentencing policy was observed only for a small group of high-risk offenders, that is convicted drink-drivers. Drinkdrivers with a previous conviction, being less constrained by informal social controls, may have modified their driving behaviour to a greater extent than did other NSW drivers once the new, more severe drink-driving penalties were introduced. Even amongst this group of high-risk offenders, however, the marginal deterrent efficacy of increases in the severity of formal punishment could still be somewhat limited, as the relatively small change in reoffending observed in Study 2 would suggest. Previous research has shown that impulsive individuals, being more present- 227

238 oriented, tend to discount future contingencies to a greater extent than other people (Nagin & Pogarsky 2001; Pogarsky, Piquero & Paternoster 2004) and that these same individuals are also more likely to commit crime (Grasmick et al. 1993; Keane, Maxim & Teevan 1993; Nagin & Paternoster 1993). As such, many offenders would place greater weight on the immediate benefits that could be gained from committing a crime than on the discounted formal costs threatened by the criminal justice system, when assessing the utility of different offending opportunities. Thus, even if many drivers with a previous conviction or a prior history of drink-driving were to perceive an increase in the formal costs associated with this offence, the fact that these increased costs would be substantially delayed, if indeed they are incurred at all (given that formal punishment is contingent on apprehension), means that a substantial proportion of this offender group might still choose to engage in the proscribed behaviour. A further constraint on the efficacy of punishment severity, relevant to both highrisk offenders and the broader offender population, is the possibility of diminishing returns at the extreme ends of the punishment severity continuum. According to subjective expected utility theory any increment in the severity of criminal penalties should raise perceptions of punishment severity, thereby increasing the disutility of offending and helping to tip the cost/benefit ratio. However, the relationship between actual penalty severity and perceived penalty severity is potentially non-linear. For example, two months in gaol may be perceived as twice as severe as one month but two years in gaol perceived as only 50 per cent more severe than one year. This would imply that the marginal deterrent effect of penalty changes on rates of offending will not simply be a function of the magnitude of the change. Instead, the response function relating punishment severity and crime rates might resemble more of a concave shape, like that shown in Figure 8.1. In this case, more severe penalties would still be associated with reductions in crime levels but the response to an incremental increase (the marginal effect) will vary depending upon the severity of the sanctions being altered. Small increments in less severe penalties would result in large changes in crime rates (P 0 P 1 ), but the same increment in very severe penalties would result in lesser reductions (P 1 P 2 ). 228

239 Figure 8.1: The marginal deterrent effect of changes in penalty severity Crime levels P 0 P 1 P 2 Penalty magnitude This issue of diminishing marginal returns for sentencing policies has particular relevance to drink-driving offences in New South Wales. The penalties stipulated for this offence are the most severe of any driving offence in this State and are tougher than those legislated by many other Australian jurisdictions. Thus, if potential offenders already regarded the penalties as exceedingly severe before 1998, then doubling these penalties is unlikely to have much additional, observable effect on their behaviour. A further point to note here is the possibility that this relationship between actual and perceived severity of formal punishment also is not continuous. Von Hirsch and his colleagues (1999) suggest that there may be a point or limit beyond which punishment severity no longer matters. Once this severity threshold has been reached or surpassed, further changes in penalty severity will no longer affect perceptions and people will begin to focus more on the risk of being caught when deciding to offend than on the formal costs that would be imposed upon conviction. The results from Study 3 provide some evidence that this severity threshold is being approached for drink-driving penalties, at least for the young people sampled in the this survey. When asked to indicate how much of a problem the current penalties for drink-driving offences would be for them, most of the survey participants judged the penalties to be very severe. Over two-thirds of the respondents rated the penalties (on a scale of 0 to 100) at 70 or above, one-third rated 229

240 them at 100. While this is not a representative sample of drink-drivers from New South Wales, if a similar distribution of perceptions of penalty severity exists in the broader offender population then policies aiming to increase perceptions of sanction severity will have limited potential to further influence offending decisions. Future directions for deterrence research and drink-driving policy This thesis has highlighted several areas of deterrence that require further investigation and has also offered some insight into effective, and less effective, strategies for addressing drink-driving and related issues. These future directions for research and policy are summarised in the following sections in conjunction with other non deterrence-based approaches that could also be employed to reduce the harm arising from drink-driving behaviours. Future research As several deterrence scholars have noted, very little is known about the factors influencing perceptions of risks, in particular the relationship between these perceptions and sanction-based policy (Miller & Iovanni 1994; Nagin 1998; von Hirsch et al. 1999; Pogarsky, Piquero & Paternoster 2004). The scenario-based research presented in this thesis examined one possible determinant of these perceptions, namely knowledge of legal penalties. As described previously, this study presents some evidence for a significant relationship between legal knowledge and stated offending likelihood but the correlations between perceptions of punishment severity and knowledge were weak, and only in the case of an individual s knowledge of actual penalties was this relationship significant. Further research validating the measures used in this scenario-based research is therefore needed in order to verify whether, controlling for other potentially confounding variables, knowledge of criminal justice policies and practices has the potential to affect risk perceptions and subsequent behaviour. This type of research should, however, be extended to include active offenders. With just a few exceptions (e.g. Horney & Marshall 1992; Wright et al. 2004), one of the major limitations of the work conducted in perceptual deterrence research thus far is 230

241 the use of student-based samples or general population surveys to examine deterrence hypotheses. Much of this work has found that informal social controls play a major role in decisions to offend. This is not surprising given that these respondents would typically have many close personal relationships and strong ties to conventional society and therefore have more to lose if their transgression is publicly exposed. On the other hand, there is a proportion of the population for whom these broader social controls are absent or ineffective and who, it has been argued in this thesis, may be more responsive to incentives and disincentives. These marginal offenders are the target group for deterrence-based interventions and, therefore, should also be the major focus of research investigating the link between sanctioning policy, risk perceptions and offending likelihood or intentions to offend. Further work is also needed on the extent to which improved communication of the current threat posed by the criminal justice system can serve to enhance deterrent outcomes. Research across various jurisdictions has shown that public knowledge of the penalties applicable to proscribed behaviours is generally quite low (Williams, Gibsb & Erickson 1980; Snortum & Berger 1989; Hough & Roberts 1998; Kenkel & Koch 2001). It is possible, as demonstrated in Study 3 of this thesis and in other work conducted in the drink-driving domain (see Kenkel & Koch 2001), that people who are being targeted by such laws have a better understanding of the legal consequences associated with different criminal behaviours. However, media reporting of atypical cases where penalties are light or of the inconsistency in sentencing practice could serve to undermine the perceived certainty that these penalties will be applied if apprehended for the offence. Thus, research needs to establish the extent to which the threat of formal punishment can be made more credible by communicating, directly to those at most risk of engaging in the behaviour, the actual penalties that will be imposed if one appears, or reappears, before the court for a particular offence. Improving offender knowledge of the legal penalties imposed by the courts will, however, only succeed in achieving a reduction in offending if perceptions of punishment severity increase and, further, increase enough to persuade some offenders that the costs associated with the behaviour have become too great. It is not 231

242 yet clear, however, from the available research on punishment severity and deterrence, just how much more punishment or what type of punishment is needed for this to occur. Small increases in the severity of formal punishment may have little effect on perceptions and offending levels. Large changes may have a greater impact on subjective utilities but might also depend on the severity level of the penalties being modified. If the current penalties applicable to an offence were already perceived as severe, only minimal, marginal deterrent effects might be achieved and if the penalties have already surpassed a severity threshold then it is possible that no change in crime levels would be observed. Alternatively, particular types of formal punishment may be perceived as more severe or more inconvenient than others and as such, certain penalties might be expected to be more efficacious in deterring offenders. Future deterrence research should therefore aim to more accurately quantify the relationship between penalty severity/type and crime levels. Policy recommendations Keeping in mind these limitations of the existing research on deterrence and formal punishment, the current work suggests two deterrence-based approaches that could be used to minimise the harm associated with drink-driving. Firstly, there is strong evidence for the marginal deterrent efficacy of increasing the perceived risk of drink-driving punishment. The enforcement of drink-driving offences has, in the words of Sherman and his colleagues (1998), been one of the great success stories of world policing (Sherman et al. 1998, p. 17). This success has been achieved not by increasing the actual probability of arrest for this offence (the size of the potential offender population would significantly reduce the credibility of such an approach) but by increasing the perceived risk of detection and apprehension. According to deterrence theorists, these risk perceptions are not stable but are continually updated on the basis of new experiences, which are both personal and vicarious in nature (e.g. Ross & LaFree 1986; Pogarsky, Piquero & Paternoster 2004). Encountering enforcement activity, or learning about it from others, will cause people to initially overestimate the risk of apprehension, but with repeat successful offending episodes there is a chance that people will begin to lower their perceptions of the probability of arrest over time. Sherman (1990) suggests that the 232

243 way to prevent this decline in risk perceptions is to keep people continually guessing about what the actual risk of detection might be at any one moment. In these situations of uncertainty, he anticipates that people will tend to play it safe and overestimate the risk of detection, thus decreasing the likelihood of offending if and when such an opportunity arises. The original version of RBT, adopted in Australia, was modelled on these basic principles of deterrence, with police enforcement of drink-driving consisting of breath testing operations situated in highly visible locations. Most importantly, the police continually varied the places where the testing sites were located and the times that the testing was undertaken, thus making it difficult for drivers to predict the chance of encountering RBTs on the roads at any particular point in time. The current research and literature review suggests that this strategy needs to continue as the dominant force in this area of drink-driving policy, if enforcement is to succeed in achieving a general deterrent effect. It is, as Homel (1986) warned almost 20 years ago, essential that policy makers and police authorities do not allow this general deterrence model of RBT to be superseded by the tendency of law enforcement officers to want to catch more offenders, as appeared to have happened in New South Wales toward the end of the 1990s. One way by which this could be achieved is through routine monitoring of breath testing operations and assessment of their effect on driver perceptions and/or behaviours. This information could then be provided on a regular basis to all officers involved in the enforcement of drink-driving to reinforce the preventative value of undertaking random breath tests of drivers and to, ultimately, shape future planning of enforcement activities in this area. Secondly, the formal threat posed by the criminal justice system should be communicated more effectively to high-risk offenders, who may be less influenced by the informal social sanctions attached to a drink-driving arrest. This need not be through increases in the severity of statutory sanctions, particularly in light of possible countervailing effects that could result from the implementation of such a strategy, but rather transforming the current threat of formal punishment into one which is more credible. Improving knowledge of the actual penalties that will be applied for drink-driving offences and increasing the perceived certainty that these 233

244 harsh penalties will be imposed if caught, particularly amongst individuals at-risk of drink-driving (e.g. young, heavy drinkers and previous offenders), could curtail the frequency of offending. Even so, the consistency with which licence sanctions are imposed by magistrates for drink-driving offences also would need to be improved, through sentencing guidelines or even the introduction of administrative penalties, for this approach to be most convincing. As discussed previously, strategies focusing on the severity of legal sanctions could be relatively limited in their ability to reduce offending levels, particularly if penalties are already severe. Therefore other drink-driving countermeasures that target hard-core or persistent offenders should also be considered. One such strategy is to combine traditional punishments for repeat offenders with alcohol treatment programs. Wells-Parker et al. (1995), who undertook a meta-analysis of 215 independent evaluations of drink-driving remediation programs, estimates that alcohol treatment (e.g. psychotherapy, counselling, education) combined with licence suspensions could reduce recidivism by between seven and nine per cent compared to licence sanctions alone. Alcohol treatment programs identified as the most effective were those that were well-structured, combined several treatment modalities and provided follow-up services after treatment completion (Wells-Parker 2000). These findings should not, however, be interpreted as evidence that treatment programs should be mandatory for all convicted drink-drivers. Instead, given the relatively low recidivism rate overall amongst these offenders, alcohol treatment should be considered simply as an another option for dealing with the hard core drink-driver for whom other more traditional methods may prove ineffective. Another drink-driving intervention that has been shown to be relative successful in reducing the offending of repeat, high-risk offenders is alcohol ignition interlocks. These devices are fitted to the ignition of a drink-driver s vehicle and prevent a vehicle from being started until a low alcohol or alcohol-free breath sample has been provided. Once the ignition has been started further tests are required every few minutes, thereby preventing a sober person from starting the engine on behalf of the impaired driver. Voas et al. (1999) reviewed eight studies evaluating the effectiveness of these sanctions and found that offenders who were ordered to use 234

245 alcohol interlocks were less likely to be reconvicted for a drink-driving offence than were offenders who received only a licence sanction upon conviction. Despite these promising findings, however, these reductions in offending were evident only during the period in which the device was operational. Once the interlock was removed, there were no significant differences in recidivism between those offenders who used an interlock and those who received more conventional punishments from the courts. Though it should also be noted that the relative risk of crash involvement for recidivist drink-drivers is more than double that for other drivers (Fetherson, Lenton & Cercarelli 2002). Therefore any modification in their drink-driving behaviour would be of public health significance. Finally, as discussed at the outset of this thesis, punishment is not the only mechanism by which policy affects offending behaviours, and accordingly criminal justice interventions should not be considered as the only means by which drinkdriving can be controlled. Other more macro-level policies also have the potential to affect levels of drink-driving through their impact on alcohol consumption and should therefore be considered alongside police enforcement and court-imposed punishments. One such policy shown to be particularly effective in reducing alcohol-related road crashes is taxation on alcoholic beverages. Ecological studies conducted in the USA have demonstrated that, net of other factors, increases in alcohol excises are associated with substantial reductions in alcohol-related road fatality rates (Chaloupka, Saffer & Grossman 1993), and, particularly, motor vehicle fatalities involving young people (Saffer & Grossman 1987). Kenkel (1993) estimates from his research that a 10 per cent increase in the price of alcohol would reduce the probability of drink-driving in the USA by about seven per cent for males and eight per cent for females, and that the impact of this change on young drivers would be even greater. Other empirical research has also shown that increases in the physical availability of alcohol are associated with increased levels of alcohol-related harm, including harms associated with drink-driving. Gruenewald and his colleagues (1996), for example, demonstrated in their cross-sectional research conducted in the USA that alcohol-related road crashes are more likely to occur where there is a high 235

246 density of alcohol outlets. Meanwhile, Australian quasi-experimental research has shown that alcohol-related crashes and drink-driving incidents increase with longer hours (Chikritzhs, Stockwell & Masters 1997) and more days (Smith 1988) of alcohol sale. Thus, measures designed to decrease the physical availability of alcohol could also have an additional beneficial effect on the level of drink-driving in a given community. The research presented throughout this thesis suggests that deterrence-based strategies, particularly those focusing on the severity of formal punishment, may have limited potential to achieve further reductions in alcohol-related road injuries and fatalities, at least within the Australian context. As such, the non-deterrence based strategies, briefly outlined above, should also be given due consideration when identifying future priorities in this area of public policy. 236

247 References Achen, C.H. 1982, Interpreting Regression, Sage Publications, Beverley Hills. Akers, R.L., Krohn, M.D., Lanza-Kaduce, L. & Radosevich, M. 1979, Social learning and deviant behavior, American Sociological Review, vol. 44, pp Andenaes, J. 1966, The general preventive effects of punishment, University of Pennsylvania Law Review, vol. 114, no. 7, pp Andenaes, J. 1971, Deterrence and specific offenses, The University of Chicago Law Review, vol. 38, pp Andenaes, J. 1974, Punishment and Deterrence, University of Michigan Press, Ann Arbor. Australian Institute of Health and Welfare 2002a, 2001 National Drug Strategy Household Survey: First results, Australian Institute of Health and Welfare, Canberra. Australian Institute of Health and Welfare 2002b, 2001 National Drug Strategy Household Survey: State and Territory supplement, Australian Institute of Health and Welfare, Canberra, ACT. Australian Transport Safety Bureau 2001, Alcohol and Road Fatalities, Australian Transport Safety Bureau, Canberra, ACT. Bachman, R., Paternoster, R. & Ward, S. 1992, The rationality of sexual offending: Testing a deterrence/rational choice conception of sexual assault Law and Society Review, vol. 26, pp Barber, R. N. & Wilson, P. 1968, Deterrent aspect of capital punishment and its effect on conviction rates: The Queensland experience, Australian and New Zealand Journal of Criminology, vol. 1, no. 2, pp Baum, S. 1999, Self-reported drink driving and deterrence, The Australian and New Zealand Journal of Criminology, vol. 32, no. 3, pp Baum, S. 2000, Drink driving as a social problem: Comparing the attitudes and the knowledge of drink driving offenders and the general community, Accident, Analysis and Prevention, vol. 32, pp

248 Becker, G. S. 1968, Crime and punishment: An economic approach, Journal of Political Economy, vol. 76, pp Beha, J. A. 1977, "And nobody can get you out"; The impact of a mandatory prison sentence for the illegal carrying of a firearm on the use of firearms and on the administration of criminal justice in Boston - Part 1, Boston University Law Review, vol. 57, pp Beirness, D. J., Simpson, H.M. & Mayhew, D.R. 1993, Assessment of the Impact of the 1985 Amendments to the Drinking and Driving Section of the Criminal Code of Canada, Department of Justice Canada, Ottawa. Benson, B. L., Mast, B.D. & Rasmussen, D.W. 2000, Can police deter drunk driving?, Applied Economics, vol. 32, pp Benson, B. L., Rasmussen, D.W. & Mast, B.D. 1999, Deterring drunk driving fatalities: An economics of crime perspective, International Review of Law and Economics, vol. 19, pp Beyleveld, D. 1979, Identifying, explaining, and predicting deterrence, British Journal of Criminology, vol. 19, no. 3, pp Birrell, J. L. 1974, Drinking, Driving and You, Sun Books, Melbourne, Victoria. Blomberg, R. D., Preusser, D.F. & Ulmer, R.G. 1987, Deterrent Effects of Mandatory Licence Suspension for DWI Conviction, National Technical Information Service, Springfield, Virginia. Blumstein, A., Cohen, J. & Nagin, D. 1978, Deterrence and Incapacitation: Estimating the Effects of Criminal Sanctions on Crime Rates, Report prepared for the National Research Council's Panel on Research on Deterrent and Incapacitative Effects 1978, National Academy of Sciences, Washington DC. Borschos, B. 2000, An evaluation of the Swedish drunken driving legislation implemented on 1/02/94, paper presented at the 15th International Conference on Alcohol, Drugs and Traffic Safety, May 22-26, Stockholm, Sweden. Box, G. E. P., Jenkins, M. & Reinsel, G.C. 1994, Time-Series Analysis: Forecasting and Control, Prentice Hall, New Jersey. Braga, A. A., Kennedy, D.M., Waring, E.J. & Piehl, A.M. 2001, Problem-oriented policing, deterrence, and youth violence: An evaluation of Boston's operation ceasefire, Journal of Research in Crime and Delinquency, vol. 38, no. 3, pp

249 Briscoe, S. 2004a, Raising the bar: Can increased statutory penalties deter the drinkdriver?, Accident, Analysis and Prevention, vol. 36, no. 5, pp Briscoe, S. 2004b, The impact of increased drink-driving penalties on recidivism rates in NSW, Alcohol Studies Bulletin, no. 5, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Bryant, P., Hawkins, A. & ACNielsen 2004, Drink-driving attitudes and selfreported behaviours in NSW drivers, paper presented at the Road Safety Research, Policing and Education Conference, Nov , Perth, WA. Carroll, J. S. 1978, A psychological approach to deterrence: The evaluation of crime opportunities, Journal of Personality and Social Psychology, vol. 36, no. 12, pp Cashmore, J. 1985, The Impact of Random Breath Testing in NSW, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Chaloupka, F. J., Saffer, H. & Grossman, M. 1993, Alcohol control policies and motor-vehicle accidents, Journal of Legal Studies, vol. 22, pp Chaloupka, F. J. & Wechsler, H. 1996, Binge drinking in college: The impact of price, availability and alcohol control policies, Contemporary Economic Policy, vol. 14, no. 4, pp Chambliss, W. 1967, Types of deviance and the effectiveness of legal sanctions, Wisconsin Law Review, Summer Issue, pp Chan, J. B. L. & Oxley, D. 2004, The deterrent effect of capital punishment: A review of the research evidence, Contemporary Issues in Crime and Criminal Justice, no. 84, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Chikritzhs, T., Stockwell, T., Heale, P., Dietze, P. & Webb, M. 2000, National Indicators Project Technical Report No. 2: Trends in Alcohol-Related Road Injury in Australia, , National Drug Research Institute and Turning Point Alcohol and Drug Centre Inc., Curtin University of Technology, Perth, WA. Chikritzhs, T., Catalano, P. Stockwell, T., Donath, S., Young, D. & Matthews, S. 2003, Australian Alcohol Indicators, ; Patterns of alcohol Use and Related Harms for Australian States and Territories, National Drug Research Institute, Curtin University of Technology, Perth, WA. 239

250 Chikritzhs, T. N., Stockwell, T.R. & Masters, L. 1997, Evaluation of the public health and safety impact of extended trading permits for Perth hotels and night-clubs, National Drug Research Institute, Curtin University of Technology, Perth, WA. Clarke, R. V. & Cornish, D.B. 1985, Modeling Offenders' Decisions: A Framework for Research and Policy, in M. Tonry & N. Morris (eds), Crime and Justice: A Review of Research, vol. 6, The University of Chicago Press, Chicago, pp Clarke, R. V. & Felson, M. 1993, Introduction: Criminology, Routine Activity, and Rational Choice in R.V. Clarke & M. Felson (eds), Routine Activity and Rational Choice: Advances in Criminological Theory, vol. 5, Transaction Publishers, New Brunswick, New Jersey, pp Collins, D. J. & Lapsley, H.M. 2001, Counting the cost: Estimates of the Social Cost of Drug Abuse in Australia in , Commonwealth Department of Health and Ageing, Canberra, ACT. Cook, P. J. 1980, Research in criminal deterrence: Laying the groundwork for the second decade, in N. Moms & M. Tonry (eds), Crime and Justice: A Review of Research, vol. 2, University of Chicago Press, Chicago pp Cook, T. D. & Campbell, D.T. 1979, Quasi-Experimentation: Design and Analysis Issues for Field Settings, Houghton Mifflin Company, Boston. Corbett, C. & Simon, F. 1992, Decision to break or adhere to the rules of the road, viewed from the rational choice perspective, British Journal of Criminology, vol. 32, no. 4, pp Decker, S., Wright, R. & Logie, R. 1993, Perceptual deterrence among active residential burglars: A research note, Criminology, vol. 31, no. 1, pp DeJong, W. & Hingson, R. 1998, Strategies to reduce driving under the influence of alcohol, Annual Review of Public Health, vol. 19, pp DeYoung, D. J. 1997, An evaluation of the effectiveness of alcohol treatment, driver license actions and jail terms in reducing drunk driving recidivism in California, Addiction, vol. 92, no. 8, pp

251 Douglas, M. & Wildavsky, A. 1982, Risk and Culture: An Essay on the Selection of Technical and Environmental Dangers, University of California Press, London, England. Durkheim, E. 1964, The Rules of Sociological Method, Free Press, New York. Ehrlich, I. 1973, Participation in illegitimate activities: A theoretical and empirical investigation, The Journal of Political Economy, vol. 81, no. 3, pp Ehrlich, I. 1975, The deterrent effect of capital punishment: A question of life and death, The American Economic Review, vol. 65, no. 3, pp Elliott, B. 2003, Deterrence theory revisited, paper presented at the Road Safety Research, Policing and Education Conference, Sept , Sydney, N.S.W.. Enders, W. 1995, Applied Econometric Time Series, Wiley & Sons, New York. Evans, L. 1991, Traffic Safety and the Driver, Van Nostrand Reinhold, New York. Evans, W. N., Neville, D. & Graham, J.D. 1991, General deterrence of drunk driving: Evaluation of recent American policies, Risk Analysis, vol. 11, no. 2, pp Exum, M. L. 2002, The application and robustness of the rational choice perspective in the study of intoxicated and angry intentions to aggress, Criminology, vol. 40, pp Farrington, D. 1983, Randomized experiments on crime and justice, in M. Tonry & N. Morris (eds), Crime and Justice: A Review of Research, vol. 4, University of Chicago, Chicago, pp Farrington, D. & Langan, P.A. 1992, Changes in crime and punishment in England and America in the 1980s, Justice Quarterly, vol. 9, no. 1, pp Fetherson, J., Lenton, S. & Cercarelli, L.R. 2002, The Repeat Drink Drivers Study, National Drug Research Institute, Curtin University of Technology, Perth, WA. Gibbs, J. P. 1975, Crime, Punishment and Deterrence, Elsevier Scientific Publishing Company, New York. Gottfredson, M. & Hirschi, T. 1990, A General Theory of Crime, Stanford University Press, Stanford, California. Grasmick, H. G. & Bryjak, G.J. 1980, The Deterrent Effect of Perceived Severity of Punishment, Social Forces, vol. 59, no. 2, pp

252 Grasmick, H. G. & Bursik, R.J. 1990, Conscience, significant others, and rational choice: Extending the deterrence model, Law and Society Review, vol. 24, no. 3, pp Grasmick, H. G., Bursik, R. J. & Arneklev, B.J. 1993, Reduction in drunk driving as a response to increased threats of shame, embarrassment, and legal sanctions, Criminology, vol. 31. no. 1, pp Grasmick, H. G., Tittle, C.R., Bursik, R.J. & Arneklev, B.J. 1993, Testing the core empirical implications of Gottfredson and Hirschi's general theory of crime, Journal of Research in Crime and Delinquency, vol. 30, no. 1, pp Grosvenor, D., Toomey, T.L. & Wagenaar, A.C. 1999, Deterrence and the Adolescent Drinking Driver, Journal of Safety Research, vol. 30, no. 3, pp Gruenewald, P.J., Miller, A.B. & Roeper, P. 1996, Access to alcohol: geography and prevention for local communities, Alcohol Health and Research World, vol. 20, pp Heckathorn, D.D. 1990, Collective sanctions and compliance norms: A formal theory of group-mediated social control, American Sociological Review, vol. 55, pp Henstridge, J., Homel, R. & Mackay, P. 1997, The long term effects of RBT in 4 Australian states: A time series analysis, Federal Office Of Road Safety, Canberra, ACT. Hingson, R., Heeren, T., Kovenock, D., Mangione, T., Meyers, A., Morelock, S., Lederman, R. & Scotch, N. A. 1987, Effects of Maine's 1981 and Massachusetts' 1982 Driving-Under-the-Influence legislation, American Journal of Public Health, vol. 77, no. 5, pp Hirschi, T. 1969, Causes of Delinquency, University of California Press, Berkeley. Holden, R. T. 1983, Rehabilitative sanctions for drunk driving: An experimental evaluation, Journal of Research in Crime and Delinquency, vol. 20, pp Homel, R. 1980, Penalties and the Drink Driver: A Study of One Thousand Offenders, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Homel, R. 1986, Policing the Drinking Driver: Random Breath Testing and the Process of Deterrence, Federal Office of Road Safety, Canberra, ACT. 242

253 Homel, R. 1988, Policing and Punishing the Drinking Driver: A Study of General and Specific Deterrence, Springer-Verlag, New York. Homel, R. 1994, Drink-driving law enforcement and the legal blood alcohol limit in NSW, Accident, Analysis and Prevention, vol. 26, no. 2, pp Homel, R. 2004, Drinking and driving, in T. Babor, R. Caetano, S. Casswell, G. Edwards, N. Giesbrecht, K. Graham, J. Grube, P. Gruenewald, L. Hill, H. Holder, R. Homel, E. Osterberg, J. Rehm, R. Room & I. Rossow (eds), Alcohol: No Ordinary Commodity, Oxford University Press, London, pp Homel, R., Carseldine, D. & Kearns, I. 1988, Drink-driving countermeasures in Australia, Alcohol, Drugs and Driving, vol. 4, no. 2, pp Horney, J. & Marshall, I.H. 1992, Risk perceptions among serious offenders: The role of crime and punishment, Criminology, vol. 30, no. 4, pp Hough, M. & Roberts, J. 1998, Attitudes to punishment: findings from the British Crime Survey, Home Office, London, England. Howe, E. S. & Brandau, C.J. 1988, Additive effects of certainty, severity and celerity of punishment on judgments of crime deterrence scale value, Journal of Applied Social Psychology, vol. 18, no. 9, pp Howe, E. S. & Loftus, T.C. 1996, Integration of certainty, severity and celerity information in judged deterrence value: Further evidence and methodological equivalence, Journal of Applied Social Psychology, vol. 26, no. 3, pp Johnson, E. & Payne, J. 1986, The decision to commit a crime: An informationprocessing analysis, in D. B. Cornish & R. V. Clarke (eds), The Reasoning Criminal: Rational Choice Perspectives on Offending, Springer-Verlag, New York. Jones, R. K. & Lacey, J.H. 2001, Alcohol and Highway Safety 2001: A Review of the State of Knowledge, National Highway and Traffic Safety Administration, Washington, DC. Keane, C., Maxim, P.S. & Teevan, J.J. 1993, Drinking and driving, self-control and gender: testing a general theory of crime, Journal of Research in Crime and Delinquency, vol. 30, no. 1, pp

254 Kenkel, D. S. 1993, Drinking and driving and deterrence: the effectiveness and social costs of alternative policies, Journal of Law and Economics, vol. 36, pp Kenkel, D. S. & Koch, S.F. 2001, Deterrence and knowledge of the law: The case of drunk driving, Applied Economics, vol. 33, pp Kleinbaum, D. G. 1996, Survival Analysis: A Self-Learning Text, Springer-Verlag, New York. Kleinbaum, D. G., Kupper, L.L. & Muller, K.E. 1988, Applied Regression Analysis and Other Multivariable Methods, Kent Publishing Company, Boston. Klepper, S. & Nagin, D.S. 1989a, Tax compliance and perceptions of the risks of detection and criminal prosecution, Law and Society Review, vol. 23, no. 2, pp Klepper, S. & Nagin, D.S. 1989b, The deterrent effect of perceived certainty and severity of punishment revisited, Criminology, vol. 27, no. 4, pp Koop, G. 2000, Analysis of Economic Data, Wiley & Sons, Chichester. Lanza-Kaduce, L. 1988, Perceptual deterrence and drinking and driving among college students, Criminology, vol. 26, no. 2, pp Legge, J. S. & Park, J. 1994, Policies to reduce alcohol-impaired driving: Evaluating elements of deterrence, Social Science Quarterly, vol. 75, no. 3, pp Levitt, S. D. 1996, The effect of prison population size on crime rates: Evidence from prison overcrowding litigation, The Quarterly Journal of Economics, vol. 111, May Issue, pp Lind, B., Weatherburn, D., Chen, S., Shanahan, M., Lancsar, E., Haas, M. & De Abreu Lourenco, R. 2002, New South Wales Drug Court Evaluation: Cost- Effectiveness, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Liu, S., Siegal, P., Brewer, R., Mokdad, A., Sleet, D. & Serdula, M. 1997, Prevalence of alcohol-impaired driving: Results from a national self-reported survey of health behaviors, Journal of American Medical Association, vol. 277, no. 2, pp Luke, G. & Lind B. 2002, Reducing juvenile crime: Conferencing versus court, Crime and Justice Bulletin, no. 69, NSW Bureau of Crime Statistics and Research, Sydney, NSW. 244

255 MacCoun, R. J. 1993, Drugs and the law: A psychological analysis of drug prohibition, Psychological Bulletin, vol. 113, no. 3, pp Mann, R. E., Smart, R. G., Stoduto, G., Adlaf, E., Vingilis, E. R., Beirness, D. J. & Lamble, R. 2000, Changing drinking-and-driving behaviour: The effects of Ontario's administrative driver's licence suspension law, Canadian Medical Association Journal, vol. 162, no. 8, pp Mann, R. E., Vingilis, E. R., Gavin, D., Adlaf, E. & Anglin, L. 1991, Sentence severity and the drinking driver; relationships with traffic safety outcomes, Accident, Analysis and Prevention, vol. 23, no. 6, pp Martin, S. E., Annan, S. & Forst, B. 1993, The special deterrent effects of a jail sanction on first-time drunk drivers: A quasi-experimental study, Accident, Analysis and Prevention, vol. 25, no. 5, pp Mayhew, D. R., Beirness, D.J. & Simpson, H.M. 1995, Impact of the 1985 Amendments to the Drink Driving Section of the Criminal Code of Canada, Proceedings of the 13th International Conference on Alcohol, Drugs and Traffic Safety, Aug 13-18, Adelaide, SA. McArthur, D. L. & Kraus, J.F. 1999, The specific deterrence of administrative per se laws in reducing drunk driving recidivism, American Journal of Preventive Medicine, vol. 16, supp. 1, pp. S68-S75. McLean, A. J., Kloeden, C.N. & McCaul, K.A. 1991, Drink-driving in the general night-time driving population, Adelaide 1989, Australian and New Zealand Journal of Public Health, vol. 15, no. 3, pp Meier, R. F. & Johnson, W.T. 1977, Deterrence as social control: The legal and extralegal production of conformity, American Sociological Review, vol. 42, April Issue, pp Miller, S. L. & Iovanni, L. 1994, Determinants of perceived risk of formal sanction for courtship violence, Justice Quarterly, vol. 11, no. 2, pp Mitchell-Taverner, P. 2002a, Community Attitudes to Road Safety: Community Attitudes Survey Wave 14, 2001, Australian Transport Safety Bureau, Canberra, ACT. Mitchell-Taverner, P. 2002b, Community Attitudes to Road Safety: Community Attitudes Survey Wave 15, 2002, Australian Transport Safety Bureau, Canberra, ACT. 245

256 Mitchell-Taverner, P. 2000, Community Attitudes to Road Safety: Community Attitudes Survey Wave 13, 2000, Australian Transport Safety Bureau, Canberra, ACT. Mitchell-Taverner, P. 1999, Community Attitudes to Road Safety: Community Attitudes Survey Wave 12, Australian Transport Safety Bureau, Canberra, ACT. Mitchell-Taverner, P. 1998, Community Attitudes to Road Safety: Community Attitudes Survey Wave 11, Federal Office of Road Safety, Canberra, ACT. Mitchell-Taverner, P. 1997, Community Attitudes to Road Safety: Community Attitudes Survey Wave 10, Federal Office of Road Safety, Canberra, ACT. Mitchell-Taverner, P., Adams, K. & Hejtmanek, S. 1996, Community Attitudes to Road Safety: Community Attitudes Survey Wave 9, Federal Office of Road Safety, Canberra, ACT. Mitchell-Taverner, P., Adams, K. & Hejtmanek, S. 1995, Community Attitudes to Road Safety: Community Attitudes Survey Wave 8, Federal Office of Road Safety, Canberra, ACT. Mitchell-Taverner, P., & Adams, K. 1994, Community Attitudes to Road Safety: Community Attitudes Survey Wave 7, Federal Office of Road Safety, Canberra, ACT. Moffatt, S., Fitzgerald, J. & Weatherburn, D. 2004, Sentencing drink-drivers: The use of dismissals and conditional discharges, Contemporary Issues in Crime and Criminal Justice, no. 81, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Morris, N. & Zimring, F.E. 1969, Deterrence and corrections, The Annals of the American Academy of Political and Social Science, vol. 381, pp Nagin, D. S. 1998, Criminal Deterrence Research at the Outset of the 21st Century, in M. Tonry (ed) Crime and Justice: A Review of Research, vol. 23, University of Chicago Press, Chicago, pp Nagin, D. S. & Paternoster, R. 1993, Enduring individual differences and rational choice theories of crime, Law and Society Review, vol. 27, no. 3, pp

257 Nagin, D. S. & Paternoster, R. 1994, Personal capital and social control: The deterrence implications of a theory of individual differences, Criminology, vol. 32, no. 4, pp Nagin, D. S. & Pogarsky, G. 2001, Integrating celerity, impulsivity, and extralegal sanction threats into a model of general deterrence: Theory and evidence, Criminology, vol. 39, no. 4, pp Nagin, D. S. & Pogarsky, G. 2003, An experimental investigation of deterrence: Cheating, self-serving bias and impulsivity, Criminology, vol. 41, no. 1, pp Nathens, A., Jurkovich, G., Jurkovich, G., Cummings, P., Rivara, F. & Maier, R. 2000, The effect of organized systems of trauma care on motor vehicle crash mortality, Journal of American Medical Association, vol. 283, no. 15, pp National Health and Medical Research Council 2001, Australian Alcohol Guidelines: Health Risks and Benefits, Commonwealth of Australia, Canberra, ACT. National Highway Traffic Safety Administration 2000, On DWI Laws in Other Countries, National Highway Traffic Safety Administration, Washington DC. Nichols, J. L. & Ross, H.L. 1990, The effectiveness of legal sanctions in dealing with drinking drivers, Alcohol, Drugs and Driving, vol. 6, no. 2, pp NSW Bureau of Crime Statistics and Research 2003, New South Wales Recorded Crime Statistics 2002, NSW Bureau of Crime Statistics and Research, Sydney, NSW. NSW Bureau of Crime Statistics and Research 2002a, New South Wales Recorded Crime Statistics 2001, NSW Bureau of Crime Statistics and Research, Sydney, NSW. NSW Bureau of Crime Statistics and Research 2002b, New South Wales Criminal Courts Statistics 2001, NSW Bureau of Crime Statistics and Research, Sydney, NSW. NSW Bureau of Crime Statistics and Research 2001, New South Wales Recorded Crime Statistics 2000, NSW Bureau of Crime Statistics and Research, Sydney, NSW. 247

258 NSW Bureau of Crime Statistics and Research 2000, New South Wales Recorded Crime Statistics 1999, NSW Bureau of Crime Statistics and Research, Sydney, NSW. NSW Bureau of Crime Statistics and Research 1999, New South Wales Recorded Crime Statistics 1998, NSW Bureau of Crime Statistics and Research, Sydney, NSW. NSW Bureau of Crime Statistics and Research 1998, New South Wales Recorded Crime Statistics 1997, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Paternoster, R. 1989, Decisions to participate in and desist from four types of common delinquency: Deterrence and the rational choice perspective, Law and Society Review, vol. 23, no. 1, pp Paternoster, R. & Iovanni, L. 1986, The Deterrent Effect of Perceived Severity: A Reexamination, Social Forces, vol. 64, no. 3, pp Paternoster, R. & Piquero, A. 1995, Reconceptualizing deterrence: An empirical test of personal and vicarious experiences, Journal of Research in Crime and Delinquency, vol. 32, no. 3, pp Paternoster, R., Saltzman, L.E., Chiricos, T. G. & Waldo, G. P. 1982, Perceived risk and deterrence: Methodological artefacts in perceptual deterrence research, The Journal of Criminal Law and Criminology, vol. 73, no. 3, pp Paternoster, R., Saltzman, L.E., Waldo, G. P. & Chiricos, T. G. 1983, Perceived risks and social control: Do sanctions really deter?, Law and Society Review, vol. 17, no. 3, pp Paternoster, R. & Simpson, S. 1996, Sanction threats and appeals to morality: Testing a rational choice model of corporate crime, Law and Society Review, vol. 30, no. 3, pp Peck, R. C. 1991, The general and specific deterrent effects of DUI sanctions: A review of California's experience, Alcohol, Drugs and Driving, vol. 7, no. 1, pp Peterson, R. D. & Bailey, W. C. 1988, Murder and capital punishment in the evolving context of the Post-Furman Era, Social Forces, vol. 66, pp

259 Piliavin, I., Gartner, R., Gartner, R., Thornton, C. & Matseuda, R. L. 1986, Crime deterrence and rational choice, American Sociological Review, vol. 51, no. 1, pp Piquero, A. & Paternoster, R. 1998, An application of Stafford & Warr's reconceptualisation of deterrence to drinking and driving, Journal of Research in Crime and Delinquency, vol. 35, no. 1, pp Piquero, A. & Pogarsky, G. 2002, Beyond Stafford and Warr's reconceptualization of deterrence: Personal and vicarious experiences, impulsivity, and offending behavior, Journal of Research in Crime and Delinquency, vol. 39, no. 2, pp Piquero, A. & Rengert, G.F. 1999, Studying deterrence with active residential burglars, Justice Quarterly, vol. 16, no. 2, pp Pogarsky, G. 2002, Identifying "deterrable" offenders: Implications for research on deterrence, Justice Quarterly, vol. 19, no. 3, pp Pogarsky, G. & Piquero, A. 2003, Can punishment encourage offending? Investigating the "resetting" effect, Journal of Research in Crime and Delinquency, vol. 40, no. 1, pp Pogarsky, G. & Piquero, A. 2004, Studying the reach of deterrence: Can deterrence theory help explain police misconduct?, Journal of Criminal Justice, vol. 32, pp Pogarsky, G., Piquero, A. & Paternoster, R. 2004, Modeling change in perceptions about sanction threats: The neglected linkage in deterrence theory, Journal of Quantitative Criminology, vol. 20, no. 4, pp Robinson, P. H. & Darley, J.M. 2004, Does criminal law deter? A behavioural science investigation, Oxford Journal of Legal Studies, vol. 24, no. 2, pp Rogers, P. N. & Schoenig, S.E. 1994, A time-series evaluation of California's 1982 driving-under-the-influence legislative reforms, Accident, Analysis and Prevention, vol. 26, no. 1, pp Ross, H. L. 1984, Deterring the Drinking Driver: Legal Policy and Social Control (revised and updated edition), Lexington Books, Lexington. Ross, H. L. & Blumenthal, M. 1975, Some problems in experimentation in a legal setting, The American Sociologist, vol. 10, August Issue, pp

260 Ross, H. L. & Gonzales, P. 1988, Effects of license revocation on drunk-driving offenders, Accident, Analysis and Prevention, vol. 20, no. 5, pp Ross, H. L. & LaFree, G.D. 1986, Deterrence in criminology and social policy, in N. J. Smesler & D. R. Gerstein (eds), Behavioral and Social Science: Fifty Years of Discovery, National Academy Press, Washington, DC, pp Ross, H. L. & McCleary R. 1983, Methods for studying the impact of drunk driving laws, Accident, Analysis and Prevention, vol. 15, no. 6, pp Ross, H. L. & Voas, R.B. 1990, The New Philadelphia story: The effects of severe punishment for drunk driving, Law and Policy, vol. 12, no. 1, pp Ryan, G. A., Ferrante, A.M., Loh, N. & Cercarelli, L. R. 1996, Repeat Drink-Driving Offenders in Western Australia, 1984 to 1994, Federal Office of Road Safety, Canberra, ACT. Saffer, H. & Chaloupka, F.J. 1989, Breath testing and highway fatalities, Applied Economics, vol. 21, no. 7, pp Saffer, H. & Grossman, M. 1987, Beer taxes, the legal drinking age and youth motor vehicle fatalities, Journal of Legal Studies, vol. 16, pp Sanson, A., Montgomery, B., Gault, U., Gridley, H. & Thomson, D. 1996, Punishment and behaviour change: An Australian Psychological Society Position Paper, Australian Psychologist, vol. 31, pp Sherman, L. W. 1990, Police Crackdowns: Initial and Residual Deterrence, in M. Tonry & N. Morris (eds), Crime and Justice: A Review of Research, vol. 12, University of Chicago Press, Chicago, pp Sherman, L. W., Gottfredson, D., MacKenzie, D., Eck, J., Reuter, P. & Bushway, S. 1998, Preventing Crime: What Works, What Doesn't, What's Promising?, National Institute of Justice, Washington, DC. Sherman, L. W., Smith, D.A., Schmidt, J. D. & Rogan, D. P. 1992, Crime, punishment and stake in conformity: Legal and informal control of domestic violence, American Sociological Review, vol. 57, pp Shults, R. A., Elder, R.W., Shults, R. A., Elder, R. W., Sleet, D. A., Nichols, J. L., Alao, M. O., Carande-Kulis, V. G., Zaza, S., Sosin, D. M., Thompson, R. S. & Task Force on Community Preventive Services 2001, Reviews of evidence regarding intervention to reduce alcohol-impaired driving, American Journal of Preventive Medicine, vol. 21, supp. 4, pp. S66-S

261 Simon, H.A. 1983, Reason in Human Affairs, Blackwell, Oxford. Siskind, V. 1996, Does license disqualification reduce reoffence rates, Accident, Analysis and Prevention, vol. 28, no. 4, pp Sloan, F. A., Reilly, B.A., & Schenzler, C. 1994, Effects of prices, civil and criminal sanctions, and law enforcement on alcohol-related mortality, Journal of Studies on Alcohol, vol. 55, pp Smith, D. I. 1988, Effect on traffic accidents of introducing Sunday alcohol sales in Brisbane, Australia, International Journal of the Addictions, vol. 23, no. 10, pp Snortum, J. R. & Berger, D.E. 1989, Drinking-driving compliance in the United States: Perceptions and behavior in 1983 and 1986, Journal of Studies on Alcohol, vol. 50, vol. 4, pp South, D. 1998, General deterrence and behaviour change: A comment on the Australian Psychological Society position paper on punishment and behaviour change, Australian Psychologist, vol. 33, no. 1, pp Span, D. 1995, Research on Knowledge, Attitudes and Reported Behaviour on Drink- Driving in New South Wales, Proceedings of the 13th International Conference on Alcohol, Drugs and Traffic Safety, Aug , Adelaide, SA. Span, D. & Stanislaw, H. 1995, Evaluation of the Long Term Impact of a Deterrence-Based Random Breath Testing Program in New South Wales, Proceedings of the 13th International Conference on Alcohol, Drugs and Traffic Safety, Aug , Adelaide, SA. Spegilman, J.J. 2005, A new way to sentence for serious crime, Judicial Officers Bulletin, vol. 17, no. 1, pp Spelman, W. 2000, What Recent Studies Do (and Don't) Tell Us about Imprisonment and Crime, in M. Tonry (ed), Crime and Justice: A Review of Research, vol. 27, University of Chicago Press, Chicago. Spohn, C. & Holleran, D. 2002, The effect of imprisonment on recidivism rates of felony offenders: A focus on drug offenders, Criminology, vol. 40, no. 2, pp Stafford, M. C. & Warr, M. 1993, A reconceptualization of general and specific deterrence, Journal of Research in Crime and Delinquency, vol. 30, no. 2, pp

262 Stewart, K., Gruenewald, P.J., & Parker, R. N. 1992, Assessing legal change: Recidivism and administrative per se laws, Journal of Quantitative Criminology, vol. 8, no. 4, pp Stoduto, G., Mann, R.E., Smart, R. G., Adlaf, E. M., Vingilis, E. R., Beirness, D. J. & Lamble, R. 2000, The Impact of the Administrative Driver's Licence Suspension Law in Ontario, paper presented at the 15th International Conference on Alcohol, Drugs and Traffic Safety, May 22-26, Stockholm, Sweden. Stover, R.V. & Brown, D. W. 1975, Understanding compliance and non-compliance with the law, Social Science Quarterly, vol. 56, pp Tibbetts, S. G. 1997, Shame and rational choice in offending decisions, Criminal Justice and Behavior, vol. 24, no. 2, pp Tittle, C. R. 1977, Sanction fear and the maintenance of social order, Social Forces, vol. 55, no. 3, pp Tittle, C. R. 1980, Evaluating the deterrent effects of criminal sanctions, in M. W. Klein & K. S. Teilman (eds), Handbook of Criminal Justice Evaluation, Sage Publications, Beverley Hills. Tonry, M. 1996, Sentencing Matters, Oxford University Press, New York. Tversky, A. & Kahneman, D. 1981, The framing of decisions and the psychology of choice, Science, vol. 211, January Issue, pp Tyler, T. R. 1990, Why People Obey the Law, Yale University Press, New Haven. Voas, R. B. 2001, Have the courts and the motor vehicle departments adequate power to control the hard-core drunk driver?, Addiction, vol. 96, pp Voas, R. B. & Hause, J.M. 1987, Deterring the drinking driver: The Stockton experience, Accident, Analysis and Prevention, vol. 19, no. 2, pp Voas, R. B., Holder, H. & Gruenewald, P. 1997, The effect of drinking and driving interventions on alcohol-involved traffic crashes within a comprehensive community trial, Addiction, vol. 92, supp. 2, pp. S221-S236. Voas, R. B., Marques, P.R., Tippetts, A. S. & Beirness, D. J. 1999, The Alberta interlock program: The evaluation of a province-wide program on DUI recidivism, Addiction, vol. 94, no. 12, pp

263 Voas, R. B. & Tippetts, A.S. 1999, The Relationship of Alcohol Safety Laws to Drinking Drivers in Fatal Crashes, National Highway Traffic Safety Administration, Washington, DC. von Hirsch, A., Bottoms, A.E., Burney, E. & Wikstrom, P. O. 1999, Criminal Deterrence and Sentence Severity, Hart Publishing, Oxford. Votey, H. L. & Shapiro, P. 1983, Highway accidents in Sweden: Modelling the process of drunken driving behaviour and control, Accident, Analysis and Prevention, vol. 15, no. 6, pp Waldo, G. P. & Chiricos, T.G. 1972, Perceived penal sanctions and self-reported criminality: A neglected approach to deterrence research, Social Problems, vol. 19, no. 4, pp Weatherburn, D., Lind, B. & Hua, J. 2003, Contact with the New South Wales court and prison systems: The influence of age Indigenous status and gender, Contemporary Issues in Crime and Criminal Justice, no. 78, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Weatherburn, D., Topp, L., Midford, R. & Allsopp, S. 2000, Drug Crime Prevention and Mitigation: A Literature Review and Research Agenda, NSW Bureau of Crime Statistics and Research, Sydney, NSW. Weinwrath, M. & Gartrell, J. 2001, Specific deterrence and sentence length, Journal of Contemporary Criminal Justice, vol. 17, no. 2, pp Wells-Parker, E. 2000, Assessment and screening of impaired driving offenders: An analysis of underlying hypotheses as a guide for development of validation strategies, paper presented at the 15th International Conference on Alcohol, Drugs and Traffic Safety, May 22 26, Stockholm, Sweden. Wells-Parker, E., Bangert-Downs, R., McMillen, R. & Williams, M. 1995, Final results from a meta-analysis of remedial interventions with drink/drive offenders, Addiction, vol. 90, pp Wilkinson, J. T. 1987, Reducing drunken driving: Which policies are most effective?, Southern Economic Journal, vol. 54, pp Williams, A. F., Ferguson, S.A., & Cammisa, M. X. 2000, Self-reported Drinking and Driving Practices and Attitudes in Four Countries and Receptions of Enforcement, paper presented at the 15 th International Conference on Alcohol Drugs and Traffic Safety, May 22 26, Stockholm, Sweden. 253

264 Williams, K. R. & Gibbs, J.P. 1981, Deterrence and knowledge of statutory penalties, The Sociological Quarterly, vol. 22, Autumn Issue, pp Williams, K. R., Gibbs, J.P., & Erickson, M. L. 1980, Public knowledge of statutory penalties: The extent and basis of accurate perception, Pacific Sociological Review, vol. 23, no. 1, pp Williams, K. R. & Hawkins, R. 1986, Perceptual research on general deterrence: A critical review, Law and Society Review, vol. 20, no. 4, pp Wilson, J.Q. & Hernstein, R. 1985, Crime and Human Nature, Simon Schuster, New York. Wright, B. R. E., Caspi, A., Moffit, T. E. & Paternoster, R. 2004, Does the perceived risk of punishment deter criminally prone individuals? Rational choice, self-control, and crime, Journal of Research in Crime and Delinquency, vol. 41, no. 2, pp Yin, R. 1994, Case Study Research: Design and Methods, Sage, Thousand Oaks, California. Yu, J. 1994, Punishment celerity and severity: Testing a specific deterrence model on drunk driving recidivism, Journal of Criminal Justice, vol. 22, no. 4, pp Zaal, D. 1994, Traffic Law Enforcement: A Review of the Literature, Monash University Accident Research Centre, Melbourne, Victoria. Zador, P. L. 1991, Alcohol-related relative risk of fatal driver injuries in relation to driver age and sex, Journal of Studies on Alcohol, vol. 52, no. 4, pp Zador, P. L., Krawchuk, S.A. & Voas, R. B. 2000, Alcohol-related relative risk of driver fatalities and driver involvement in fatal crashes in relation to driver age and gender: An update using 1996 data, Journal of Studies on Alcohol, vol. 61, pp Zimring, F. E. & Hawkins, G.J. 1973, Deterrence: The Legal Threat in Crime Control, University of Chicago Press, Chicago. 254

265 Appendices Appendix I Summary of drink-driving studies reviewed in Chapter 3 The tables below present summary data from each of the studies reviewed in Chapter 3. Included in these tables is information on the jurisdiction studied (i.e. State, Territory or Province), the type of study design employed and the specific outcome measures analysed (e.g. self-reported drink driving or road fatalities). In the case of recidivism studies on penalty severity, Table A3 includes additional information on the type of penalty examined. If the study described found a statistically significant reduction, in the outcome examined, this is also indicated in the table Voas, Holder & Gruenewald 1997 Henstridge, Homel & Mackay 1997 USA California, USA NSW, QLD, WA & Tasmania, Australia Table A1: Aggregate-level studies on punishment certainty Study Jurisdiction Design Outcome measure Statistically significant reduction? Wilkinson 1987 USA Ecological Road fatalities No Evans, Neville USA Ecological Road fatalities No & Graham 1991 Sloan, Reilly & USA Ecological Road fatalities Yes Schenzler 1994 Saffer & USA Ecological Road fatalities Yes Chaloupka 1994 Benson, USA Ecological Driver involvement rate Yes Rasmussen & Mast 1999 Kenkel 1993 USA Ecological Self reported drink-driving Self reported drinking Yes Yes Chaloupka & USA Ecological Self-reported drinking Yes Weschler 1996 Ross 1984 Britain Quasiexperimental Road fatalities Yes Serious injury accidents Voas & Hause Stockton, CA, Quasi- Night-time accidents Yes experimental Quasiexperimental Quasiexperimental Roadside breath tests SVNT accidents & positive BACs Self reported drink-driving Road fatalities SVNT accidents NB: SVNT accidents are those where only one vehicle is involved in an accident occurring at night Yes Yes Yes 255

266 Table A2: Aggregate-level studies on punishment severity Study Jurisdiction Design Outcome measure Statistically significant reduction? Rogers & Shoenig 1994 California, USA Quasiexperimental Peck 1991 California, USA Quasiexperimental Borschos 2000 Sweden Quasiexperimental Beirness, Canada Quasiexperimental Simpson & Mayhew 1993 Ross 1984 Finland Quasiexperimental Hingson et al Homel 1986 Ross & Voas 1990 Sloan, Reilly & Schenzler 1994 Chaloupka, Saffer & Grossman 1993 Votey & Shapiro 1983 Night-time accidents SVNT accidents Accidents designated by police as alcohol-related Recidivism Yes Yes USA Ecological Road fatalities Yes Sweden Ecological Road fatalities Serious injury accidents Yes Table A3: Individual-level studies on punishment severity Study Jurisdiction Design Outcome measure Penalty type Holden 1983 USA Experimental Rearrest rates Supervised probation, Education & Bloomberg et al DeYoung 1997 Homel 1980 Martin, Annan & Forst 1993 Wisconsin, USA California, USA NSW, Australia Minnesota, USA Road fatalities Yes Accidents w/ severe injury Road fatalities & arrests No Self-reported drink-driving & positive BACs Road fatalities No Serious injury accidents Single-vehicle fatalities Massachusetts Quasiexperimental Road fatalities No & Maine, USA Self-reported drink-driving NSW, Quasiexperimental Self-reported drink-driving Yes Australia New Quasiexperimental Roadside breath tests No Philadelphia, Self-reported drink-driving USA USA Ecological Road fatalities No Quasiexperimental Cross-sectional Cross-sectional Cross-sectional Rearrest rates Road accident rates Reconviction rates Reconviction rates Reconviction rates Therapy 3-month minimum licence disqualification 2-day gaol sentences Licence sanctions Gaol sentences Licence sanctions Fines Probation 2-day gaol sentences Stat. sig. reduction? No Yes No (gaol) Yes No No 256

267 Siskind 1996 Yu 1994 Mann et al 1991 Qld, Australia New York, USA Ontario, Canada Cross-sectional Cross-sectional Cross-sectional Rearrest rate Road accident rate Reconviction rate Rearrest rate Road accident rate Licence sanctions Licence sanctions Fine Gaol sentences Licence sanctions Fines Yes No Yes (fines) No (gaol) Yes No (fines) Table A4: Aggregate and Individual-level studies on punishment celerity Study Jurisdiction Design Outcome measure Statistically significant reduction? Nagin & Pogarsky Students, Experimental Likelihood of committing No 2001 USA a drink-driving offence Simon, Cleary & Minnesota, Cross-sectional Reconviction rates Yes Storkamp 2000 USA Yu 1994 New York, Cross-sectional Reconviction rates No Mann et. al Voas & Tippets 1999 Stewart, Gruenewald & Parker 1992 USA Ontario, Canada Proportion of fatalities w/ positive BACs Self-reported drink-driving Yes USA Ecological Road fatalities Yes Louisiana, Mississippi & North Dakota, USA Quasiexperimental Quasiexperimental Recidivism rates Yes 257

268 Appendix II Participant information sheet and drink-driving questionnaire 258

269 SCHOOL OF SOCIAL SCIENCE & POLICY Approval No HREC03163 THE UNIVERSITY OF NEW SOUTH WALES PARTICIPANT INFORMATION STATEMENT The deterrent effect of increased penalties for drink-driving You are invited to participate in a study of drink-driving. From this study we hope to learn what impact the penalties for drink-driving have on decisions to offend. We are recruiting students from various disciplines at the University of New South Wales to ensure that we have responses from a wide range of young people. If you decide to participate, we will ask you to answer a series of questions related to drink-driving, which will take you approximately five minutes to complete. To ensure anonymity, we would ask that you do not write your name or any other identifying information on any part of the questionnaire. If you give us your permission by completing this questionnaire, we plan to publish the results as part of a PhD thesis to be submitted to The University of New South Wales. In any publication, information will be provided in such a way that you cannot be identified. Your participation in this study is voluntary and your decision whether or not to participate will not prejudice your future relations with The University of New South Wales. If you decide to participate, you are free to discontinue participation at any time without prejudice. Your self-completion of this questionnaire indicates that, having read the information provided above, you have decided to participate. If you have any questions, please feel free to ask us. If you have any additional questions later, Suzanne Briscoe (ph ) will be happy to answer them. Any complaints about this study may be directed to the Ethics Secretariat, The University of New South Wales, SYDNEY 2052 AUSTRALIA (ph , fax , ethics.sec@unsw.edu.au). You can keep this form. 259