This work is distributed as a Discussion Paper by the STANFORD INSTITUTE FOR ECONOMIC POLICY RESEARCH. SIEPR Discussion Paper No.

Transcription

1 This work is distributed as a Discussion Paper by the STANFORD INSTITUTE FOR ECONOMIC POLICY RESEARCH SIEPR Discussion Paper No Evaluating the Effectiveness of a 0.08% BAC Limit and Other Policies Related to Drunk Driving Daniel Eisenberg Stanford University January 2001 Stanford Institute for Economic Policy Research Stanford University Stanford, CA (650) The Stanford Institute for Economic Policy Research at Stanford University supports research bearing on economic and public policy issues. The SIEPR Discussion Paper Series reports on research and policy analysis conducted by researchers affiliated with the Institute. Working papers in this series reflect the views of the authors and not necessarily those of the Stanford Institute for Economic Policy Research or Stanford University.

2 1 Evaluating the effectiveness of a 0.08% BAC limit and other policies related to drunk driving 1 Abstract During the 1980s and 1990s in the United States, a large number of new policies have been adopted with the aim of reducing drunk driving. The value of many of these policies is controversial in policymaking circles. Recently the illegal per se blood alcohol content (BAC) limit, which is either 0.10% or 0.08% in nearly all states, has come under scrutiny from the federal government. This study presents new findings on the effectiveness of the 0.08 BAC limit, and drunk driving policies in general. The results indicate that the incremental effect of moving from a 0.10% BAC limit to a 0.08% standard lowers fatal traffic crash rates by 2.6%. This study analyzes the timing of the effects of the 0.08% limit, and finds that conventional empirical models do not capture an interesting pattern, in which the strongest effects of the law do not appear to set in until several years after enactment. Also, this paper sketches a welfare analysis of the change from a 0.10% limit to a 0.08% limit. In the context of the drunk driving policy literature in general, the paper evaluates two factors which have previously been ignored in this type of analysis: graduated licensing programs and the Mothers Against Drunk Driving (MADD) organization. Both initiatives are estimated to reduce fatal traffic crash rates. Graduated licensing, in fact, appears to be one of the most effective initiatives, reducing fatal crash rates for all drivers by 4.7%, and fatal crash rates for drivers under 21 by 11.5%. Other apparently significant policies are open container laws (5.1% reduction in fatal crash rates for all drivers), dram shop laws (5.8%), seat belt laws (4.9%), and beer taxes (0.8% per 10 cents). This paper also provides new evidence on zero tolerance BAC limits for drivers under 21, indicating that these laws reduce fatal crash rates for under 21 drivers by 6.1%, but only in the presence of administrative license revocation (ALR) laws. Daniel Eisenberg Department of Economics Stanford University daniel.eisenberg@stanford.edu January, I thank Mark McClellan, Doug Bernheim, Antonio Rangel, Joanna Campbell, Neva Kerbeshian, Feliza Guidero, and seminar participants at Stanford for helpful comments and advice.

3 2 I. Introduction The intense lobbying by interest groups and wrangling by legislators that took place over a particular provision in the 2000 Federal Transportation Appropriations Bill epitomizes the controversy surrounding the issue of how to address drunk driving in the United States. On one side, organizations such as Mothers Against Drunk Driving and the American Medical Association, along with Congressmen including Senator Frank Lautenberg (D - NJ), supported a provision that would require states to adopt 0.08% blood alcohol content (BAC) standard for drunk driving, or else lose federal highway funding. On the other side, restaurant and bar advocacy groups like the American Beverage Institute (ABI) and legislators including House Majority Whip Tom Delay (R - TX), opposed any such provision. The compromise solution which was signed by President Clinton in October 2000 specified that states which do not adopt the 0.08 limit will lose 2% of federal highway funds in 2003, 4% in 2004, 6% in 2005, and 8% in 2006 and in years thereafter. However, states which adopt the limit by 2007 will recoup any previously lost funds. The division in the U.S. over the 0.08 law is mirrored in the fact that states have had for many years significantly different laws related to drunk driving, and traffic safety more generally. For this reason, among others, an important question is how effective are the various policies in reducing drunk driving and its adverse consequences? In the tradition of a substantial body of economic, public health, and traffic safety literature, this paper takes up the question, with particular emphasis on the currently controversial 0.08 BAC law. Specifically, this paper examines the extent to which the 0.08 BAC law and other initiatives reduce fatal traffic crashes. This study presents new findings for the literature on the 0.08 BAC law, and the drunk driving policy literature in general. In evaluating the BAC law, the analysis controls for two factors previously ignored in other studies: graduating licensing programs and the presence of MADD. These additional controls, in combination with the most recent data, which is essential due to the fact that over half of the 0.08 states adopted the limit in 1994 or later, yield results that indicate that the incremental effect of moving from a 0.10 BAC limit (which most states still have) to a 0.08 standard lowers

4 3 traffic crash rates by 2.6%. This estimate is lower than most previous estimates, but still statistically significant. In addition, this study presents a novel analysis of the timing of the effects of the 0.08 limit, and finds that the strongest effects of the law do not appear to set in until several years after enactment. Based on these estimates, this paper sketches a welfare analysis of the change from a.10% limit to a.08% limit, with a view towards the decision faced by policymakers in states which will have to decide when, if ever, they wish to comply with the federal incentive program that was recently passed. In the context of the drunk driving policy literature in general, the analysis provides new evidence that both MADD and graduated licensing programs are important factors in reducing fatal traffic crashes. Graduated licensing, in fact, appears to be one of the most effective initiatives, reducing fatal crash rates by 4.7%. Other apparently significant policies are open container laws (5.1%), dram shop laws (5.8%), seat belt laws (4.9%), and beer taxes (0.8% per 10 cents). The paper also provides new evidence on zero tolerance BAC limits for drivers under 21, indicating that these laws reduce fatal crash rates for under 21 drivers by 6.1%, but only in the presence of administrative license revocation (ALR) laws. The paper is organized as follows. Section II provides background information on drunk driving and policies related to this problem in the United States. Section III summarizes the existing literature on the effect of the 0.08 BAC law and other initiatives, and describes how this study relates to the literature. Section IV details the empirical model, data, and variables used in the analysis. Section V then presents the results. Finally, Section VI sketches a welfare analysis of the 0.08 BAC law, and Section VII concludes the paper. II. Background Policies related to drunk driving The last twenty years in the United States have witnessed a barrage of public policy measures related to drunk driving, with hundreds of laws being passed on the state level, as well as several bills on the federal level. The various measures can be categorized into the following groups (see Table 2.1): direct deterrence of drunk driving;

5 4 indirect deterrence of drunk driving; general alcohol control; and general vehicle safety. Measures in this last category, general vehicle safety, are related to drunk driving in the sense that they might be expected to mitigate the adverse consequences of drunk driving. The first category, direct deterrence measures, includes the illegal per se BAC limit which continues to generate controversy in public policy debates. In 1980 most states either had no per se limit at all or had a limit higher than 0.10, but by 2000 all states except for Massachusetts 2 had a limit of either 0.10 or For drivers under 21 years old, the per se limit has been set at 0.02 or lower in all states during the last twenty years. In most states, these so-called zero tolerance laws for young drivers were adopted during the mid or late 1990s. While the per se limit sets a standard for what is considered illegal drunk driving, other direct deterrence measures prescribe the consequent penalties. Administrative license revocation (ALR) laws, which allow for immediate revocations of driving licenses upon failure of a BAC test (as opposed to allowing the offender to retain the license at least until court proceedings), appeared in states beginning in the early 1980s, and were present in 42 states by Other penalties which have been adopted largely in the past two decades include mandatory minimum jail time and mandatory minimum fines for driving under the influence (DUI) offenses. Table 2.1: Summary of initiatives related to drunk driving and traffic safety Direct deterrence Indirect deterrence General alcohol control General vehicle safety Other per se BAC limit Prelim. breath tests open container seat belts grassroot (MADD) zero tolerance Implied consent MLDA speed limits ALR Sob. checkpoints alcohol taxes grad.licensing Mandatory jail Anti-plea bargains Mandatory fines Comparative negl. Dram shop A handful of polices fall into the second category, indirect deterrence measures, because, while they do not designate standards nor penalties for drunk driving, they are designed to affect the expected costs faced by a potential drunk driving offender in less 2 Although MA has no illegal per se limit, it has a 0.08 limit for presumptive evidence and for ALR.

6 5 direct ways. Three such measures increase the probability that a potential offender will be apprehended by police officers: preliminary breath test (PBT) laws, implied consent laws, and sobriety checkpoints. All three of these measures have become increasingly prevalent over the last twenty years. PBT laws allow police officers to give drivers preliminary breath tests with handheld BAC sensor devices; the officers can then use the results of these tests to decide whether to arrest the driver (and then administer a more accurate chemical test). Implied consent laws require drivers to submit to BAC tests, or else face prescribed penalties (which normally include ALR). Sobriety checkpoints are programs where police officers randomly pull over cars at a particular location and administer BAC tests to the drivers. Other indirect deterrence measures do not involve the probability of apprehension, but rather the probability of high penalties, given that an offense has occurred. First, anti-plea bargaining laws prevent DUI offenders from reducing their offenses to lesser crimes. Next, in cases of traffic crashes involving multiple parties, states have differing standards for establishing tort liability. For example, a general trend in the United States during the 1980s was a switch from contributory negligence to comparative negligence (Sloan et. al 1994). Finally, dram shop laws operate as deterrents on an even more indirect level than the above laws by targeting not the drivers but the people serving alcohol to them. They provide for the possibility that establishments serving alcohol to intoxicated patrons can be held liable for subsequent crashes caused by the patrons. Along with anti-plea bargaining laws and comparative negligence laws, dram shop laws have proliferated in the U.S. during the last twenty years. The third category of policy measures mentioned at the outset of this section, general alcohol control measures, comprises laws which are directed at alcohol consumption, but not drunk driving specifically. This category includes open container (in vehicles) laws, minimum legal drinking age (MLDA) laws, and alcohol taxes, all of which have changed significantly since the early 1980s. Open container laws prohibit open containers of alcohol in moving motor vehicles (regardless of whether the driver is consuming alcohol). The number of states with such a law has increased from 18 in 1982 to 32 in MLDA laws underwent a great amount of change during a more narrow

7 6 period, from 1980 to 1988, during which 38 states raised their MLDAs to 21 years. By 1988 all 50 states plus D.C. had MLDAs of 21, largely due to a 1986 federal incentive program designed to establish a uniform MLDA of 21. Alcohol taxes, meanwhile, might be considered the one outlier among all of the measures mentioned here, in that they have generally moved in recent years in a direction that would be expected to lead to more, not less, drunk driving. In many states the real values of the beer tax, for example, have fallen as the nominal values have remained unchanged. General vehicle safety measures constitute a fourth category of policies that have seen a flurry of action during the last two decades. As mentioned above, these policies relate to drunk driving because they might be expected in general to mitigate the harm caused by drunk driving crashes. Seat belt laws, for example, arrived in the 1980s, and by 2000 every state except New Hampshire had adopted one. Speed limit laws have also undergone substantial changes during the 1980s and 1990s, mostly due to bills from the federal level. In 1987, Congress relaxed the 1974 National Maximum Speed Limit (NMSL) of 55 mph, allowing states to post 65 mph speed limits on rural interstates. Then in 1995 Congress repealed the NMSL altogether, leading many states to raise rural highway limits to above 65 mph, and some urban highway limits to 65 mph. Finally, in contrast to speed limits, a policy measure referred to as graduated licensing has seen a lot of recent activity at the state level. Graduated licensing refers to programs in which teenage drivers gain privileges over the course of a number of steps with prescribed time intervals, assuming that the drivers maintain clean records at each step. The National Committee on Uniform Traffic Laws and Ordinances (NCUTLO) defines a specific model for the program, with a number of core provisions: a learner s phase of at least six months, which can begin at age 16; an intermediate phase of at least six months, during which unsupervised driving is prohibited at night; and a requirement that applicants for intermediate and full licenses have no safety belt or zero tolerance violations, nor any other major traffic violations. States began adopting such programs largely in the latter half of the 1990s; by 2000, 18 states had programs which the NCUTLO considered full graduated licensing, and 6 had what the committee considered partial graduated licensing.

8 7 In addition to the many policy initiatives over the last twenty years, one other important force designed to address drunk driving arrived on a large scale in the United States: non-profit, grassroots organizations. The most prominent of these organizations, Mothers Against Drunk Driving (MADD), began in 1980 in California, and has since grown to over 300 chapters, with at least one in 49 different states, in addition to hundreds of less formal affiliates called community action teams. MADD and other organizations have been active on a number of fronts related to drunk driving, including public awareness, legislative lobbying, victim support, and educational programs. How do we measure the effectiveness of these policies? In the process of evaluating the effectiveness of the initiatives described above, it is not obvious how exactly to define the outcome measure of interest. Should we be concerned simply with the prevalence of drunk driving, i.e. the percentage of drivers on the road who exceed a certain BAC level? Or should we take into account the entire distribution of BACs among drivers on the road? Alternatively, one could make a reasonable argument that BAC levels only matter to the extent that they lead to crashes, and thus we should focus only on drunk driving-related crashes. Section III discusses which outcome measures have received attention in previous literature, and Section IV discusses the measures chosen for this study. This section provides background on the general trends in various outcomes related to drunk driving, which by any measure have shown improvement over the last two decades in the United States. First, the little data on drunk driving prevalence that is available indicates that the proportion of drivers with high BAC levels has fallen substantially since the 1970s. The only source of national data based on actual BAC testing of drivers is the National Roadside Survey, which has been conducted on three occasions: 1973, 1986, and Table 2.2 summarizes the results from the surveys. The results indicate substantial decreases from 1973 to 1986 in the proportion of drivers, both male and female, with BACs above.05% or.10%. There appears to have been a further slight decrease in drunk driving among males from 1986 to 1996, in contrast to a slight increase

9 8 among females during that time. Voas et. al (1998), however, point out that the changes between 1986 and 1996 are not statistically significant. Table 2.2: National Roadside Surveys on DUI prevalence (weekend nights) BAC results, males BAC results, females BAC results, total N >=.05 (%) >=.10 (%) N >=.05 (%) >=.10 (%) N** >=.05 (%) >=.10 (%) year *table adapted from data presented in Voas et. al (1998) ** the total number of people does not add up exactly to the sum of the males and females, because gender was not recorded in a small handful of cases Other data on the prevalence of drunk driving comes from surveys which ask people questions regarding drunk driving. The Health Promotion and Disease Prevention supplement to the 1985 National Health Interview Survey, for example, asks respondents During the past year, how many times did you drive when you had perhaps too much to drink?. In recent years, the CDC s Behavioral Risk Factor Surveillance System (BRFSS) has asked respondents a question that is identical except that they ask about the past month rather than year. Of course, the value of these self-reported data is limited due to the subjective nature of the questions. The National Survey of Drinking and Driving Attitudes and Behavior, conducted by the NHTSA in 1991, 1993, 1995, and 1997, provides somewhat more objective information by asking respondents whether they have driven in the last year after drinking at all within the two preceding hours. The percentage of respondents who said yes fell from 28% to 24% between 1991 and Like the DUI prevalence data from the National Roadside Surveys, drunk driving crash data show a general downward trend over the past couple decades. The National Highway Traffic Safety Administration (NHTSA) collects data on crashes of all different severity levels, but provides by far the most comprehensive and accurate data for those crashes which involve fatalities. NHTSA s Fatal Accident Report System (FARS) constitutes a complete census of fatal traffic crashes, with a large amount of information regarding each crash, including the BAC level of a majority of the drivers

10 9 involved. For those drivers whose BACs were not tested, NHTSA uses a discriminant analysis to impute the probability that they had a BAC of 0, between 0.01 and 0.10, or greater than 0.10, based on factors such as the police report, the time of day, and how many vehicles were involved (Klein 1986). Table 2.3 displays trends in various fatal traffic crash rates for the United States between 1982 and The three lower trend lines, which are for drivers of all ages, show that the total fatal crash rate has fallen substantially, but most, if not all, of this decrease, can be accounted for by the large drop in high BAC fatal crash. The upper trend line, on the other hand, indicates that the fatal crash rate for year olds rose between 1982 and 1988, fell between 1988 and 1992, and then rose back to the level of the early 1980s by Furthermore, in each year the total fatal crash rate for year olds is about twice as high as the rate for all ages. Table 2.3: Fatal accidents per 10,000 people total fatal crash rate, yr olds total fatal crash rate, all ages positive BAC (>0.01) crash rate, all ages high BAC (>0.10) crash rate, all ages year

11 10 Why would we expect the policies to be effective? In order to consider explicitly how the 0.08 BAC law and other policies might be expected to reduce drunk driving (and consequently, drunk driving crashes), it is useful to look at a simple version of a model of crime first discussed by Becker (1968). We can think of a supply function for drunk driving offenses for an individual j: O j = O j ( p j, f j, u j ) where O j is the number of drunk driving offenses in a given time period, p j is the probability of conviction per offense, f j is the punishment per conviction, and u j is a vector of parameters capturing tastes, the price and availability of alcohol, and other factors. O j is decreasing in both p j and f j,aswellasthepriceofalcohol,and increasing in the taste for drunk driving and the availability of alcohol. To relate drunk driving offenses to traffic crashes, we write the following aggregate traffic crash production function, following Benson et. al (1999): R = R( O, T) where R istheaggregatetrafficcrashrate,o is the aggregate number of drunk driving offenses (summed over all individuals j = 1,, N), and T is a vector of measures of traffic safety, vehicle safety, and driver safety (e.g. lighting of roads, prevalence of airbags, use of seatbelts). R is increasing in O, and decreasing in elements of T. In the context of the simple functions above, one can easily imagine a number of channels through which the policies discussed in this section could operate to reduce the traffic crash rate. The policies categorized as direct deterrence or indirect deterrence measures increase alcohol control policies influence p j, or increase f j, or both, and thus reduce O. General u j ; alcohol taxes increase the price of alcohol, while MLDA and open container laws restrict the availability of alcohol. The general vehicle safety measures increase T. Finally, organizations such as MADD might influence the taste parameter embodied in u j and influence the adoption of the various policies. Due to the particular emphasis of this paper on the.08% BAC level, a more detailed consideration of how this particular policy could affect drunk driving

12 11 outcomes is warranted. Perhaps the most obvious mechanism is the increase in p j,the probability of conviction for a potential offender. Of course, given that people do not have perfect information, the important value for deterrence is the perceived probability of conviction. Surveys suggest that lack of information might be a serious issue in this context. For example, the 1997 National Survey on Drunk Driving Attitudes and Behavior found that only 29% of respondents knew the correct BAC limit level in their states in However, this fact alone does not preclude the possibility that a lowering of the BAC level to 0.08 can increase the perceived probability of conviction for a large fraction of drivers. For example, it is entirely possible that a much greater proportion of people are aware of the fact that the BAC limit has been lowered, even when they do not know the exact level. Furthermore, drivers who are more likely to drink and drive might also be more likely to be aware of the BAC limit. What are some specific mechanisms through which the lowering of the BAC limit can lead to an increase in a driver s perceived probability of conviction? Media publicity surrounding the debate over the legislation before it is passed, and after it is passed, are prime examples. In some cases, states which have lowered the BAC limit have sponsored a publicity campaign to make drivers aware of the lower limit (NHTSA 2000). Also, prosecution of drivers under the new law could increase awareness of the change, both through media coverage of the prosecution and through word-of-mouth. Beyond increasing the perceived probability of conviction, the 0.08 BAC limit could lead to reduced drunk driving in at least two other ways. First, suppose many drivers initially do not change their behavior at all. Out of that group of drivers, the lower BAC limit would lead to more people being convicted, and thus more of these people would be temporarily removed from the road, either through license suspension or jail time. The result would be that a lot of risky drivers would no longer be on the road under the 0.08 BAC law, whereas they would be under the previous, higher BAC limit. Second, the lowering of the BAC limit could influence people s tastes for drunk driving. For example, for many people it might lower the the threshold for what they consider an acceptable amount of drinking before driving.

13 12 III. Literature review Literature on the effect of a.08% BAC law Out of the various outcome measures discussed in the previous section, the 0.08 BAC literature has focused exclusively on fatal traffic crashes. The main reason, as one might have gathered from the discussion in the last section, is that these data are by far the most accurate and comprehensive out of the data available for different outcome measures of interest. Furthermore, fatal traffic crashes are perhaps more relevant outcomes to policymakers than DUI prevalence rates. Non-fatal crashes are obviously also outcomes of interest to policymakers, but they are not recorded in comprehensive state-by-state data. On the whole the literature has produced mixed results regarding the effect of the 0.08 BAC limit on fatal traffic crashes. Table 3.1 summarizes the 10 published studies on this topic. The conflicting results among the various studies is not surprising, given the extreme limitations of most of the studies. The foremost of these limitations include smallness of samples, lack of appropriate controls for unobserved state heterogeneity, and lack of controls for concurrent policy measures. With the exception of Dee (2001), each of the studies has at least one of these three drawbacks, and most of the studies have more than one of them. Dee s results indicate that the 0.08 BAC limit does in fact reduce fatal traffic crashes. He employs an OLS model with state and year fixed effects, with panel data from the 48 continental states and years Dee estimates that the 0.08 limit reduces the fatal traffic crash rate by 7.2%, whereas the 0.10 limit reduces the rate by 5.3%. However, he does not report any test of the most relevant policy question: whether the difference between the 0.08 BAC limit effect and the 0.10 BAC limit effect is statistically significant. In the course of this study, I replicated his results and confirmed that the 0.08 effect is in fact greater than the 0.10 effect at the 0.05 significance level. Thus, Dee s results, when supplemented by this hypothesis test, indicate that a 0.08 BAC limit reduces fatal traffic crash rates by about 2% beyond the rates achieved under a 0.10 BAC limit.

14 13 While the literature summarized in Table 3.1 provides direct evidence on the effect of a 0.08 BAC policy on traffic crashes, a second group of studies provides indirect evidence for the issue (Levitt and Porter 1999; Zador et. al (2000); Zador (1991); Moskowitz and Fiorentino (2000)). These studies look at the relationship between crash risk and driver BAC levels (as opposed to BAC laws). First, Levitt and Porter (1999) observe that under reasonable assumptions, the number of crashes involving two drinking drivers is proportional to the square of the number of drinking drivers on the road, whereas the number of crashes involving one drinking driver and one sober driver is linearly related to the number of drinking drivers on the road. The authors exploit this observation to estimate the relative risk for drinking drivers. They estimate that drinking drivers are eight times more likely to be in a fatal crash than sober drivers, and legally drunk drivers are 15 times more likely. Zador et. al (2000) and Zador (1991) calculate relative risks at a particular BAC level by comparing the relative proportion of drivers at this BAC level in fatal accidents, as determined from FARS data, to the relative proportion of drivers at this BAC level who are on the road, as determined from the National Roadside Surveys (1986 and 1996 respectively). The more recent study examines BAC levels between 0.08 and 0.10, and estimates relative risks ranging from 6 to 24, depending on the age-gender group. The later study s findings are largely in line with those of the earlier study, which finds for year old males, for example, relative risks of 1.2 for BACs, 18.3 for BACs, 39.7 for BACs, and over 600 for.15+ BACs. Moskowitz and Fiorentino (2000) review the medical evidence on the effect of different BAC levels on a variety of functions related to driving ability, such as reaction time and tracking ability. They conclude that this literature provides strong evidence that impairment of some driving-related skills begins with any departure from zero BAC. Their review finds that the majority of studies report some level of impairment from alcohol at a 0.05 or lower BAC, and 94% of the studies reviewed report impairment at a 0.08 or lower BAC. A potentially important caveat for the Levitt and Porter (1999) and both Zador studies which the authors do not mention is that the entire estimated relative risks cannot necessarily be attributed to the effects of alcohol. It is possible that those who drink and

15 14 Table 3.1: Previous studies on the effectiveness of 0.08% BAC law Study Method Results Main Limitations NHTSA (1991) before/after comparison for California 12% decline in alcohol related fatalities -short post-law period: 1 year -ALR policy introduced 6 months after.08 in CA -nocomparisontootherstates NHTSA (1994) Office of Traffic Safety, State of California (1995) Hingson, R., T. Heeren and M. Winter (1996). Foss, R.D., J.R. Stewart, and D.W. Reinfurt (1998) Apsler, R., A.R. Char, W.M. Harding, and T.M. Klein (1999) Jonah, B., R. Mann, S. McDonald, G. Stoduto, S. Bondy, and A. Shaikh (2000). Voas, R., E. Taylor,T.Baker, and A.S. Tippetts (2000) Voas, R., A. Tippetts, and J. Fell (2000) Dee, T. (2001) before/after comparison for 6 different outcomes measures for each of 5 early adopters (OR, UT, ME, CA, VT) time series analysis to look at CA before/after comparison for same 5 states as in NHTSA (1994), but with respect to comparison states before/after comparison for NC, with respect to the rest of the nation time series analysis (Box- Tiao intervention model) to look at 11 states which adopted 0.08 by 1994 review of studies from Australia, France, Sweden, Canada, and the U.S. (some adopted lower limits than 0.08) before/after comparison, with respect to 5 surrounding states, for IL s adoption of 0.08 law in 1997 regression analysis for all 51 states, regression analysis for all 51 states, , with state and year fixed effects 9 significant decreases out of 30 total measures significant decrease in nighttime fatal accidents, but not in total fatal accidents or alcohol-related fatal accidents 4of5states experienced a greater reduction in estimated alcohol involvement in fatal accidents than respective control states decreases in various outcome measures after the law were not significantly greater than national decreases significant effect in 5 states reviewed studies and found that all show reductions in various measures of drunk driving, or alcoholrelated crashes 13.7% in alcoholrelated fatal crashes in IL in 1.5 yrs following.08, vs. no significant decline in surrounding states 0.08 law associated with reduced proportion of alcohol-related fatal accidents (out of total fatal accidents) 7% reduction in total fatal accidents associated with.08 law (5% for.10 law) -only 1 state (CA) -no control for other policies (e.g. ALR adopted w/in 10 months of.08 in 3 of 5 states -nocomparisontootherstates -only 5 states -ALR introduced 6 months after.08 in CA -nocomparisontootherstates -only 1 state (CA) -no control for other policies -comparison states somewhat arbitrary (based on size and proximity) -only 5 states -only 1 state (NC) -no control for other policies -nocomparisontootherstates -none of the studies from foreign countries have control states -U.S. studies reviewed are listed above -short post-law period: 1.5 yrs -only 1 state (IL) -under 21 drivers excluded -no state fixed effects -lacking controls for many policies -no major limitations (see discussion at end of this section for how my study relates to Dee s)

16 15 drive tend to be less than average drivers when they are sober, even after controlling for observable factors such as age and gender. In such a scenario, the actual effect of alcohol on crash risks would be smaller than estimated. To the extent that this scenario is true, a BAC limit policy would be a somewhat blunt instrument in reducing traffic crashes. In the extreme, if alcohol served only as a signal of risky drivers, as opposed to a cause of risky driving, then the main beneficial effect of BAC limits would be to deter inherently risky drivers from driving, not deterring drivers from drinking. BAC limits would help to identify risky drivers and remove them from the road through licensing and jail penalties. Literature on the effects of other policies related to drunk driving Like the literature on the 0.08 limit summarized in Table 3.1, the literature on other drunk driving-related policies (see Section II for descriptions of these policies) focuses on fatal traffic crashes as outcome measures. However, this literature exhibits a wide variety in the particulars of the data and methodologies used, and, not surprisingly, conflicting conclusions have been reached about nearly all of the policies. Dram shop law appears to be perhaps the only policy measure which the literature unanimously finds to be significantly effective. The table in Appendix A summarizes a number of studies from the literature published in the last ten years. Ruhm (1995) and Benson et. al (1999) are two of the most comprehensive. One policy which has drawn particular attention from economists is the beer tax. Ruhm (1995) and Chaloupka, Saffer, and Grossman (1991) find significant effects, in which higher beer taxes are associated with lower crash rates. However, more recent evidence from Dee (1999b) and Young and Likens (2000) do not find strong effects, although Dee (1999a) suggests at least that the effect is significant for drivers under 21 years old. The relationship of this study to the literatures This study presents findings that are novel in a number of respects for the 0.08 BAC literature and the more general drunk driving policy. In the context of the 0.08 BAC literature, this study conducts a baseline analysis similar to that of Dee (2001), but with the addition of two potentially important controls: graduated licensing programs and

17 16 the presence of MADD. This study then examines the timing of the effect of the 0.08 limit, an issue which has been ignored in drunk driving policy literature. Using the estimates of the policy effects, a welfare analysis is then sketched of the lowering of the BAC limit from 0.10 to.08, another issue which has received virtually no attention previously. Regarding the literature on drunk driving policies in general, this study represents the first analysis of the effects of graduated licensing programs and of the presence of MADD chapters with this type of empirical model 3. The most recent data available is necessary to analyze graduated licensing programs, because they are such a recent phenomenon. This study constructs unique data on the number of MADD chapters in each state-year in order to estimate the impact of the organization s presence. This study also presents new evidence on the effect of zero tolerance laws, which, like the 0.08 law, have previously been analyzed mostly using only one or a handful of different states. IV. Empirical model, data and variables Empirical model Like many of the studies discussed in the previous section, this study employs an OLS regression with state and year fixed effects to evaluate the effectiveness of policies. The unit of analysis is the state-year, and the data covers all 50 states plus D.C. and the years The empirical model is represented by the following equation: Y it =B 0 +X 1it B 1it +X 2it B 2it +S i +T t + ε it In the equation above, Y is a traffic crash rate variable; B 0 aconstant;x 1 a vector of policy variables (e.g. BAC limit); X 2 a vector of other controls (e.g. unemployment rate); S a vector of state dummy variables (state fixed effects); T a vector of year dummy variables (year fixed effects); and ε an error term. The inclusion of fixed effects helps to control for heterogeneity in traffic crash rates across states and time which is unrelated to the policies and other control variables. For example, the crash rate in Utah is significantly lower than in most other states. We 3 McCarthy and Ziliak (1990) study the effects of MADD on fatal crashes and non-fatal crashes in California counties They do not use a fixed effects model such as the one in this analysis.

18 17 would not want necessarily to attribute this lower rate to the particular policies in Utah, because they could well be due to some unobserved, persistent feature of the state, such as the attitude towards drinking. Similarly, the crash rates for later years in the panel of states are generally significantly lower than those in the earlier years. We would not want necessarily to attribute these lower rates to the policies which exist in the later years, as the differential could be due to some unobserved feature of the later years which is consistent across states, such as enhanced automobile safety features (compared to the earlier years). The net result of the state and year fixed effects is to focus attention on changes in crash rates over time within particular states. Policy effects are only be observed if these changes in crash rates within states are significantly associated with changes in policies within these states. The empirical strategy is thus similar to differences-in-differences methods used often in the health economics literature, with the main difference being that the fixed effects regression controls for a number of explanatory factors simultaneously (whereas the diff-in-diff approach examines just one explanatory factor). Data and variables: outcome measures Table 4.1 lists the various fatal traffic crash rate variables that serve as outcome measures in this study. As discussed previously, all of the measures involve fatal traffic crashes due to the quality of such data, which comes from the NHTSA s Fatal Accident Reporting System (FARS), a complete annual census of all fatal traffic crashes in the U.S.. The FARS includes substantial information about the circumstances of each crash and the people and vehicles involved. A variety of measures is used because there is no single measure that is clearly ideal. The first variable listed, ALL, represents the total fatal crash rate, including both alcohol-related and non-alcohol-related crashes. This variable has the advantage of being measured with virtually no error, and perhaps represents the outcome that is most relevant to policymakers. On the other hand, it is a somewhat blunt measure for measuring the impact of policies related to drunk driving, since it includes more than just alcohol-related crashes (which represent less than half of all crashes in most state-years).

19 18 The variables HIBAC and ANYBAC are fatal crash measures which do include only alcohol-related crashes. HIBAC includes fatal crashes with at least one driver at or above a BAC of 0.10, while ANYBAC includes crashes with at least one driver at or above 0.01 BAC. Thus these two variables have the advantage of being sharper instruments for detecting the effects of drunk driving related policies. However, they also have the potentially serious drawback of being measured imprecisely. The BACs of a substantial portion of drivers in fatal crashes were not tested (more than half in many state-years), and in these cases FARS imputes the probabilities that drivers had BACs within the following ranges: 0.00; ; and This imputation is based on a discriminant analysis, using information including police reports on alcohol involvement, the time of the crash, and the type of crash (Klein 1986). Given that the first three outcome measures listed in Table 4.1 each have their drawbacks, the next two, WKND NIGHT and WKND NIGHT/SV provide valuable robustness checks for results involving the the first three. WKND NIGHT includes fatal crashes that occur on weekend nights (7 pm 5 am on Friday or Saturday nights), while WKND NIGHT/SV includes a subset of those weekend night crashes: those which involve only one vehicle. As previous authors have pointed out (e.g. Dee 2001), crashes on weekend nights and single vehicle crashes have particularly high proportions of alcohol involvement. Thus, we would expect for policy effects that are estimated to be significant for ALL, HIBAC, and ANYBAC to be also significant for WKND NIGHT and WKND NIGHT/SV. These last two variables are similar to ALL in the sense that they are both measured with virtually no error, but include many non-alcohol-related crashes and thus might be somewhat blunt measures of what is affected by policies. The final outcome variable, UNDER21, includes all fatal crashes involving at least one driver under 21 years old. This outcome measure is used to explore possible policy effects for this particular age group.

20 19 Table 4.1: Outcome measures Var. name mean description ALL total fatal crash rate: # crashes per 10K licensed drivers HIBAC hi BAC fatal crash rate: estimated # hi-bac (>=0.10) crashes per 10K licensed drivers ANYBAC any positive BAC fatal crash rate: estimated # hi-bac (>=0.01) crashes per 10K licensed drivers WKND NIGHT WKND NIGHT/SV weekend/nighttime fatal crash rate: # fatal crashes occuring between 7 pm- 5 am on Friday or Saturday night, per 10K licensed drivers weekend/nighttime/single vehicle fatal crash rate: # single vehicle fatal crashes between 7 pm and 5 am on Friday or Saturday night, per 10K licensed drivers UNDER # fatal crashes which involved at least one driver aged years old, per 10K population of that age Data and variables: policies Table 4.2 below summarizes the policy variables included in this study. With the exceptions of BTAXR and MADD, all of the policy variables are binary, equal to 1 if the policy is in effect and equal to 0 if not. In cases in which a policy is adopted mid-year in a state, the value of the variable is set to the fraction of the year for which it was in effect. The source for most of the policy data is the NHTSA, either through its annual publication Digest of State Alcohol-Highway Safety Related Legislation, private correspondence, the annual publication Traffic Safety Facts, or the organization s web site, The beer tax data was drawn from The Brewer s Almanac published by The Beer Institute, and the data for the MADD variable was collected via private correspondence with MADD representatives in state offices and the national headquarters. The empirical analysis does not include all of the policies mentioned in Section II (Table 2.1). In the interest of preserving degrees of freedom, a handful of policies were omitted (mandatory fines, implied consent, sobriety checkpoints, anti-plea bargaining, comparative negligence, and speed limits), each for one or more of the following reasons: prior studies have shown that the policy does not have signficant effects on traffic crash

21 20 rates; the policy cannot be adequately quantified; or the policy does not have much variation over the sample period at the state level. Table 4.2: Policy variables Note: all variables below are binary dummies, which equal 1 if the policy is in effect and 0 if not, except for btaxr and madd Var. name mean description BAC illegal per se to drive with a BAC >= 0.08 BAC illegal per se to drive with a BAC >= 0.10 ZERTOL zero tolerance : drivers under 21 are guilty of DUI if their BAC exceeds.02% (.01% or.00% in some states) ALR administrative license revocation: license is automatically revoked or suspended if BAC exceeds a certain amount (usually the same as the illegal per se limit) JAIL a mandatory jail sentence for 1 st conviction of DUI (can usually be replaced by community service) PBT law enforcement officer can give driver a preliminary breath test prior to arrest DRMSHP dram shop law: establishments serving alcohol can be held at least partially liable for ensuing crashes involving alcohol OPCONT restrictions or bans on open containers of alcohol in vehicles MLDA the minimum legal drinking age is 21 years old BTAXR real state excise tax (in 1999 cents) on 1 gallon of beer BELT primary or secondary enforcement of seatbelt use GRADLI graduated licensing program for young drivers MADD # of MADD chapters existing For a fuller description of the policies listed in Table 4.2, see Section II. Note that the policies are grouped in the table by the categories discussed in that section: direct deterrence measures (BAC08, BAC10, ZERTOL, ALR, JAIL1); indirect deterrence measures (PBT, DRMSHP, OPCONT); general alcohol control measures (MLDA21, BTAXR); general safety measures (BELT, GRADLI); and other initiatives (MADD). Data and variables: other controls In addition to the policy variables discussed above, right hand side variables in the analysis include a handful of other factors which might be expected to affect traffic crash rates. These controls are summarized in Table 4.3 below. INCOMR and UNEMPL control for economic conditions: real per capita income and the unemployment rate, respectively. These data were taken from the Bureau of Economic Analysis (BEA) and

22 21 the Bureau of Labor Statistics (BLS) respectively, both via In general, more favorable economic conditions, as represented by higher income and lower unemployment, are expected to lead to more driving, and thus higher traffic crash rates. Next, AVGAGE represents the average age of people over 16. These data were collected from the U.S. Census Bureau, via Since young drivers tend to have higher crash rates, lower average ages should produce higher crash rates. Finally, VMTPER represents a measure of how many thousands of miles people drive on average. The Department of Transportation s annual publication Highway Statistics provides this information. Of course, we would expect that higher levels of VMTPER would be associated with higher crash rates. The reason that both VMTPER and the economic indicators, INCOMR and UNEMPL, are included is that VMTPER is not necessarily measured perfectly (the data are projected based on limited samples) and economic factors could influence crash rates independent of their effect on miles driven (e.g. higher income could lead to visits to bars instead of visits to friends houses). Table 4.3: Other control variables INCOMR real per capita income (1999$) UNEMPL unemployment rate AVGAGE average age of driving age population (16 and over) VMTPER thousands of vehicle miles traveled (VMT) per licensed driver Independent variation of policies A common concern in an empirical analysis of multiple policies is whether the policies vary independently enough in the sample to allow for sorting out the individual policy effects. Table 4.4 addresses this issue, with particular emphasis on the 0.08 BAC policy. The table shows, for each of the 18 states which adopted the 0.08 BAC law by 2000, when each of the different drunk driving-related policies included in this study were adopted by the state. Inspection of the table reveals that the 0.08 BAC law does not tend to be adopted in obvious combination with any other particular policy. Furthermore, the table suggests that none of the other policies tend to clump together consistently. This pattern generally holds for the other 33 states (not shown here).

23 Table 4.4: States which have adopted.08% BAC law: when did they adopt other laws related to drunk driving? <= none AL CA DC FL HI ID IL KS KY bac10 drmshp mlda21 bac10 drmshp opcont alr bac10 drmshp bac10 pbt* drmshp bac10 drmshp opcont bac10 drmshp opcont mlda21 bac10 drmshp opcont pbt jail1 opcont mlda21 drmshp mlda21 belt belt mlda21 alr bac 08 mlda21 belt opcont alr jail1 belt mlda21 alr belt bac 08 zertol alr pbt jail1 opcont gradli pbt zertol gradli jail1 pbt opcont zertol bac 08 belt mlda21 zertol alr bac 08 belt alr pbt zertol mlda21 pbt bac10 belt alr bac10 jail1 bac 08 belt pbt* gradli zertol zertol zertol bac 08 bac 08 zertol bac 08 bac 08 jail1 gradli jail1 pbt gradli pbt jail1 gradli drmshp gradli alr opcont gradli *FL has had pbt law from , and then 1996-