IR 2 : Daytime running lights

Transcription

1 IR 2 : Daytime running lights A systematic review of effects on road safety Rune Elvik Peter Christensen Svenn Fjeld Olsen TØI report xxx/2003 Copyright Transportøkonomisk institutt, xxxx 1

2 1 Executive Summary - Daytime running lights: A systematic review of effects on road safety This report presents a systematic review of the effects on road safety of daytime running lights (DRL) for cars and motorcycles. The objective of this review is to provide the best current estimates of the effects on road accidents of using DRL on cars or motorcycles. In addition to reviewing the safety effects of DRL, the report presents an analysis of costs and benefits of introducing mandatory use of DRL in the European Union. 1.1 Questions answered in this report The report provides answers to the following questions regarding the effects of daytime running lights on road safety: 1. What is the best estimator of the effects on accidents of daytime running lights? 2. What is the quality of the evidence provided by studies that have evaluated the safety effects of daytime running lights? Can these studies be trusted? 3. Are the effects of daytime running lights novelty effects that are likely to erode and possibly disappear completely over time? 4. What is the relationship between the effects of daytime running lights for each vehicle using it (intrinsic effects) and the effects of laws or campaigns that lead to an increased use of daytime running lights in a country or part of a country (aggregate effects)? 5. Do the effects of daytime running lights vary depending on geographical latitude? 6. Do the effects of daytime running lights vary with respect to accident severity? 7. Do the effects of daytime running light depend on season (winter/summer)? 8. Does daytime running lights have adverse effects on accidents involving pedestrians, cyclists or motorcyclists or on rear-end collisions? 2 Copyright Transportøkonomisk institutt, xxxx

3 9. What are the costs and benefits of requiring all motor vehicles in the European Union to use daytime running lights? 1.2 Approach taken to answer questions In order to answer the questions asked above, a systematic review of studies that have evaluated the effects on road safety of daytime running lights has been undertaken. A literature search was conducted in order to identify relevant evaluation studies. A total of 25 studies that have evaluated the safety effects of DRL for cars and a total of 16 studies that have evaluated the safety effects on DRL for motorcycles were found. Evidence from these studies has been summarised by means of meta-analysis. A meta-analysis is a statistical analysis of estimates of effect provided by each study, with the objective of combining all estimates of effect into a summary estimate. The summary estimate of effect should ideally be an unbiased estimate of the weighted mean effect, based on the individual estimates of effect. An appendix to the report explains the details of the meta-analysis. Results of the meta-analysis of safety effects were used as input to the cost-benefit analysis. Estimates of the costs of daytime running lights were based on previous studies. 1.3 Main findings of the study The main findings of the study, presented in the same order as the questions asked above, can be summarised as follows Estimators of the effects of daytime running lights Three estimators of the safety effects of daytime running lights have been used in evaluation studies. These three estimators are labelled the accident rate ratio, the odds ratio and the ratio of odds ratios. In the report, estimates of effect obtained by each of the estimators have been compared. In nearly all cases, the three estimators give estimates that are close to each other. The choice of estimator for measuring the safety effect of daytime running lights does therefore not decisively influence the conclusions reached. A simple simulation was conducted in order to test the performance of the three estimators in the presence of various confounding effects that could generate biased estimates of the effects of DRL. It was found that neither of the three Copyright Transportøkonomisk institutt, xxxx 3

4 estimators adequately control for confounding, and that all of them can give biased estimates of the effects of DRL. In view of this, it is necessary to critically assess the quality of studies that have evaluated the effects of DRL Quality of evaluation studies A systematic assessment of the quality of evaluation studies was made, based on four criteria of study quality: 1. The extent to which a study specifies accident severity. 2. The extent to which a study specifies various types of accident that may be affected by DRL. 3. The extent to which a study controls for confounding factors that may distort estimates of the effects of DRL. 4. Whether or not a study provides information on the use of DRL in the country or region in which the study was conducted. Based on these criteria, an overall score for study quality was developed, ranging from 0 (worthless study) to 1 (perfect study). No evaluation study obtained the maximum score for quality. Recent studies of the intrinsic effects of DRL for cars tend to be of poorer quality than older studies. Studies of the effects of DRL for motorcycles are of poorer quality than studies of the effect of DRL for cars. In general, studies do not have an ideal quality and it cannot be ruled out that some estimates of the effects of DRL are influenced by confounding factors Novelty effects There are indications that the effects of laws mandating the use of DRL tend to diminish over time. However, not all studies that have evaluated the effects of DRL-laws have found such a tendency. It is therefore concluded that further study is needed in order to determine more precisely the duration and stability over time of the effects of DRL-laws. As far as the cost-benefit analysis in this report is concerned, it has been assumed that the effects of a DRL-law are maintained for a period of at least twelve years Intrinsic and aggregate effects of DRL The best current estimates of the intrinsic and aggregate effects of DRL are as follow s: 4 Copyright Transportøkonomisk institutt, xxxx

5 Intrinsic effect of DRL for cars: 5-10% reduction of multi-party daytime accidents. Effect of DRL-laws or campaigns for cars (aggregate effects): 3-12% reduction of multi-party daytime accidents. Intrinsic effect of DRL for motorcycles: 32% reduction of multi-party daytime accidents (a very uncertain estimate, 95% CI: -64%, +28%). Effect of DRL-laws or campaigns for motorcycles (aggregate effects): 5-10% reduction of multi-party daytime accidents Variation in effects of DRL with respect to geographical latitude The relationship between geographical latitude and effects of DRL has been investigated. Only a very weak relationship was found, indicating that the effects of DRL tend to increase the further away from the Equator one moves. In the cost-benefit analysis, however, a single estimate of effect for the whole European Union has been applied Variation in effects of DRL with respect to accident severity There is evidence showing that the effects of DRL vary according to accident severity. DRL has the greatest effects on the most severe accidents. Unfortunately, evidence concerning the effects on fatal accidents is rather weak and inconsistent, and does not permit a very reliable prediction of the effect to be made. In the cost-benefit analysis, DRL has been assumed to reduce fatal multi-party daytime accidents by 15%, serious injury multi-party daytime accidents by 10%, and slight injury multi-party daytime accidents by 5%. No effect on property- accidents has been damage-only assumed Seasonal variation in effects of DRL Evidence concerning variation in the effects of DRL with respect to season is sparse and inconclusive. In the cost-benefit analysis, no seasonal variation in effects has been assumed. Copyright Transportøkonomisk institutt, xxxx 5

6 1.3.8 Potential adverse effects of DRL for certain types of accident Based on the meta-analysis, it is concluded that the DRL is unlikely to have any adverse effects on accidents involving pedestrians, cyclists or motorcyclists. Some estimates indicate an adverse effect of DRL-laws for pedestrian accidents, but the summary estimate of effect, taking all individual estimates into account, indicates a reduction in pedestrian accidents. It is likely that a DRL-law could have an adverse effect on rear-end collisions. Again, however, the evidence is somewhat mixed and alternative interpretations cannot be ruled out. The use of high mounted stop lamps is encouraged in order to counteract any adverse effect of DRL on rear-end collisions. The use of full beam headlights on motorcycles can make them more conspicuous without causing glare to other road users Costs and benefits of mandatory use of DRL in the European Union Costs and benefits of five different policy options for the introduction of mandatory use of DRL in the European Union have been assessed. These five options are: 1. Requiring drivers to turn on headlights during daytime (behavioural measure only). 2. Requiring drivers to turn on headlights in daytime and requiring new cars to have automatic DRLs, using low beam headlights as DRLs (behavioural + technical). 3. Requiring drivers to turn on headlights in daytime and requiring new cars to have dedicated DRLs (behavioural + technical). 4. Requiring new cars to have automatic DRLs (low beam headlights; technical measure only) 5. Requiring new cars to have automatic DRLs (dedicated DRLs; technical measure only) Benefits were found to exceed costs for all policy options. Policy option 1 (behavioural measure only) was found to be the most cost effective, having a benefit-cost ratio close 2.0. The benefit cost-ratio for the other policy options ranged between 1.4 and 1.7. While it cannot be entirely ruled out that benefits are smaller than costs, the probability that this is the case is small, ranging between 12% and 25% for the various policy options. The results of the cost-benefit analysis made in this report agree with the results of other cost-benefit analyses that have been reported in recent years. 6 Copyright Transportøkonomisk institutt, xxxx

7 1.4 Concerns about meta-analysis The summary estimates of effects of DRL presented in this report are based on meta-analysis of evaluation studies. There are several techniques of meta-analysis, which can result in different summary estimates of effect. Extensive sensitivity analyses have therefore been undertaken in order to test the robustness of the summary estimates of effect with respect to various analytical choices made. These sensitivity analysis include the following: 1. The preparation of funnel plots and an examination of these with respect to the statistical properties of the distribution of estimates of effect (a single mode, no skewness, no outlying data points). 2. Testing for the presence of publication bias in the set of estimates available, and adjusting for publication bias if it is found. 3. The use of both a fixed-effects model and a random-effects model in all analyses. 4. Testing the sensitivity of summary estimates of effect with respect to the statistical weighting of individual estimates of effect. 5. Testing the sensitivity of summary estimates of effect with respect to the quality score assigned to each study. 6. Testing the sensitivity of summary estimates of effect with respect to the inclusion or exclusion of a single study. The results of these tests can briefly be summarised as follows. Funnel plots show that a summary, mean estimate of effect makes sense both with respect to DRL for cars and with respect to DRL for motorcycles. A weighted summary estimate of effect correctly represents the centre of gravity in all distributions of estimates of effect. A slight publication bias was found in all data sets. Adjusting for it had only a marginal effect on the summary estimates of effect, modifying them by at most one percentage point (e. g. from 9% to 8%). Summary estimates of effect are virtually identical for the fixed-effects and random-effects models of analysis. Estimates based on the random-effects analysis are statistically the most appropriate, as all data sets contained a large element of systematic variation in estimates of effect. Summary estimates of effect are not sensitive to the statistical weighting adopted. Summary estimates of effect are not very sensitive to study quality assessment. This means that if one were to exclude the worst studies, summary estimates of effect would remain more or less unaffected. Since no study gets a full score for quality, the real issue is whether the evidence provided by these studies should be Copyright Transportøkonomisk institutt, xxxx 7

8 taken seriously or rejected altogether. This report concludes that the evidence from evaluation studies should be taken seriously, for the following reasons: 1. There are theoretical reasons, based on studies of human perception, to believe that DRL improves conspicuity and that improving conspicuity can prevent accidents. 2. Randomised controlled trials of DRL have found them to be effective in preventing accidents. These findings have been reproduced in nonexperimental studies. 3. The findings of studies that have evaluated the effects of DRL are highly consistent. Very few estimates of effect depart from the general pattern by indicating an increase of the number of accidents. Despite the fact that shortcomings can be found in virtually all evaluation studies, it is concluded that the evidence from these studies is likely to mainly reflect the effects of daytime running lights, and not of confounding factors not controlled for in the studies. With very few exceptions, a single study does not decisively influence the summary estimates of effect. The summary estimates of effect therefore reflect the combined contributions to knowledge of all reported evaluation studies. 8 Copyright Transportøkonomisk institutt, xxxx

9 2 Background and research problems Previous studies evaluating the effects of daytime running lights indicate that road safety is improved. Two recent meta-analyses of evaluation studies (Elvik 1996, Koornstra, Bijleveld and Hagenzieker 1997) both conclude that the use of daytime running lights is associated with a reduction of the number of accidents. The use of daytime running lights as a road safety measure remains, however, sensitive and controversial. The European Commission has initiated a research project with the following objective: Assessment of the crash-reducing effectiveness of passenger car daytime running lights together with their effect on crashes into pedestrians, cyclists, motorcyclists and other vulnerable road users. Also, assessment of the crashreducing effectiveness of motorcycles daytime running lights together with an assessment of their effectiveness for reducing motorcyclists injury severity. This report presents an updated review of studies that have evaluated the safety effects of daytime running lights on cars and motorcycles. The report is intended to shed light on the following questions: 1. How can one best define and measure the effect of daytime running lights on the number of accidents? 2. What is the methodological quality of available evaluation studies? Can findings be trusted? 3. Are the effects attributed to daytime running lights novelty effects that are likely to erode over time? 4. What is the relationship between the usage rate for daytime running lights and the effects on road safety (dose-response relationship)? 5. Do the effects of daytime running lights vary systematically depending on geographical latitude? 6. Do the effects of daytime running lights vary systematically with respect to accident severity? 7. Do the effects of daytime running lights vary systematically with respect to season (winter/summer)? 8. What are the effects on accidents involving motorcyclists of requiring daytime running lights for cars? 9. What are the effects on accidents involving pedestrians or cyclists of requiring daytime running lights for cars? Copyright Transportøkonomisk institutt, xxxx 9

10 10. Are there adverse effects of daytime running lights on cars for other types of accident, in particular rear-end collisions? The objective of the review is to try to clarify these issues as far as possible. The approach taken in the report is to retrieve as many studies as possible that have evaluated the effects on accidents of daytime running lights (abbreviated to DRL). Before presenting the evaluation studies the report is based on, some comments will be given with regard to each of the research problems that are listed above. 2.1 The definition and measurement of the effects on road safety of DRL The reason for asking how best to define and measure the effect on accidents of DRL is that this issue has been considerably debated. At least three estimators of the safety effects of DRL have been proposed and used in evaluation studies (Theeuwes and Riemersma 1995, Elvik 1996). These estimators have all been developed for the purpose of controlling for confounding factors when estimating the effect of DRL on accidents. The estimators are:: 1. The accident rate ratio 2. The odds ratio 3. The ratio of odds ratio An accident rate ratio is the ratio of accidents per kilometre of driving for cars using DRL to accidents per kilometre of driving for cars not using DRL: Accident rate ratio = Number of accidents involving cars with DRL Kilometres of driving for cars with DRL Number of accidents involving cars without DRL Kilometres of driving for cars without CRL If the accident rate ratio is less than 1.0, the use of DRL is associated with an improvement in road safety. If the accident rate ratio is larger than 1.0, the use of DRL is associated with a deterioration of road safety. In most studies, it has been assumed that DRL only affects accidents in daytime involving more than one road user (multi-party daytime accidents). In studies that rely on this assumption, it is the number of multi-party daytime accidents that form the numerator in each estimate of the accident rate on which the accident rate ratio is based. In the following discussion, the notation below will be adopted: MD = multi-party daytime accidents 10 Copyright Transportøkonomisk institutt, xxxx

11 SD = single-vehicle daytime accidents MN = multi-party accidents at night SN = single-vehicle accidents at night a = using DRL, or after a law requiring the use of DRL has been passed b = not using DRL, or before a law requiring the use of DRL has been passed The three estimators of the safety effects of DRL can then be defined as follows: Accident rate ratio = (MD a /KMT a )/(MD b /KMT b ) Odds ratio = [MD a /(MN a + SD a + SN a )]/[MD b /(MN b + SD b + SN b )] Ratio of odds ratios = [(MD a /SD a )/(MN a /SN a )]/[MD b /SD b )/(MN b /SN b )] where KMT denotes vehicle kilometres of travel. All three measures of effect take on values less than 1.0 if DRL reduces the number of multi party daytime accidents. If the three estimators give very different estimates of the safety effects of DRL, one should discuss which measure to prefer. If, on the other hand, the three measures of effect give rather similar estimates of the safety effects of DRL, there is less reason to be concerned about the choice of a measure of effect. In this report, the robustness of the estimated effects on road safety of DRL with respect to the choice of estimator of effect will be evaluated. 2.2 The methodological quality of evaluation studies In applied road safety research, a multiplicity of study designs tend to be used, and not all of them are equally rigorous from a methodological point of view. It is known (Elvik 1997, Hauer 1997) that poor studies can give very biased estimates of the effect of a road safety measure. In reviewing and summarising the results of road safety evaluation studies, careful attention should therefore be paid to study design and quality. Studies that have evaluated the effects of DRL are no exception to this rule. These studies span a range of designs from randomised controlled trials (experiments) to simple before-and-after studies. An attempt will be made to assess the quality of DRL-evaluation studies according to the following criteria: 1. The extent to which a study rules out the influence of potentially confounding factors on the estimates of the effects of DRL. 2. The extent to which a study specifies the effects of DRL for different types of accidents. Copyright Transportøkonomisk institutt, xxxx 11

12 3. Whether or not a study states accident severity. 4. Whether or not a study states the level of use of DRL in the traffic system to which estimates of effect apply. Confounding factors are all factors, other than the use of DRL, that influence the number of road accidents, and whose effects can be mixed up with the effect of DRL. An example of a confounding factor is the presence of a long-term trend in the number of MD-accidents, apparent in the period before the use of DRL became common. Such a long-term trend can lead to a reduction of the number of MD-accidents, even if DRL has no effect. A more precise definition of the criteria of study quality will be given in Chapter The duration of the effects of daytime running lights There has been a concern that the effects attributed to DRL in some evaluation studies were novelty effects, that is effects that would be attenuated and might disappear completely as road users become accustomed to a high level of DRL- use. In this report, an attempt will be made to test whether the effects on road safety of DRL tend to become smaller as time goes by. This will be done by comparing the estimates of effect in studies that used periods of different length. If a smaller effect of DRL is found in studies that have observed safety for a long time during which DRL was used than in studies that have observed safety for a shorter time, the novelty effect hypothesis is strengthened. If no such relationship between the length of after periods and the size of the effect of DRL is found, the novelty effect hypothesis weakened. 2.4 The relationship of the effects of daytime running lights to the level of their use An issue which has been given some attention, both from a theoretical and empirical point of view, is the relationship between the level of use of DRL and the size of the effect on safety. Basically, evaluation studies provide two different, and non-comparable, estimates of the safety effects of DRL: 1. An estimate of the effect on the safety of an individual car or motorcycle of using DRL (intrinsic effect). 2. An estimate of the effect on the total number of accidents in a country of making the use of DRL mandatory, or otherwise encouraging it (aggregate effect). 12 Copyright Transportøkonomisk institutt, xxxx

13 The effect of DRL for the safety of each vehicle using it is generally referred to as the intrinsic effect of DRL. The effect of DRL on total number of accidents in country is often referred to as the aggregate effect of DRL. Koornstra et al (1997) have tried to model statistically the relationship between the intrinsic and aggregate effects of DRL. The objective of developing statistical models of this relationship, is to make all estimates of the safety effects of DRL comparable, by converting aggregate effects to intrinsic effects. Stone (1999) has criticised the models developed by Koornstra et al (1997) and suggested that these models rely on implausible assumptions and may give biased estimates of the effects of DRL. This report will not try to resolve this debate, but takes the position that estimates of the intrinsic and aggregate effects of DRL are not directly comparable. The functions proposed by Koornstra et al (1997) to describe the relationship between the aggregate and intrinsic effects of DRL can nevertheless serve as a theoretical framework for assessing this relationship. When assessing the validity of studies that have evaluated the effects of DRLlaws, it is important to test for the presence of a dose-response relationship. There is a dose-response relationship if a large increase in the use of DRL (e. g. from 20% to 90%) is associated with a greater effect on accidents than a small increase in the use of DRL (e. g. from 60% to 80%). As part of the review of evaluation studies, a test for such a relationship will be made. 2.5 Variation in effects of daytime running lights according to geographical latitude Previous studies (Elvik 1996, Koornstra et al 1997) suggest that there is a relationship between geographical latitude and the effect of DRL on road accidents. Larger effects are found the further away from the equator one moves. For the purpose of trying to predict the effect of DRL in countries where it is still not in wide use, knowing such a relationship is very useful. This report will therefore, once more, probe the relationship between the effect of DRL and latitude. 2.6 Variation in effects of daytime running lights according to accident severity Koornstra et al (1997) suggest that the effects of DRL on accidents vary according to accident severity. Few studies have evaluated whether there is an accident severity gradient with respect to the effects of DRL, and the estimates produced by Koornstra et al (1997) are based on rather few data points. It is important to know whether the safety effects of DRL vary according to accident severity, since this can greatly influence the balance of costs and benefits Copyright Transportøkonomisk institutt, xxxx 13

14 of this measure. Therefore, this report will use available data to test for an accident severity gradient in the effect of DRL. There is, however, a chance that such a test will be inconclusive, due to the paucity of evidence from evaluation studies. 2.7 Variation in effects of daytime running lights with respect to season It has been suggested (Koornstra et al 1997) that daytime running lights are more effective in reducing accidents in winter than in summer. This proposition is reasonable, in view of the often poorer conditions for visibility and contrast in winter than in summer. It should be recognised, though, that the length and severity of the winter is strongly correlated with latitude, so one should avoid double counting factors that influence the effects of DRL. Latitude may be one of the factors influencing the effect of DRL, and the additional effect of wintry conditions may be minor. To the extent that relevant data are available, the report will test whether the effects of DRL vary according to season. 2.8 Variation in the effects of daytime running lights with respect to types of accident Questions 8, 9, and 10 above all refer to variation in the effects of daytime running lights with respect to different types of accident. These questions refer to types of accidents, or groups of road users, for which there has been a concern that DRL may have an adverse effect. The first type of accident for which an adverse effect of DRL has been suspected is accidents involving motorcycles. In some countries, motorcycles have been required to use DRL long time before it became mandatory for cars. During the time when motorcycles only were using DRL, this gave them a conspicuity advantage compared to cars not using DRL. When cars start using DRL, this conspicuity advantage vanishes. This may adversely affect the number of accidents involving motorcycles. A second type of accident for which adverse effects have been hypothesised, is accidents involving pedestrians or cyclists. It has been suggested that pedestrians and cyclists become relatively less visible when motor vehicles have their headlights on. On the other hand, the enhanced conspicuity of motor vehicles make them easier to observe for pedestrians and cyclists. A third type of accident that may be adversely affected by DRL is rear-end collisions. When low beam headlights are used as daytime running lights, the rear ligh ts are lit. It has been claimed that this can make it more difficult to detect brake lights, thus possibly leading to more rear-end collisions. 14 Copyright Transportøkonomisk institutt, xxxx

15 In order to test these hypotheses regarding the effects of DRL, accidents have, whenever the source reports make it possible, been broken down into the categories listed in Figure 1. The target group of accidents is multi-party daytime accidents, broken down into: 1. Frontal- or side impacts 2. Rear-end collisions 3. Pedestrian accidents 4. Bicyclist accidents 5. Motorcycle accidents Copyright Transportøkonomisk institutt, xxxx 15

16 All accidents Single daytime Multiparty daytime Single nighttime Multiparty nighttime Front/ side Rearend Pedestrian Cyclist Motorcyclist Figure 1: Breakdown of accidents for evaluating the effects of daytime running lights 16 Copyright Transportøkonomisk institutt, xxxx

17 3 Study retrieval, data extraction and study quality assessment 3.1 Study retrieval Several studies that have evaluated the effects of daytime running lights have been retrieved from previous reviews, in particular Elvik (1996) and Koornstra et al (1997). In addition to identifying relevant studies on the basis of these reviews, a search was made of the TRANSPORT literature data base, using daytime running lights as the search term. The search resulted in a list of 136 references. An examination of the list revealed that 31 of the entries were duplicates or triplicates of other entries on the list. The true number of independent publications is, therefore, at most 105 (there may be duplicates that were not detected when the list was examined). When an abstract was provided, it was read and studies that were relevant were identified. A study was treated as relevant if it was an evaluation of the effects on road safety of daytime running lights. Studies discussing the use of DRL in general or theoretical terms, or primarily discussing the design of evaluation studies, were not treated as relevant for the purposes of this report. No restrictions on language or country of origin were used. Most of the publications listed in TRANSPORT are, however, in English. A total of 25 studies that have evaluated and quantified the effects on accidents of DRL for cars have been retrieved. A total of 16 studies that have evaluated and quantified the effects on accidents of DRL for motorcycles have been retrieved. This chapter will give a brief presentation of the information extracted from each study. 3.2 Information extracted from each study From each study, the following information was extracted: 1. Authors of the study 2. Country of origin for the data presented in the study 3. Year of publication 4. Type of publication (paper in scientific journal, conference paper, report from research institute, report from public agency) Copyright Transportøkonomisk institutt, xxxx 17

18 5. Type of estimate of effects of DRL (intrinsic or aggregate) 6. A code for study design (explained below) 7. A code for accident severity (fatal accidents, injury accidents, property damage only accidents or accidents of unspecified severity) 8. Codes for the type of accident to which estimates of effect refer (confer the typology in Figure 1, Chapter 1) 9. The number of accidents on which estimates of effect are based (if provided) 10. The level of use of DRL before a law was passed or a campaign conducted 11. The level of use of DRL after a law was passed or a campaign conducted 12. Accident rate ratio estimate of effect of DRL (if it can be derived) 13. Odds ratio estimate of effect of DRL (if it can be derived) 14. Ratio of odds ratios estimate of effect (if it can be derived) For some studies, some of this information will be missing or not relevant. For example, not all the definitions of the effect of DRL (accident rate ratio, odds ratio, ratio of odds ratios) will make sense for all studies. Broadly speaking, the following study designs have been used in studies that have evaluated the safety effects of DRL: 1. Randomised controlled trials, that is experiments in which vehicles are randomly allocated to one group using DRL and one group not using DRL. 2. Before-and-after studies using comparison groups not exposed to DRL (external comparison group). 3. Before-and-after studies using accidents that are assumed not to be influenced by DRL as comparison group (internal comparison group). 4. Before-and-after studies in which DRL was first introduced and then repealed, allowing for reverse effects to be observed. 5. Before-and-after studies providing data on some potentially confounding variables, in particular annual driving distance. 6. Simple before-and-after studies, not using any comparison group and not providing data on any potentially confounding variables. 7. Case-control studies, that is non-experimental comparisons of vehicles using and not using DRL. A randomised controlled trial is clearly the best study design. In principle, it controls for the effects of all confounding factors. Non-experimental study designs exist in many forms, that embody more or less rigorous control of a limited number of confounding factors. The most important potentially 18 Copyright Transportøkonomisk institutt, xxxx

19 confounding factors that need to be controlled in studies evaluating the road safety effects of DRL are: 1. Regression-to-the-mean. This could influence study results if DRL is introduced because of an abnormally high number of accidents. In such cases, one may expect the number of accidents to decline even if DRL has no effect. 2. Long-term trends. This refers to systematic changes in the number of accidents during several years. In order to identify a long-term trend, accident data for at least three, preferably four years, is needed. 3. Other road safety measures. In some cases, other road safety measures have been introduced almost at the same time as DRL, making it difficult to estimate the specific contribution of DRL (see e.g. Hollo 1998). 4. Changes in accident reporting. Some studies have noted that changes in accident reporting may have influenced their results (see e.g. Elvik 1993). 3.3 Study quality assessment A formal assessment of the quality of each evaluation study has been made. The assessment of quality is based on these criteria: Criterion A. Specification of accident severity (relative weight = 0.15) B. Specification of types of accident (relative weight = 0.15) C. Control for confounding factors (relative weight = 0.60) Points assigned 3 = fatal, injury and property-damage-only accidents are specified 2 = injury (including fatal) and propertydamage-only accidents are specified 1 = study is confined to injury accidents only (with no additional information on severity) 0 = accident severity is not specified 3 = study specifies all types of accident that are of interest (see Figure 1) 2 = study specifies some, but not all types of accident that are of interest 1 = study refers to MD-accidents as a whole (not subdivision) 0 = study does not make clear what type of accident estimates of effect refer to 3 = study controls for regression-to-themean, long-term trends, and other events Copyright Transportøkonomisk institutt, xxxx 19

20 D. Information on DRL-use (relative weight = 0.10) (measures) 2 = study controls for some, but not all, of these confounding factors 1 = study controls for just one confounding factor 0 = study does not control for any confounding factor 2 = detailed information is given on DRL-use before and after measure 1 = summary information is given on DRLuse before and after measure 0 = no information is given on DRL-use This assessment system consists of four items that are scored on ordinal scales. The single most important item is control of confounding factors. Hence, the following relative weights, summing to 1.0 are assigned to each of the items used to assess study quality: Item Relative weight A. Specification of accident severity 0.15 B. Specification of types of accident 0.15 C. Control for confounding factors 0.60 D. Information on DRL-use 0.10 An overall quality score is computed by multiplying the score for each item and summing. Hence the maximum score a study can get is: (0.15 x 3) + (0.15 x 3) + (0.60 x 3) + (0.10 x 2) = = 2.9 The minimum score is, of course, 0. The range of scores from 2.9 to 0.0 are divided by 2.9 so that the value 2.9 (maximum score) is given a value of 1.0 and the minimum score is given a value of 0.0. All other scores (between 2.9 and 0.0) are then converted to the scale spanning the interval (0, 1). In computing the overall quality score this way, the scale is in effect treated as an approximation to an interval scale. It can be seen that control for confounding factors contributes 62.1% to the maximum possible score (1.8 points out of 2.9). Scoring studies for quality is easy with respect to items A, B, and D. For item C, control of confounding factors, studies have been scored as controlling for regression-to-the mean if: 20 Copyright Transportøkonomisk institutt, xxxx

21 1. Cars using DRL were randomly assigned to this condition, or evidence is given in the report showing that cars using DRL did not have an accident rate that differed substantially from cars not using DRL during the beforeperiod, or 2. The study refers to introduction of a DRL-law in a country (with the rationale that the total number of accidents in country typically several thousand per year displays great stability over time, and that the introduction of a law is unlikely to be strongly influenced by an abnormal count of accidents in a given year). If neither of these two conditions obtain, a study has been scored as not controlling for regression-to-the-mean. Studies have been scored as controlling for long-term trends if: 1. Cars using DRL were randomly assigned to that condition (random assignment ensures that past trends should not differ between the DRLgroup and the control group), or 2. The study provides at least three years of before-data, enabling long-term trends to be discerned and controlled for. Studies that have at least three years of before-data, but did not control for an apparent trend, have been reanalysed so as to control for the trend present in the data. If neither of these two conditions obtain, studies have been scored as not controlling for long-term trends. Studies have been scored as controlling for other events, including the introduction of other road safety measures, if: 1. Cars using DRL were randomly assigned to that group, and the experiment involved the use of DRL only (and not other, additional measures), or 2. A study explicitly mentions that other road safety measures have been introduced and seeks to control for their impacts by way of study design or data analysis. The latter criterion means that the scoring of a study depends on whether or not it reports that other measures were introduced. If a study does not report on this, and if the study did not employ an experimental study design, it does not get the benefit of doubt and is scored as not having controlled for other events. Studies were scored for quality by the principal author of the report. Ideally speaking, scoring studies for quality should be done by at least two researchers independently, and preferably without knowing the results of the studies scored for quality. In the present report, however, the quality scores are used for sensitivity analysis only and are not intended as an exhaustive assessment of study quality. It was therefore decided to use the resources available for the project for doing more careful meta-analyses, rather than developing a very rigorous means of scoring studies for quality. Copyright Transportøkonomisk institutt, xxxx 21

22 3.4 Systematic presentation of studies that have evaluated DRL for cars A chronological and standardised presentation of studies that have evaluated the effects on road safety of DRL for cars is given below. The presentation is confined to studies that have been included in the meta-analysis presented in Chapter 3. A total of 25 studies have been included in the meta-analysis of the effects of daytime running lights on cars. The data extracted from each study is listed in Appendix 1, along with examples explaining how estimates of effect have been extracted. In chronological order, these studies are: 22 Copyright Transportøkonomisk institutt, xxxx

23 Study number 1 Authors Allen, M. J.; Clark, J. R. Country Year 1964 Publication Type of estimate Study design Intrinsic effects Aggregate effects Use of DRL United States American Journal of Optometry and Archives of American Academy of Optometry, Intrinsic (for Greyhound bus company) and aggregate (for the state of Oklahoma) Simple before-and-after study Reductions of between 7% and 24% in daytime accidents given for various divisions of Greyhound. Neither accident severity nor the number of accidents is stated. The following statistics are given for Oklahoma. The DRL-campaign was in Year Fatal accidents Injury accidents PDOaccidents A maximum of 80% is stated for No before data. Quality assessment Intrinsic effects: 0 (severity), 1 (accident type), 0 (confounding), 0 (DRL-use); overall (0, 1). Inclusion in metaanalysis Re-analysis of data Aggregate effects: 3 (severity), 0 (accident type), 2 (confounding; controls for regression and trends), 0 (DRL-use); overall (0, 1). Estimates of the intrinsic effect cannot be included in meta-analysis. For the aggregate effects, the following estimates, adjusted to control for long-term trends, are included (odds ratio estimator): Fatal and injury accidents: +21.7% Property-damage-only accidents: +8.0% The data presented above were re-analysed by fitting linear functions for long-term trends to each data series. A predicted number of accidents for 1961 was then Copyright Transportøkonomisk institutt, xxxx 23

24 compared to the actual number. Other data were not available for re-analysis. Study number 2 Authors Cantilli, E. Country Year 1965 United States Publication Traffic Engineering, 1965, 17 Type of estimate Study design Intrinsic effects Use of DRL Quality assessment Inclusion in metaanalysis Intrinsic Experimental (controlled randomised trial) Accident rate reduction of 42% for frontal- and side impacts, and 54% for rear-end collisions Vehicles (passenger cars and station wagons) with DRL were compared to those without. Overall use of DRL in experimental area is not stated. 0 (accident severity), 2 (accident types), 3 (confounding; controls for all factors), 0 (DRL-use); overall (0, 1) The two results given above have been included (accident rate ratio). Study number 3 Authors Cantilli, E. Country Year 1970 United States Publication Highway Research Record, 332, Type of estimate Study design Intrinsic effects Use of DRL Quality assessment Intrinsic Experimental (randomised controlled trial) 13% reduction of accident rate for frontal- and side impacts, 27% reduction of accident rate for rear-end collisions. Vehicles with DRL (same types as for study 2 above) were compared to those without. Overall use of DRL in experimental area is not stated. 0 (accident severity), 2 (accident types), 3 (confounding), 24 Copyright Transportøkonomisk institutt, xxxx

25 Inclusion in metaanalysis 0 (DRL-use); overall (0, 1). The two results presented above have been included in the meta-analysis (accident rate ratio). Copyright Transportøkonomisk institutt, xxxx 25

26 Study number 4 Authors Andersson, K.; Nilsson, G.; Salusjärvi, M. Country Finland Year 1976 Publication Type of estimate Study design Aggregate effects Use of DRL Report from Swedish Road and Transport Research Institute Aggregate Before-and-after with internal comparison group The following aggregate effects are presented: Type of accident Before to recommendation Before to mandatory Frontal/side -11% -11% Rear-end +2% +9% Pedestrian -35% -34% These estimates are all crude ratios of odds ratios (i. e. not adjusted for confounding by means of a multivariate analysis). Use of DRL was estimated at 50% in the before period, 84% in the recommendation period and 97% in the mandatory period Quality assessment 1 (accident severity), 2 (types of accident), 2 (confounding; regression-to-mean and other events), 1 (DRL-use); overall (0, 1). Inclusion in metaanalysis The results presented above have been included in the meta-analysis. In addition, six similar results based on the odds ratio and six similar results based on the accident rate ratio have been included. A total of 18 results are included for this study. 26 Copyright Transportøkonomisk institutt, xxxx

27 Study number 5 Authors Andersson, K.; Nilsson, G. Country Sweden Year 1981 Publication Type of estimate Study design Aggregate effects Use of DRL Quality assessment Inclusion in metaanalysis Report from Swedish Road and Transport Research Institute Aggregate Before-and-after with internal comparison group The following aggregate effects are presented: Type of accident Estimate of effect Frontal- and side impacts -9% Rear-end collisions -2% Pedestrian accidents -21% Cyclist accidents -17% These estimates are adjusted ratios of odds ratios derived from a log-linear model. Use of DRL was estimated to 50% in the before period and 95% in the after period 1 (accident severity), 2 (accident types), 2 (confounding; regression-to-mean and other events), 2 (DRL-use); overall (0, 1). The results presented above have been included in the meta-analysis. In addition, similar results were derived for the odds ratio and accident rate ratio, yielding a total of 12 estimates of effect extracted from this study. Copyright Transportøkonomisk institutt, xxxx 27

28 Study number 6 Authors Attwood, D. Country Canada Year 1981 Publication SAE Technical Paper Type of estimate Study design Intrinsic effects Use of DRL Quality assessment Inclusion in metaanalysis Intrinsic Experimental (randomised controlled trial) 96% increase of injury accident rate for all multi-party daytime accidents, 35% reduction of property-damageonly accident rate for all multi-party daytime accidents. Vehicles (passenger cars) with DRL were compared to those without. Overall use of DRL in experimental area is not stated. 2 (accident severity), 1 (accident types), 3 (confounding; all factors), 0 (DRL-use); overall (0, 1). The two results presented above have been included in the meta-analysis (accident rate ratio). Study number 7 Authors Stein, H. Country Year 1985 United States Publication SAE Technical Paper Type of estimate Study design Intrinsic effects Use of DRL Quality assessment Inclusion in metaanalysis Intrinsic Experimental (randomised controlled trial) 26% reduction of frontal- and side impact accident rate, 18% increase of rear-end collision rate per vehicle. Vehicles (passenger cars) with DRL were compared to those without. Overall use of DRL in experimental area is not stated. 0 (accident severity), 2 (accident types), 3 (confounding; all factors), 0 (DRL-use); overall (0, 1). The two results presented above have been included in the meta-analysis. In addition, two similar results based 28 Copyright Transportøkonomisk institutt, xxxx

29 on the odds ratio have been derived, for a total of 4 results based on this study. Copyright Transportøkonomisk institutt, xxxx 29

30 Study number 8 Authors Vaaje, T. Country Norway Year 1986 Publication Type of estimate Study design Aggregate effects Use of DRL Unpublished paper, Institute of Transport Economics Aggregate Before-and-after with internal comparison group The following aggregate effects are presented: Type of accident All multi-day, except pedestrian accidents Estimate of effect -17% Pedestrian accidents -20% These estimates are crude ratios of odds ratios (i. e. not adjusted for confounding). Use of DRL was estimated to 35% in the before period and 65% in the after period Quality assessment 1 (accident severity), 2 (types of accident), 1 (confounding; regression-to-mean), 1 (DRL-use); overall (0, 1). Inclusion in metaanalysis The results presented above have been included in the meta-analysis. In addition, similar estimates based on the accident rate ratio and odds ratio have been derived, for a total of 6 estimates of effect. 30 Copyright Transportøkonomisk institutt, xxxx

31 Study number 9 Authors Sparks, G. A.; Neudorf, R. D.; Smith, A. E. Country Canada Year 1989 Publication Type of estimate Study design Intrinsic effects Use of DRL Quality assessment Inclusion in metaanalysis Report from consultancy firm Intrinsic Before-and-after with internal comparison group There was a 33% reduction in daytime frontal- or side impacts resulting in personal injury, and a 23% reduction in rear-end collisions involving personal injury. For PDO-accidents, the corresponding reductions were 17% and 10%, respectively. These estimates are crude ratios of odds ratios. Presumably the use of DRL rose from 0% to 100% for the fleet studied. The general level of DRL use in the study area is not stated. 3 (accident severity), 2 (accident types), 2 (confounding; trend), 0 (DRL-use); overall (0, 1). The results stated above have been included in the metaanalysis. In addition, similar estimates of effect have been derived for the odds ratio and accident rate ratio, making for a total of 12 estimates based on this study. Copyright Transportøkonomisk institutt, xxxx 31

32 Study number 10 Authors Schützenhöfer, A.; Knoch, U.; Henökl, H. Country Austria Year 1990 Publication Type of estimate Study design Intrinsic effects Use of DRL Quality assessment Inclusion in metaanalysis Report from Kuratorium für Verkehrssicherheit Intrinsic Before-and-after with removal of measure Daytime injury accidents went down by 22% when DRL was used, and went further down by 17% when DRL was no longer used. For PDO-accidents, a reduction of 22% in daytime accidents was found when DRL was used. When DRL was no longer used, PDO accidents in daytime went up by 23%. Presumably all vehicles used DRL when it was required. Overall use in the study area is not stated. 2 (accident severity,) 2 (accident types), 2 (confounding; regression), 0 (DRL-use); overall (0, 1). The four results given above have been included in the meta-analysis (accident rate ratios). 32 Copyright Transportøkonomisk institutt, xxxx