Bicycle Tracks and Lanes: a Before-After Study

This paper presents a before-after crash, injury and traffic study of constructing bicycle tracks and marking bicycle lanes in Copenhagen, Denmark. Corrections factors for changes in traffic volumes and crash / injury trends are included using a general comparison group in this non-experimental observational study. Analysis of long-term crash trends points towards no significant abnormal crash counts in the before period. The safety effects of bicycle tracks in urban areas are an increase of about 10 percent in both crashes and injuries. The safety effects of bicycle lanes in urban areas are an increase of 5 percent in crashes and 15 percent in injuries. Bicyclists’ safety has worsened on roads, where bicycle facilities have been implemented. Design of bicycle facilities and parking conditions for motor vehicles clearly seems to have safety implications, especially at intersections. The study has revealed a few points in relation to this. Construction of bicycle tracks resulted in a 20 percent increase in bicycle / moped traffic mileage and a decrease of 10 percent in motor vehicle traffic mileage, whereas marking of bicycle lanes resulted in a 5 percent increase in bicycle / moped traffic mileage and a decrease of 1 percent in motor vehicle traffic mileage. The changes in traffic do result in health benefits due to more physical activity, less air pollution and less traffic noise. The positive benefits may well be much higher than the negative consequences caused by new safety problems. Søren Underlien Jensen 3 INTRODUCTION The traditional Danish bicycle track with a curb to the carriageway and a curb to the sidewalk is depictured in Figure 1 along with a bicycle lane. The first bicycle tracks in Denmark were introduced in Copenhagen as early as 1910. Since then about 8,000 km of bicycle tracks and paths with a dividing verge to the carriageway have been constructed so about every ninth km of road have these bicycle facilities in rural and urban areas in Denmark. FIGURE 1 Photos of bicycle track (left) and bicycle lane (right). Many studies of bicycle tracks have been undertaken in Northern Europe. A metaanalysis of 11 studies shows a reduction of 4 percent in crashes, and the crash reduction is almost the same for pedestrians, bicyclists and motorists respectively (1). Danish results show that construction of bicycle facilities leads to fewer and less severe crashes in rural areas, but more crashes in urban areas primarily due to increasing crash rates at intersections (2). Studies show that constructing bicycle tracks and paths increase bicycle traffic volumes (1). Three studies of marking bicycle lanes in urban areas indicate an increase in crashes of about 10 percent primarily due to more crashes at intersections (3-5). No reliable studies of bicycle lanes impact on traffic volumes have been found. The before-after traffic, crash and injury study, which is presented in the following, includes construction of one-way bicycle tracks on both road sides along 20.6 km of road and marking of one-way bicycle lanes on both road sides along 5.6 km of road in Copenhagen, Denmark. These bicycle tracks were constructed during the years 1978-2003 and the bicycle lanes were marked 1988-2002. The width of bicycle tracks are about 2-2.5 meters, whereas bicycle lanes are about 1.5-2 meters. The volume of motor vehicles 6-18 o’clock on a weekday on the studied roads varies from 5,000 to 28,000 and the corresponding volumes of bicyclists are 1,000-17,000. A Danish report describes the study in detail (6). SECOND-BEST METHODOLOGY Randomized experiments (7), where the experimental units like roads are randomized to treatment like bicycle lanes, are often viewed as the best way to study road safety effects. In our case, a randomized experiment has not been undertaken. The safety effects of bicycle facilities are therefore studied using an observational study methodology. The Empirical Bayes method (8) is viewed by many as the best of the non-experimental observational methods. The Empirical Bayes method accounts for three Søren Underlien Jensen 4 major possible biases in before-after crash studies; regression-to-the-mean effects, crash trends and traffic volumes. However, the Empirical Bayes method has not been used in this study. One thing is that using this method includes a very time-consuming effort to calculate many crash models, which is needed in this case because the bicycle facilities have been applied over a long period, and hence many different before and after periods are part of the study. Such crash models include relationships between crashes / injuries and traffic volumes for different types of intersections and road links. A second but much more important thing is that some of the roads, where bicycle facilities have been applied, are the most trafficked in Copenhagen in terms of bicyclists and pedestrians. The crash models that need to be developed if the Empirical Bayes method were to be used could be of the kind shown in general in Formula 2 and 3 later in this paper. Such crash models are relatively reliable to use in order to predict the number of crashes, if traffic volumes on the road or at the intersection, where you wish to predict crash figures, are pretty normal compared to the traffic volumes that the crash models are based upon. In the Copenhagen case, many of the studied roads / intersections are in the far end of the traffic volume axis, i.e. much trafficked, and we are therefore close to or outside the boundaries of the possible crash models’ valid area. The prediction of crash figures for these much trafficked roads / intersections are unreliable, because the beta figures of the crash models becomes crucial for the prediction, and these beta figures change from model to model primarily due to uncertainty, because the models are based on a relatively low number of roads / intersections. The prediction results for regression-to-the-mean effects and figures for expected crashes and consequently safety effects will therefore be relatively unreliable, because most of the crashes in this study actually take place on the much trafficked roads. Instead a stepwise methodology is used. First, a general comparison group is used to account for crash trends. Second, changes in traffic volumes are taken into account. And third, an analysis of long-term crash trends is made in order to check for abnormally high or low crash counts, i.e. regression-to-the-mean, in the before period. It was chosen to use equally long before and after periods, which for the individual studied roads was of 1-5 years duration. The expected number of crashes in the after period is calculated based on a formula, here shown in the general form: , ) 1 ( RTM Traffic Trend Before Expected C C C A A where AExpected is the number of crashes / injuries expected to occur in the after period if bicycle facilities were not applied, ABefore is the number of crashes / injuries that occurred in the before period, CTrend, CTraffic and CRTM are correction factors for crash trends, traffic volumes and regression-to-the-mean respectively. The study of bicycle facilities is part of project including studies of reconstructions, markings, signal-control and traffic calming schemes in the City of Copenhagen. A major effort was made in order to register all physical changes to the road network in the period 1976-2004. Several hundred schemes were identified. Several intersections and links had undergone more than one reconstruction or other scheme. Only “clean” schemes are studied, meaning that the roads, where bicycle facilities have been applied, no other scheme has been implemented in before and after periods and in the year(s), when the bicycle facility was applied. Søren Underlien Jensen 5 Unchanged roads with known developments in traffic volumes were used to set up a general comparison group. The Copenhagen Police District covers the entire area of the City of Copenhagen, and there is no indication that crashes are registered differently in one city quarter compared to another. The general comparison group consists of 110 km of roads with 170 locations, where motor vehicle and bicycle / moped traffic is counted yearly or every fourth to sixth year. A total of 24,369 crashes and 8,648 injuries occurred on the 110 km of roads in the period 1976-2004. Since a general comparison group has been chosen instead of a matched comparison group, an effort was made in order to avoid consequences of larger differences between general comparison group and treated roads, where bicycle facilities were applied. Trends for different types of crashes and injuries of the general comparison group were compared. Trends for intersection and link crashes are very similar, and hence no need for sub-grouping. However, trends for different crash / injury severities and transport modes exhibit rather different developments. It was found reasonable to describe trends by 7 crash sub-comparison groups and 5 injury sub-comparison groups. These sub-groups are defined in Table 1. TABLE 1 Definition of 12 Sub-comparison Groups (in Brackets: Number of Crashes / Injuries 1976-2004) Bicycle/moped a Pedestrian b Motor vehicle c Crashes with killed / severe injuries 1 (2,197) 2 (1,445) 3 (1,584) Crashes with minor injuries and no killed / severe injuries 4 (1,289) 5 (1,228) Property damage only crashes 6 (3,316) 7 (13,310) Killed and severe injuries 8 (2,235) 9 (1,477) 10 (1,907) Minor injuries 11 (1,359) 12 (1,670) a Crashes involving cyclists / moped riders and injuries in these crashes, b Crashes between pedestrians and motor vehicles and injuries in these crashes, c Crashes only with motor vehicles involved and injuries in these crashes. So the correction factor CTrend is actually 12 different correction factors, which is the number of crashes / injuries in the sub-comparison group in the after period divided by the number of crashes / injuries in the sub-comparison group in the before period. The individual correction factor, e.g. CTrend,1, is then multiplied with the same sub-group of crashes, which occurred on the treated road in the before period, ABefore,1, as part of Formula 1. The correction factor CTraffic is based on changes in traffic volumes on the treated road and in the general comparison group. The relationship between traffic flow and crashes / injuries is non-linear. Danish crash prediction models for links (Formula 2) and intersections (Formula 3) are most often of the following kinds: , ) ( ) 3 ( , ) ( ) 2 (