Safety Effects of Converting Intersections to Roundabouts

Transcription

1 Jensen 1 Safety Effects of Converting Intersections to Roundabouts Initial Submission Date: 26 July 2012 Submission Date of Revised Paper: 17 October 2012 Word count: 4,248 words + 1 figure + 11 tables = 7,248 words Author: Søren Underlien Jensen Trafitec ApS Research Park Scion-DTU Diplomvej Kongens Lyngby Denmark Tel.: (+45) Fax: (+45) suj@trafitec.dk

2 Jensen 2 ABSTRACT This paper presents a before-after crash study of converting intersections to roundabouts in Denmark. General crash trends and regression-to-the-mean effects are accounted for by using correction factors estimated on the basis of 32 general comparison groups in this nonexperimental observational study. 332 converted sites, 57 fatalities, 1,271 other injuries and 2,497 crashes are included in the study. Conversions have resulted in decreases in the numbers of crashes and injuries of 27% and 60% respectively. Crashes become less severe e.g. fatalities decreased by 87%, whereas property-damage-only (PDO) crashes only decreased by 16%. Safety effects vary between sites. The safety effect becomes better as the speed limit on the roundabout arms becomes higher. As the share of crashes at intersections, which are left-turn and angle crashes, becomes higher the safety effect becomes better. As the share of crashes at intersections, which are bicycle crashes, becomes lower the safety effect becomes better. Central islands more than two meters (6.6 feet) high produce better safety effects compared to lower central islands. Triangle or trumpet splitter islands produce better effects than no or parallel splitter islands. The numbers of bicycle crashes and injured cyclists have increased by 65% and 40% respectively due to the conversions. Cycle lanes next to the circulating lane have produced the worst safety effects for cyclists, whereas cycle paths without priority to cyclists resulted in the best effects. Colored cycle lanes and blue cycle crossings have produced worse safety effects for cyclists than comparable bicycle facilities without color.

3 Jensen 3 INTRODUCTION A roundabout in Denmark is defined as having a one-way counterclockwise circulation for motor vehicles and yield or signal control for entering vehicles. Only sites that are converted to this sort of modern roundabout are included in the presented study. This means that traffic circles, where circulating vehicles have to yield for entering vehicles, are excluded. The before-after crash study includes 332 sites, where intersections have been converted to roundabouts in the years These converted sites are located in 61 of Denmark s 98 municipalities. A site was only included if the number of roads going into the intersection(s) before was the same as the number of roundabout arms. Figure 1 shows an included conversion of a 4-armed signalized intersection to a 4-armed roundabout in the two aerial photographs at the top and an included conversion of two 3-armed staggered intersections to a 4-armed roundabout. FIGURE 1 Aerial Photographs of two Converted Sites Before and After. Modern roundabouts have been a popular type of crossroad in the past three decades in Denmark. At the beginning of 2010 there existed about 1,450 roundabouts in Denmark, which is rather many for a country with only 73,000 kilometers (45,000 miles) of road and 5.5 million inhabitants. A recent systematic literature study estimating safety effects based on meta-analyses of 19 before-after crash studies of converting intersections to modern roundabouts shows that these conversions have resulted in decreases of injury and PDO crashes of 60% and 25% respectively (1). The literature study shows decreases of fatalities, severe and slight injuries

4 Jensen 4 of 87%, 75% and 66% respectively. It also shows an increase of 21% in bicycle crashes due to the conversions. Roundabouts with cycle lanes next to the circulating lane have the worst safety effects for cyclists, whereas roundabouts with cycle paths without priority to cyclists crossing roundabout arms have the best safety effects. Roundabouts without bicycle facilities are safer for cyclists than roundabouts with cycle lanes, but more dangerous than roundabouts with cycle paths. The study presented in the following was initiated due to the safety problems that cyclists encounter at roundabouts. These problems have been well-known for years, but acceptable and safe designs for cyclists at roundabouts have not been identified in previous Danish studies. The objectives of the presented study are to quantify safety effects for cyclists and other road users of converting intersections to roundabouts, and reveal the influence that intersection and roundabout designs have on these safety effects, and also quantify both the short- and long-term safety effects of the conversions. The study has been published in a Danish report with English summary (2). METHODOLOGY Safety effects of converting intersections to roundabouts are studied using an observational study methodology, where the observed number of crashes in a period after the conversions is compared to the expected number of crashes for the same period. The expected number of crashes is estimated on the basis of the number of crashes in a period before the conversions and corrections for confounding factors. The before period is 5 years long (1 January to 31 December) for all converted sites, whereas the after period is 1-5 years long. Usually it is good practice to use a methodology, which accounts for three major possible biases in beforeafter crash studies; crash trends, regression-to-the-mean effects and traffic volumes. Traffic volumes before and after the conversions have not been measured at most of the converted sites. Therefore, it is not possible to account for changes in traffic volumes. A stepwise methodology is used. First, general comparison groups are used to account for crash trends. Second, analyses of long-term crash trends are made in order to check for abnormally high or low crash counts, i.e. regression-to-the-mean, in the before period. The expected number of crashes in the after period is calculated based on a formula, here shown in the general form: ( 1) AExpected ABefore CTrend CRTM, where A Expected is the number of crashes / injuries expected to occur in the after period if the conversions were not implemented, A Before is the number of crashes / injuries that occurred in the before period, C Trend and C RTM are correction factors for crash trends and regression-tothe-mean respectively. Correction for General Crash Trends Crashes occurring in the 61 municipalities, in which the conversions took place, were used to set up a general comparison group. In the years a total of 448,465 crashes and 176,373 injuries occurred in the general comparison group excluding crashes and injuries at converted sites. Since a general comparison group was chosen instead of a matched comparison group, an effort was made in order to avoid consequences of larger differences between the general comparison group and converted sites. Trends for the municipalities and

5 Jensen 5 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 and injury severities, transport modes and municipalities exhibit rather different developments. It was found reasonable to describe crash trends by 32 sub-comparison groups. These sub-groups are defined in Table 1. PDO crashes are recorded by Danish Police with or without a police report. A police report includes solid descriptions of the crash site, witness statements, vehicle inspections, etc. A PDO crash without a police report is most often a police recording based on phone calls to the police, and the police has not necessarily been a the crash site, called witnesses or seen vehicles. Broken bones, burns and concussions are viewed as severe injuries by Danish Police, whereas minor bruises, cuts and abrasions are not viewed as injuries. TABLE 1 Definition of 32 Sub-comparison Groups. The per cent is the Average Yearly Change and in Brackets are Numbers of Crashes or Injuries Type of crashes or injuries City of Copenhagen Municipalities with good crash Municipalities with poor crash developments developments Injury crashes with fatalities/severe injuries in urban areas -5.4% (12,540) -5.9% (21,815) -4.5% (21,329) Injury crashes with only slight injuries in urban areas +1.2% (5,218) -5.3% (15,816) -1.9% (15,754) PDO crashes with vulnerable road users in urban areas -4.0% (12,873) +1.0% (9,590) +3.2% (9,785) PDO crashes without vulnerable road users in urban areas -4.6% (39,730) -3.8% (44,557) -1.7% (43,731) PDO crashes without police report in urban areas % (25,819) +3.7% (26,589) Fatalities and severe injuries in urban areas -5.4% (13,363) -6.1% (15,988) -4.5% (15,492) Slight injuries in urban areas +1.4% (6,111) -5.2% (8,401) -1.9% (9,063) Injury crashes with fatalities/severe injuries in rural areas % (22,838) -3.3% (22,365) Injury crashes with only slight injuries in rural areas % (18,775) +0.1% (20,197) PDO crashes with police report in rural areas % (23,563) +0.4% (23,221) PDO crashes without police report in rural areas % (19,525) +2.7% (20,090) Fatalities and severe injuries in rural areas % (19,707) -3.6% (20,120) Slight injuries in rural areas % (15,112) +0.1% (15,541) So the correction factor C Trend is actually 32 different correction factors, which is the number of crashes or injuries in the sub-comparison group in the after period divided by the number of crashes or injuries in the sub-comparison group in the before period. Correction for Regression-to-the-Mean The analyses of long-term crash trends are made in order to check for abnormal crash counts, i.e. regression-to-the-mean, in the before period. Analyses are made using a before-before period, which is a 4-year period 9 to 12 years before conversions. The before-before period is chosen because it most often will be prior to an eventual black spot identification period or other type of systematic crash investigation period that may have lead to converting intersections to roundabouts. This before-before period is then used to calculate an expected number of crashes and injuries in the before period of the treated roads by making corrections for crash trends: ( 2) A Expected, Before A Before Before C Trend

6 Jensen 6 The C RTM correction factor is then calculated as the expected number of crashes in the before period divided by the observed number of crashes in the before period, and likewise for injuries. However, because not all converted sites can undergo this type of analysis, the C RTM is set to be the same for similar conversions and is only used, if the difference between the expected and observed numbers of crashes or injuries in the before period is statistically significant. Of the 332 converted sites it is possible to make the calculation of A Expected, Before for 324 sites. Several analyses were then made in order to find the best way to calculate the correction factors for regression-to-the-mean, C RTM. Many intersections have been converted to roundabouts as a black-spot treatment by county road administrations in Denmark. State and municipal road administrations do not as often use roundabouts as black-spot treatment. It is widespread not to use PDO crashes without police report in black-spot identification or other systematic crash analyses in Denmark. Therefore no statistically significant C RTM was found for PDO crashes without police report. Correction factors for regression-to-the-mean, which are used and statistically significant on a 5% level, are listed in Table 2. TABLE 2 Expected and Observed Crashes and Injuries in the Before-Before and Before Period at Converted Sites Road administration Type of crashes or injuries Observed before-before Expected before Observed before County Injury crashes with fatalities/severe injuries Injury crashes with only slight injuries PDO crashes with police report Fatalities and slight injuries Severe injuries State or municipal C RTM Injury crashes with fatalities/severe injuries Injury crashes with only slight injuries PDO crashes with police report Fatalities and slight injuries Severe injuries The results of the analyses of long-term accident trends, which are shown in Table 2, indicate abnormally high counts of injuries and injury and PDO crashes with police report, i.e. regression-to-the-mean effects, in the before period. The insignificant C RTM for PDO crashes without police report was estimated to 1.01, but set to 1. Safety effects for All Converted Sites Safety effects for all converted sites are estimated using a Danish methodology by Jørgensen (3). Here the measure of effectiveness, θ, is the sum of observed crashes in the after period for the sites divided by the sum of expected crashes for the after period for the sites, in other words:

7 Jensen 7 Safety effects are described in per cent in the following way 100 (θ 1), so if θ is 0.70 then the safety effect is -30%. The safety effects are tested in two ways. First the homogeneity of the safety effect across the converted sites is tested. This is done in order to find whether the changes in the numbers of crashes at sites are random variations of one and same effect. Changes in the numbers of crashes have to be rather uniform in order to be homogeneous. The chi-square test is calculated in the following way: Sites, where no crashes have occurred in the before and after periods, are not part of the test for homogeneity. If no crashes occurred in the before period then a part of the denominator of formula 4 A Expected, i / A Before, i is replaced by the overall correction factor, C, see formula 6. The calculated χ 2 value in formula 4 has N 1 degrees of freedom. If the χ 2 value is above a 5% significance level then the safety effect is homogeneous otherwise it is heterogeneous. Heterogeneous safety effects should not be generalized. Second the safety effect across the converted sites is tested for statistical significance. This is also a chi-square test and is calculated in the following way: where C is the overall correction factor, which is calculated this way: The calculated χ 2 value in formula 5 has one degree of freedom. If the χ 2 value is above 3.84, i.e. p 0.05, then the safety effect is statistically significant, which means the likelihood that the difference between observed and expected number of crashes in the after period is due to random variations is less than 5%. If the χ 2 value is above 2.71 but less than 3.84, i.e > p 0.10, then the safety effect is indicate a tendency but is a bit uncertain, which means the likelihood that the difference between observed and expected number of crashes in the after period is due to random variations is between 5 and 10%. If the χ 2 value is less than 2.71 then the safety effect is not statistically significant meaning that the difference between observed and expected number of crashes in the after period may well be due to random variations or the numbers of crashes are simply too small to prove any effect. RESULTS The 332 conversions from intersections to roundabouts resulted in the safety effects given in Table 3. Most safety effects in Table 3 are heterogeneous and should not be generalized.

8 Jensen 8 TABLE 3 Safety Effects of Converting Intersections to 332 Roundabouts Type of crashes or injuries Before Expected After Effect Significant? Homogeneous? Injury crashes with police report % Yes No PDO crashes with police report % Yes No PDO crashes without police report % Yes Yes All crashes 1,729 1, % Yes No All crashes with police report 1, % Yes No Fatalities % Yes Yes Severe injuries % Yes No Slight injuries % Yes No All injuries 1, % Yes No The safety effects of converting intersections to roundabouts are basically good, because crashes decrease in numbers and become less severe. The numbers of injury crashes and injuries have decreased by 47 and 60% respectively. The conversions have prevented 20 fatalities in the after period corresponding to an 87% decrease. Reported PDO crashes have decreased by 29% and PDO crashes without police report have increased 25% due to the conversions. In total, PDO crashes have decreased by 16%. Three causes to large deviations in changes in the numbers of crashes between sites have been documented. One cause is the strong relation between safety effect and motor vehicle speed at the intersection prior to conversion. Motor vehicle speeds have not been measured, but indicated by the highest speed limit on entry roads about 100 meters (328 feet) away from the roundabout. Crash records show that speed limits only have changed at a few sites from before to after. The numbers of crashes have increased by 1% at converted sites with km/h (19-31 mph) speed limits and decreased by 14%, 33%, 43% and 67% with 60, 70, 80 and km/h (37, 43, 50, mph) speed limits respectively, see Table 4. The numbers of injuries have decreased by 1%, 55%, 63%, 81% and 81% in numbers at 30-50, 60, 70, 80 and km/h (19-31, 37, 43, 50, mph) speed limits respectively. The background for the strong relation may be that crashes at the intersections occur at higher collision speeds than at roundabouts, and the difference in these collision speeds primarily is dependent of the speed at intersections. TABLE 4 Safety Effects for All Crashes of Converting Intersections to Roundabouts Split by the Highest Speed Limit at the Roundabout. N is the Number of Converted Sites with Crashes in Before and/or After Period Highest speed limit N Before Expected After Effect Significant? Homogeneous? km/h (19-31 mph) % No No 60 km/h (37 mph) % No Yes 70 km/h (43 mph) % Tendency Yes 80 km/h (50 mph) % Yes No km/h (56-81 mph) % Yes Yes Another cause to the large deviations is that the safety effect depends on left-turn and angle crashes share of all crashes at intersections in the before period. As this share of crashes increases the safety effect improves. At a share of 0-39% the number of crashes increased by 1%, whereas the numbers of crashes decreased by 16%, 26% and 38% at shares of 40-59%, 60-79% and % respectively. The numbers of injuries decreased by 1%, 51%, 66% and

9 Jensen 9 69% at shares of 0-39%, 40-59%, 60-79% and % respectively. The background for this relation may be that risky left-turns and traffic crossing the main road at intersections are replaced by not so risky right-turns at roundabouts. Safety Effects for Cyclists and Other Road Users A third cause is that the safety effect depends on bicycle crashes share of all crashes at intersections in the before period. As this share decreases the safety effect improves. At a share of % the total number of crashes increased by 12%, whereas the number of crashes decreased by 36%, 13% and 9% at shares of 0-14%, 15-29% and 30-49% respectively. The total number of injuries increased by 18% at a share of %, while injuries decreased by 77%, 32% and 28% in numbers at shares of 0-14%, 15-29% and 30-49% respectively. The background for this is that the safety effects for cyclists are considerably worse than for pedestrians and motor vehicle occupants, see Table 5. The safety effects for moped and motorcycle riders are also worse than those for pedestrians and motor vehicle occupants. TABLE 5 Safety Effects of Converting Intersections to Roundabouts Split by Crashes and injuries of various severity and with different road users or objects involved. Note: Safety Effects in Gray are statistically Significant on a 5% Level and Effects in Quotes are Heterogeneous Type of crashes or injuries Pedestrian crashes, pedestrian injuries Bicycle crashes, cyclists injuries Moped/motorcycle crashes, moped and motorcycle rider injuries Motor vehicle crashes, motor vehicle occupant injuries Crashes with objects and injuries in these crashes Injury crashes with police report -36% +31% +30% -54% -18% PDO crashes with police report -30% +108% +78% -30% +186% PDO crashes without police report -100% +143% +38% +20% +58% All crashes -39% +65% +46% -31% +86% All crashes with police report -34% +59% +47% -40% +103% Fatalities -100% -49% -62% -100% -42% Severe injuries +2% +10% +25% -86% -74% Slight injuries -6% +80% +50% -83% +9% All injuries -15% +40% +30% -85% -36% The reason for the poor safety effects for cyclists seems to be that a high share of bicycle crashes at intersections in the before period involve right turning vehicles, and rightturn crashes increase tremendously in numbers due to the conversions. The number of rightturn crashes has also increased among other road users than cyclists, but right-turn crashes are less common among other road users compared to cyclists. Safety effects for cyclists also depend heavily on speed limits. When the highest speed limit is above 60 km/h (37 mph) then safety effects are favorable for cyclists, whereas

10 Jensen 10 effects are very unfavorable at lower speed limits. Bicycle crashes constitute 10% of all crashes in the before period and 21% in the after period. Bicycle crashes constitute larger shares at low speed limits. The speed limits also influence other crashes than those with cyclists involved, see Table 6. However, the safety effect for crashes without cyclists depends less on the speed limit as the safety effect for all crashes. Table 6 shows that the safety effects are pretty good for crashes without cyclists even at low speed limits. In total, injury and PDO crashes without cyclists decreased by 62% and 24% due to the conversions. TABLE 6 Safety Effects for Crashes without Cyclists of Converting Intersections to Roundabouts Split by the Highest Speed Limit at the Roundabout. N is the Number of Converted Sites with Crashes in Before and/or After Period Highest speed limit N Before Expected After Effect Significant? Homogeneous? km/h (19-31 mph) % Yes No 60 km/h (37 mph) % Yes Yes 70 km/h (43 mph) % No Yes 80 km/h (50 mph) % Yes No km/h (56-81 mph) % Yes Yes Short- and Long-Term Safety Effects Converting intersections to roundabouts has more than doubled the number of single-vehicle crashes. A crash in Denmark has to involve one vehicle like a bicycle, moped, motorcycle or motor vehicle. Pedestrians are not involved in single-vehicle crashes. A multi-element crash may involve a pedestrian or two or more vehicles. An adaptation effect has been documented where the increase in single-vehicle crashes diminishes as time goes, see Table 7. TABLE 7 Safety Effects of Converting Intersections to Roundabouts Split by Single- Vehicle Crashes and Multi-Element Crashes and by Urban and Rural Areas Type of crashes Years after conversion First year Second year Third-fifth year Sixth-ninth year Short term, first-second year Long term, third-ninth year Single-vehicle Expected crashes, After urban areas Effect +162% +141% +126% +75% +152% +101% Single-vehicle crashes, rural areas Multi-element crashes, urban areas Multi-element crashes, rural areas Expected After Effect +378% +305% +196% +171% +341% +183% Expected After Effect -11% -28% -25% -15% -20% -20% Expected After Effect -71% -79% -72% -77% -75% -75% A huge increase in single-vehicle crashes occurred just after converting intersections to roundabouts. The increase has since diminished. The safety effect for the number of multielement crashes is stable, while the effect for injuries in multi-element crashes improves as time goes. This means that there also is an adaptation effect where multi-vehicle crashes

11 Jensen 11 become less severe as time goes, however, this adaptation effect is rather small. Long-term safety effects 3-9 years after converting intersections to roundabouts are percentage points better than short-term safety effects in the first and second year after the conversions both in urban and rural areas and both for crashes and injuries. Long-term safety effects for cyclists are also better than short-term safety effects both in urban and rural areas. The number of bicycle crashes increased 105% in the first and second year after the conversions, whereas it only increased by 63% 3-9 years after. Cyclist injuries increased 88% in the first and second year after converting intersections to roundabouts, but increased only 26% 3-9 years after. Influence of Intersection Design on Safety Effects The converted intersections were designed rather different. The safety effects depend on intersection design. Converting non-signalized intersections have resulted in better safety effects compared to the effects of conversions of signalized intersections, see Table 8. The safety effects of converting non-signalized intersections with yield or stop signs respectively seems to be very similar, and has therefore been compiled to one group in Table 8. Safety effects are better when 4-armed non-signalized intersections are converted to roundabouts compared to effects of converting 3-armed non-signalized intersections. The safety effects of conversions of staggered intersections are similar to the effects of converting 4-armed nonsignalized intersections. Conversions of signalized and 3-armed non-signalized intersections with km/h (19-31 mph) speed limits to roundabouts have increased the number of crashes and injuries, whereas these numbers have decreased in all other cases. TABLE 8 Safety Effects for All Crashes (for All Injuries in Brackets) of Converting Different Types of Intersections to Roundabouts Split by the Highest Speed Limit at the Roundabout Highest speed limit Non-signalized intersections Signalized intersections Staggered intersections 3-armed 4-armed 3- and 4-armed 2-3 intersections km/h (19-31 mph) +35% (+37%) -15% (-19%) +18% (+70%) -5% (-34%) km/h (37-43 mph) -32% (-53%) -20% (-67%) -2% (-44%) -26% (-54%) km/h (50-81 mph) -46% (-83%) -51% (-84%) -36% (-75%) -4% (-85%) Bicycle safety overall worsened when intersections were converted to roundabouts with km/h (19-31 mph) speed limits, whereas bicycle safety improved at conversions at km/h (50-81 mph) speed limits. However at all speed limits, safety effects for cyclists and other road users are good when intersections with dual-way paths along one or more entry roads are converted to roundabouts. Influence of Roundabout Design on Safety Effects The 332 roundabouts are also designed differently. The safety effects depend on roundabout design, see Table 9. The numbers of crashes and injuries have increased and decreased respectively when intersections are converted to mini-roundabouts. Conversions to single-lane roundabouts have most often resulted in safety benefits, however, not when converting to 3-armed roundabouts with km/h (19-31 mph) speed limit. Conversions to multilane roundabouts with 3 arms

12 Jensen 12 have resulted in more accidents and injuries, while conversions to multilane roundabouts and signalized roundabout with more than 3 arms have produced safety benefits. TABLE 9 Safety Effects for All Crashes (for All Injuries in Brackets) of Converting Intersections to Different Types of Roundabouts Split by the Highest Speed Limit (km/h) at the Roundabout Speed limit Mini-roundabouts Single-lane roundabouts Multilane roundabouts Signalized roundabout 3-armed 4-armed 3-armed 4-7 arms 3-armed 4-5 arms % (-16%) +15% (-8%) +37% (+41%) -14% (-1%) +29%(+248%) -86%(-100%) % (-41%) -13% (-62%) -49% (+166%) +3% (-35%) % (-86%) -49% (-82%) +93%(+137%) -27% (-95%) -18% (-74%) Nor shunts or central island diameter seem to influence safety effects, however, the height of the central island does. Central islands higher than 2 meters (6.6 feet) at the center of the island produce better safety effects compared to lower central islands, see Table 10. This height is measured from the pavement of the circulating lane(s) to the top of an item at the center of the central island. The item has to be so wide that a passenger car may hide behind it, and such item may be e.g. sculptures, dense planting or landscaping. The relation between the safety effect and the height of the central island is present at all speed limits, and safety effects are homogeneous at individual speed limits when effects are split into various heights of the central island. TABLE 10 Safety Effects for All Crashes and All Injuries of Converting Intersections to Single-Lane Roundabouts with Central Islands of Various Heights. N is the Number of Converted Sites with Crashes in Before and/or After Period Type Central island height N Before Expected After Effect Significant? Homogeneous? Crashes m (0-3.2 feet) % Yes No m ( feet) % Yes Yes m ( feet) % Yes No Injuries m (0-3.2 feet) % Yes No m ( feet) % Yes No m ( feet) % Yes No Safety effects are best when the width of the circulating lane and cycle lane together is 4-6 meters (13-20 feet) wide at km/h (19-31 mph) speed limits, 6-8 meters (20-26 feet) wide at km/h (37-43 mph) and more than 5 meters (16 feet) wide at km/h (50-81 mph). Roundabouts with triangle or trumpet splitter islands produce better safety effects than roundabouts without splitter islands or with parallel splitter islands. One- and dual-way cycle paths at roundabouts without priority to cyclists crossing arms have resulted in the best safety effects for cyclists, whereas cycle lanes marked next to the circulating lane at roundabouts resulted in the worst safety effects for cyclists, see Table 11. This is to some extent due to different designs at different speed limits. Cycle lanes at roundabouts are the most common design for cyclists at roundabouts in Denmark. There are only small differences in safety effects between roundabouts without bicycle facilities and roundabouts with cycle tracks next to the circulating lane with priority

13 Jensen 13 to cyclists. Colored cycle lanes and blue cycle crossings have resulted in worse safety effects compared to comparable bicycle facilities without color. TABLE 11 Safety Effects for Bicycle Crashes of Converting Intersections to Roundabouts with Different Design for Cyclists. N is the Number of Converted Sites with Crashes in Before and/or After Period Bicycle facility N Before Expected After Effect Significant? Homogeneous? None, priority to cyclists % No Yes Cycle lane, priority to cyclists % Yes Yes Colored cycle lane, priority to % Yes Yes cyclists Cycle track, priority to cyclists % No Yes Cycle track with blue cycle crossings, priority to cyclists % No Yes Cycle path, no priority to cyclists % Yes Yes DISCUSSION Meta-analyses of earlier before-after crash studies of converting intersections to roundabouts show decreases of injury and PDO crashes of 60% and 25% respectively. The presented before-after crash study shows smaller decreases of 47% and 16% respectively. However, when bicycle crashes are excluded from the presented study then the decreases are 62% and 24% respectively. The differences in safety effects between earlier studies and the presented study may very well be the poor safety effects for cyclists of the conversions in Denmark. The poor safety effects seem predominantly to stem from a very unsafe design for cyclists, namely cycle lanes next to the circulating lanes. Other results of the meta-analyses and the presented study are rather similar when differences in safety effects between urban and rural areas, conversions to 3- and 4-armed roundabouts, conversions of non-signalized and signalized intersections, and also short- and long-term effects are compared. Examples are that earlier studies show that converting nonsignalized intersections to roundabouts produce safety effects, which are 15 percentage points better than safety effects of conversions of signalized intersections to roundabouts (1). Earlier studies also show that long-term safety effects are 12 percentage points better than short-term effects (1). Both examples are in line with results from the presented before-after crash study. CONCLUSIONS The main conclusions of the research reported in this paper can be summarized in the following points: 1. A before-after crash study of converting intersections to roundabouts has been completed taking into account changes in general crash and injury trends and regression-to-the-mean effects in the before period. 2. The safety effects of the conversions are decreases in injury and PDO crashes of 47% and 16% respectively. The conversions have also resulted in decreases in fatalities, severe and slight injuries of 87%, 58% and 59% respectively. These safety effects should not be generalized, because they are heterogeneous due to large deviations in changes in the numbers of crashes and injuries between converted sites.

14 Jensen The safety effects are mainly heterogeneous due to three variables. The safety effect becomes better as the speed limit on the roundabout arms becomes higher. As left-turn and angle crashes share of all crashes at intersections becomes higher the safety effect becomes better. As bicycle crashes share of all crashes at intersections becomes lower the safety effect becomes better. 4. Long-term safety effects 3-9 years after converting intersections to roundabouts are percentage points better than short-term safety effects in the first and second year after the conversions both in urban and rural areas and both for crashes and injuries. This is mainly due to an adaptation effect for single-vehicle crashes. 5. Central islands more than two meters (6.6 feet) high produce better safety effects compared to lower central islands. If a roundabout is designed with a central island of two meters (6.6 feet) or more in height then the safety effect is about twice as good compared to lower central islands. 6. The numbers of bicycle crashes and injured cyclists have increased by 65% and 40% respectively due to the conversions. Cycle lanes next to the circulating lane at roundabouts have produced the worst safety effects for cyclists, whereas cycle paths without priority to cyclists crossing the arms of the roundabouts resulted in the best effects. REFERENCES 1. Jensen, S. U. and P. B. Madsen. Rundkørsler, sikkerhed og cyklister (roundabouts, safety and cyclists). Trafitec, Lyngby, Denmark, Jensen, S. U. Sikkerhedseffekter af rundkørsler (safety effects of roundabouts). Trafitec, Lyngby, Denmark, Jørgensen, E. Sikkerhedsmæssig effekt (road safety effects). Road Directorate, Næstved, Denmark, 1981.