- Email: [email protected]
Traffic signal phasing at intersections to improve safety for alcohol-affected pedestrians
Accident Analysis and Prevention 39 (2007) 751–756
Traffic signal phasing at intersections to improve safety for alcohol-affected pedestrians Michael G. Lenn´e ∗ , Bruce F. Corben, Karen Stephan Monash University Accident Research Centre, Building 70, Monash University, Victoria 3800, Australia Received 23 March 2006; received in revised form 30 October 2006; accepted 15 November 2006
Abstract Alcohol-affected pedestrians are among the highest-risk groups involved in pedestrian casualty crashes. This paper investigates the opportunities to use a modified form of traffic signal operation during high-risk periods and at high-risk locations to reduce alcohol-affected pedestrian crashes and the severity of injuries that might otherwise occur. The ‘Dwell-on-Red’ treatment involves displaying a red traffic signal to all vehicle directions during periods when no vehicular traffic is detected, so that drivers approach high-risk intersections at a lower speed than if a green signal were displayed. Vehicle speed data were collected before and after treatment activation at both a control and treatment site. Speed data were collected both 30 m prior to and at the intersection stop line. The treatment was associated with a reduction in mean vehicle speeds of 3.9 kph (9%) and 11.0 kph (28%) at 30 m and stop line collection points, respectively, and substantial reductions in the proportion of vehicles travelling at threatening speeds with regard to the severity of pedestrian injury. Other important road safety concerns may also benefit from this form of traffic signal modification, and it is recommended that other areas of application be explored, including the other severe trauma categories typically concentrated around signalised intersections. © 2006 Elsevier Ltd. All rights reserved. Keywords: Vehicle speeds; Crash risk; Pedestrian injury severity; Intoxication
1. Introduction Motor vehicle crashes involving pedestrians represent a substantial proportion of all road deaths and serious injuries world-wide, comprising between 10% and 20% of road fatalities in many western countries (Australian Transport Safety Bureau [ATSB], 2005; Davies, 1999; National Highway Traffic Safety Administration [NHTSA], 2003). Pedestrians using alcohol are at greater risk of being fatally or seriously injured (e.g., Lee and Abdel-Aty, 2005; Martinez and Porter, 2004; Miles-Doan, 1996; Zajac and Ivan, 2003), and a substantial proportion of pedestrian deaths and serious injuries involve intoxicated pedestrians. Blood alcohol concentration (BAC) levels at or above 0.10% have been reported in 30–40% of pedestrian fatalities and casualties (Clayton and Colgan, 2001; NHTSA, 2003). In 2001, 88% of those pedestrians killed in the US who had been drinking had a BAC over 0.08% (NHTSA, 2003). A review of fatal
∗
Corresponding author. Tel.: +61 3 9905 1389; fax: +61 3 9905 4363. E-mail address: [email protected] (M.G. Lenn´e).
0001-4575/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2006.11.006
pedestrian crashes in Australia from 1997–1999 showed 70% of males 18–54 years had a BAC over 0.05%, and about 80% of those who were alcohol affected had BAC over 0.15% (ATSB, 2003). The late evening and early morning hours account for a substantial proportion of alcohol-involved pedestrian crashes. In the US almost 50% of pedestrian fatalities occur between 6 p.m. and midnight. Around half of these pedestrian fatalities involved alcohol (BAC > 0.01%), and this proportion rose to two-thirds between midnight and 3 a.m. (NHTSA, 2003). Similarly, from 1977–1995 in Ume˚a, Sweden, half of the pedestrian deaths between 9 p.m. and 5 a.m. were alcohol positive (Ostr¨om and Eriksson, 2001). Recent Victorian data show that 57% of pedestrian fatalities occurred during high alcohol hours (HAH), defined as 6 p.m. through to 6 a.m. (Transport Accident Commission, 2005). Furthermore, 85% of male alcohol-involved pedestrian fatalities occurred during these times (ATSB, 2003). Strategies to reduce the number of crashes involving alcohol-affected pedestrians should aim to prevent pedestrians reaching high BACs and to prevent alcohol-affected pedestrians
752
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
being exposed to traffic (Victorian Parliamentary Road Safety Committee, 1999). While responsible alcohol service courses have an effect on the behaviour of personnel serving alcohol, studies continue to find that a large proportion of clearly intoxicated individuals are served at licensed premises even when servers have attended such courses (Buka and Birdthistle, 1999; Donnelly and Briscoe, 2003; Stockwell, 2001; Wallin et al., 2002). Lowering vehicle speeds in areas where alcohol-affected pedestrians may be present will reduce the risk for this group. Vehicle speed is directly related to crash risk (Aarts and van Schagen, 2006), and to the severity of pedestrian injury (Anderson et al., 1997; Ashton and Mackay, 1979; Garder, 2004; Matsui, 2005). At an impact speed of 20 kph there is about a 5% chance of serious injury to a pedestrian, which increases to about 25% at an impact speed around 30 kph (Ashton and Mackay, 1979; Leaf and Preusser, 1999). For an impact speed of 40 kph there is approximately a 25% chance of pedestrian fatality, and this figure increases markedly to about an 80% chance of death at impact speeds around 50 kph (Anderson et al., 1997). These comparisons illustrate why it is important to examine the proportion of vehicles travelling at speeds likely to cause serious injury and fatalities to pedestrians. There is little information on the effects of commonly used infrastructure-based treatments, such as pedestrian fencing, on pedestrian safety. Displaying a red traffic signal to all vehicle directions at signalised intersections during periods when no vehicular traffic is detected has been suggested as a potential method of reducing vehicle speeds at intersections in order to reduce the risk of collision with a pedestrian and the severity of injuries should a collision occur (Corben et al., 1998). While modifications to traffic signal operation reduce the risk of crashes involving pedestrians and bicyclists (Retting et al., 2002), the present study evaluated the effects of a new traffic signal phasing on safety for alcohol-affected pedestrians.
2. Method 2.1. Experimental design This study utilised a controlled before-and-after design, commonly defined as a trial in which data are collected at both a treatment and an appropriate control site, before and after the intervention at the treatment site (Elvik, 2002). The use of a control site helps to increase the likelihood that observed effects at the treatment site are attributable to the treatment itself rather than other extraneous influences. 2.2. Site selection A number of metropolitan and regional sites with a high number of police-reported pedestrian casualty crashes during HAH were identified. HAH are those time periods when the proportion of drivers or riders killed or seriously injured with a known BAC over 0.05% was greater than 15% (Shtifelman et al., 1998), and broadly covers the period between 6 p.m. and 6 a.m. for each day of the week. A similar methodology has also been used in analyses of alcohol-involved pedestrian crashes in Baltimore (Blomberg and Cleven, 2000). HAH were used as a surrogate indicator of alcohol involvement in pedestrian crashes because BAC cannot be relied upon as an accurate indicator of alcohol-related pedestrian crashes in Victoria as pedestrians are not routinely tested when involved in a crash (Victorian Parliamentary Road Safety Committee, 1999). The City of Ballarat in regional Victoria (population around 65,000) was selected for the trial. The configuration of the treatment site is presented in Fig. 1. The treatment site is the main route through the city, running east–west, and is a divided road with two through-lanes in each direction and dedicated left- and right-turn lanes. The north–south road is also two through-lanes in each direction. The control site is 300–350 m to the west along the same east–west main route, with a similar
Fig. 1. Representation of the treatment intersection with location of vehicle detectors.
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
north–south cross-road. The speed limit was 50 kph for both sites. Vehicle speed data were collected for traffic moving in an easterly direction, meaning that drivers travelled through the control site before the treatment site. Consequently, behaviour at the control site would not be influenced by prior exposure to the treatment. 2.3. Measure of pedestrian safety The aim of the Dwell-on-Red (DOR) treatment is to reduce vehicle speeds and thus increase safety for pedestrians. Vehicle speed is the most critical factor in determining pedestrian safety and was chosen as the dependent variable. Loop vehicle classifiers installed below the road surface were used to measure vehicle speed at two locations: 30 m on approach to the intersection and close to the stop line. For vehicles travelling at around the posted speed limit of 50 kph, the minimum distance required to brake and stop safely before the intersection is approximately 30 m (Anderson et al., 1997). Speed at 30 m before the intersection is therefore indicative of crash risk, while speed at the stop line is indicative of potential injury severity for pedestrians. 2.4. The Dwell-on-Red treatment The signal phasing for the trial had four phases: phase A, green signal to through-vehicles and pedestrians on the east–west route (minimum time on green 10 s, minimum time on yellow 3.5 s, and minimum time on all-red of 2.5 s); phase B, green signal to through-vehicles and pedestrians on the north–south route (minimum green, yellow, and all-red times of 8.0, 3.5, and 2.5 s, respectively); phase C, which could be skipped, involved a green signal to through and turning traffic in the north–south median (minimum green, yellow, and all-red times of 6.0, 3.0, and 1.5 s, respectively); and phase D, red signal to vehicles and pedestrians in all directions (DOR phase). The DOR phase was only included in the traffic signal sequence between 10 p.m. and 5 a.m. at the treatment site in the second data collection period. Within these times the sequence was DABC, whereby phase D was the recall phase, that is, the D phase was called every cycle when no vehicles or pedestrians were detected for a 15-s period, and the sequence rested in phase D when there were no other phase demands. Phases A, B and C were only called when demanded by a vehicle or pedestrian. Phase D was programmed to run for a minimum of 15 s. The duration of operation of the DOR (phase D) was dependent on the presence/absence of vehicles and pedestrians at the treatment site. The pedestrian signals were red in all directions during the DOR phase and only changed to green if activated by any pedestrian or vehicle arriving on any leg of the intersection. Normal traffic signal practices were activated once a vehicle or pedestrian was detected while DOR was activated.
(Tabachnick and Fidell, 1996). This customised model targeted the analysis of the main effects and interactions of interest, and used the differences observed across treatment and control sites in the before period as a basis to calculate the relative difference across sites in the after period. Four-way Univariate Analyses of Variance (ANOVA) were conducted with the variables being site (control and treatment), sampling period (before and after), day of week, and time of day. Time of day and day of week were included in the analyses as potential confounding variables but are not reported here. Logistic regression was used to compare the proportion of vehicles travelling below specified speeds at the treatment and control site, before and after the DOR signal phasing was introduced. When a significant interaction between site and sampling period was present, a comparison of the proportion of vehicles below the specified speed between the before period and after period was made for the control and treatment sites separately. 3. Results 3.1. Sampling characteristics All data were collected between February and August 2005. Data were collected prior to treatment activation (before period) between 10 p.m. on February 22 through to 5 a.m. on March 16. After treatment activation (after period) data were collected between 10 p.m. on August 11 through to 5 a.m. on August 23. The DOR treatment was active for a ‘settling in’ period of 4 weeks prior to the after data collection period. Table 1 shows the total number of speed measurements collected at the two points. The smaller number of observations in the after period reflects the shorter duration of this data collection period. Approximately half of the data collected at the treatment site at the 30 m point in the before period were erroneous due to a system error and were excluded from the analyses. Table 2 shows the number of DOR activations that occurred across the period of DOR operation, as well as the mean duration per activation. In periods of low traffic operations there was more scope for activation of the treatment with longer durations per activation, while increased traffic volume was associated with more frequent activation of shorter duration. In areas of very high traffic volume there is limited scope for activation of the treatment. Preliminary analyses considered the potential influence of traffic volume on the effectiveness of DOR in reducing vehicle speeds however there was no noticeable effect. Presumably this was because DOR activation is influenced by traffic flow rather than volume. As traffic flow was not measured in this study, Table 1 The number of speed measurements collected at the two collection points at the treatment and control site Site
2.5. Data analysis A customised general linear model was used to analyse the speed data for the two locations (30 m and at the stop line)
753
Control Treatment Total
30 m point
Stop line point
Before
After
Total
Before
After
Total
16,668 9,080 25,748
7,285 8,191 15,476
23,953 17,271 41,224
16,210 16,404 32,614
7,374 7,121 14,495
23,584 23,525 47,109
754
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
Table 2 Characteristics of Dwell-on-Red activation across the hours of treatment operation Time of day
22:00–22:59 23:00–23:59 00:00–00:59 01:00–01:59 02:00–02:59 03:00–03:59 04:00–04:59
Mean activations/h
Mean duration per activation (s)
Mean total duration of activation/h (min)
Traffic counts
71 72 60 49 42 44 41
14.9 17.8 22.0 36.8 56.3 58.2 63.5
17.6 21.4 22.0 30.0 39.0 42.2 43.2
2769 1830 1075 622 330 263 232
Table 3 Mean speed (kph ± S.E.) for control and treatment sites at 30 m and the stop line Site
30 m
Control Treatment
Stop line
Before
After
Before
After
42.2 (±0.1) 41.7 (± 0.2)
47.3 (±0.3) 37.8 (±0.3)
36.9 (±0.2) 39.0 (±0.2)
44.6 (±0.3) 28.0 (±0.3)
this was not analysed further and the data are presented pooled across all traffic volumes. 3.2. Vehicle speed Table 3 presents mean vehicle speeds collected at the two points at the control and treatment intersections. At both points vehicle speeds dropped after the activation of the DOR phasing at the treatment site but not at the control site (30 m point, F(1, 41,170) = 373.9, p < 0.001; stop line, F(1, 46,582) = 1086.8, p < 0.001). Mean speed for the control site increased in the after period compared to the before period at both collection points despite no differences in signal phasing. 3.2.1. Speed in the potential pedestrian impact zone At the stop line, mean speed was 11 kph (28%) lower at the treatment site in the after period compared to the before period. Taking into account the difference in speed at control and treatment sites in the before period, further analysis showed that speed was significantly lower in the after period at the treatment site compared to the control site (p < 0.001), resulting in a net reduction in mean speed at the treatment site of around 40%. For reasons relating to the speed-dependent risk of pedestrian injury discussed earlier, the speed data collected at the stop line were categorised into vehicles travelling ≤30 or >30 kph (see Table 4). The proportion of vehicles travelling ≤30 kph at the treatment site increased in the after period (OR = 3.77, 95% Table 4 Proportion (%) of vehicles travelling below specified speeds Site
Control Treatment
Stop line,≤30 kph
30 m point, ≤40 kph
30 m point, ≤50 kph
Before
After
Before
After
Before
After
41 39
30 71
46 49
33 73
78 81
68 94
CI = 3.54–4.00, p < 0.0001), while at the control site the proportion ≤30 kph was reduced in the after period (OR = 0.61, 95% CI = 0.58–0.65, p < 0.0001). 3.2.2. Speed on approach to the intersection At the 30 m point, mean speed was 3.9 kph (9%) lower at the treatment site in the after period when DOR was activated compared to the before period. Taking into account the difference in speed between control and treatment sites in the before period, further analysis showed that speed was significantly lower in the after period at the treatment site compared to the control site (p < 0.05), resulting in a net reduction in mean speed at the treatment site of around 19%. For reasons related to pedestrian fatality risk and vehicle crash risk, respectively, the proportion of vehicles travelling below/above 40 and 50 kph at the 30 m point were analysed (see Table 4). The proportion of vehicles travelling at ≤40 kph at the treatment site increased in the after period (OR = 2.83, 95% CI = 2.66–3.02, p < 0.0001), but decreased at the control site in the after period (OR = 0.57, 95% CI = 0.54–0.61, p < 0.0001). The treatment was therefore indicative of a reduction in potential risk of injury to pedestrians crossing between the 30 m and stop line points. The proportion of vehicles travelling at ≤50 kph at the treatment site also increased in the after period (OR = 3.36, 95% CI = 3.03–3.73, p < 0.0001), but again decreased at the control site in the after period (OR = 0.61, 95% CI = 0.58–0.65, p < 0.0001). The treatment was therefore suggestive of a reduction in crash risk. 4. Discussion This study evaluated the effect of DOR signal phasing during HAH on vehicle speeds at an intersection with a high risk of crashes involving alcohol-affected pedestrians. It was necessary to perform this research to determine if the speed reductions expected upon activation of DOR signal phasing did occur in practice, and to estimate the magnitude of any speed reduction. It was also of interest to observe the number of times per hour that the DOR signal phasing was activated and how this related to traffic volume, as no previous data were available. We demonstrated measurable reductions in mean speeds with DOR signal phasing. Taking a conservative approach and considering the treatment site only, the mean speed at the stop line decreased from 39 to 28 kph, which represents a 28% reduction. There was a concomitant 52% reduction in the proportion of vehicles travelling faster than 30 kph. If speed at the stop line is considered as the potential impact speed with pedestrians on the crossing, this suggests significant reductions in the risk of fatal and serious injury to pedestrians after activation of the DOR phasing. Additionally, we found that mean speeds measured 30 m prior to the intersection decreased from 41.7 to 37.8 kph (a 9% reduction) with a 47% reduction in the proportion of vehicles travelling faster than 40 kph, and a 68% reduction in the proportion travelling over 50 kph. This represents a significant
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
reduction in the proportion of vehicles travelling too fast to brake and stop safely before the intersection. Having found a measurable impact of DOR signal phasing on vehicle speeds during HAH, it is our intention to consider the broader effects and potential application of the DOR signal phasing. Of prime importance is the assessment of the effect of DOR on the number of pedestrian–vehicle conflicts and the rates of crashes involving pedestrians. It is also of interest to determine if the rate of vehicle–vehicle crashes is affected, as it has been suggested that the DOR signal phasing might increase the number of stop–start maneuvers that vehicles make which could have a detrimental effect on rear-end collision rates. While the current study was able to investigate the relationship between the number and duration of DOR activations and traffic volume, the most important predictor of DOR activation is traffic flow. Further research will measure this relationship. There has been some concern that DOR signal phasing could increase vehicle travel times, however this is unlikely to be a major problem as it is envisaged that DOR will be operated selectively at the small number of sites that have a high alcohol-pedestrian crash problem during HAH. Traffic volumes are generally lower during these hours. The effect on travel times could be assessed by undertaking number plate surveys in conjunction with the traffic flow measurements. While this evaluation of DOR signal phasing was conducted at one intersection in a regional city during HAH to target alcohol-involved pedestrian crashes, the system-wide effects and other potential areas of application need to be explored. Further research should explore the potential to translate the speed reductions and the associated safety benefits gained through the application of DOR signal phasing to other regional and metropolitan locations, including intersections where other vulnerable groups (e.g., children and the elderly) at high risk of road trauma would benefit from a reduction in vehicle speeds. The acceptability of DOR signal phasing to drivers and pedestrians should also be assessed. The authors have received funding to develop the concept further. In conclusion, alcohol-affected pedestrians are a group that is at high risk of being involved in pedestrian casualty crashes. A range of treatments and programs have been implemented to increase safety for alcohol-affected pedestrians with limited widespread success. The results reported here demonstrate that the DOR modification to traffic signal operation in a regional setting significantly reduced vehicle speeds within the assumed HAH for pedestrians. Such speed reductions are associated with substantial reductions to the potential risk of fatal and serious injury to pedestrians, and a reduction in crash risk. Acknowledgements This project was undertaken under contract for VicRoads. The authors wish to thank VicRoads, in particular Elizabeth Knight and Linda Ivett, for advice received in conducting this project and for approval to publish the research. The authors would also like to thank MUARC colleagues Ashley Verdoorn for his technical assistance and Christine Mulvihill for her work in an earlier phase of this research. Funding for the preparation of
755
this paper was provided through the MUARC SDP Publication Award. References Aarts, L., van Schagen, I., 2006. Driving speed and the risk of road crashes: a review. Accid. Anal. Prev. 38 (2), 215–224. Anderson, R., McLean, A., Farmer, M., Lee, B., Brooks, C., 1997. Vehicle travel speeds and the incidence of fatal pedestrian crashes. Accid. Anal. Prev. 29, 667–674. Ashton, S., Mackay, G., 1979. Some characteristics of the population who suffer trauma as pedestrians when hit by cars and some resulting implication. In: Proceedings of the IRCOBI Conference, Gothenburg, Sweden. Australian Transport Safety Bureau, 2003. Male Pedestrian Fatalities. Retrieved February 17, 2006, from http://www.atsb.gov.au/publications/2003/Ped Male 1.aspx. Australian Transport Safety Bureau, 2005. Road Deaths Australia: 2004 Statistical Summary. Australian Transport Safety Bureau, Canberra, ACT. Blomberg, R.D., Cleven, A.M., 2000. Development, Implementation and Evaluation of a Countermeasure Program for Alcohol-involved Pedestrian Crashes. US Department of Transportation, National Highway Traffic safety Administration, Washington, DC. Buka, S.L., Birdthistle, I.J., 1999. Long-term effects of a community-wide alcohol server training intervention. J. Stud. Alcohol 60, 27–36. Clayton, A., Colgan, M., 2001. Alcohol and pedestrians. A final report to Road Safety Division (Road Safety Research Report No. 20). Department of the Environment, Transport and the Regions, London, UK. Corben, B.F., Diamantopoulou, K., Mullan, N., 1998. Environmental countermeasures for alcohol-related pedestrian crashes. In: Proceedings of the Ninth Road Engineering Association of Asia and Australasia, Wellington, New Zealand. Davies, D., 1999. Research, development and implementation of pedestrian safety facilities in the United Kingdom (Report No. FHWA-RD-99-089). Federal Highway Administration, US Department of Transportation, Washington, DC. Donnelly, N., Briscoe, S., 2003. Signs of intoxication and server intervention among 18 to 39-year-olds drinking at licensed premises in New South Wales, Australia. Addiction 98 (9), 1287–1295. Elvik, R., 2002. The importance of confounding in observational before-andafter studies of road safety measures. Accid. Anal. Prev. 34 (5), 631–635. Garder, P.E., 2004. The impact of speed and other variables on pedestrian safety in Maine. Accid. Anal. Prev. 36 (4), 533–542. Leaf, W.A., Preusser, D.F., 1999. Literature Review on Vehicle Travel Speeds and Pedestrian Injuries (Report No. DTNH22-97-D-05018). National Highway Traffic Safety Administration, Washington, DC. Lee, C., Abdel-Aty, M., 2005. Comprehensive analysis of vehicle–pedestrian crashes at intersections in Florida. Accid. Anal. Prev. 37 (4), 775–786. Martinez, K.L.H., Porter, B.E., 2004. The likelihood of becoming a pedestrian fatality and drivers’ knowledge of pedestrian rights and responsibilities in the Commonwealth of Virginia. Transport. Res., Part F: Traffic Psychol. Behav. 7 (1), 43–58. Matsui, Y., 2005. Effects of vehicle bumper height and impact velocity on type of lower extremity injury in vehicle–pedestrian accidents. Accid. Anal. Prev. 37 (4), 633–640. Miles-Doan, R., 1996. Alcohol use among pedestrians and the odds of surviving an injury: evidence from Florida law enforcement data. Accid. Anal. Prev. 28 (1), 23–31. National Highway Traffic Safety Administration, 2003. Pedestrian Roadway Fatalities (DOT HS 809 456). US Department of Transportation, Washington, DC. Ostr¨om, M., Eriksson, A., 2001. Pedestrian fatalities and alcohol. Accid. Anal. Prev. 33 (2), 173–180. Retting, R.A., Chapline, J.F., Williams, A.F., 2002. Changes in crash risk following re-timing of traffic signal change intervals. Accid. Anal. Prev. 34 (2), 215–220. Shtifelman, M., Cameron, M., Diamantopoulou, K., 1998. Update of Alcohol Times as a Surrogate Measure of Alcohol-involvement in Accidents
756
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
in Melbourne and Country Victoria During 1990–1997. Monash University Accident Research Centre, Melbourne. Stockwell, T., 2001. Responsible alcohol service: Lessons from evaluations of server training and policing initiatives. Drug Alcohol Rev. 20, 257–265. Tabachnick, B.G., Fidell, L.S., 1996. Using multivariate statistics, 3rd ed. HarperCollins, New York. Transport Accident Commission, 2005. Pedestrian Statistics. Retrieved November 30, 2005, from http://www.tacsafety.com.au/jsp/homepage/home.jsp.
Victorian Parliamentary Road Safety Committee, 1999. Walking Safely: Inquiry into the Incidence and Prevention of Pedestrian Accidents. Victorian Government Printer, Melbourne. Wallin, E., Gripenberg, J., Andreasson, S., 2002. Too drunk for a beer? A study of overserving in Stockholm. Addiction 97 (7), 901–907. Zajac, S.S., Ivan, J.N., 2003. Factors influencing injury severity of motor vehicle–crossing pedestrian crashes in rural Connecticut. Accid. Anal. Prev. 35 (3), 369–379.
Traffic signal phasing at intersections to improve safety for alcohol-affected pedestrians Michael G. Lenn´e ∗ , Bruce F. Corben, Karen Stephan Monash University Accident Research Centre, Building 70, Monash University, Victoria 3800, Australia Received 23 March 2006; received in revised form 30 October 2006; accepted 15 November 2006
Abstract Alcohol-affected pedestrians are among the highest-risk groups involved in pedestrian casualty crashes. This paper investigates the opportunities to use a modified form of traffic signal operation during high-risk periods and at high-risk locations to reduce alcohol-affected pedestrian crashes and the severity of injuries that might otherwise occur. The ‘Dwell-on-Red’ treatment involves displaying a red traffic signal to all vehicle directions during periods when no vehicular traffic is detected, so that drivers approach high-risk intersections at a lower speed than if a green signal were displayed. Vehicle speed data were collected before and after treatment activation at both a control and treatment site. Speed data were collected both 30 m prior to and at the intersection stop line. The treatment was associated with a reduction in mean vehicle speeds of 3.9 kph (9%) and 11.0 kph (28%) at 30 m and stop line collection points, respectively, and substantial reductions in the proportion of vehicles travelling at threatening speeds with regard to the severity of pedestrian injury. Other important road safety concerns may also benefit from this form of traffic signal modification, and it is recommended that other areas of application be explored, including the other severe trauma categories typically concentrated around signalised intersections. © 2006 Elsevier Ltd. All rights reserved. Keywords: Vehicle speeds; Crash risk; Pedestrian injury severity; Intoxication
1. Introduction Motor vehicle crashes involving pedestrians represent a substantial proportion of all road deaths and serious injuries world-wide, comprising between 10% and 20% of road fatalities in many western countries (Australian Transport Safety Bureau [ATSB], 2005; Davies, 1999; National Highway Traffic Safety Administration [NHTSA], 2003). Pedestrians using alcohol are at greater risk of being fatally or seriously injured (e.g., Lee and Abdel-Aty, 2005; Martinez and Porter, 2004; Miles-Doan, 1996; Zajac and Ivan, 2003), and a substantial proportion of pedestrian deaths and serious injuries involve intoxicated pedestrians. Blood alcohol concentration (BAC) levels at or above 0.10% have been reported in 30–40% of pedestrian fatalities and casualties (Clayton and Colgan, 2001; NHTSA, 2003). In 2001, 88% of those pedestrians killed in the US who had been drinking had a BAC over 0.08% (NHTSA, 2003). A review of fatal
∗
Corresponding author. Tel.: +61 3 9905 1389; fax: +61 3 9905 4363. E-mail address: [email protected] (M.G. Lenn´e).
0001-4575/$ – see front matter © 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.aap.2006.11.006
pedestrian crashes in Australia from 1997–1999 showed 70% of males 18–54 years had a BAC over 0.05%, and about 80% of those who were alcohol affected had BAC over 0.15% (ATSB, 2003). The late evening and early morning hours account for a substantial proportion of alcohol-involved pedestrian crashes. In the US almost 50% of pedestrian fatalities occur between 6 p.m. and midnight. Around half of these pedestrian fatalities involved alcohol (BAC > 0.01%), and this proportion rose to two-thirds between midnight and 3 a.m. (NHTSA, 2003). Similarly, from 1977–1995 in Ume˚a, Sweden, half of the pedestrian deaths between 9 p.m. and 5 a.m. were alcohol positive (Ostr¨om and Eriksson, 2001). Recent Victorian data show that 57% of pedestrian fatalities occurred during high alcohol hours (HAH), defined as 6 p.m. through to 6 a.m. (Transport Accident Commission, 2005). Furthermore, 85% of male alcohol-involved pedestrian fatalities occurred during these times (ATSB, 2003). Strategies to reduce the number of crashes involving alcohol-affected pedestrians should aim to prevent pedestrians reaching high BACs and to prevent alcohol-affected pedestrians
752
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
being exposed to traffic (Victorian Parliamentary Road Safety Committee, 1999). While responsible alcohol service courses have an effect on the behaviour of personnel serving alcohol, studies continue to find that a large proportion of clearly intoxicated individuals are served at licensed premises even when servers have attended such courses (Buka and Birdthistle, 1999; Donnelly and Briscoe, 2003; Stockwell, 2001; Wallin et al., 2002). Lowering vehicle speeds in areas where alcohol-affected pedestrians may be present will reduce the risk for this group. Vehicle speed is directly related to crash risk (Aarts and van Schagen, 2006), and to the severity of pedestrian injury (Anderson et al., 1997; Ashton and Mackay, 1979; Garder, 2004; Matsui, 2005). At an impact speed of 20 kph there is about a 5% chance of serious injury to a pedestrian, which increases to about 25% at an impact speed around 30 kph (Ashton and Mackay, 1979; Leaf and Preusser, 1999). For an impact speed of 40 kph there is approximately a 25% chance of pedestrian fatality, and this figure increases markedly to about an 80% chance of death at impact speeds around 50 kph (Anderson et al., 1997). These comparisons illustrate why it is important to examine the proportion of vehicles travelling at speeds likely to cause serious injury and fatalities to pedestrians. There is little information on the effects of commonly used infrastructure-based treatments, such as pedestrian fencing, on pedestrian safety. Displaying a red traffic signal to all vehicle directions at signalised intersections during periods when no vehicular traffic is detected has been suggested as a potential method of reducing vehicle speeds at intersections in order to reduce the risk of collision with a pedestrian and the severity of injuries should a collision occur (Corben et al., 1998). While modifications to traffic signal operation reduce the risk of crashes involving pedestrians and bicyclists (Retting et al., 2002), the present study evaluated the effects of a new traffic signal phasing on safety for alcohol-affected pedestrians.
2. Method 2.1. Experimental design This study utilised a controlled before-and-after design, commonly defined as a trial in which data are collected at both a treatment and an appropriate control site, before and after the intervention at the treatment site (Elvik, 2002). The use of a control site helps to increase the likelihood that observed effects at the treatment site are attributable to the treatment itself rather than other extraneous influences. 2.2. Site selection A number of metropolitan and regional sites with a high number of police-reported pedestrian casualty crashes during HAH were identified. HAH are those time periods when the proportion of drivers or riders killed or seriously injured with a known BAC over 0.05% was greater than 15% (Shtifelman et al., 1998), and broadly covers the period between 6 p.m. and 6 a.m. for each day of the week. A similar methodology has also been used in analyses of alcohol-involved pedestrian crashes in Baltimore (Blomberg and Cleven, 2000). HAH were used as a surrogate indicator of alcohol involvement in pedestrian crashes because BAC cannot be relied upon as an accurate indicator of alcohol-related pedestrian crashes in Victoria as pedestrians are not routinely tested when involved in a crash (Victorian Parliamentary Road Safety Committee, 1999). The City of Ballarat in regional Victoria (population around 65,000) was selected for the trial. The configuration of the treatment site is presented in Fig. 1. The treatment site is the main route through the city, running east–west, and is a divided road with two through-lanes in each direction and dedicated left- and right-turn lanes. The north–south road is also two through-lanes in each direction. The control site is 300–350 m to the west along the same east–west main route, with a similar
Fig. 1. Representation of the treatment intersection with location of vehicle detectors.
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
north–south cross-road. The speed limit was 50 kph for both sites. Vehicle speed data were collected for traffic moving in an easterly direction, meaning that drivers travelled through the control site before the treatment site. Consequently, behaviour at the control site would not be influenced by prior exposure to the treatment. 2.3. Measure of pedestrian safety The aim of the Dwell-on-Red (DOR) treatment is to reduce vehicle speeds and thus increase safety for pedestrians. Vehicle speed is the most critical factor in determining pedestrian safety and was chosen as the dependent variable. Loop vehicle classifiers installed below the road surface were used to measure vehicle speed at two locations: 30 m on approach to the intersection and close to the stop line. For vehicles travelling at around the posted speed limit of 50 kph, the minimum distance required to brake and stop safely before the intersection is approximately 30 m (Anderson et al., 1997). Speed at 30 m before the intersection is therefore indicative of crash risk, while speed at the stop line is indicative of potential injury severity for pedestrians. 2.4. The Dwell-on-Red treatment The signal phasing for the trial had four phases: phase A, green signal to through-vehicles and pedestrians on the east–west route (minimum time on green 10 s, minimum time on yellow 3.5 s, and minimum time on all-red of 2.5 s); phase B, green signal to through-vehicles and pedestrians on the north–south route (minimum green, yellow, and all-red times of 8.0, 3.5, and 2.5 s, respectively); phase C, which could be skipped, involved a green signal to through and turning traffic in the north–south median (minimum green, yellow, and all-red times of 6.0, 3.0, and 1.5 s, respectively); and phase D, red signal to vehicles and pedestrians in all directions (DOR phase). The DOR phase was only included in the traffic signal sequence between 10 p.m. and 5 a.m. at the treatment site in the second data collection period. Within these times the sequence was DABC, whereby phase D was the recall phase, that is, the D phase was called every cycle when no vehicles or pedestrians were detected for a 15-s period, and the sequence rested in phase D when there were no other phase demands. Phases A, B and C were only called when demanded by a vehicle or pedestrian. Phase D was programmed to run for a minimum of 15 s. The duration of operation of the DOR (phase D) was dependent on the presence/absence of vehicles and pedestrians at the treatment site. The pedestrian signals were red in all directions during the DOR phase and only changed to green if activated by any pedestrian or vehicle arriving on any leg of the intersection. Normal traffic signal practices were activated once a vehicle or pedestrian was detected while DOR was activated.
(Tabachnick and Fidell, 1996). This customised model targeted the analysis of the main effects and interactions of interest, and used the differences observed across treatment and control sites in the before period as a basis to calculate the relative difference across sites in the after period. Four-way Univariate Analyses of Variance (ANOVA) were conducted with the variables being site (control and treatment), sampling period (before and after), day of week, and time of day. Time of day and day of week were included in the analyses as potential confounding variables but are not reported here. Logistic regression was used to compare the proportion of vehicles travelling below specified speeds at the treatment and control site, before and after the DOR signal phasing was introduced. When a significant interaction between site and sampling period was present, a comparison of the proportion of vehicles below the specified speed between the before period and after period was made for the control and treatment sites separately. 3. Results 3.1. Sampling characteristics All data were collected between February and August 2005. Data were collected prior to treatment activation (before period) between 10 p.m. on February 22 through to 5 a.m. on March 16. After treatment activation (after period) data were collected between 10 p.m. on August 11 through to 5 a.m. on August 23. The DOR treatment was active for a ‘settling in’ period of 4 weeks prior to the after data collection period. Table 1 shows the total number of speed measurements collected at the two points. The smaller number of observations in the after period reflects the shorter duration of this data collection period. Approximately half of the data collected at the treatment site at the 30 m point in the before period were erroneous due to a system error and were excluded from the analyses. Table 2 shows the number of DOR activations that occurred across the period of DOR operation, as well as the mean duration per activation. In periods of low traffic operations there was more scope for activation of the treatment with longer durations per activation, while increased traffic volume was associated with more frequent activation of shorter duration. In areas of very high traffic volume there is limited scope for activation of the treatment. Preliminary analyses considered the potential influence of traffic volume on the effectiveness of DOR in reducing vehicle speeds however there was no noticeable effect. Presumably this was because DOR activation is influenced by traffic flow rather than volume. As traffic flow was not measured in this study, Table 1 The number of speed measurements collected at the two collection points at the treatment and control site Site
2.5. Data analysis A customised general linear model was used to analyse the speed data for the two locations (30 m and at the stop line)
753
Control Treatment Total
30 m point
Stop line point
Before
After
Total
Before
After
Total
16,668 9,080 25,748
7,285 8,191 15,476
23,953 17,271 41,224
16,210 16,404 32,614
7,374 7,121 14,495
23,584 23,525 47,109
754
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
Table 2 Characteristics of Dwell-on-Red activation across the hours of treatment operation Time of day
22:00–22:59 23:00–23:59 00:00–00:59 01:00–01:59 02:00–02:59 03:00–03:59 04:00–04:59
Mean activations/h
Mean duration per activation (s)
Mean total duration of activation/h (min)
Traffic counts
71 72 60 49 42 44 41
14.9 17.8 22.0 36.8 56.3 58.2 63.5
17.6 21.4 22.0 30.0 39.0 42.2 43.2
2769 1830 1075 622 330 263 232
Table 3 Mean speed (kph ± S.E.) for control and treatment sites at 30 m and the stop line Site
30 m
Control Treatment
Stop line
Before
After
Before
After
42.2 (±0.1) 41.7 (± 0.2)
47.3 (±0.3) 37.8 (±0.3)
36.9 (±0.2) 39.0 (±0.2)
44.6 (±0.3) 28.0 (±0.3)
this was not analysed further and the data are presented pooled across all traffic volumes. 3.2. Vehicle speed Table 3 presents mean vehicle speeds collected at the two points at the control and treatment intersections. At both points vehicle speeds dropped after the activation of the DOR phasing at the treatment site but not at the control site (30 m point, F(1, 41,170) = 373.9, p < 0.001; stop line, F(1, 46,582) = 1086.8, p < 0.001). Mean speed for the control site increased in the after period compared to the before period at both collection points despite no differences in signal phasing. 3.2.1. Speed in the potential pedestrian impact zone At the stop line, mean speed was 11 kph (28%) lower at the treatment site in the after period compared to the before period. Taking into account the difference in speed at control and treatment sites in the before period, further analysis showed that speed was significantly lower in the after period at the treatment site compared to the control site (p < 0.001), resulting in a net reduction in mean speed at the treatment site of around 40%. For reasons relating to the speed-dependent risk of pedestrian injury discussed earlier, the speed data collected at the stop line were categorised into vehicles travelling ≤30 or >30 kph (see Table 4). The proportion of vehicles travelling ≤30 kph at the treatment site increased in the after period (OR = 3.77, 95% Table 4 Proportion (%) of vehicles travelling below specified speeds Site
Control Treatment
Stop line,≤30 kph
30 m point, ≤40 kph
30 m point, ≤50 kph
Before
After
Before
After
Before
After
41 39
30 71
46 49
33 73
78 81
68 94
CI = 3.54–4.00, p < 0.0001), while at the control site the proportion ≤30 kph was reduced in the after period (OR = 0.61, 95% CI = 0.58–0.65, p < 0.0001). 3.2.2. Speed on approach to the intersection At the 30 m point, mean speed was 3.9 kph (9%) lower at the treatment site in the after period when DOR was activated compared to the before period. Taking into account the difference in speed between control and treatment sites in the before period, further analysis showed that speed was significantly lower in the after period at the treatment site compared to the control site (p < 0.05), resulting in a net reduction in mean speed at the treatment site of around 19%. For reasons related to pedestrian fatality risk and vehicle crash risk, respectively, the proportion of vehicles travelling below/above 40 and 50 kph at the 30 m point were analysed (see Table 4). The proportion of vehicles travelling at ≤40 kph at the treatment site increased in the after period (OR = 2.83, 95% CI = 2.66–3.02, p < 0.0001), but decreased at the control site in the after period (OR = 0.57, 95% CI = 0.54–0.61, p < 0.0001). The treatment was therefore indicative of a reduction in potential risk of injury to pedestrians crossing between the 30 m and stop line points. The proportion of vehicles travelling at ≤50 kph at the treatment site also increased in the after period (OR = 3.36, 95% CI = 3.03–3.73, p < 0.0001), but again decreased at the control site in the after period (OR = 0.61, 95% CI = 0.58–0.65, p < 0.0001). The treatment was therefore suggestive of a reduction in crash risk. 4. Discussion This study evaluated the effect of DOR signal phasing during HAH on vehicle speeds at an intersection with a high risk of crashes involving alcohol-affected pedestrians. It was necessary to perform this research to determine if the speed reductions expected upon activation of DOR signal phasing did occur in practice, and to estimate the magnitude of any speed reduction. It was also of interest to observe the number of times per hour that the DOR signal phasing was activated and how this related to traffic volume, as no previous data were available. We demonstrated measurable reductions in mean speeds with DOR signal phasing. Taking a conservative approach and considering the treatment site only, the mean speed at the stop line decreased from 39 to 28 kph, which represents a 28% reduction. There was a concomitant 52% reduction in the proportion of vehicles travelling faster than 30 kph. If speed at the stop line is considered as the potential impact speed with pedestrians on the crossing, this suggests significant reductions in the risk of fatal and serious injury to pedestrians after activation of the DOR phasing. Additionally, we found that mean speeds measured 30 m prior to the intersection decreased from 41.7 to 37.8 kph (a 9% reduction) with a 47% reduction in the proportion of vehicles travelling faster than 40 kph, and a 68% reduction in the proportion travelling over 50 kph. This represents a significant
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
reduction in the proportion of vehicles travelling too fast to brake and stop safely before the intersection. Having found a measurable impact of DOR signal phasing on vehicle speeds during HAH, it is our intention to consider the broader effects and potential application of the DOR signal phasing. Of prime importance is the assessment of the effect of DOR on the number of pedestrian–vehicle conflicts and the rates of crashes involving pedestrians. It is also of interest to determine if the rate of vehicle–vehicle crashes is affected, as it has been suggested that the DOR signal phasing might increase the number of stop–start maneuvers that vehicles make which could have a detrimental effect on rear-end collision rates. While the current study was able to investigate the relationship between the number and duration of DOR activations and traffic volume, the most important predictor of DOR activation is traffic flow. Further research will measure this relationship. There has been some concern that DOR signal phasing could increase vehicle travel times, however this is unlikely to be a major problem as it is envisaged that DOR will be operated selectively at the small number of sites that have a high alcohol-pedestrian crash problem during HAH. Traffic volumes are generally lower during these hours. The effect on travel times could be assessed by undertaking number plate surveys in conjunction with the traffic flow measurements. While this evaluation of DOR signal phasing was conducted at one intersection in a regional city during HAH to target alcohol-involved pedestrian crashes, the system-wide effects and other potential areas of application need to be explored. Further research should explore the potential to translate the speed reductions and the associated safety benefits gained through the application of DOR signal phasing to other regional and metropolitan locations, including intersections where other vulnerable groups (e.g., children and the elderly) at high risk of road trauma would benefit from a reduction in vehicle speeds. The acceptability of DOR signal phasing to drivers and pedestrians should also be assessed. The authors have received funding to develop the concept further. In conclusion, alcohol-affected pedestrians are a group that is at high risk of being involved in pedestrian casualty crashes. A range of treatments and programs have been implemented to increase safety for alcohol-affected pedestrians with limited widespread success. The results reported here demonstrate that the DOR modification to traffic signal operation in a regional setting significantly reduced vehicle speeds within the assumed HAH for pedestrians. Such speed reductions are associated with substantial reductions to the potential risk of fatal and serious injury to pedestrians, and a reduction in crash risk. Acknowledgements This project was undertaken under contract for VicRoads. The authors wish to thank VicRoads, in particular Elizabeth Knight and Linda Ivett, for advice received in conducting this project and for approval to publish the research. The authors would also like to thank MUARC colleagues Ashley Verdoorn for his technical assistance and Christine Mulvihill for her work in an earlier phase of this research. Funding for the preparation of
755
this paper was provided through the MUARC SDP Publication Award. References Aarts, L., van Schagen, I., 2006. Driving speed and the risk of road crashes: a review. Accid. Anal. Prev. 38 (2), 215–224. Anderson, R., McLean, A., Farmer, M., Lee, B., Brooks, C., 1997. Vehicle travel speeds and the incidence of fatal pedestrian crashes. Accid. Anal. Prev. 29, 667–674. Ashton, S., Mackay, G., 1979. Some characteristics of the population who suffer trauma as pedestrians when hit by cars and some resulting implication. In: Proceedings of the IRCOBI Conference, Gothenburg, Sweden. Australian Transport Safety Bureau, 2003. Male Pedestrian Fatalities. Retrieved February 17, 2006, from http://www.atsb.gov.au/publications/2003/Ped Male 1.aspx. Australian Transport Safety Bureau, 2005. Road Deaths Australia: 2004 Statistical Summary. Australian Transport Safety Bureau, Canberra, ACT. Blomberg, R.D., Cleven, A.M., 2000. Development, Implementation and Evaluation of a Countermeasure Program for Alcohol-involved Pedestrian Crashes. US Department of Transportation, National Highway Traffic safety Administration, Washington, DC. Buka, S.L., Birdthistle, I.J., 1999. Long-term effects of a community-wide alcohol server training intervention. J. Stud. Alcohol 60, 27–36. Clayton, A., Colgan, M., 2001. Alcohol and pedestrians. A final report to Road Safety Division (Road Safety Research Report No. 20). Department of the Environment, Transport and the Regions, London, UK. Corben, B.F., Diamantopoulou, K., Mullan, N., 1998. Environmental countermeasures for alcohol-related pedestrian crashes. In: Proceedings of the Ninth Road Engineering Association of Asia and Australasia, Wellington, New Zealand. Davies, D., 1999. Research, development and implementation of pedestrian safety facilities in the United Kingdom (Report No. FHWA-RD-99-089). Federal Highway Administration, US Department of Transportation, Washington, DC. Donnelly, N., Briscoe, S., 2003. Signs of intoxication and server intervention among 18 to 39-year-olds drinking at licensed premises in New South Wales, Australia. Addiction 98 (9), 1287–1295. Elvik, R., 2002. The importance of confounding in observational before-andafter studies of road safety measures. Accid. Anal. Prev. 34 (5), 631–635. Garder, P.E., 2004. The impact of speed and other variables on pedestrian safety in Maine. Accid. Anal. Prev. 36 (4), 533–542. Leaf, W.A., Preusser, D.F., 1999. Literature Review on Vehicle Travel Speeds and Pedestrian Injuries (Report No. DTNH22-97-D-05018). National Highway Traffic Safety Administration, Washington, DC. Lee, C., Abdel-Aty, M., 2005. Comprehensive analysis of vehicle–pedestrian crashes at intersections in Florida. Accid. Anal. Prev. 37 (4), 775–786. Martinez, K.L.H., Porter, B.E., 2004. The likelihood of becoming a pedestrian fatality and drivers’ knowledge of pedestrian rights and responsibilities in the Commonwealth of Virginia. Transport. Res., Part F: Traffic Psychol. Behav. 7 (1), 43–58. Matsui, Y., 2005. Effects of vehicle bumper height and impact velocity on type of lower extremity injury in vehicle–pedestrian accidents. Accid. Anal. Prev. 37 (4), 633–640. Miles-Doan, R., 1996. Alcohol use among pedestrians and the odds of surviving an injury: evidence from Florida law enforcement data. Accid. Anal. Prev. 28 (1), 23–31. National Highway Traffic Safety Administration, 2003. Pedestrian Roadway Fatalities (DOT HS 809 456). US Department of Transportation, Washington, DC. Ostr¨om, M., Eriksson, A., 2001. Pedestrian fatalities and alcohol. Accid. Anal. Prev. 33 (2), 173–180. Retting, R.A., Chapline, J.F., Williams, A.F., 2002. Changes in crash risk following re-timing of traffic signal change intervals. Accid. Anal. Prev. 34 (2), 215–220. Shtifelman, M., Cameron, M., Diamantopoulou, K., 1998. Update of Alcohol Times as a Surrogate Measure of Alcohol-involvement in Accidents
756
M.G. Lenn´e et al. / Accident Analysis and Prevention 39 (2007) 751–756
in Melbourne and Country Victoria During 1990–1997. Monash University Accident Research Centre, Melbourne. Stockwell, T., 2001. Responsible alcohol service: Lessons from evaluations of server training and policing initiatives. Drug Alcohol Rev. 20, 257–265. Tabachnick, B.G., Fidell, L.S., 1996. Using multivariate statistics, 3rd ed. HarperCollins, New York. Transport Accident Commission, 2005. Pedestrian Statistics. Retrieved November 30, 2005, from http://www.tacsafety.com.au/jsp/homepage/home.jsp.
Victorian Parliamentary Road Safety Committee, 1999. Walking Safely: Inquiry into the Incidence and Prevention of Pedestrian Accidents. Victorian Government Printer, Melbourne. Wallin, E., Gripenberg, J., Andreasson, S., 2002. Too drunk for a beer? A study of overserving in Stockholm. Addiction 97 (7), 901–907. Zajac, S.S., Ivan, J.N., 2003. Factors influencing injury severity of motor vehicle–crossing pedestrian crashes in rural Connecticut. Accid. Anal. Prev. 35 (3), 369–379.