Traffic Injury Prevention
ISSN: 1538-9588 (Print) 1538-957X (Online) Journal homepage: http://www.tandfonline.com/loi/gcpi20
Comparative Analysis of Risky Behaviors of Electric Bicycles at Signalized Intersections Lu Bai, Pan Liu, Yanyong Guo & Hao Yu To cite this article: Lu Bai, Pan Liu, Yanyong Guo & Hao Yu (2015) Comparative Analysis of Risky Behaviors of Electric Bicycles at Signalized Intersections, Traffic Injury Prevention, 16:4, 424-428, DOI: 10.1080/15389588.2014.952724 To link to this article: http://dx.doi.org/10.1080/15389588.2014.952724
View supplementary material
Accepted author version posted online: 18 Aug 2014.
Submit your article to this journal
Article views: 130
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=gcpi20 Download by: [University of Lethbridge]
Date: 06 November 2015, At: 21:37
Traffic Injury Prevention (2015) 16, 424–428 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2014.952724
Comparative Analysis of Risky Behaviors of Electric Bicycles at Signalized Intersections LU BAI, PAN LIU, YANYONG GUO, and HAO YU Jiangsu Key Laboratory of Urban ITS, Jiangsu Collaborative Innovation Center of Modern Urban Traffic Technologies, Southeast University, Nanjing, China
Downloaded by [University of Lethbridge] at 21:37 06 November 2015
Received 11 February 2014, Accepted 4 August 2014
Objective: The primary objective of this study was to compare the risky behaviors of e-bike, e-scooter, and bicycle riders as they were crossing signalized intersections. Methods: Pearson’s chi-square test was used to identify whether there were significant differences in the risky behaviors among e-bike, e-scooter, and bicycle riders. Binary logit models were developed to evaluate how various variables affected the behaviors of 2-wheeled vehicle riders at signalized intersections. Field data collection was conducted at 13 signalized intersections in 2 cities (Nanjing and Kunming) in China. Results: Three different types of risky behaviors were identified, including stop beyond the stop line, riding in motorized lanes, and riding against traffic. Two-wheeled vehicle riders’ gender and age and traffic conditions were significantly associated with the behaviors of 2-wheeled vehicle riders at the selected signalized intersections. Conclusions: Compared to e-bike and bicycle riders, e-scooter riders are more likely to take risky behaviors. More specifically, they are more likely to ride in motorized lanes and ride against traffic. Keywords: e-scooter, e-bike, bicycle, risky behavior, signalized intersection, binary logit model
Introduction Electric bicycles (e-bicycles) have been increasingly used in the urban areas in China over the past 2 decades. In 2012, the number of e-bicycles exceeded 160 million, which was up from 58,000 in 1998 (Ma et al. 1998–2012). The number of e-bicycles is expected to continue to grow over the next few years. E-bicycles have been considered an environmental friendly alternative to automobiles, with convenient, flexible, and affordable mobility. Despite these advantages, the use of ebicycles has also raised some safety concerns. In 2012, 5,314 riders of e-bicycles were killed and 26,966 were injured, mostly in crashes with automobiles. The fatalities and injuries of ebicycle riders accounted for 49.6 and 61.4% of the nonmotorized traffic fatalities and injuries in 2012 in China (Yang et al. 2012). In the past, numerous studies have investigated the behavioral characteristics of e-bicycle and bicycle riders as they are crossing signalized intersections (Bernhoft and Carstensen 2008; Johnson et al. 2011; Ling and Wu, 2004; Wang et al. 2012; Wu et al. 2012). Several studies were conducted to shed
Associate Editor Clay Gabler oversaw the review of this article Address correspondence to Lu Bai, Jiangsu Key Laboratory of Urban ITS, Jiangsu Collaborative Innovation Center of Modern Urban Traffic Technologies, Southeast University, Si Pai Lou #2, Nanjing 210096, China. E-mail: [email protected]
light on how the behaviors of e-bicycle riders influenced safety (Elliott et al. 2007; Reason et al. 1990; Steg and Brussel 2009; Yao and Wu 2012). The undesirable behaviors of riders were found to be the main predictors of crash risks (Elliott et al. 2007; Reason et al. 1990; Yao and Wu 2012). It is reasonable to believe that a strong relationship exists between the behaviors of e-bicycle riders and crash risks. Two major types of e-bicycles are currently being used in China, including bicycle- and scooter-style e-bikes. For convenience, the scooter-style e-bike is termed an e-scooter, and the term e-bike is used to denote the bicycle-style e-bike. The e-bicycles studied in this article included both e-bikes and escooters. Our previous studies have investigated the behaviors of e-bicycle and bicycle riders and their effects on safety at signalized intersections (Bai et al. 2013; Guo et al. 2014). We have also investigated the red light–running behaviors of the riders of e-bikes, e-scooters, and bicycles as they were crossing signalized intersections (Guo et al. 2014). In the present study, comparative analyses were conducted to investigate the differences in the behavioral characteristics of e-bike, e-scooter, and bicycle riders as they were crossing signalized intersections. The focus was on the risky behaviors that were not examined by Guo et al. (2014). It is expected that the research results will help transportation engineers design a safety environment optimized for different proportions of e-bike, e-scooter, and conventional bicycle traffic. Quantitatively knowing relative likelihoods of risky behaviors would allow cost–benefit optimization of intersection design, given expected traffic
Risky Behaviors of Electric Bicycles
425
composition and the estimates of the effects of different design features on risky behaviors.
Data and Methods
Downloaded by [University of Lethbridge] at 21:37 06 November 2015
Field Data Collection Field data collection was conducted at 13 signalized intersections in 2 cities (Nanjing and Kunming) in China. The characteristics of the selected sites are given in Table A1 (see online supplement). The selected sites had similar geometric design and traffic control features, which represented the most typical intersections in major cities in China (shown in Figure A1, see online supplement). More specifically, all selected sites were 4-leg intersections and installed with pedestrian signals, of which 10 sites were installed with flashing pedestrian signals that used flashing green lights to preempt the termination of the green phase; the other 3 sites were installed with countdown timers that displayed the remaining time of the red and green phases. Four video cameras were used to cover the study area from different angles. The cameras were placed on top of a roadside building to achieve an adequate viewing height. Before we started recording, the 4 cameras were synchronized such that the data extracted from different videos could be matched. Data were recorded during weekday peak periods under fine weather conditions, and live traffic enforcement was not present. The purpose of doing so was to minimize the impacts of confounding factors, such that the differences in the observed crossing behaviors can be more readily attributed to e-bike, e-scooter, and bicycle riders’ behavioral characteristics. For each selected site the research team recorded 6 h of data. The observation time for each selected site was from 6:00 to 9:00 a.m. and from 5:00 to 8:00 p.m. In total, 78 h of data were recorded at the selected sites. The recorded videos were reviewed in the laboratory for data reduction. For each individual rider, the research team recorded the gender, estimated age, type of the vehicle, group size, and risky behaviors. The age of the riders was divided into 3 groups. “young” corresponded to riders younger than approximately 25 years old; “middle-aged” corresponded to riders between 25 and 60 years old; and “older” corresponded to riders older than approximately 60 years old. The group size associated with each rider was recorded to identify the differences in the crossing behaviors between individuals and groups. The group size was defined as the number of riders who ride side-by-side within a distance of approximately 2 times the length of 2-wheeled vehicles (Wu et al. 2012). The group size was divided into 3 types according to the number of riders in a group (10). The volume of the traffic flow that was in conflict with the bicycle flow was counted in 5min time intervals. The conflicting flow included the through traffic in both directions on the crossing street, the opposing left-turning traffic, and the right-turning traffic from the same approach. The crossing behaviors of 6,169 individuals, including the riders of 632 e-bikes, 3,990 e-scooters, and 1,547 bicycles, were observed and recorded at the selected sites. Three different
Fig. 1. Proportion histogram and standard error bars of risky behaviors of 2-wheeled vehicle riders.
types of risky behaviors were identified by reviewing videos, including stopping beyond the stop line, riding in motorized lanes, and riding against the traffic (shown in Figures A2 and A3, see online supplement). Stopping beyond the stop line was defined as the phenomenon in which the front wheel of the stopped 2-wheeled vehicles crossed the stop line; riding in motorized lanes was defined as the phenomenon in which 2wheeled vehicles illegally used the motorized lane in the same direction; riding against traffic was defined as the phenomenon in which 2-wheeled vehicles used the nonmotorized lanes in the wrong direction. The data shown in Figure 1 represent the proportions of different types of risky behaviors that were displayed by e-bike, e-scooter, and bicycle riders.
Data Analysis Methods The data were analyzed using the following two methods. 1. The Pearson’s chi-square test was used to identify whether there were significant differences in the risky behaviors among e-bike, e-scooter, and bicycle riders. Multiple comparisons were applied to compare the risky behaviors among multiple groups. Caution should be exercised, however, when applying Pearson’s chi-square test on small sample sizes. If any of the expected cell counts are less than 5, Fisher’s exact test was applied instead of Pearson’s chisquare test (Washington et al. 2003). 2. The binary logit model has been widely used for predicting a binary dependent variable as a function of explanatory variables. In this study, 4 binary logit models were developed to evaluate how influential factors affected the behaviors of 2-wheeled vehicle riders at signalized intersections. The base level of the target variable in each model was the nonoccurrence of one or more of the risky behaviors. The odds ratio (OR) was estimated to quantitatively evaluate the impacts of various explanatory variables on the chance of a rider taking risky behaviors. It indicates the relative amount by which the odds of the outcome increases (OR > 1) or decreases (OR < 1) when the value of the corresponding explanatory variable increases by one unit (Washington et al. 2003).
426
Bai et al.
Table 1. Odds ratios of risky behavior for e-bike, e-scooter, and bicycle riders Risky behaviors
Variable
Stop beyond the stop line
Male vs. female Young vs. older Middle-aged vs. older Group size 2 vs. group size 1 Group size 3 vs. group size 1 Volume of 2-wheelers in 5 min Conflicting traffic volume in 5 min Constant Young vs. older Middle-aged vs. older E-scooter vs. bicycle Volume of 2-wheelers in 5 min Constant Male vs. female Young vs. older Middle-aged vs. older E-scooter vs. bicycle Morning vs. afternoon Roadway width (m) Volume of 2-wheelers in 5 min Constant
Riding in motorized lanes
Downloaded by [University of Lethbridge] at 21:37 06 November 2015
Riding against traffic
Model specification started from calculating the Pearson correlation coefficients to identify possible correlations between various candidate variables. The correlation was found to exist between the type of automobile signal phase and the width of the intersection. The type of automobile signal phase was excluded from the candidate explanatory variables in order to keep the variables independent in the regression model. Stepwise variable selection was conducted to select the explanatory variables that should be included in the binary logit model. The model with the highest log likelihood at convergence was considered the best. The best model had 8 explanatory variables, including 6 binary variables. To better understand different types of risky behaviors, binary logit models were developed to evaluate how various variables affect different types of behaviors of 2-wheeled vehicle riders at signalized intersections. All of the variables in the best model were statistically significant at the 95% confidence level.
Results of Comparative Analyses Results of Pearson’s Chi-square Test The results of the Pearson’s chi-square test are shown in Tables A2 and A3 (see online supplement). More than 20% of ebike and e-scooter riders displayed at least one type of the identified risky behaviors as they were crossing intersections. The proportion of the riders taking risky behaviors was 1.05 and 1.19 times as large as that of the riders of the bicycles for e-bikes and e-scooters, respectively. The proportions of ebike, e-scooter, and bicycle riders engaging in risky behaviors were significantly different at the 95% confidence level. The proportion of e-scooter riders engaging in risky behaviors was significantly higher than that of bicycle riders. This finding indicated that compared to bicycle riders, e-scooter riders were more inclined to take risky behaviors. The differences were found to be statistically significant among e-bike, e-scooter, and bicycle riders for the risky
Coefficient 0.521 1.247 0.676 1.060 0.749 0.011 −0.014 −3.015 1.915 0.878 0.948 0.016 −5.767 2.147 1.736 1.258 1.792 2.129 −0.060 −0.011 −8.313
SE 0.090 0.146 0.144 0.102 0.111 0.001 0.002 0.246 0.428 0.438 0.231 0.003 0.523 1.015 0.739 0.735 0.604 0.416 0.029 0.005 1.519
P value
ISSN: 1538-9588 (Print) 1538-957X (Online) Journal homepage: http://www.tandfonline.com/loi/gcpi20
Comparative Analysis of Risky Behaviors of Electric Bicycles at Signalized Intersections Lu Bai, Pan Liu, Yanyong Guo & Hao Yu To cite this article: Lu Bai, Pan Liu, Yanyong Guo & Hao Yu (2015) Comparative Analysis of Risky Behaviors of Electric Bicycles at Signalized Intersections, Traffic Injury Prevention, 16:4, 424-428, DOI: 10.1080/15389588.2014.952724 To link to this article: http://dx.doi.org/10.1080/15389588.2014.952724
View supplementary material
Accepted author version posted online: 18 Aug 2014.
Submit your article to this journal
Article views: 130
View related articles
View Crossmark data
Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalInformation?journalCode=gcpi20 Download by: [University of Lethbridge]
Date: 06 November 2015, At: 21:37
Traffic Injury Prevention (2015) 16, 424–428 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2014.952724
Comparative Analysis of Risky Behaviors of Electric Bicycles at Signalized Intersections LU BAI, PAN LIU, YANYONG GUO, and HAO YU Jiangsu Key Laboratory of Urban ITS, Jiangsu Collaborative Innovation Center of Modern Urban Traffic Technologies, Southeast University, Nanjing, China
Downloaded by [University of Lethbridge] at 21:37 06 November 2015
Received 11 February 2014, Accepted 4 August 2014
Objective: The primary objective of this study was to compare the risky behaviors of e-bike, e-scooter, and bicycle riders as they were crossing signalized intersections. Methods: Pearson’s chi-square test was used to identify whether there were significant differences in the risky behaviors among e-bike, e-scooter, and bicycle riders. Binary logit models were developed to evaluate how various variables affected the behaviors of 2-wheeled vehicle riders at signalized intersections. Field data collection was conducted at 13 signalized intersections in 2 cities (Nanjing and Kunming) in China. Results: Three different types of risky behaviors were identified, including stop beyond the stop line, riding in motorized lanes, and riding against traffic. Two-wheeled vehicle riders’ gender and age and traffic conditions were significantly associated with the behaviors of 2-wheeled vehicle riders at the selected signalized intersections. Conclusions: Compared to e-bike and bicycle riders, e-scooter riders are more likely to take risky behaviors. More specifically, they are more likely to ride in motorized lanes and ride against traffic. Keywords: e-scooter, e-bike, bicycle, risky behavior, signalized intersection, binary logit model
Introduction Electric bicycles (e-bicycles) have been increasingly used in the urban areas in China over the past 2 decades. In 2012, the number of e-bicycles exceeded 160 million, which was up from 58,000 in 1998 (Ma et al. 1998–2012). The number of e-bicycles is expected to continue to grow over the next few years. E-bicycles have been considered an environmental friendly alternative to automobiles, with convenient, flexible, and affordable mobility. Despite these advantages, the use of ebicycles has also raised some safety concerns. In 2012, 5,314 riders of e-bicycles were killed and 26,966 were injured, mostly in crashes with automobiles. The fatalities and injuries of ebicycle riders accounted for 49.6 and 61.4% of the nonmotorized traffic fatalities and injuries in 2012 in China (Yang et al. 2012). In the past, numerous studies have investigated the behavioral characteristics of e-bicycle and bicycle riders as they are crossing signalized intersections (Bernhoft and Carstensen 2008; Johnson et al. 2011; Ling and Wu, 2004; Wang et al. 2012; Wu et al. 2012). Several studies were conducted to shed
Associate Editor Clay Gabler oversaw the review of this article Address correspondence to Lu Bai, Jiangsu Key Laboratory of Urban ITS, Jiangsu Collaborative Innovation Center of Modern Urban Traffic Technologies, Southeast University, Si Pai Lou #2, Nanjing 210096, China. E-mail: [email protected]
light on how the behaviors of e-bicycle riders influenced safety (Elliott et al. 2007; Reason et al. 1990; Steg and Brussel 2009; Yao and Wu 2012). The undesirable behaviors of riders were found to be the main predictors of crash risks (Elliott et al. 2007; Reason et al. 1990; Yao and Wu 2012). It is reasonable to believe that a strong relationship exists between the behaviors of e-bicycle riders and crash risks. Two major types of e-bicycles are currently being used in China, including bicycle- and scooter-style e-bikes. For convenience, the scooter-style e-bike is termed an e-scooter, and the term e-bike is used to denote the bicycle-style e-bike. The e-bicycles studied in this article included both e-bikes and escooters. Our previous studies have investigated the behaviors of e-bicycle and bicycle riders and their effects on safety at signalized intersections (Bai et al. 2013; Guo et al. 2014). We have also investigated the red light–running behaviors of the riders of e-bikes, e-scooters, and bicycles as they were crossing signalized intersections (Guo et al. 2014). In the present study, comparative analyses were conducted to investigate the differences in the behavioral characteristics of e-bike, e-scooter, and bicycle riders as they were crossing signalized intersections. The focus was on the risky behaviors that were not examined by Guo et al. (2014). It is expected that the research results will help transportation engineers design a safety environment optimized for different proportions of e-bike, e-scooter, and conventional bicycle traffic. Quantitatively knowing relative likelihoods of risky behaviors would allow cost–benefit optimization of intersection design, given expected traffic
Risky Behaviors of Electric Bicycles
425
composition and the estimates of the effects of different design features on risky behaviors.
Data and Methods
Downloaded by [University of Lethbridge] at 21:37 06 November 2015
Field Data Collection Field data collection was conducted at 13 signalized intersections in 2 cities (Nanjing and Kunming) in China. The characteristics of the selected sites are given in Table A1 (see online supplement). The selected sites had similar geometric design and traffic control features, which represented the most typical intersections in major cities in China (shown in Figure A1, see online supplement). More specifically, all selected sites were 4-leg intersections and installed with pedestrian signals, of which 10 sites were installed with flashing pedestrian signals that used flashing green lights to preempt the termination of the green phase; the other 3 sites were installed with countdown timers that displayed the remaining time of the red and green phases. Four video cameras were used to cover the study area from different angles. The cameras were placed on top of a roadside building to achieve an adequate viewing height. Before we started recording, the 4 cameras were synchronized such that the data extracted from different videos could be matched. Data were recorded during weekday peak periods under fine weather conditions, and live traffic enforcement was not present. The purpose of doing so was to minimize the impacts of confounding factors, such that the differences in the observed crossing behaviors can be more readily attributed to e-bike, e-scooter, and bicycle riders’ behavioral characteristics. For each selected site the research team recorded 6 h of data. The observation time for each selected site was from 6:00 to 9:00 a.m. and from 5:00 to 8:00 p.m. In total, 78 h of data were recorded at the selected sites. The recorded videos were reviewed in the laboratory for data reduction. For each individual rider, the research team recorded the gender, estimated age, type of the vehicle, group size, and risky behaviors. The age of the riders was divided into 3 groups. “young” corresponded to riders younger than approximately 25 years old; “middle-aged” corresponded to riders between 25 and 60 years old; and “older” corresponded to riders older than approximately 60 years old. The group size associated with each rider was recorded to identify the differences in the crossing behaviors between individuals and groups. The group size was defined as the number of riders who ride side-by-side within a distance of approximately 2 times the length of 2-wheeled vehicles (Wu et al. 2012). The group size was divided into 3 types according to the number of riders in a group (10). The volume of the traffic flow that was in conflict with the bicycle flow was counted in 5min time intervals. The conflicting flow included the through traffic in both directions on the crossing street, the opposing left-turning traffic, and the right-turning traffic from the same approach. The crossing behaviors of 6,169 individuals, including the riders of 632 e-bikes, 3,990 e-scooters, and 1,547 bicycles, were observed and recorded at the selected sites. Three different
Fig. 1. Proportion histogram and standard error bars of risky behaviors of 2-wheeled vehicle riders.
types of risky behaviors were identified by reviewing videos, including stopping beyond the stop line, riding in motorized lanes, and riding against the traffic (shown in Figures A2 and A3, see online supplement). Stopping beyond the stop line was defined as the phenomenon in which the front wheel of the stopped 2-wheeled vehicles crossed the stop line; riding in motorized lanes was defined as the phenomenon in which 2wheeled vehicles illegally used the motorized lane in the same direction; riding against traffic was defined as the phenomenon in which 2-wheeled vehicles used the nonmotorized lanes in the wrong direction. The data shown in Figure 1 represent the proportions of different types of risky behaviors that were displayed by e-bike, e-scooter, and bicycle riders.
Data Analysis Methods The data were analyzed using the following two methods. 1. The Pearson’s chi-square test was used to identify whether there were significant differences in the risky behaviors among e-bike, e-scooter, and bicycle riders. Multiple comparisons were applied to compare the risky behaviors among multiple groups. Caution should be exercised, however, when applying Pearson’s chi-square test on small sample sizes. If any of the expected cell counts are less than 5, Fisher’s exact test was applied instead of Pearson’s chisquare test (Washington et al. 2003). 2. The binary logit model has been widely used for predicting a binary dependent variable as a function of explanatory variables. In this study, 4 binary logit models were developed to evaluate how influential factors affected the behaviors of 2-wheeled vehicle riders at signalized intersections. The base level of the target variable in each model was the nonoccurrence of one or more of the risky behaviors. The odds ratio (OR) was estimated to quantitatively evaluate the impacts of various explanatory variables on the chance of a rider taking risky behaviors. It indicates the relative amount by which the odds of the outcome increases (OR > 1) or decreases (OR < 1) when the value of the corresponding explanatory variable increases by one unit (Washington et al. 2003).
426
Bai et al.
Table 1. Odds ratios of risky behavior for e-bike, e-scooter, and bicycle riders Risky behaviors
Variable
Stop beyond the stop line
Male vs. female Young vs. older Middle-aged vs. older Group size 2 vs. group size 1 Group size 3 vs. group size 1 Volume of 2-wheelers in 5 min Conflicting traffic volume in 5 min Constant Young vs. older Middle-aged vs. older E-scooter vs. bicycle Volume of 2-wheelers in 5 min Constant Male vs. female Young vs. older Middle-aged vs. older E-scooter vs. bicycle Morning vs. afternoon Roadway width (m) Volume of 2-wheelers in 5 min Constant
Riding in motorized lanes
Downloaded by [University of Lethbridge] at 21:37 06 November 2015
Riding against traffic
Model specification started from calculating the Pearson correlation coefficients to identify possible correlations between various candidate variables. The correlation was found to exist between the type of automobile signal phase and the width of the intersection. The type of automobile signal phase was excluded from the candidate explanatory variables in order to keep the variables independent in the regression model. Stepwise variable selection was conducted to select the explanatory variables that should be included in the binary logit model. The model with the highest log likelihood at convergence was considered the best. The best model had 8 explanatory variables, including 6 binary variables. To better understand different types of risky behaviors, binary logit models were developed to evaluate how various variables affect different types of behaviors of 2-wheeled vehicle riders at signalized intersections. All of the variables in the best model were statistically significant at the 95% confidence level.
Results of Comparative Analyses Results of Pearson’s Chi-square Test The results of the Pearson’s chi-square test are shown in Tables A2 and A3 (see online supplement). More than 20% of ebike and e-scooter riders displayed at least one type of the identified risky behaviors as they were crossing intersections. The proportion of the riders taking risky behaviors was 1.05 and 1.19 times as large as that of the riders of the bicycles for e-bikes and e-scooters, respectively. The proportions of ebike, e-scooter, and bicycle riders engaging in risky behaviors were significantly different at the 95% confidence level. The proportion of e-scooter riders engaging in risky behaviors was significantly higher than that of bicycle riders. This finding indicated that compared to bicycle riders, e-scooter riders were more inclined to take risky behaviors. The differences were found to be statistically significant among e-bike, e-scooter, and bicycle riders for the risky
Coefficient 0.521 1.247 0.676 1.060 0.749 0.011 −0.014 −3.015 1.915 0.878 0.948 0.016 −5.767 2.147 1.736 1.258 1.792 2.129 −0.060 −0.011 −8.313
SE 0.090 0.146 0.144 0.102 0.111 0.001 0.002 0.246 0.428 0.438 0.231 0.003 0.523 1.015 0.739 0.735 0.604 0.416 0.029 0.005 1.519
P value
