On the causal effect of proximity to school on pedestrian safety at signalized intersections: A heterogeneous endogenous econometric model

Highlights

• We developed an endogenous model to estimate the causal effect of proximity to school on safety.

• We found a strong safety-in-numbers effect at intersections in the vicinity of schools.

• Proximity to school decreases pedestrian safety at nearby intersections.

• Unobserved factors that increase likelihood of presence of school nearby intersections decrease pedestrian crash risk propensity.

Abstract

Pedestrian safety in proximity to schools is a major concern of transportation authorities, local governments, and residents. In fact, several countermeasures (e.g., school-zone speed limits) are usually in place around schools to provide a safer environment, especially for school-age children. Two questions arise here: (i) are transportation facilities in proximity to schools truly safer than other facilities given a variety of implemented road safety interventions around schools? and (ii) how can we answer the previous question properly using a reliable approach that accounts for possible confounding? While previous literature has mixed results and does not provide clear methodological/empirical guidelines in this regard, we propose an approach that answers the above questions. We illustrate our method on a sample of intersections in Montreal, Canada. Specifically, to underpin a causal interpretation, for the first time in the extent of transportation literature, we develop a heterogeneous endogenous econometric model that estimates the causal effect of proximity to school on pedestrian safety, addressing a complex endogenous relationship between the two. Various built environment, traffic exposure, and road geometric/operational characteristics are considered. The results indicate that if endogeneity is not accounted for, the effect of proximity to school is underestimated, while not being significant at a 5% level of significance. However, after accounting for confounding factors, the proposed endogenous model indicates that proximity to school deteriorates pedestrian safety. Therefore, traffic safety countermeasures and policies in place (if any) during the study period have not been sufficient and/or effective in improving pedestrian safety at intersections near schools. Our heterogeneity in mean and variance formulation provided more insights. For example, we found that, interestingly, as pedestrian volume increases at intersections around schools, the adverse effect of proximity to school on pedestrian safety decreases, a possibility not previously explored in the extent of road safety literature, confirming a strong safety-in-numbers effect.

Introduction

Pedestrian safety in proximity to schools is a major concern of transportation authorities, local governments, and residents. This could be mainly attributed to a relatively large proportion of child and adolescent pedestrians around schools. In fact, local authorities, by designing and implementing safety improvement programs, often aim to enhance traffic safety in proximity to schools, particularly for pedestrians, which in turn helps to promote commuting to school by walking. An exemplar of such programs is the federal safe route to school initiative in the US, which aimed to improve safety around schools and promote active commute to school (Yu and Zhu, 2016). This initiative was found to be effective in improving pedestrian and cyclist safety as discussed, for example, by DiMaggio and Li, 2013, Ragland et al., 2014.

There is a whole body of literature discussing various issues (urban planning, health and safety, etc.) relating to school siting (Ewing and Greene, 2003, McDonald, 2010, Guliani et al., 2015, Yu and Zhu, 2016, Vitale et al., 2019 Sersli et al., 2019). With respect to traffic safety, a relatively limited number of studies, as part of their larger analyses, have examined the association between pedestrian safety and proximity to school (or the relationship between safety and the number of schools in an area) at both micro and macro levels (Clifton and Kreamer-Fults, 2007, Clifton et al., 2009, Zahabi et al., 2011, Ukkusuri et al., 2012, Xin et al., 2017, Bhat et al., 2017). Previous studies have identified some common attributes associated with school siting that increase or decrease crash risk.

For instance, exposure to traffic increases the likelihood of crashes involving pedestrians, so reducing the interaction between pedestrians and vehicles through higher levels of more connected sidewalks contributes to a reduction in pedestrian crash risks (Yu and Zhu, 2016, Hwang et al., 2017). Schools located within a network of more local roads are safer than those near highways where the volume and speed of traffic is higher (Dumbaugh and Li, 2011, Yu, 2015, Yu and Zhu, 2016). Commercial land use tends to increase crash risks, as not only are there higher levels of traffic (Clifton and Kreamer-Fults, 2007), but entryways tend to be set back, resulting in more potential conflicts (Dumbaugh and Li, 2011). The presence of transit stops locally, which are likely to involve higher pedestrian activity and to be sited along major arterial highways for greater accessibility, also increase crash risk near schools (Miranda-Moreno et al., 2011, Ukkusuri et al., 2012, Yu and Zhu, 2016, Briz-Redón et al., 2019, Heydari et al., 2017).

However, these and other studies have mixed results (Merlin et al., 2020). In fact, there are inconsistencies in previous research regarding whether pedestrian crash and injury risk propensities increase or decrease in proximity to schools or by the number of schools, say, in a geographic area. For example, Zahabi et al. (2011) found that proximity to school decreases the likelihood of fatality or serious injury for pedestrians while Clifton et al. (2009) found the contrary. Similarly, there are inconsistencies in whether a higher proportion of children or adolescents in a specific area (which is often the case around schools) decreases pedestrian risk propensity for injuries. For instance, Bhat et al. (2017) found that a higher proportion of young individuals decreases pedestrian risk propensity for serious injuries. In contrast, a study conducted by Amoh-Gyimah et al. (2016) is not in accordance with Bhat et al.’s finding.

We speculate that such mixed results may be mostly because of the fact that proximity to school is often endogenous, as we describe below. Endogeneity issues are shown to play an important role in delineating inferences in the context of traffic safety analysis (Shankar and Mannering, 1998, Carson and Mannering, 2001, Kim and Washington, 2006, Winston et al., 2006, Eluru and Bhat, 2007, Bhat et al., 2014, Roesel, 2017, Sarwar et al., 2017, Xu et al., 2017, Afghari et al., 2018). For example, Kim and Washington (2006) addressed selection bias in estimating the effect of left-turning lanes on angle crashes. In their study, a simple negative binomial model indicated that the presence of left-turning lanes at intersections has an increasing effect on angle crashes; however, the authors found the contrary after controlling for confounding. Endogeneity issues could easily arise in the context of traffic safety around schools since safety treatments are not assigned randomly, and several interrelated confounding factors may affect crash mechanisms.

Some examples of “conflicting” factors affecting safety or the perception of safety (perceived safety importance) around schools are as follows: (i) speed limits are generally lower in proximity to schools, which is expected to reduce the risk of crash and injury severity sustained by pedestrians; (ii) land-use characteristics such as the area of commercial land-use could be different in proximity to schools increasing traffic exposure and conflicts between pedestrians and vehicles (Heydari et al., 2017, Bhat et al., 2017); (iii) school zones are supposed to be safer, for instance, due to an increased driver awareness because of the presence of traffic signs and school-age children, but other risk factors would be in place as well.

For example, school zones could be more dangerous due to a high proportion of children and teenagers who, due to their natural propensity for inattentiveness, could be more likely to be involved in risk-taking behaviors (e.g., unsafe crossing, disobeying signs, etc.) compared to adults. In fact, due to their limited cognitive abilities, very young individuals are more likely to be involved in traffic conflicts (Stoker et al., 2015, Amoh-Gyimah et al., 2016, Bhat et al., 2017). In this regard, a study conducted by Rothman et al., considering child pedestrian safety around one hundred elementary schools in Toronto, indicated that dangerous pedestrian behaviors and/or drop-off were pervasive around schools (Rothman et al., 2016); and finally (iv) perception of safety and risk compensation (offset hypothesis) could also affect safety mechanisms around schools. For instance, pedestrians could trade off enhanced safety in proximity to schools due to lower speed limits for less attentive behaviors resulting in an increased risk. A detailed discussion related to risk compensation is provided in Winston et al. (2006), who estimated the effect of airbags and antilock brakes on safety. Another exemplar of risk compensation can be found in Lv et al. (2015).

To summarize, safety mechanisms at intersections located in proximity to schools may be different from those away from schools in three major aspects: type of road users, their behavior, and safety features at nearby transportation facilities. Different safety improvement programs may be in place around schools so that transportation facilities (e.g., intersections or road segments) in proximity to schools are often subject to certain safety treatments. Since data are usually scarce in terms of details relating to type of interventions and their effects, one could focus on proximity to school to estimate the joint effect of various countermeasures that might be in place around schools in a jurisdiction. This could be justified due to the fact that safety interventions and policies, regardless of treatment types and their individual effectiveness, are collectively expected to enhance the safety of transportation facilities around schools.

Despite the general consensus about the importance of traffic safety around schools, studies that examine the causal effect of proximity to school on safety are surprisingly rare. While previous literature does not provide clear methodological or empirical guidelines, this paper introduces an econometric framework that quantifies the causal effect of proximity to school on pedestrian safety. This in turn allows the analyst to, indirectly, assess whether implemented safety policies and countermeasures around schools, if any, are truly effective in enhancing traffic safety.

Based on the above discussion, the goal of this paper is to estimate the causal effect of proximity to school on pedestrian safety at signalized intersections. To this end, we answer the following questions:

• Are intersections in proximity to schools truly safer than those away from schools given a variety of implemented safety improvement programs and policies?

• How can we answer the previous question properly using a reliable approach that accounts for possible confounding?

We therefore develop a framework to answer the above questions, focusing on intersection safety. However, our approach can be readily applied to examine safety of other transportation facilities around schools or where self-selection bias exists. Our model is an extension of the approach proposed initially by Terza (1998) in econometrics literature, and its later treatment by Kozumi (2002) from a Bayesian perspective. Specifically, a heterogeneous (random parameter) Bayesian endogenous econometric model is developed to identify safety correlates of pedestrian safety in urban areas using a host of variables, including built environment characteristics and traffic exposure measures. In accordance with real-world scenarios, the proposed approach allows the causal effect of proximity to school to vary across intersections (capturing its heterogeneous effect on pedestrian safety) by employing a heterogeneous endogenous model. We then extend to a heterogeneous endogenous model with heterogeneity in mean and variance. This helps us to provide further interesting insights with the same set of variables available in the data.

Due to computational complexities involved in the presence of endogeneity, most endogenous modeling applications, especially in traffic safety research, employed an instrumental variables approach (see Washington et al. (2011) for a discussion on this approach). However, a system-equation method is more valuable since it explicitly accounts for the contemporaneous correlation between equations (and error terms) (Washington et al., 2011). In this regard, a limited number of traffic safety studies, providing valuable methodological and empirical insights, have employed system-equation methods in the extent of the crash literature mostly during the past five to ten years (Eluru and Bhat, 2007; Eluru et al., 2010; Paleti et al., 2010; Abay et al., 2013; Bhat et al., 2014; Lavieri et al., 2016). The estimation of system-equations is often computationally expensive and requires simulation-based approaches. An important study conducted by Bhat et al. (2014), however, estimates model parameters analytically. Of course, analytical approaches are often superior to simulation-based techniques. In this paper, similar to the latter examples, we directly estimate a system-equation model, explicitly accommodating the correlation between error terms. Note that in contrast to the propensity score method, which is also employed in traffic safety literature to address issues relating to confounding and selection bias (e.g., Li et al., 2013), the system-equation approach does not require a control group, avoiding sensitivity of inferences to the choice of a control group, thereby reducing data requirements.

While most instances of traffic safety studies that deal with endogeneity issues use classical statistics, we adopt a Bayesian approach – as a viable alternative – in estimating our endogenous specification. Alternative densities such as lognormal can be assumed for varying regression coefficients and can be easily estimated in our setting. More importantly, we employ a heterogeneity in mean and variance approach the use of which is rare if non-existent in investigating the (varying) effect of treatments, especially while considering a system-equation approach. Our model formulation therefore allows us to explain the varying effect of proximity to school in our data, revealing risk factors causing heterogeneity in the effect of proximity to school on pedestrian safety. A limited number of previous traffic safety studies have implemented and discussed different versions of such models, highlighting their advantages (Venkataraman et al., 2014, Seraneeprakarn et al., 2017, Heydari et al., 2018). These models are particularly appealing in assessing treatments and in the presence of endogeneity as they provide further empirical insights using the same set of variables available in the data. For example, we would be able to identify factors that minimize or maximize the beneficial effect of a safety intervention. This is while previous traffic safety research rarely allows the effect of treatments to vary across the sample. This is the first instance of such an endogenous model in the extent of road safety literature.

This is also the first study that examines safety-in-numbers (Elvik and Bjørnskau, 2017, Murphy et al., 2017 Elvik and Goel, 2019) within a causal modeling framework, explicitly exploring how pedestrian safety in proximity to school varies according to pedestrian volume. Such an investigation, while being different from examining safety-in-numbers across the entire sample, has not been explored in the extent of road safety research previously and is of major empirical importance. That is, in this research, we investigate safety-in-numbers at two different levels. Further empirical contribution relates to the fact that, while many studies use proxy exposure measures (e.g., population density) for non-motorized and/or motorized flows in modeling pedestrian safety, we use direct pedestrian counts and disaggregated motorized (right-turning, non-turning, and left-turning) volume, providing more accurate and detailed information for safety policy design and analysis.