Pedestrian exposure measures: A time-space framework

Abstract

Modeling pedestrian-vehicle crashes is a spatially complex and temporally dynamic process. Examining the probability of and degree to which pedestrians are exposed to pedestrian-vehicle crash risk has important implications for formulating effective road safety measures. Pedestrian exposure can be a useful explanatory variable for modeling crashes but this piece of information is often difficult and costly to collect. The study attempts to take advantage of time geography and travel activity data to propose a new pedestrian exposure metric. Making use of the concept of potential path tree (PPT), this paper developed an individual-based and network constrained pedestrian exposure measure. Using negative binomial regressions to examine crash frequency with exposure, roadway and environmental variables, the proposed metric is compared with other existing pedestrian exposure methods to examine its applicability and potential in road safety analysis.

Introduction

Pedestrian-vehicle crash is a significant public health concern but it is a preventable cause of death and hospitalization. Worldwide, over 400,000 pedestrians lost their lives in pedestrian-vehicle crashes annually (Naci et al., 2009). In Hong Kong, around 55% of all registered land transport-related deaths between 2001 and 2007 involved pedestrians. The overrepresentation of pedestrians in road traffic crashes is a huge concern that merits further scrutiny. Among all victims, the causality rate was the highest among the elderly population. Both international and local statistics in Hong Kong show that the elderly is a susceptible population at risk of dying in pedestrian road trauma. In Europe, a significant portion of the elderly (30–50%) relied on walking in conducting their daily trips. However, research had found out that almost half of all pedestrian fatalities in Europe involved elderly pedestrians (Hakamies-Blomqvist, 2003, Organisation for Economic Co-operation and Development, 2001). Similarly in Hong Kong, crash statistics reveal that about one in every two of the pedestrian deaths in the city involved an elderly pedestrian aged 65 and above (Centre for Health Protection, 2009). Based on the latest city-wide Travel Characteristics Survey (TCS2002) in 2002, the elderly respondents spent nearly 80% of their travel time on walking. Older people represent a key pedestrian group; yet they are more prone to traffic injury with their reduced physical capacity and strength (Loo and Tsui, 2009). Therefore, effective preventive strategies have to be formulated to improve pedestrian safety. Novel approaches to analyze pedestrian crashes are essential to meet the challenges of an ageing population.

Models that identify and characterize high pedestrian crash risk factors and locations for mitigation measures can improve road safety. Factors like the physical characteristics of the built environment as well as the behavioral factors of drivers and pedestrians have been extensively studied in pedestrian-vehicle collisions. To establish effective road safety countermeasures, it is essential to have information not only on the number and nature of the collisions but also the exposure of road users to crash risk. The term ‘exposure’ originates from the field of epidemiology. It is defined as the rate of contact with a potentially harmful agent or event (Raford and Ragland, 2004). Pedestrian exposure therefore refers to the rate of contact that a pedestrian has with vehicular traffic. It is a crucial component of assessing risk (Chapman, 1973). In principle, pedestrians are exposed to crash risk whenever they are walking in the vicinity of vehicular traffic. The volume of pedestrian activities was found to have a positive relationship with the occurrence of pedestrian-vehicular collisions (Davis and Braaksma, 1988). Pedestrian exposure can be a useful explanatory variable for modeling traffic crashes; however this piece of information is often difficult and costly to collect (Qin and Ivan, 2001).

Pedestrian exposure is difficult to quantify as pedestrian route choices are complex. Pedestrians are more maneuverable than vehicles as they can change their routes more easily based on the surrounding environments. They can pause and turn direction abruptly and are not forced to use a particular pathway like that of vehicles with specified lanes. Numerous metrics can be used to measure pedestrian exposure but none of them have been widely accepted and adopted. Generally, these measures can be grouped into two levels – aggregated and disaggregated. At the aggregate level, place-based and trip-based measures have been widely used to estimate exposure (Wundersitz and Hutchinson, 2008, Greene-Roesel et al., 2007). Examples of place-based methods include the number of population living within a certain predefined areal units like census blocks (Wier et al., 2009, Chakravarthy et al., 2010) and population density computed at census tract level (Cottrill and Thakuriah, 2010). An advantage of using such methods is that one can make efficient use of readily available data sources. But an analysis based on areal distributions can easily lead to erroneous conclusions by obscuring the variability of pedestrian activities within the area. This raises the issue of the Modifiable Areal Unit Problem (MAUP). Apart from the areal size, the analysis depends greatly on the shape and orientation of the areal units chosen (Openshaw, 1983). Such disparities bias the statistical results and substantive interpretations of potentially dangerous crash locations. Some commonly used trip-based measures are also aggregate in nature; and they include distance travelled and time spent walking (Jonah and Engel, 1983). But these measures usually examine one trip type at a time and do not consider trip chaining effects. Moreover, samples of pedestrian volume are often gauged by counting the number of persons passing through designated measurement points during the observation periods (Davis and Braaksma, 1988). Places like intersection crossings are most commonly used as the points for data collection. Nonetheless, mid-block crossings are ignored. Moreover, this method does not differentiate pedestrians by their individual factors like age. Also, it cannot be easily adapted to a wide area due to the high cost involved in large-scale data collection. In an effort to compensate the deficiencies found in these previous measures, activity-based approaches which use travel diary data of individuals’ activity-travel patterns have been proposed (Lam et al., 2012). Travel surveys which collect people’s travel behavior provide invaluable data as better alternatives for estimating pedestrian exposure. It has the strengths to reflect the disaggregate route choices of people. Nonetheless, the method is time consuming and computationally expensive. It is crucial to note that there is no one-size fits all method in measuring pedestrian exposure, what is important is that the methods used are reliable as compared to alternative methods. Hence, a comparative analysis between different methods can provide valuable insights into pedestrian-vehicle crash analysis, and this study provides an opportunity for comparison and discussion.

The modeling of pedestrian movement and their resultant likelihood to be involved in a crash is a spatially complex and temporally dynamic process. Efforts to examine the probability of and degree to which travelers are exposed to risk have important implications for the health and well-being of people from different walks of life. The ability to formulate appropriate measures to fill the gap in collecting pedestrian exposure information is crucial. But little attempt has been made to model the distribution of pedestrians moving through the spatiotemporally changing risk surface. In light of this, the study attempts to take advantage of time geography and the travel activity information to propose an individual-based and network-constrained exposure measure to portray people’s exposure scenarios. In examining the likelihood of future crashes, one is concerned with multiple patterns of movement of the population at risk, which are constantly changing with space and time. This study on pedestrian-vehicle crashes is inspired by the time-geographic principles which emphasize the inseparable spatiotemporal bound of entities (Hägerstrand, 1970). Estimating pedestrian exposure at the disaggregate level involves finding the number and routes of the susceptible population and the possible environment through which the exposed population transverse. Time geography has rarely been applied in pedestrian safety analysis, despite its obvious potential as a tool of uncovering people’s exposure to traffic risk on a network. Previous studies have made use of time geography to integrate experienced or potential exposure to environmental conditions in health research. Examples include exposure to foodscape (Kestens et al., 2010, Rainham et al., 2010) and environmental pollution (Gulliver and Briggs, 2005). It has also been applied in examining individual accessibility in transport studies (Loo and Lam, 2011). Along the same line, pedestrian exposure to crash risk can also be modeled based on the time-geographic notions.

The time geography framework provides an ontological base within which individuals interact with their surrounding spatiotemporally dynamic environment. A number of time geography concepts can be applied in road safety studies. The first one is the space-time path (STP). It traces the walking path that a person took within the given constraints such as time budget to perform different tasks. The sequential movement in space and time is represented as x and y coordinate-locations and extruded at time-point t. The slope of the path at different time locations represents the velocity of the walking speed. Relating to this, the STP can be used as an exposure measure as it represents the activity-travel pattern of people moving through the changing risk field. It mediates the individual-level exposure to crash risk. A previous attempt to compute pedestrian exposure (measured as pedestrian kilometers travelled, PKT) by Lam et al. (2012) represented an experimental design using the STP.

Two other useful time-geographic concepts for road safety analysis include the potential path area (PPA) and the potential path tree (PPT). Fig. 1 shows a hypothetical PPA as an envelope on the two-dimensional plane with a single anchor point in a home-destination-home tour. It is a means to analyze potential movement area which encloses all potentially reachable locations an individual can feasibly reach through the specified constraints, given knowledge of their locations, travel speed and time budget. As it shows the locations that an individual could occupy given the constraints, it represents locations where exposure events might occur (Miller, 2004). Contained within the PPA is a network of all accessible paths that this person can take and these paths are the PPT. It is a useful concept in examining pedestrian crashes since exposure to risk refers to the probability where a person and a car meet in his/her PPT, in this case where the person is walking alongside the roadways. However, the concept of PPT has never been fully applied in road safety analysis. Overall, space-time methods can be viewed as providing more refined measures in capturing pedestrian experiences at the micro-level. These space-time exposures take into consideration pedestrian crossings as well as travelling along the roads. The metrics capture the importance in highlighting the concept of shared facilities of road users in increasing the likelihood of crash occurrence. They are also cost-effective especially in comparison with the pedestrian count volume method which also follows the disaggregate approach.

This paper compares three different pedestrian exposure measures to provide a more robust design in search for more suitable exposure measures for preventive interventions. By so doing, the present study aims to provide in-depth discussion to establish a way forward in this important and essential field of endeavor for improved road safety analysis. Specifically, this paper provides a demonstration and evaluation of the newly-defined PPT exposure concept. By making use of the time-geographic principles, this study makes explicit not only the spatial but also the temporal ordering and magnitude of the exposure as well as its inter-relationships with the changing attributes of the surrounding environment. Since pedestrians can move comparatively freely in their environment, their route choices are often less well-defined. This study attempts to provide a more cost effective and versatile method to collect this missing data. The main research question is: “Can space-time methods reveal useful information about human vulnerability to road crashes?” One is interested to explore whether the new piece of information on pedestrian exposure can be used to answer questions in vulnerability mitigation and eventually be used as a useful measure for road safety research. To evaluate the applicability of the proposed time-geographic based exposure metric, it will be tested and illustrated using travel activity and crash records of elderly pedestrians in a selected district in Hong Kong. The paper is organized with the following structure. In the following section, the study design and methods are presented. Specific discussion on the development of the PPT time-geographic exposure measure is given. Following this, statistical models are built to evaluate the performance of different exposure methods. Their potential advantages and limitations in estimating pedestrian crash occurrence will be compared and discussed. The results of the most optimal model are presented in section “Results and discussions” and their implications and applications are assessed. Several conjectures and recommendations for further studies are put forward in the conclusion section in section “Conclusion and further research”.