Around the world, more than 1.2 million people are killed on the roads every year and as many as 50 million others are injured, and over 90% of the deaths occur in the low and middle income countries [1]. Low and middle income countries also have a higher road traffic fatality rate (20.2 deaths per 100,000 population) than high income countries (12.6 deaths per 100,000 population). Moreover, the total number of road traffic deaths and injuries is forecasted to rise by 80% in low-income and middle-income countries between 2000 and 2020 compared to 65% worldwide. Also, the majority of traffic deaths in developing countries are vulnerable road users (pedestrians, cyclists and motorcyclists). Therefore, it is imperative that more effort be devoted to understanding the challenges and reducing the traffic injuries in developing countries, especially among the vulnerable road users.

Vehicle-pedestrian crash is a major road safety problem in many developing countries, including Bangladesh [2]. Dhaka, the capital of Bangladesh, is one of the most densely populated cities in the world and the state of pedestrian safety can be considered as alarming [3]. In one study, it is observed that pedestrians represent up to 72 percent of road traffic fatalities in the Dhaka Metropolitan area [4]. In Dhaka, pedestrians at intersections are more likely to be involved in these fatal crashes since a significant share of the total crashes occurred near the intersections [5-7]. It is observed that intersection crashes account for around 40 percent of total accidents occurring in the Metropolitan City of Dhaka [8]. Therefore, it is important to explore the factors affecting vehicle-pedestrian crashes to provide evidence based recommendations for improving the safety of these vulnerable road users.

The objective of this study is to identify the factors contributing to the number of vehicle-pedestrian crashes at intersections in a developing country. This study will also contribute to advancing knowledge in the field because relatively few studies have been conducted in the developing country context, compared to developed countries [9]. Since the road environment and user behavior are very different, some of the risks factors identified in developed countries may have different effects in a developing country context.

In addition, this study will also explore many factors that have thus far received relatively little attention in the literature, including manual traffic controls and the presence of waste disposal facilities, speed breakers, solar panels, etc. Many of these features are quite unique to developing countries and are not found in most developed countries.

The literature on vehicle-pedestrian crashes is extensive. For example, one study collected pedestrian safety statistics at the global, regional and national levels, and studied the effect of several factors related to driver, roadway, vehicle and demography [10]. Another study examined the monthly pedestrian crashes patterns on urban roads in the American state of Connecticut [11]. Cottrill and Thakuriah [12] found that pedestrian crash frequency in the city of Chicago was related to exposure, crime rates and general population demographics. Aziz et al. [13] and Yasmin et al. [14] found road characteristics, traffic attributes and land use to be statistically significant in determining the severity of pedestrian crashes in New York City.

Several studies have specifically examined vehicle-pedestrian crashes at intersections. For example, Schneider et al. [15] examined the association between roadway characteristics and pedestrian crash risk in Alameda County in California while Pulugurtha and Sambhara [16] examined the influences of demographic, socioeconomic and land use characteristics on the frequency of pedestrian crashes at intersections in the American city of Charlotte. Femandes et al. [17] and Miranda-Moreno et al. [18] examined the frequency of vehicle-pedestrian crashes at intersections in the city of Montreal while Lyon and Persaud [19] examined the pedestrian crashes in the city of Toronto. Kim et al. [20] examined the factors determining pedestrian injury severity in the American state of North Carolina while Lee and Abdel-Aty [21] examined the severity of pedestrian crashes in Florida.

Although many studies have been conducted to understand the factors influencing vehicle-pedestrian crashes at intersections, most of these studies are conducted in developed countries where the roads are better designed, the traffic is better separated and controlled, and traffic rules are generally obeyed. On the contrary, road environment and user behaviors are very different in many cities in the developing countries [22-28].

Like many large cities in the developing world, the roadway and traffic conditions in Dhaka city are significantly different from many cities in the developed world. For example, there is a much higher share of non-motorized traffic (pedal bicycles, rickshaws, pushcarts, animals, etc.) on many roads (Fig. 1a) and restrictions on non-motorized traffic have to be imposed on some of the intersections. Most of the pedestrian footpaths are occupied by vendors and extension of roadside shops (Fig. 1b), illegally parked vehicles, refuse, other roadside furniture and motorcycles. Thus, pedestrians are often forced to use the road instead of the allotted pedestrian pathways, which will increase the chances of conflict with vehicles. Pedestrian crossing facilities (e.g., zebra crossing, underpass and overpass) are inadequate, and when available, rarely used due to inconvenience and poor law enforcement.

Fig. (1). Road and traffic conditions in Dhaka city.

Also, the sight distance at the approach roads are often restricted by sculptures, advertisement billboards, solar panels (Fig. 1c) and waste disposal facilities (Fig. 1d). Moreover, side roads and gas stations are often present close to intersections, and many of them carry heavy traffic, which contribute to significant traffic jams and congestion. Since traffic congestion is a common phenomenon in Dhaka, drivers often have to take more risks due to insufficient clearance time at intersections. Interestingly, the traffic flow in many intersections at Dhaka city, even signalized intersections, is manually controlled by traffic police (Fig. 1e). It should also be noted that many of the drivers are not educated, not properly trained and are often ignorant about the road rules and regulations [29]. Similarly, many pedestrians often ignore safety rules and exhibit risky walking and crossing behaviors (Fig. 1f).

The aim of this study is to explore the factors contributing to the frequency of crashes at intersections in a developing country. Researchers have used various statistical techniques to model accident frequencies (Lord and Mannering [30] for a review of the various techniques). Since the number of crashes is a count variable, the Poisson regression model is an appropriate technique to use [31-37]. However, the Poisson model, assumes that the mean is equal to its variance and this assumption is often violated in practice. Hence, a more general model, namely the negative binomial model, has been employed by many researchers to account for over-dispersion in the data [18, 38-41] while some researchers have used the Sichel model [42]. In this study, the Poisson regression model was used because the dependent variable was a count variable. The Poisson model was selected over the negative binomial model because over-dispersion was found to be statistically insignificant (p=0.500) in the data.

Let 'y' be the number of crashes at an intersection, then the Poisson regression model is given by:

(1)

(2)

where P is the probability of observing y crashes at an intersection

λ is the mean of the distribution

β ₀, β₁,..., β_k are parameters to be estimated

x₁,..., x_k are explanatory variables

The above model was estimated using the maximum likelihood method. It should be noted that several mixed-Poisson and random parameters Poisson models have recently been applied to account for heterogeneity [42-46]. The random parameters model assumed that some or all of the parameters may be random because of unobserved variations or unobserved heterogeneity [43, 44]. However, preliminary analyses found no significant unobserved heterogeneity (no statistically significant scale parameter or variance for all coefficients). Hence, the fixed parameters Poisson regression model was used to estimate the final model.

Vehicle-pedestrian crash data from 2011 to 2013 at 45 intersections in Dhaka city were provided by the Accident Research Institute at the Bangladesh University of Engineering and Technology. These intersections have been termed as the hazardous intersections by traffic police in Dhaka city corporation road network. In this study, an intersection crash included all crashes within 20m of an intersection. In order to collect relevant information on geometric features, roadside environment, traffic operations and neighborhood land use, a thorough in-situ investigation was performed by the research team at every intersection.

The descriptive statistics of the 13 variables used in the final model were summarized and presented in Table 1. Correlations between variables have been tested and no significant relationship was observed between the independent variables. In this study, any variable with a 90% confidence level was considered to be marginally significant and retained in the models.