Roadway safety in rural and small urbanized areas

Abstract

Police Accident Reports (PAR) reveal that in a 5-year period between 1993 and 1997, there were 892 crashes at 87 two lane, undivided roadway sites in Strafford County, NH, a county consisting of suburban and rural communities. The purpose of this paper is to describe: (1) logistic regression model building efforts to identify statistically significant factors that predict the probabilities of crashes and injury crashes; and (2) to use these models to perform a risk assessment of the study region. The models are functions of factors that describe a site by its land use activity, roadside design, use of traffic control devices and traffic exposure. Comparative risk assessment results show village sites to be less hazardous than residential and shopping sites. Residential and shopping sites, which are distinctly different from village sites, reside in single-purpose, land-use zones consisting mostly of single-family dwelling units and roadside shopping units with ample off-street parking. Village sites reside in multi-purpose, land-use zones permitting a combination of activities found in residential, shopping and commercial areas. They are pedestrian friendly, that is, have sidewalks and crosswalks, permit onstreet parking, have speed limits and other amenities that promote walking. Adjusted odds ratios and other comparative risk measures are used to explain why one site is more hazardous than another one. For example, the probability of a crash is two times more likely at a site without a sidewalk than at a site with one. The implications on roadway design to improve safety are discussed.

Introduction

We hypothesize that land use activity, roadside design and other site characteristics are important factors in predicting roadway risk, the probability of a crash or injury. In order to test the hypothesis, a data set consisting of 87 sites in Strafford County, NH was compiled. The crash and traffic flow data are linked for each site and logistic regression is used to calibrate and test crash and injury crash models. Model predictions and other risk measures derived from models help explain why one site is more hazardous than another one. The model results suggest the actions to be taken to reduce roadway risk. The effectiveness of these actions are evaluated in terms of information gained from on-site evaluation, research literature and discussions with local and regional planning boards and law enforcement officials.

A two-person team took photographs and made an on-site measurement of paved shoulder width at each site. More importantly, they studied the nature of the land use activity, observed the street life, and evaluated the interaction between motor-vehicles and pedestrians. This qualitative information and the information obtained from the linked crash data set gives a better insight into to the nature of a site rather than just the information contained in the linked data set alone. For example, since drivers were frequently observed waiting for motor vehicles and pedestrians to clear their paths, we initially generalized that sites with high motor-vehicle and pedestrian traffic flow rates would be among the most hazardous in the study area. While traffic exposure proves to be a critical variable for prediction, we discovered that many of these high traffic exposure sites, having a ‘pedestrian friendly’ street atmosphere, were the least hazardous! The quantitative crash and injury crash predictions and the qualitative information obtained from the field complement one another in explaining the results.