Feasibility study of hazard-specific and vehicle-specific crash modification factors for safety performance functions: application to rural highways

Author(s): RILLAGODAGE, Mehran, Regehr, Grande

Slidedeck Presentation:

Slide deck link



Safety performance functions (SPF) attempt to model average crash frequency (ACF): the number of crashes per year, as a function of crash exposure factors such as annual average daily traffic (AADT) and roadway geometric characteristics. Although exposure to transportation hazards such as wildlife and adverse road-weather is well-documented to trigger crashes, the current SPFs do not consider the impact of such hazards on ACF; thus, limiting the applicability of SPFs to a certain extent. Moreover, crash vulnerability is often perceived to be vehicle type-specific (e.g., passenger cars vs. heavy vehicles). Thus, we recognize the need to introduce additional crash exposure factors: transportation hazards and vehicle type, to SPFs. We argue that hazard- and vehicle-specific crash modification factors (CMF)-i.e., multiplicative factors applied to SPF respectively reflecting the impact of transportation hazards and vehicle type, could effectively estimate the effect of such additional crash exposure factors on ACF. Accordingly, we intend to evaluate the feasibility of introducing relevant CMFs to SPF.


The goal of this study is to enhance crash prediction in rural highways. The following objectives are set to achieve this goal: (i) examining the impact of different hazards and vehicle types on ACF at different severity levels in rural highways, and (ii) evaluating the suitability of different distribution functions to explain crash data in rural highways under different conditions.


Crash data reported on Saskatchewan rural highways are used in this study. First, the crash data were grouped to produce three-way contingency tables describing ACF under three crash concomitants: (i) transportation hazards (e.g., wildlife, adverse road-weather), (ii) vehicle type (light and heavy), and (iii) crash severity levels (e.g., fatal, injury), for two types of rural highways: two-lane and multilane. Chi-squared test was used to evaluate the (i) statistical significance of the relationships inferred by the contingency tables, and (ii) assess the suitability of Poisson and negative binomial distributions to explain crash data under the crash concomitants for both highway types considered.


Relationships between ACF and most of the studied crash concomitants were found to be statistically significant for both two-lane and multilane highways. Adverse road-weather and wildlife were identified as significant hazards jeopardizing rural road safety. Negative binomial distribution (basis for SPF) satisfactorily explained crash data under most of the crash concomitant combinations studied for both two-lane and multilane highways.


The study results (i) highlight the importance of incorporating hazard and vehicle type in SPF, and (ii) imply the feasibility of incorporating hazard and vehicle type in SPF. Yet, this study only focuses on uniform rural highway segments with similar geometric characteristics. Therefore, further research intends to (i) incorporate geometric conditions as a crash concomitant, and (ii) expand the descriptive analysis presented in this study to a predictive analysis to propose hazard- and vehicle-specific CMF associated with SPF for rural highways.


This study, by conducting a preliminary crash analysis, recommends incorporating transportation hazards and vehicle types in SPF for rural highways. The study results help transport authorities to understand the combined impact of different crash concomitants on AFC, and by extension, to select appropriate safety countermeasures.

link to full paper