Incorporating Demographic Proportions into Crash Count Models by Quasi-Induced Exposure Method

Abstract

Quasi-induced exposure (QIE) is an effective technique for estimating the exposure of a specific driving or vehicle population when real exposure data are not available. Typically crash prediction models are carried out at the site level, that is, segment or intersection. Driving population characteristics are generally not available at this level, however, and thus are omitted from count models. Because of the sparsity of traffic crashes, estimating driving population distributions at the site level using crash data at individual sites is challenging. This study proposes a technique to obtain demographic proportions to incorporate in the count models as an exposure at each site by aggregating similar adjacent sites until significant demographic proportions are obtained. Information on driver gender, age, and vehicle type are obtained by QIE using five years (2010–2014) of crash data; and road inventories are obtained for 1,264 urban four-lane divided highway segments in California. Count models including only site level factors were compared with models including both crash level and site level factors. The latter outperformed the former in relation to mean prediction bias and mean absolute deviation statistics on holdout sample predictions. Results indicate that teen drivers are more crash prone in total and in fatal plus injury severity crashes. For senior drivers, crash risk increases with the increase in severity level. The presence of vehicles other than passenger cars and trucks reduces total and property damage only crash counts. Female drivers are associated with higher total and fatal plus injury crash counts.