Appropriate Regression Model Types for Intersections in SafetyAnalyst

Transportation agencies in the United States have started implementing SafetyAnalyst, highway safety management software. Some agencies have been developing or considering developing safety performance functions (SPFs), a core of the software, using local data to maximize the benefits of the software. With multiple years of data on hand, several regression model types are available for developing the SPFs, yet no reliable guide has been provided in selecting appropriate model types. This study examined 11 model types including 8 panel and 3 cross-sectional models to recommend appropriate model types for intersection SPFs for use with SafetyAnalyst. The models were developed using the data collected from 18,356 intersections in Virginia from 2003 through 2008 and were evaluated using 4 comparison approaches. Appropriate model types for each of the 8 intersection subtypes were found; four all-way stop control subtypes were excluded because of the lack of sufficient observations. In general, a cross-sectional model with summed or averaged crash frequencies was found to underestimate a dispersion parameter that plays a critical role in the empirical Bayes method. Depending on intersection subtype, appropriate model types were found to vary. A panel model with independent correlation and a pooled cross-sectional model were found to be appropriate across all 8 subtypes; if the same model type is desired for the 8 subtypes, these 2 models are recommended for developing SPFs using local data for SafetyAnalyst. Although the conclusions and recommendations were based on sufficient data, they are valid for Virginia and may not be applicable to other states.