Quantile Effects of Causal Factors on Crash Distributions

Abstract

Crash data are heterogeneous because they are collected from different sources and locations at different times. This data heterogeneity may cause a significant bias in the estimation of standard errors for the coefficients as well as the coefficients' statistical inferences. In the past decade, several promising modeling strategies have been proposed to handle overdispersed crash data, most of which have focused on estimating the conditional mean crash count. This paper applies an alternative crash modeling approach: quantile regression (QR) in the context of a count data model. The application of QR to model crash frequency is illustrated, and empirical results are interpreted. Poisson gamma, the benchmark statistical model for crash counts, is referenced to estimate the covariate coefficients for the mean crash count. Focusing on the mean may result in important aspects of the data being missed. A more detailed analysis, using a QR model for crash count data, confirms that crash predictors have varying impacts on the different areas of the crash distribution. Moreover, the marginal effects of covariates provide a more direct observation of changes in the quantity, rather than the percentage, of crash frequency when responding to one-unit changes in regressors.