Temporal instability and the analysis of highway accident data

Virtually every statistical analysis of highway safety data is predicated on the assumption that the estimated model parameters are temporally stable. That is, the assumption that the effect of the determinants of accident likelihoods and resulting accident-injury severities do not change over time. This paper draws from research previously conducted in fields such as psychology, neuroscience, economics, and cognitive science to build a case for why we would not necessarily expect the effects of explanatory variables to be stable over time. The review of this literature suggests that temporal instability is likely to exist for a number of fundamental behavioral reasons, and this temporal instability is supported by the findings of several recent accident-data analyses. The paper goes on to discuss the implications of this temporal instability for contemporary accident-data modeling methods (unobserved heterogeneity, data driven, traditional, and causal inference methods) and concludes with a discussion of how temporal instability might be addressed and how its likely presence can be accounted for to better interpret accident data-analysis findings. (C) 2017 Elsevier Ltd. All rights reserved.