Improving reliability of road safety estimates based on high correlated accident counts

Calibrating a safety performance function (SPF) with many years ofaccident data creates a temporal correlation that traditional model calibrationprocedures cannot deal with. It is well known that generalizedestimating equations (GEE) models are able to incorporate trends intoaccident data and thus overcome difficulties in accounting for correlation;the usual application of GEEs to safety analysis uses robust (or sandwich)estimates of regression coefficients under the independence hypothesis forthe working correlation matrix. This practice is justified by the robustnessof the GEE procedure against misspecification of the response correlationstructure. Nevertheless, with this method, one has to renounce the entiretyof the advantages of GEE estimates, and—especially when correlationwithin the subject is high—significant losses in efficiency and misleadingconclusions in model interpretation can occur. In such a case, losses inefficiency of the estimates will be transferred to the reliability of the finalsafety estimation, for example, by the empirical Bayes method. On thebasis of these considerations, the main idea of this study is that, in safetymodeling, additional effort to obtain the true data correlation structurewill result in better precision in the estimation of SPF parameters. Anexample to illustrate the methodological aspects of the proposed approachis included.