Discrete valued time series models for examining weather effects in daily accident counts

Abstract

In this paper we aim at examining the effect of weather conditions in daily accident counts. In order to account for the serial correlation and the overdispersion present to the data, we make use of two models for discrete valued time series using covariate information. The models considered are the model of Zeger and the Integer Autoregressive model including covariates. Estimation procedures and possible extensions of the models are discussed. Data from 27 major cities roads in the Netherlands are examined. We make use of a meta-analysis approach in order to combine the effects retrieved for each site with site-specific covariate information.