Multilevel data and bayesian analysis in traffic safety

Background: Traditional crash prediction models, such as generalized linear regression model, are incapable of taking into account multilevel data structure. Therefore they suffer from a common underlying limitation that each observation (e.g. a crash or a vehicle involvement) in the estimation procedure corresponds to an individual situation in which the residuals exhibit independence.

Problem: However, this "independence" assumption may often not hold true since multilevel data structures exist extensively because of the traffic data collection and clustering process. Disregarding the possible within-group correlations may lead to production of models with unreliable parameter estimates and statistical inferences. PROPOSED THEORY: In this paper, a 5 x ST-level hierarchy is proposed to represent the general framework of multilevel data structures in traffic safety, i.e. [Geographic region level-Traffic site level-Traffic crash level-Driver-vehicle unit level-Occupant level] x Spatiotemporal level. The involvement and emphasis for different sub-groups of these levels depend on different research purposes and also rely on the heterogeneity examination on crash data employed. To properly accommodate the potential cross-group heterogeneity and spatiotemporal correlation due to the multilevel data structure, a bayesian hierarchical approach that explicitly specifies multilevel structure and reliably yields parameter estimates is introduced and recommended.

Case studies: Using bayesian hierarchical models, the results from several case studies are highlighted to show the improvements on model fitting and predictive performance over traditional models by appropriately accounting for the multilevel data structure.