Evaluating Long-Range Regional Safety Using Scenario Planning Analysis

This study attempted to evaluate the long-range regional safety of alternative growth scenarios. In doing so, a scenario planning process was undertaken for the midregion of New Mexico. This paper discusses the manifold aspects of data preparation that went into a scenario planning process. The study used land use and travel demand models to parameterize future scenarios and applied these parameters as covariates to model total and severe crashes separately for the future planning horizon with the use of three statistical modeling techniques: a negative binomial model, a Bayesian Poisson-lognormal model without accounting for spatial heterogeneity, and a spatial Bayesian Poisson lognormal model. These models were investigated and compared for their forecasts for three alternative scenarios that were developed during the scenario planning process. The method adopted for generating and assembling future covariates from the land use and travel demand model accounted for their intricate nature and complexity to be geographically appropriate so that they might reliably synthesize future scenarios. The widest band of increase in crashes was observed from nonspatial Bayesian models. Despite their better predictive fit compared with the negative binomial models, they provided similar or worse safety forecasts for the alternative scenarios. The spatial models provided the least variability in their forecasts for all the alternative scenarios and had the best goodness of fit values. It was concluded that the approach of a safety forecast (short term versus long term) can be dictated by the accuracy of the predictability needed and expected from a regional safety model.