A Poisson-lognormal conditional-autoregressive model for multivariate spatial analysis of pedestrian crash counts across neighborhoods

摘要: Abstract This work examines the relationship between 3-year pedestrian crash counts across Census tracts in Austin, Texas, and various land use, network, demographic attributes, such as use balance, residents’ access to commercial uses, sidewalk density, lane-mile densities (by roadway class), population employment type). The model specification allows for region-specific heterogeneity, correlation response types, spatial autocorrelation via a Poisson-based multivariate conditional auto-regressive (CAR) framework is estimated using Bayesian Markov chain Monte Carlo methods. Least-squares regression estimates of walk-miles traveled per zone serve exposure measure. Here, Poisson-lognormal CAR outperforms an aspatial (without cross-severity correlation), both terms fit inference. Positive emerges neighborhoods, expected (due latent heterogeneity or missing variables that trend space, resulting clustering counts). In comparison, positive aspatial, bivariate cross severe (fatal incapacitating) non-severe rates reflects covariates have impacts severity levels but are more local nature (such lighting conditions sight obstructions), along with spatially lagged correlation. Results also suggest greater mixing residences uses associated higher risk different levels, ceteris paribus, presumably since produces potential conflicts vehicle movements. Interestingly, network show variable effects, provision lower severe-crash rates.