The Fifty American States in Space and Time: Applying Conditionally Autoregressive Models

Spatial conditionally autoregressive (CAR) models in a hierarchical Bayesian framework can be informative for understanding state politics, or any similar population of border-defined observations. This article explains how a hierarchical CAR model is specified and estimated and then uses Monte Carlo analyses to show when the CAR model offers efficiency gains. We apply this model to data structures common to state politics: A cross-sectional example replicates Erikson, Wright & McIver's (1993) Statehouse Democracy model, and a multilevel panel model example replicates Margalit's (2013) study of social welfare policy preferences. The CAR model fits better in each case and some inferences differ from models that ignore geographic correlation.