Spatial Econometrics and Political Science

Many theories in political science predict the spatial clustering of similar behaviors among neighboring units of observation. This spatial autocorrelation poses implications for both inference and modeling that are distinct from the more familiar serial dependence in time series analysis. In this paper, I examine how political scientists can diagnose and model the spatial dependence that is predicted by our theories. First, global and local measures of spatial autocorrelation are estimated to determine whether the data are spatially autocorrelated. If the data are spatially autocorrelated, the researcher attempts to model this dependence with a standard econometric specification and applies diagnostics to determine whether the covariates model the spatial autocorrelation. If the variables do not fully model the spatial dependence, the researcher estimates the spatial econometric model indicated by the diagnostics. Monte Carlo results and an empirical application to voting during the New Deal realignment highlight the importance of diagnosing and modeling spatial autocorrelation.