Spatial, Temporal, and Spatiotemporal Autoregressive Probit Models of Binary Outcomes: Estimation, Interpretation, and Presentation

摘要: Spatial/Spatiotemporal interdependence - i.e., that the outcomes, actions, or choices of some unit-times depend on those others is substantively and theoretically ubiquitous central in binary outcomes interest across social sciences. However, most empirical applications omit spatial and, at best, treat temporal dependence as nuisance to be “kludged”; indeed, even theoretical substantive discussion usually ignores (inter)dependence. Moreover, few contexts where has been acknowledged emphasized, such social-network policy-diffusion literatures, models either do not fully reflect simultaneity units, they recognize endogeneity lags which are used (appropriately) model interdependence. This paper notes explains severe challenges posed by spatiotemporal binary-outcome then follows recent spatial-econometric advances suggest two simulation-based approaches for surmounting computational intensiveness these models: classical recursive-importance-sampling (RIS) Bayesian Markov-chain Monte-Carlo (MCMC). Serial autocorrelation raises essentially same challenges, so strategies offer effective approach well. We provide comparisons performance alternative estimators probit, including estimation-strategies blind naive about (inter)dependence omitting them but treating exogenous regressors standard probit estimation we show how apply related simulation methods calculate estimated effects hypothetical shocks terms probabilities (with associated confidence/credibility regions) rather than only parameter-estimate latent-variable all prior spatial-probit applications. illustrate with U.S. states’ adoptions legislative term-limits great-power decisions enter World War I.