Social Science Seminar - Beyond Causal Inference, toward Causal-Model and Causal-Response Estimation: Empirical Modeling for Positive Theorists

Current methodological research and applied work focuses on the design of empirical tests for proper causal inference in "observational" data, and much progress has been made in those directions. However, beyond testing whether causal effects exist, there are at least 3 other things scientists aim to do with empirical analyses: measurement/description/summary, prediction/forecasting, and most focally here: estimate causal responses, which requires estimating causal models, i.e. useful empirical simplifications of causal process. Effective and appropriate empirical-methodological strategies differ across these different aims. This talk discusses central roles for several empirical models and estimators that are especially useful for Empirical Modelling of Positive Social Science Theory: the maximum likelihood and Bayesian frameworks; nonlinear least-squares and multilevel models; temporal, spatial, and spatiotemporal autoregressive models of (inter)dependence and dynamics; and systems estimation, including dynamic systems estimation (VAR).