Graphs, Causality, and Structural Equation Models

Structural equation modeling (SEM) has dominated causal analysis in the social and behavioral sciences since the 1960s. Currently, many SEM practitioners are having diiculty articulating the causal content of SEM and are seeking foundational answers. Recent developments in the areas of graphical models and the logic of causality show potential for alleviating such diiculties and thus for revitalizing structural equations as the primary language of causal modeling. This paper summarizes several of these developments, including the prediction of vanishing partial correlations, model testing, model equivalence, parametric and nonparametric identiiability, control of confounding , and covariate selection. These developments clarify the causal and statistical components of structural equation models and the role of SEM in the empirical sciences .