Chapter 24 Causal Inference

A principal aim of many sciences is to model causal systems well enough to provide insight into their structures and mechanisms and to provide reliable predictions about the effects of policy interventions. To succeed in either of these aims, in general, one must specify a model at least approximately correctly. In the social and behavioral sciences, causal models are often described within any of a variety of statistical formalisms: categorical data models, logistic regression models, linear regression models, factor analysis models, principal components models, structural equation models, and so on. In practice, these models are obtained by a variety of methods: experimentation, investigator’s convictions, uncontested background knowledge, automated search procedures such as stepwise regression and factor analysis, and any of a wide range of ad hoc model selection procedures. To understand the assumptions built into these and other classes of models, as well as their limitations, one