Skip to main content

Assessing Causal Quantities from Experimental and Nonexperimental Data

In a bloodless uprising against the statistical establishment of the 1940-1960's, economists have developed a language for causation -- structural equation models (SEM). Today, this language serves as a basic modeling tool in econometrics and the social and behavioral sciences, though it is still a mystery to statisticians and a semi-mystery to its practitioners. In this talk, I will show how the SEM language, enriched with a few concepts from logic and graph theory, offers a powerful conceptual and mathematical framework for managing difficult nonlinear problems in causal and counterfactual analysis. I will survey three such problems: (1) the identification of policy effects, (2) the assessment of causal attributions in specific scenarios, and (3) the separation and assessment of direct and indirect effects.

For background information, see Causality (Cambridge University Press, 2000), or, or the following papers:

Paper 1
Paper 2
Paper 3