The causal inference revolution has been one of the most important developments of the social sciences and public policy in the last fifty years. However, the basic causal inference toolkit is often ill equipped to deal with real world observational data within the social sciences and public policy because of implicit clustering or heterogeneity. Modern causal inference tools can be readily extended to more complicated structures for things like heterogeneous treatment effects, but the intuition is more complex when building reasonable identification strategies. This workshop will cover the basics of the modern causal inference toolkit but explicitly from the standpoint of heterogeneous effects and complex structure (e.g. in panel, multilevel, and mixture populations) and the generalizability of inferences to new populations. We will focus on the development of identification strategies through the potential outcomes framework and DAGs for both simple models and models with complex structure. We will discuss matching, inverse probability weighting, selection effects, difference in difference models, synthetic control functions, and instrumental variables designs. This workshop combines work from machine learning, mixed effects modeling, fixed effects modeling, and causal inference to help participants build better designs.

Presented as part of the ICPSR Summer Program in Quantitative Methods of Social Research. Instructor: John Poe, University of Kentucky. To register and for further information, visit this page or contact Eva Seto, Associate Director at Social Science Matrix, at evaseto@berkeley.edu.