7.1 Basics

  • Data: Normally cross-sectional (Q: What?)

  • Assumption(s): “there is some set of [pre-treatment] covariates such that treatment assignment is random conditional on these covariates” (Barnow, Cain, and Goldberger 1980; Keele 2015b, 8)

    • ..also called conditional ignorability, no omitted variables or conditional on observables

  • Pre-treatment: A priori known to be unaffected by treatment assignment (permanent or took on value before)

  • Strategy: Control/condition for covariates using adjustment methods such as regression or matching

  • Bold assumption → ideally combine with other strategy!

  • Q: Why can this assumption not be verified with observed data? (plausibility)

  • Q: Some covariates may implicitly control for unobserved variables? What do I mean by that?

References

Barnow, Burt S, Glen George Cain, and Arthur Stanley Goldberger. 1980. Issues in the Analysis of Selectivity Bias. University of Wisconsin, Inst. for Research on Poverty.

Keele, Luke. 2015b. “The Statistics of Causal Inference: A View from Political Methodology.” Polit. Anal. 23 (3): 313–35.