Session 3: Regression and Causality

The CIA makes the CEF causal; when the CIA is credible; from a causal CEF to a linear regression; omitted variables bias; good and bad controls; propensity scores and the efficiency paradox.

When: TBA · Where: TBA

Materials

Overview

  • The conditional independence assumption (CIA): within covariate cells, selection bias vanishes and the CEF is causal
  • When the CIA is credible: lotteries within groups and well-understood institutional screening
  • From a causal CEF to a linear regression: the extra assumptions linearity needs
  • The omitted variables bias formula, and using it to sign bias
  • What controls are for: include what makes treatment as good as random, exclude what treatment causes (bad control)
  • Propensity scores: identification and dimension reduction; matching, weighting, and regression; three efficiency puzzles (Hahn; Angrist and Hahn; Hirano, Imbens, and Ridder)

Readings

  • Angrist and Pischke (2009), Sections 3.2 and 3.3.2.
  • Rosenbaum, P. R., and D. B. Rubin (1983). “The Central Role of the Propensity Score in Observational Studies for Causal Effects.” Biometrika, 70(1), 41–55.
  • Hahn, J. (1998). “On the Role of the Propensity Score in Efficient Semiparametric Estimation of Average Treatment Effects.” Econometrica, 66(2), 315–331.
  • Angrist, J. D., and J. Hahn (2004). “When to Control for Covariates? Panel Asymptotics for Estimates of Treatment Effects.” Review of Economics and Statistics, 86(1), 58–72.
  • Hirano, K., G. W. Imbens, and G. Ridder (2003). “Efficient Estimation of Average Treatment Effects Using the Estimated Propensity Score.” Econometrica, 71(4), 1161–1189.