8.2 Why Matching?

  • Pure regression approach is increasingly questioned (e.g. Aronow and Samii 2015; Ho et al. 2007)
  • Matching methods (Stuart 2010, 2)
    • Complementary to regression adjustment
    • Reduce imbalance
    • Highlight areas of covariate distribution without sufficient overlap/common support between treatment/control (extrapolation)33
    • (Treatment effects without heroic parametric assumptions34)
    • Straightforward diagnostics to assess performance
    • Makes you think about selection
  • Q: If I match on a set of variables, do I still need to make the selection-on-observables assumption?

References

Aronow, Peter M, and Cyrus Samii. 2015. “Does Regression Produce Representative Estimates of Causal Effects?” American Journal of Political Science.

Ho, D E, K Imai, G King, and E A Stuart. 2007. “Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference.” Political Analysis: An Annual Publication of the Methodology Section of the American Political Science Association.

Stuart, E A. 2010. “Matching Methods for Causal Inference: A Review and a Look Forward.” Stat. Sci.

  1. Extrapolation: Imagine you don’t have data for the education category “3”. Then you may use observations of the other categories to infer the conditional mean at education = 3. Bascially, that’s what we do, e.g. with a linear model.↩

  2. See slide on parametric statistics at the end.↩