7.4 Covariates: Confounding/overcontrol bias
Common-cause confounding bias
- Q: Should we control for race (U.S. example). Why or why not?
- idea: “use whatever information you have about why treatment and control [… ] [units] are not readily comparable and adjust for these differences statistically […] works well to the extent that you can figure out and measure the confounders and how they are related to treatment […] not good if you don’t know what the confounders are” (Egap)
- idea: “use whatever information you have about why treatment and control [… ] [units] are not readily comparable and adjust for these differences statistically […] works well to the extent that you can figure out and measure the confounders and how they are related to treatment […] not good if you don’t know what the confounders are” (Egap)
Overcontrol/post-treatment bias
- Q: Should we control for voting intentions measured five minutes before voting/vote choice?
- Q: Should we control for voting intentions measured five minutes before voting/vote choice?
See Keele (2015b), Elwert and Winship (2014a) and Montgomery and Nyhan (2016)
References
Elwert, Felix, and Christopher Winship. 2014a. “Endogenous Selection Bias: The Problem of Conditioning on a Collider Variable.” Annual Review of Sociology 40 (1): 31–53.
Keele, Luke. 2015b. “The Statistics of Causal Inference: A View from Political Methodology.” Polit. Anal. 23 (3): 313–35.
King, Gary. 2010b. “A Hard Unsolved Problem? Post-Treatment Bias in Big Social Science Questions.” In Hard Problems in Social Science” Symposium, Harvard University.
Montgomery, Jacob M, and Brendan Nyhan. 2016. “How Conditioning on Post-Treatment Variables Can Ruin Your Experiment and What to Do About It.”