Problem:
Conditioning on covariates is standard practice in political research, but scholars often overlook a critical issue: unobserved variables and their interactions with observed ones.
Key Insight:
This paper demonstrates that when two confounding factors act in opposition (one strengthens the bias while the other reduces it), adding an extra variable can actually increase bias in your results. This seemingly counterintuitive finding occurs even when standard balance tests suggest improvement.
Implications & Findings:
- We show this phenomenon isn't rare, appearing frequently in real political data sets
- Common approaches like balance testing and sensitivity analysis do NOT protect against this potential overadjustment bias
- Researchers need to reconsider their variable inclusion strategies more carefully
This research highlights a crucial nuance about omitted variables that challenges conventional wisdom.
