Three IV Myths We See in Field Memos

Instrumental variables estimates are local: they recover treatment effects for compliers under stated assumptions. A strong first stage is necessary but not sufficient for a credible exclusion restriction. We often read memos that treat overidentification tests as proof of validity. These tests check consistency across instruments, not whether instruments are exogenous. Pair test results with a narrative that could convince a sceptical colleague. When IV and OLS diverge, report both with transparent discussion of threats. A significant Hausman test is not a personality contest between estimators — it is a prompt to investigate endogeneity pathways.