Working Paper: NBER ID: w29417
Authors: Joshua Angrist; Michal Kolesar
Abstract: We revisit the finite-sample behavior of just-identified instrumental variables (IV) estimators, arguing that in most microeconometric applications, just-identified IV bias is negligible and the usual inference strategies likely reliable. Three widely-cited applications are used to explain why this is so. We then consider pretesting strategies of the form t1 > c, where t1 is the first-stage t-statistic, and the first-stage sign is given. Although pervasive in empirical practice, pretesting on the first-stage F-statistic exacerbates bias and distorts inference. We show, however, that median bias is both minimized and roughly halved by setting c = 0, that is by screening on the sign of the estimated first stage. This bias reduction is a free lunch: conventional confidence interval coverage is unchanged by screening on the estimated first-stage sign. To the extent that IV analysts sign-screen already, these results strengthen the case for a sanguine view of the finite-sample behavior of just-ID IV.
Keywords: Instrumental Variables; Bias; Inference; Microeconometrics
JEL Codes: C21; C26; C31; C36; J08
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| just-identified IV bias is negligible (C26) | conventional inference strategies are likely reliable (C20) |
| screening on the sign of the estimated first-stage can mitigate bias (C24) | median bias of just-identified IV can be minimized (C26) |
| endogeneity does not exceed 0.47 (C20) | bias is not severe (C46) |
| conventional IV estimates and t-tests are not significantly compromised (C36) | finite-sample distribution approaches OLS probability limits (C46) |
| theoretical results strengthen the case for a positive view of finite-sample behavior of just-identified IV (C26) | more optimistic perspective on IV inference (C26) |