TY - JOUR
T1 - One instrument to rule them all
T2 - The bias and coverage of just-ID IV
AU - Angrist, Joshua
AU - Kolesár, Michal
N1 - Funding Information:
We thank Ahmet Gulek and Luther Yap for expert research assistance. Thanks also go to Tim Armstrong, Isaiah Andrews, Brigham Frandsen, Guido Imbens, Mike Keane, Dave Lee, Whitney Newey, and Steve Pischke for helpful discussions and insightful comments. Kolesár acknowledges support from a Sloan Research Fellowship and by National Science Foundation, United States Grant SES-22049356 . The views expressed here are our own.
Publisher Copyright:
© 2023 Elsevier B.V.
PY - 2023
Y1 - 2023
N2 - We revisit the finite-sample behavior of single-variable just-identified instrumental variables (just-ID IV) estimators, arguing that in most microeconometric applications, the usual inference strategies are 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.
AB - We revisit the finite-sample behavior of single-variable just-identified instrumental variables (just-ID IV) estimators, arguing that in most microeconometric applications, the usual inference strategies are 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.
KW - Bias
KW - Confidence intervals
KW - Instrumental variables
KW - Weak instruments
UR - http://www.scopus.com/inward/record.url?scp=85147203053&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85147203053&partnerID=8YFLogxK
U2 - 10.1016/j.jeconom.2022.12.012
DO - 10.1016/j.jeconom.2022.12.012
M3 - Article
AN - SCOPUS:85147203053
SN - 0304-4076
JO - Journal of Econometrics
JF - Journal of Econometrics
ER -