TY - JOUR
T1 - Valid t-Ratio Inference for IV
AU - Lee, David S.
AU - McCrary, Justin
AU - Moreira, Marcelo J.
AU - Porter, Jack
N1 - Funding Information:
* Lee: Princeton University and NBER (email: davidlee@princeton.edu); McCrary: Columbia University and NBER (email: jmccrary@law.columbia.edu); Moreira: FGV EPGE (email: mjmoreira@fgv.br); Porter: University of Wisconsin (email: jrporter@ssc.wisc.edu). Isaiah Andrews was the coeditor for this article. We are grateful to Charlie Fefferman for his generous spirit and interest in our problem, and to Peter Ozsváth for connecting us with him. We thank Josh Angrist and Jim Stock for their comments and suggestions. We also thank Orley Ashenfelter, Marinho Bertanha, Stéphane Bonhomme, Janet Currie, Michal Kolesár, Alex Mas, José Montiel-Olea, Ulrich Mueller, Zhuan Pei, Mikkel Plagborg-Møller, Chris Sims, Eric Talley, Mark Watson, and participants of the joint Industrial Relations/Oskar Morgenstern Memorial Seminar at Princeton, the applied econometrics workshop at FGV, seminars at UC Davis and UQAM, the California Econometrics Conference, and the World Congress, for feedback on earlier iterations of this project. We are also grateful to Camilla Adams, Victoria Angelova, Cate Brock, Santiago Deambrosi, Colin Dunkley, Jared Grogan, Bailey Palmer, and Myera Rashid, and especially Sarah Frick and Katie Guyot for extraordinary research assistance. This study was financed in part by the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior - Brasil (CAPES) - Finance Code 001. For supplementary material, including updates to the original online Appendix and a STATA package to compute tF critical values/standard error adjustments, please visit https://www.princeton.edu/~davidlee/wp/SupplementarytF.html.
Publisher Copyright:
© 2022 American Economic Association. All rights reserved.
PY - 2022/10
Y1 - 2022/10
N2 - In the single-IV model, researchers commonly rely on t-ratio-based inference, even though the literature has quantified its potentially severe large-sample distortions. Building on Stock and Yogo (2005), we introduce the tF critical value function, leading to a standard error adjustment that is a smooth function of the first-stage F-statistic. For one-quarter of specifications in 61 AER papers, corrected standard errors are at least 49 and 136 percent larger than conventional 2SLS standard errors at the 5 percent and 1 percent significance levels, respectively. tF confidence intervals have shorter expected length than those of Anderson and Rubin (1949), whenever both are bounded.
AB - In the single-IV model, researchers commonly rely on t-ratio-based inference, even though the literature has quantified its potentially severe large-sample distortions. Building on Stock and Yogo (2005), we introduce the tF critical value function, leading to a standard error adjustment that is a smooth function of the first-stage F-statistic. For one-quarter of specifications in 61 AER papers, corrected standard errors are at least 49 and 136 percent larger than conventional 2SLS standard errors at the 5 percent and 1 percent significance levels, respectively. tF confidence intervals have shorter expected length than those of Anderson and Rubin (1949), whenever both are bounded.
UR - http://www.scopus.com/inward/record.url?scp=85139542870&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85139542870&partnerID=8YFLogxK
U2 - 10.1257/aer.20211063
DO - 10.1257/aer.20211063
M3 - Article
AN - SCOPUS:85139542870
SN - 0002-8282
VL - 112
SP - 3260
EP - 3290
JO - American Economic Review
JF - American Economic Review
IS - 10
ER -