Inference for Losers
AbstractResearchers frequently report league tables ranking units (neighborhoods or firms, for instance) based on estimated coefficients. Since the rankings are formed based on estimates, however, the coefficients reported in league tables suffer from selection bias, with estimates for highly ranked units biased upward and those for low-ranked units biased downward. Further, conventional confidence intervals can undercover. This paper introduces corrected estimators and confidence intervals that address these biases, ensuring that estimates and confidence intervals reported for each position in a league table are median unbiased and have correct coverage, respectively.
CitationAndrews, Isaiah, Dillon Bowen, Toru Kitagawa, and Adam McCloskey. 2022. "Inference for Losers." AEA Papers and Proceedings, 112: 635-42. DOI: 10.1257/pandp.20221065
- C13 Estimation: General