Abstract
We show theoretically and empirically that measurement error can bias in favor of falsely rejecting a true null hypothesis (i.e., a “false positive”) and that regression models with high-dimensional fixed effects can exacerbate measurement error bias and increase the likelihood of false positives. We replicate inferences from prior work in a setting where we can directly observe the amount of measurement error and show that the combination of measurement error and fixed effects materially inflates coefficients and distorts inferences. We provide researchers with a simple diagnostic tool to assess the possibility that the combination of measurement error and fixed effects might give rise to a false positive, and encourage researchers to triangulate inferences across multiple empirical proxies and multiple fixed effect structures.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 959-995 |
| Number of pages | 37 |
| Journal | Review of Accounting Studies |
| Volume | 29 |
| Issue number | 2 |
| DOIs | |
| State | Published - Jun 2024 |
Funding
We thank Beth Blankespoor, Ted Christensen, Ed deHaan, Patricia Dechow (editor), Ian Gow, Wayne Landsman, Alexander Ljungqvist, Nathan Marshall, Brian Miller, Robbie Moon, Robert Resutek, Cathy Schrand, Sarah Zechman, Frank Zhou, Christina Zhu, two anonymous reviewers, and seminar participants at the JFR mini-method conference, the City University of Hong Kong, University of Georgia, University of Iowa, University of Maryland, Michigan State University, New York University, University of Washington, and The Wharton School for helpful comments. We thank our schools for financial support. Headquarters data used in the paper are available on Josh Lee\u2019s website.
Keywords
- Accounting research
- C18
- Causal models
- Fixed effects
- G17
- Measurement error
ASJC Scopus subject areas
- Accounting
- General Business, Management and Accounting