Full Citation

Title: Shift-Share Designs: Theory and Inference

Citation Type: Miscellaneous

Publication Year: 2018

ISBN:

ISSN:

DOI:

NSFID:

PMCID:

PMID:

Abstract: Since Bartik (1991), it has become popular in empirical studies to estimate regressions in which the variable of interest is a shift-share, such as when a regional labor market outcome is regressed on a weighted average of observed sectoral shocks, using the regional sector shares as weights. In this paper, we discuss inference in these regressions. We show that standard economic models imply that the regression residuals are likely to be correlated across regions with similar sector shares, independently of their geographic location. These correlations are ignored by inference procedures commonly used in these regressions, which can lead to severe undercoverage. In regressions studying the effect of randomly generated placebo sectoral shocks on actual labor market outcomes in U.S. commuting zones, we find that a 5% level significance test based on standard errors clustered at the state level rejects the null of no effect in up to 45% of the placebo interventions. We derive novel confidence intervals that correctly account for the potential correlation in the regression residuals.

Url: https://arxiv.org/pdf/1806.07928.pdf

User Submitted?: No

Authors: Adao, Rodrigo; Kolesar, Michal; Morales, Eduardo

Publisher: Princeton University

Data Collections: IPUMS USA

Topics: Labor Force and Occupational Structure, Other

Countries: