Working Paper: NBER ID: w2135
Authors: Daniel S. Hamermesh
Abstract: A large and growing line of research has used longitudinal data to eliminate unobservable individual effects that may bias cross-section parameter estimates. The resulting estimates, though unbiased, are generally quite imprecise. This study shows that the imprecision can arise from the measurement error that commonly exists in the data used to represent the dependent variable in these studies. The example of economists' salaries, which are administrative data free of measurement error, demonstrates that estimates based on changes in longitudinal data can be precise. The results indicate the importance of improving the measurement of the variables to which the increasingly high-powered techniques designed to analyze panel data are applied. The estimates also indicate that the payoff to citations to scholarly work is not an artifact of unmeasured individual effects that could be biasing previous estimates of the determinants of academic salaries.
Keywords: No keywords provided
JEL Codes: No JEL codes provided
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| measurement error in the dependent variable (C20) | imprecise estimates in fixed-effects models (C23) |
| administrative data on academic salaries (free from measurement error) (J30) | more precise estimates (C13) |
| increase in citations (A14) | increase in salary (J31) |
| citations (A14) | salary (J31) |
| longitudinal data (C23) | control for unobserved individual heterogeneity (C21) |
| autocorrelation of measurement errors (C22) | affect estimates (C51) |