Phantom Results from Panel Regressions
49 Pages Posted: 15 May 2011 Last revised: 28 Nov 2012
Date Written: May 12, 2011
This paper examines current practice with respect to serial correlation in studies that rely on regression analysis of panel datasets. It first examines how frequently various techniques for dealing with serial correlation are used. Next, it runs a horse race using Monte Carlo simulation to establish how well each works in this context. Test sizes based on clustered standard errors are considerably more accurate than those of any of the common alternatives. Finally, it applies the results of the simulation to about one-quarter of the articles from the survey that can be grouped into four research programs. This leads to severe pruning in each line and to a dramatically changed assessment of the overall findings in two of the four.
Keywords: Panel Regressions, Biased Standard Errors, Monte Carlo Simulation, Statistical Inference
JEL Classification: C12, C23
Suggested Citation: Suggested Citation