Space and time in macroeconomic panel data: young workers and state-level unemployment revisited

A provocative paper by Shimer (2001) finds that state-level youth shares and unemployment rates are negatively correlated, in contrast to conventional assumptions about demographic effects on labor markets. This paper updates Shimer's regressions and shows that this surprising correlation essentially disappears when the end of the sample period is extended from 1996 to 2005. This shift does not occur because of a change in the underlying economy during the past decade. Rather, the presence of a cross-sectional (that is, spatial) correlation in the state-level data sharply reduces the precision of the earlier estimates, so that the true standard errors are several times larger than those originally reported. Using a longer sample period and some controls for spatial correlation in the regression, point estimates for the youth-share effect on unemployment are positive and close to what a conventional model would imply. Unfortunately, the standard errors remain very large. The difficulty of obtaining precise estimates with these data illustrates a potential pitfall in the use of regional panel data for macroeconomic analysis.