The urbanization effect is insensitive to the inclusion of the mean skills level by
school district as an explanatory variable. The positive estimated elasticity for the
covariate implies matching efficiency decreases as the average skill level of principals
rises.8 This effect on skills variation is strong - the comparison of R2 in models I and II
suggests the covariate accounts for over half the explained variation in the dependent
variable. The point estimate and standard error of the urban size effect, however,
changes very little across the two specifications.
Controlling for school district characteristics helps identify the effect urbanization
has on the quality of the final labor matches. The statistically significant effect school
district size has on skills variation contrasts intuitively with the effect urban size is found
to have. While the scale of the local labor market is hypothesized to decrease skills
disparities, school district size, measuring within-district heterogeneity, is hypothesized
to increase skills variation. The results indicate larger school districts employ a more
heterogeneous set of principals, while larger labor markets induce greater homogeneity.
The proportion of students non-white, a more specific indicator of student district
heterogeneity, however, is not found to be a significant determinant.9
Population density, a second standard measure of urbanization, is found
insignificant across models. The proportion of adults with an advanced degree
significantly determines the variation in teaching experience but the direction of its effect
is counterintuitive. Its positive estimated effect suggests labor match efficiency decreases
as the proportion of the urban population with minimum education qualifications
8 The estimated elasticity less than one implies the coefficient of variation in teaching experience would be
inversely related to the mean skills level by school district. This is confirmed in models (not shown)
estimating the coefficient of variation in skills.
9 The proportion of the student body in poverty was also tested in the regression models with similar
insignificant results.
10