Labor force, L it, is present in (32) as the denominator for the other variables (except
investment and physicians per capita). The theoretical model deals with the numbers of
activists and bosses as shares of labor force, and the wage (retail divided by labor).
Enrollment and housing also should be measured with respect to the total labor force.
Therefore, the net effect of the labor force is given by:
(33) βNL = β1 + β3+ β4 + вб+ β8 - 1.
This net effect should equal zero if the linearized model (29) is sufficiently accurate and the
proxies for the change in capital-labor ratio are adequate.
Predictions of the promotion contract model with respect to the empirical specification
(32) are summarized in Table 2.
4.2. Estimation results
The model (32) is estimated using the feasible GLS method with three error term
specifications: random effects, cross-sectional heteroskedasticity, and cross-sectional
correlation.31 Specifications with salaries and numbers of bosses imputed using the two
methods described in Section 3.3 yield similar results, although the second method
(regression) produces coefficients with higher levels of significance. Only the results from the
estimations that rely on the second method are reported. Both proxies for the change in
capital-labor ratio produce similar results (reported in Table 3a), with the exception of the
effect of investment itself, which has higher significance if log change in investment rate is
31 First-differencing practically removes cross-sectional variation: between-variance accounts for less
than two percent of the overall variance in ∆ln(Na it). If (32) is estimated using fixed effects, the null
hypothesis (all fixed effects equal zero) cannot be rejected. I also find no evidence of first-order serial
correlation. These two types of specifications are not reported in this paper.
35