used (columns (4-6) in Table 3a). All these estimates are robust against sample selection. In
particular, truncation of the time period on either or both sides or removal of one to four
cross-sections at random does not affect the sign and the order of magnitude of any parameter
estimate. Experimenting with various subperiod dummies, I find that an additional constant
for the years after 1961 improves the results most significantly. Estimates with the After_1961
dummy are reported in Table 3b.
The signs and magnitudes (whenever applicable) of the parameter estimates are
consistent with the promotion contract model. The number of bosses, Nb, has a positive
coefficient as expected. Bureaucratic rent (boss salary), R. also has positive effect, although
its significance is low. Signs and magnitudes of both investment indicators are consistent with
the production technology with low elasticity of substitution between labor and capital. Net
effects of labor force, insignificantly different from zero, suggests that linearized model (29)
is appropriate. The variables that measure public consumption - enrollment in higher
education institutions, ST, new public housing construction, NH, and physicians per capita,
PH, - also have expected signs or are insignificant (NH). These effects provide additional
support for the theoretical result that the wages of the working population (broadly
conceived) should have a destimulating effect on the supply of activists. This result can be
also considered within the framework of the models of political dictatorship discussed in the
Introduction (Grossman and Noh, 1994; Wintrobe, 1998). Their prediction that the power-
maximizing dictatorship should produce public goods that are valued by the population in
order to buy loyalty of the latter is not supported by my results. The approach employed in
this paper produces the results that are in a better agreement with the data. Enrollment, ST,
can be also interpreted, as I argue earlier, as an indicator of the availability of alternative
36