bicameralism and the constructive vote of no confidence raise the incidence of
coalition governments, as expected, while the investiture vote is not statistically
significant.
Finally, the theory predicts that the electoral system influences the type of
government only through its effect on the party structure. To test this prediction,
we estimate the type of government regressions by two-stage least squares, using
the electoral rules variables as instruments for party fragmentation (or the number
of parties in government). Because we have an additional variable (threshold)
measuring the electoral rule, beyond maj , semi and district, and given that the
electoral rule variables are not highly mutually correlated (and measure different
features of electoral rules) we can test the over-identifying restriction that they are
valid instruments, i.e., that electoral rules do not exert an independent influence
in the second-stage regression for the type of government.
The results are displayed in columns 6-8 of Tables 2a and 2b. Party frag-
mentation has a strong and significant effect on the type of government, with the
expected sign, and the number of parties in the legislature have a significant effect
on the number of parties in government. The estimated coefficients of the party
structure variables on the type of government are precisely estimated, and their
size is similar across the two samples, a sign of robustness. Moreover, we cannot
reject the over-identifying restrictions that the electoral-rule variables have no
direct effect on the type of government once we allow for the effect of party struc-
ture, except in the case of coalition governments in Table 2a, where we marginally
reject at the 5% confidence level — cf. the row labelled Over-id.
This last set of results also give some comfort against the risk of misspecifi-
cation. As mentioned in connection with Table 1, the observed variation in the
type of goverment within electoral rules suggests stochastic shocks to government
formation. Such shocks might be systematically correlated with electoral rules,
through strategic decisions in government and/or party formation. Ideally, these
issues should be analyzed in the context of our model, but unfortunately the me-
chanical government formation stage does not permit such analysis. Neverthelss,
our general inability to reject the over-identifying restrictions reassures us that
neglecting these issues does not bias our inferences.
Overall, these estimates are remarkably consistent with the predictions of the
theory.
Economic effects of electoral rules We now turn to the second set of pre-
dictions, concerning the effects on the size of government spending, cgexp.
39