Throughout, we control for the economic, social, historical and geographic vari-
ables listed in Section 6.1, such as federalism, demographics, and British colonial
origin; the specification is a bit more parsimonious in the 1960-98 sample, because
we have fewer degrees of freedom and because some controls are not available over
this longer time period (see the notes to Table 3 for details). The results are very
robust to alternative specifications of the set of controls.
The reduced-form estimates are displayed in columns 1 and 2 of Table 3, for
the 1990 and the 1960-98 samples respectively. To save on degrees of freedom, we
omit the variable for mixed electoral system (semi) for the 1960-98 sample; we also
omit the constitutional variables, bicam, investiture and constructive. District
magnitude (district) has a very strong positive effect on the size of government,
as expected. Plurality rule (maj) also has the expected (negative) effect, but
the estimated coefficient is not statistically significant. We omit the electoral
threshold variable, as it is never statistically significant.
The remaining columns of Table 3 ask whether the effect of the electoral rule on
government spending operates through the type of government, and only through
that channel. As before, we measure the type of government alternatively as the
incidence of coalition governments (coalition), the incidence of single party ma-
jority government (single), and the number of parties in government (ngov). The
instruments for the type of government are the electoral-rule variables as displayed
in columns 1 and 2 of Table 3 respectively, plus the investiture vote variable when
the type of government is measured by coalition, and the electoral threshold vari-
able when it is measured by the number of parties in government.23 The estimated
effects of the type of government on government spending are strongly significant
with the predicted sign. The coefficients are not very precisely estimated, but the
point estimates are often large enough to keep them comfortably away from zero
(at 95% confidence). Finally, we cannot reject the over-identifying restrictions,
that the electoral-rule variables have no direct influence on government spending,
beyond their indirect influence through the type of government.
In summary, then, the cross-sectional estimates support all the predictions
of the theory. Altogether, they suggest that proportional electoral rules indeed
cause larger government spending than majoritarian electoral rules in the way our
model predicts: they lead to more fragmented party systems and hence to more
frequent coalition governments.
23 Given the caveats above, we prefer not to use the constructive vote variable as an instrument,
although its inclusion makes no difference. Here we also skip the intervening step of the party
structure, regressing the type of government directly on the electoral rule variables.
40