comparable over time than the Freedom House data. Specifically, we use the en-
compassing polity index, which assigns to each country and year an integer score
ranging from -10 to +10 (higher values associated with better democracies), re-
stricting the panel to countries and years with positive values of polity (censored
observations are treated as randomly missing). Persson and Tabellini (2003) pro-
vide further details on our sample selection criteria.19 We use this data set in
three ways: First, we average yearly observations throughout the whole period,
1960-98, to draw inference from cross-sectional variation. Second, we use it as a
panel, taking the theoretical model literally and defining the relevant time period
as a whole legislature (i.e., for all variables we take averages over the yearly out-
comes between two consecutive elections). Of course, this does not correspond to
the same calendar time for different countries. Third, we use the data set as a
standard panel with yearly data.
The variables The model predicts that proportional elections should induce
higher government spending. If we ignore the effect on political rents, this is
entirely an indirect effect, however, working through a larger number of parties
and, thereby, a larger incidence of coalition governments. We should thus find no
direct effect of the electoral rule on the size of government, once we account for
the different political outcomes caused by different electoral rules. In addition,
of course, the model has specific predictions about the signs of various political
and economic effects. How do we measure the variables needed to confront these
predictions with the data?
Electoral rules In the model, the most important aspect of the electoral
system (at least in the sections which abstract from heterogeneity) is the electoral
formula: plurality rule versus PR. To capture this, we rely on the binary indi-
cator variable maj, defined and discussed in Persson and Tabellini (2003). The
indicator is coded 1 for countries relying exclusively on plurality rule in the elec-
tions to the lower house, and 0 otherwise. A few parliamentary democracies rely
on a mixed-majoritarian electoral system, with some version of plurality rule in
certain districts and proportional rule in others. We code these mixed systems
with the dummy variable semi, taking a value of 1 if the electoral system is mixed,
19 For a few small countries, the Polity IV data are not available. We thus interpolate Polity
IV with the Freedom House data to make an out of sample prediction. See Persson and Tabellini
(2003).
33