Finally, when time periods are defined as the whole legislature in our panel
data, we also measure the length of the legislature, leg_length, in years, to control
for different durations of legislatures.
Economic and social variables To measure the overall size of government,
we rely on central government spending as a percent of GDP, cgexp, defined and
discussed in Persson and Tabellini (2003). This measure is based on data from the
IMF. Since government spending is affected by many other determinants, we also
control for several variables that reflect the economical, political, geographical,
and historical characteristics of the countries in the sample. These variables are
discussed and defined more extensively in Persson and Tabellini (2003). Some
of them are only used in the cross-sectional estimates, some only in the panel
estimates, some in both.
The following variables refer to economic and social determinants of fiscal pol-
icy or of political outcomes: openness to international trade, measured as exports
plus imports over GDP (trade), population size measured in logs (lpop), the per-
centage of the population above 65 years of age (prop65), the log of real per capita
income (lyp), the output gap (ygap), measured as the log-deviation of output from
the country specific trend, a measure of ethno-linguistic fractionalization (avelf).
These variables have been shown to correlate with measures of fiscal policy in pre-
vious studies, such as Cameron (1978), Rodrik (1998), and Persson and Tabellini
(2003).
To measure the influence of colonial history, and since many majoritarian
countries are also former British colonies, we typically control for British colonial
origin. Because the influence of colonial heritage is likely to fade with time,
we weigh colonial origin by the time since independence, giving more weight to
colonial history in young independent states and no weight at all to colonial rule
more than 250 years ago. The colonial history variable is called col_uka.
Finally, since spending refers to central government, we also use an indicator
variable for federal political structures (f ederal).
The results reported below are very robust to alternative specifications of these
control variables. To save on degrees of freedom, we generally include these con-
trols only when they are statistically significant, or when we have strong priors
that they really belong to the specification.
Preliminary inspection of the data Table 1 displays means and standard
deviations of the main variables of interest, in the three types of electoral sys-
36