# SCOPE OF GOVERNMENT: Evidence 4

Regression analysis

Table 1 gives a selected set of results from cross-country regressions on government size. All equations are estimated with OLS. The dependent variable is total expenditures, either by central government (columns 1-5), or by general government (columns 6 and 7). All regressions include the parsimonious set of control variables described in the previous subsection, called XB. We also exploit a set of extended controls XE, that also adds the (log) of population size and our measure of centralization to XB (column 4). Reading here Finally (in column 5), we include a set of dummy variables C for the OECD, Latin America, Asia and Africa.

The table displays the regression coefficients for dummy variables reflecting the political system; PRES is set to 1 for presidential and 0 for parliamentary regimes, MAJ is set to 1 for majoritarian and 0 for proportional elections, MAJPRES is set to 1 for presidential regime cum majoritarian elections and 0 otherwise, and so on. Numbers within brackets are t-ratios for a test of the null hypothesis that the corresponding regression coefficient is equal to zero. These are estimated using White’s (1980) consistent estimator.24 Finally, we list the number of observations and the adjusted R2 for each regression.

Our results confirm the visual impression from Graph 3. The presidential dummy is significant in all specifications. Furthermore, the implied differences are substantial; the size of government is about 10 percentage points smaller in presidential regimes, a large number given that the average size in the sample is just below 29 per cent. The majoritarian dummy, on the other hand, is smaller in absolute value and less significant, though of the expected sign. The specifications, including a finer classification of the political system, confirm these results.

Table 2 illustrates the results of regressions on public goods. Once more, all equations are estimated with OLS. The dependent variable is the sum of central government expenditures on order and safety, transport and education (columns 1-6), or a wider aggregate also incorporating expenditure on health (column 7). All regressions include the parsimonious set of control variables described in the previous subsection, ZB. We also use an extended set of controls ZE that adds to ZB the (log) of population, and the index of ethno-linguistic fractionalization (columns 5-6), or the set of dummies C for OECD, Latin America, Africa and Asia (column 6).

As in Table 1, we display the regression coefficients for dummy variables indicating the type of political system and the corresponding (White-adjusted) t-ratios. We also include (column 4) MAJORIT, a continuous measure of (inverse) proportionality of the electoral system, taking a minimum of 0 and a maximum of 1.25 The expected sign of the coefficient on this variable is therefore negative.