Here, the results are more fragile, and do not always confirm the visual impression from Graph 4. The majority dummy, as well as MAJORIT, has a negative estimated coefficient in all specifications but one. The estimate is generally around -1.5, suggesting that majoritarian elections are associated with a supply of public goods which is about 1.5 percentage points lower. This, again, is not a negligible difference, considering that the average in the sample is just above 8 per cent.
But the estimated coefficient is statistically significant only when the majoritar-ian dummy is interacted with the presidential dummy, or when the continental dummies are included. Inspection of the residuals reveals the existence of a large outlier country, Botswana, a majoritarian parliamentary system with large spending on public goods. When this country is excluded from the sample (or when the dummy for Africa is included), the estimated coefficients and the t -ratios on the MAJ dummy and on MAJORIT become more negative, as predicted by the theory, and are statistically significant at the 5% level.
The results are also sensitive to the specification, however. As illustrated in column 5, when the set of controls is expanded to ZE, the political dummy changes signs and does not significantly differ from zero. This feature of the evidence does not depend on outlier countries, and the results do not improve significantly when Botswana is excluded. Similar results hold for the broader measure of public goods, as illustrated in the last column. Finally, the estimated coefficients for the presidential dummies generally have the sign predicted by the theory. But these are only statistically significant when interacted with the electoral system. Overall, the evidence from the regressions on public good provision is not inconsistent with the theory but seems too fragile to draw any reliable inference. The predictions from our models regarding public goods should thus be investigated further, perhaps with better measures of public good provision.
Our analysis raises many questions. The empirical results are still preliminary. Their robustness should be checked more carefully with regard to sample selection, omitted variables, measurement error, and other statistical problems. Moreover, additional implications could also be tested, for instance exploiting available data on government corruption (see in particular the recent work by Mauro (1998)). We believe, however, that some of the empirical results of this paper are likely to hold up. In particular, we have confidence in the finding that presidential regimes are associated with smaller governments.
The two simple models we have studied may reflect the current state of the art, but they fail to capture important aspects of political interactions. For example, we treated the pre-election policy announcements and the post-election policy choices as entirely separate phenomena. It is not obvious how they should be linked. payday loans consolidation