multiple regressions are astonishingly similar to the one in the univari ate regression. Contrary to the theoretical predictions, Table 3.2 shows a steady rise over time of the explanatory power of the multiple regres sion model 
andthe magnitude of the size effect. It is important to classify the results properly. By comparing model (1) and (5) one can easily see that R2rises from 0.119 to 0.305, which means that simple geographic location can explain by far more variation than size alone. The size effect is clearly robust, but one should not run the risk of over-interpreting its consequences.55 As mentioned above, we do not intend to measure the relationship of openness and country size in this study. There is, however, an impor- tant theoretical expectation which relates openness to public sector size. One might presume that more open countries are more prone to exter- nal economic shocks (there is indeed a very high correlation between a vulnerability index from Briguglio (1995) and openness, which is how - ever partly due to the definition of that index) and, hence, they should have larger public sectors in order to be able to cope with those external shocks. We checked for such a rationale and did not find any of the mo- dels in Table A.6, where an openness proxy (we applied several) came out significantly. Of course, government consumption is not the appro- priate dependent variable in connection with openness (public expendi- ture would be more suitable), but given that the coefficients for open ness are always far from being significant, we are very convinced that our result would also be valid for other aggregates of public sector size. Note further that external shocks should alter public expenditure only tem- porarily, although there might be some inertia or lock-in effects. One would hence have to test for higher variances in yearly public expendi- ture in very small countries.56 56Does 
country size matter for public sector size? 55Calculating confidence intervals, we obtain –3.668 ≤ β2≤ –1.396 for the univariate regression. Taking a very small country with 250,000 inhabitants would yield a range of government consumption in % of GDP of 13.62 and 25.88 within the in- terval. For a middle-sized country with 10 million inhabitants we arrive at 7.74 and 23.65, and for a large country with 250 million inhabitants we have 2.61 and 21.69. 56We refrain from going into detail with regard to this question, because it would require other variables and a different methodological approach, and we doubt whether it is possible to single out the size effect in such a setting.