In brief, wealthier countries have significantly larger governments (relative to GDP) than poorer ones. This result is more appealing from an intuitive as well as from a theoretical viewpoint than that of Alesina and Wacziarg. Developed countries with high GDP per capitaratios are supposed to have a differentiated system of political participation, admi- nistration, and political bargaining They provide a wide range of public goods and therefore display higher public consumption expenditure than similar sized developing countries. Public consumption relative to GDP has steadily been rising since World War II in those countries60, and some theories create a direct link between economic and/or political development and public expenditure growth.61Nevertheless, it is strik - ing at first sight that the OECD dummy is not significant in any of the model specifications. An explanation for this result may lie in the recent enlargements of the OECD, to which countries with average sized pub lic sectors like Mexico or Korea have been admitted, or in the high correlation between the OECD dummy and per capita GDP (see Table A.4). The evidence of a positive relationship between per capita GDP and government size brings up an interesting presumption. Does per capita GDP affect not only public consumption, but also country size? The theoretical rationale behind such a relationship would be the conjecture that split ups are more feasible for regions in wealthy countries due to the fact that highly developed countries are generally more open than poorer ones. Is the significant correlation between government con- sumption and government size influenced by the variable welfare, repre- sented by per capita GDP? The appropriate method to answer this question is to run a partial correlation. It basically tests whether there is a correlation between va- riable A (government size) and variable B (country size) by removing the linear effects of a variable C (per capita GDP), which possibly affects both variables A and B. Variable C is often called «control variable» in the context of partial correlations. Technically, the partial correlation is estimated by regressing A on C and B on C. For the residuals for each of the two regression equations, the Pearson correlation is then com - puted. The result is a correlation of variables A and B, in which the line- 60Does 
country size matter for public sector size? 60See, for instance, Blankart (1998) or Cusack (1997). 61Baumol (1967), Cusack (1997), Timm (1961), Wagner (1892) etc.