Актуальные проблемы Европы №2 / 2014 - стр. 28

One interpretation leaps out from Model 1: the observed variation in Rule of Law scores among Mediterranean nations and non-Mediterranean EU nations is mainly attributable to differences in country wealth. That suggests that geographical location and therefore climate has no significant effect on Rule of Law ratings – notwithstanding Montesquieu’s contention. Poorer nations – e.g., EU members like Romania and Bulgaria – tend to score low on Rule of Law, regardless of their geographical location.

In social research, there often is not much opportunity for additional explanation beyond explaining 76 percent of the variance, but that was not true when expanding the analysis this time. Model 2 dropped country size (insignificant in the previous analysis) and kept country wealth while adding two party system variables: whether or not parliamentary parties existed (No Parties) and the strength of the second largest party among the existing parties.

The explained variance jumped to 83 percent, but only the absence of parliamentary parties was significant, not the strength of parliamentary party competition. Compared to countries worldwide, most of the 41 countries in this study had high levels of party competition, producing relatively little variance for explaining differences in RL scores. Including the «no parties» variable, however, brought the two cases of Lebanon and Libya closer to the regression line, improving the fit and the explained variance. Both countries had no public parliamentary parties and also had very low scores on Rule of Law.

Up to now, we have not given Montesquieu his due; we have not specifically included geography as a variable. Model 3 does this by employing a «Mediterranean» variable with values of 1 for the 22 countries bordering the sea and 0 for the 19 EU countries not on the sea12. Model 3 added the binary Mediterranean variable to the previous model containing country wealth and absence of parliamentary parties. Doing so improved the explained variance of RL only slightly, to 85 percent. However, the Mediterranean variable was statistically significant in the expanded equation, indicating that the model was more properly specified.

The regression statistics generated by the three models are summarized in Table 2. It reports standardized (beta) coefficients instead of unstandardized b-coefficients to better reflect the relative impacts of the independent variables across the equations.

Table 2

Effects of wealth, parties, and location on RL Scores for 41 Nations