Apart from the general results reported in the text, I have performed a number of robustness checks. Table C.1 displays the results of these tests for the determinants reported in the text to have statistically significant effects that do not depend on variable-specific extreme outliers (as diagnosed by dfbetas, see Appendix B above). The first test deletes ten observations that exert an extreme influence on the performance of the entire model, as diagnosed by both Cook’s D and dffits (Fox 1991, 30).

In the second test, I exclude all Western countries from the estimation sample. Since most (but not all) of these countries democratized before the third wave, one could otherwise fear that their inclusion would bias my results.