Correlation Versus Interchangeability: The Limited Robustness of Empirical Findings on Democracy Using Highly Correlated Data Sets

This article shows that highly correlated measures can producedifferent results. We identify a democratization model fromthe literature and test it in more than 120 countries from 1951to 1992. Then, we check whether the results are robust regardingmeasures of democracy, time periods, and levels of development.The findings show that measures do matter: Whereas some of thefindings are robust, most of them are not. This explains, inpart, why the debates on democracy have continued rather thanbeen resolved. More important, it underscores the need for morecareful use of measures and further testing to increase confidencein the findings. Scholars in comparative politics are drawnincreasingly to large-N statistical analyses, often using datasets collected by others. As in any field, we show how theymust be careful in choosing the most appropriate measures fortheir studies, without assuming that any correlated measurewill do.