Do "Observables" Really Produce Better Data? Problems with the Pacl Data Set for the Analysis of Regime Survival

Abstract:     

This paper revisits the coding rules Przeworski, Alvarez, Cheibub, and Limongi (PACL) use to categorize political regimes as democracies or dictatorships. Taking for granted their definition of democracy and dichotomous approach to its measurement, I show that their “consolidation” and “alternation” coding rules produce selection effects in the identification of regime type. Although the authors acknowledge this fact, they seem to have underestimated its potential impact on the utility of the resulting data for analysis of regime survival. By systematically coding as dictatorships possibly democratic regimes that fail in a certain way, or that repeatedly produce specific electoral outcomes, the authors engage in a form of direct selection that compromises both the external and internal reliability of statistical estimates produced from the resulting samples. In light of this selection effect, I argue that any reduction in misclassification error achieved through PACL’s emphasis on “observables” produces a pyrrhic victory at best. More generally, I argue that statistical research on regime survival must inevitably confront the task of assessing the competitiveness of specific election.

Number of Pages in PDF File: 22

Keywords: democracy, dictatorship, democratization, regime type, measurement