Robust Finding or Measurement Artifact? Reconsidering the Relationship between Democracy and International Conflict.

The idea that democracies are less apt to engage in conflict with each other is the central finding of contemporary international relations. Yet at the same time, the operationalization of the concept of democracy in this literature has been relatively unreflective, with little critical consideration about how to measure what we mean by democracy. Since the mid-1990s the vast majority of studies that test for the effect of democracy on conflict have relied upon Polity to measure democracy. Given that important concerns have been raised about Polity previously (Gleditsch and Ward 1997), reliance solely upon one instrument may put such findings on too thin an evidentiary basis. In this paper we raise additional concerns about Polity, most notably that there is a mismatch between conceptualization of democracy as a regime type and using an interval scalar measure like Polity. We argue that complex types are better captured by "object concepts" that are best measured as dichotomous variables. If our contention is correct, we would expect to find that models that use a dichotomous coding provide a better fit for the data than models using Polity, or, even different findings, if a binary measure is better suited to capturing the concept of democracy. In this paper we look at what happens when we substitute a binary coding of democracy (Bernhard, Nordstrom and Reenock 2001) for Polity. We find that the democratic peace is robust using BNR, and that it even improves the fit of the data to the models. Other important findings on democracy and conflict, notably Mansfield and Snyder's (1995a) contention that democratizing countries are more prone to engage in interstate conflicts, are not strongly supported by the substitution of BNR for Polity. Both of these findings give credence to our contention that a binary coding has advantages in capturing the concept of democracy. At minimum, it suggests that findings based only on interval scalar measures like Polity, should be checked for robustness with binary measures that capture democracy as a type. ..PAT.-Unpublished Manuscript [ABSTRACT FROM AUTHOR]