Case Studies and Theory Development in the Social Sciences - Alexander L. George [26]
On the first two tasks of refining the research question and testing for possible spuriousness, statistical methods greatly advanced the research program and achieved a growing consensus among researchers.93 The conflict behavior of democracies and other regime types gained attention as a research program worthy of intensive study, even among skeptics of a democratic peace. Also, a consensus emerged that democracies are not markedly more peaceful in general, although some studies continue to challenge and qualify this conclusion.94 The consensus view is also that democracies have fought wars substantially less frequently against one another than they have against other types of states, although opinions differ on the number and seriousness of exceptions to this generalization. 95 A weaker consensus emerged around the idea that democracies are less likely to engage in militarized disputes with one another short of war.96
Statistical methods proved less successful at explaining why an interdemocratic peace might exist. Researchers using statistical methods had theorized and rigorously defined several potential causal mechanisms that might explain the democratic peace, focusing on democratic institutions and democratic norms. However, the posited causal mechanisms were often contradictory, and no consensus existed on which of these variables caused an interdemocratic peace. Statistical methods proved inadequate to test these mechanisms for three reasons. First, they faced daunting measurement problems.97 One of the most methodologically sophisticated efforts to test for the normative versus institutional causes of the democratic peace, by Bruce Russett and Zeev Maoz, illustrates these problems. Maoz and Russett use well-established and straightforward measures to control for wealth, economic growth, and contiguity. They also employ careful measures of more complex variables such as alliance membership and ratios of military capabilities. Their measurement of democratic institutions is more complex, though there is at least some consensus on this issue, as many quantitative studies have joined Maoz and Russett in relying on the “Polity II” data set, or modified versions of this data set.98 The most difficult measurement problem, however, is that there is no easy way to quantify the slippery variable of “democratic norms” and no widely accepted database for this variable. Consequently, Maoz and Russett used the longevity of political regimes as a proxy for the prevalence of their norms, and they used the average number of recent deaths from domestic political violence or executions within a dyad as a measure of the “democraticness” of that dyad’s norms.99 Clearly, these proxy measures are problematic, as authoritarian and totalitarian states that persist for decades may minimize the use of domestic violence by monopolizing the instruments of force and creating powerful police and intelligence institutions that deter domestic violence and political opposition.
To some extent, even measurement problems on complex variables like democratic norms can be addressed, and statistical researchers have proven adept at devising creative ways of measuring complex variables. One study by Bear Braumoeller, for example, has developed a dedicated definition and data set for looking at democratic norms as they relate to the democratic peace. This study even measures the differences between the norms of elites and those of mass publics.100 This is a very laborintensive task and it is all but impossible to implement for states for which extensive and dedicated polling data is not available. More generally, data sets that quantify or dichotomize variables can achieve reproducible results across many cases (external validity), but only at the cost of losing some of the ability to devise measures that faithfully represent the variables that they