The Better Angels of Our Nature: Why Vio - By Steven Pinker Page 0,172

on their exercise rate. Finally, you see whether the residuals left over from the smoking-exercise relationship (the degree to which men smoke more or less than you’d predict from their exercise rate) correlate with the residuals left over from the exercise–heart attack relationship (the degree to which men have more or fewer heart attacks than you’d predict from their exercise rate). If the residuals correlate with the residuals, you can conclude that smoking correlates with heart attacks, above and beyond their joint correlation with exercise. And if you measured smoking at an earlier point in the men’s lives, and heart attacks at a later point (to rule out the possibility that heart attacks make men smoke, rather than vice versa), you can inch toward the claim that smoking causes heart attacks. Multiple regression allows you to do this not just with two tangled predictors, but with any number of them.

A general problem with multiple regression is that the more predictors you want to untangle, the more data you need, because more and more of the variation in the data gets “used up” as each nuisance variable sucks up as much of the variation as it can and the hypothesis you’re interested in has to make do with the rest. And fortunately for humanity, but unfortunately for social scientists, interstate wars don’t break out all that often. The Correlates of War Project counts only 79 full-fledged interstate wars (killing at least a thousand people a year) between 1823 and 1997, and only 49 since 1900, far too few for statistics. So Russett and Oneal looked at a much larger database that lists militarized interstate disputes—incidents in which a country put its forces on alert, fired a shot across a bow, sent its warplanes scrambling, crossed swords, rattled sabers, or otherwise flexed its military muscles.224 Assuming that for every war that actually breaks out there are many more disputes that stop short of war but have similar causes, the disputes should be shaped by the same causes as the wars themselves, and thus can serve as a plentiful surrogate for wars. The Correlates of War Project identified more than 2,300 militarized interstate disputes between 1816 and 2001, a number that can satisfy even a data-hungry social scientist.225

Russett and Oneal first lined up their units of analysis: pairs of countries in every year from 1886 to 2001 that had at least some risk of going to war, either because they were neighbors or because one of them was a great power. The datum of interest was whether in fact the pair had had a militarized dispute that year. Then they looked at how democratic the less democratic member of the pair was the year before, on the assumption that even if a democratic state is war-averse, it still might be dragged into a war by a more belligerent (and perhaps less democratic) adversary. It hardly seems fair to penalize democratic Netherlands in 1940 for getting into a war with its German invaders, so the Netherlands-Germany pair in 1940 would be assigned the rock-bottom democracy score for Germany in 1939.

To circumvent the temptation of data snooping when deciding whether a state was democratic, especially states that call themselves “democracies” on the basis of farcical elections, Russett and Oneal got their numbers from the Polity Project, which assigns each country a democracy score from 0 to 10 based on how competitive its political process is, how openly its leader is chosen, and how many constraints are placed on the leader’s power. The researchers also threw into the pot some variables that are expected to affect military disputes through sheer realpolitik: whether a pair of countries were in a formal alliance (since allies are less likely to fight); whether one of them is a great power (since great powers tend to find trouble); and if neither is a great power, whether one is considerably more powerful than the other (because states fight less often when they are mismatched and the outcome would be a foregone conclusion).

So are democracies less likely to get into militarized disputes, all else held constant? The answer was a clear yes. When the less democratic member of a pair was a full autocracy, it doubled the chance that they would have a quarrel compared to an average pair of at-risk countries. When both countries were fully democratic, the chance of a dispute fell by more than half.226

In fact, the Democratic Peace theory did even better than its advocates hoped. Not only do