Does Incomplete Democratization Risk War?

Perhaps you have come across this argument about promoting democracy: "Yes, maybe once countries achieve a liberal (mature, well-established) democracy, they don't make war on each other, BUT in the process of democratization, they make more war than do other nations, even more than dictators against each other." Therefore, it is sometimes concluded, fostering democracy is a dangerous project. And this argument has been used against our involvement in Iraq

A major source for this idea is the published research by Jack Snyder and Edward Mansfield (see my democratic peace bibliography here). I've finally been able to study in detail their research that is given in their book, Electing To Fight: Why Emerging Democracies Go To War (2005). Their data cover over a century of wars, 1816-1992, during which they found (dealing with only their composite index of democratization for simplicity here) 90 incomplete transitions to democracy for 64 nations, and 50 complete for 35 nations. Over this period, there were 79 wars, and for their sample, the probability of a nation going to war in their sample is .037, very small.

First, they further confirm that when democratized, nations do not make war on each other. This is an easy one, since in their data no two democracies made war on each other. Second, they go though extensive tests to determine whether incomplete democratized nations were most prone to war. Having done such research myself, I have much respect for the effort, time, and thought they put into this, and therefore hate to be a spoilsport. However, it is like two neighbors who build a car in their garage. It's beautiful, with glittering chrome, comfortable fake leather seats, state of the art dashboard, and a well waxed red paint job. But when they start it up, all the unseen motor will do is put, put a few times, and stop.

Since they are trying to establish whether democratizing nations went to war (1 = yes, or 0 = no) more than others, they used logistic regression analysis, but they did not check if the assumptions of their model were met in order to assess the significance of their regression coefficients. They provide no correlations between the independent variables so that one can assess their multicollinearity (see my
Little Primer" on this here), which is just as applicable to logistic regression as it is to multiple regression) and seem unaware of the problem it creates. They put much emphasis on the significance of their index of incomplete transition, but if their twelve independent variables are highly correlated, which I think they are, then the significance of their regression coefficients may be inflated. When they claim that nations with incomplete democratizations are "roughly four to fifteen times more likely to go to war," this is probably based on highly biased regression coefficients.

Also, the authors provide no justification for their applying tests of significance to a whole population. If this whole sample is meant to represent all nations at all time, then it is not random, and its distribution is unknown. Then there is the problem of the number of cases for which they calculate their significance. As the number of cases ("N") increases, smaller results become significant until what is significant is meaningless. For example, a correlation of .378 is statistically significant for 20 cases, .165 for 100 cases, and .052 for 1,000 cases. Now, square .052, which is .003 rounded off. This says that the two variables only have 0.3% of their variation in common. This is meaningless (would you buy an expensive drug that had a 3 out of a 1,000 chance it would help you? Only if you had terminal cancer), although some unwary researchers might trumpet such significant results. This misuse of significance happens all the time, since the idea is that bigger samples are always better. This is only true if one concentrates not on significance, but on the percent of variation in common. All this being said, what was the sample size in the democratization study? It was 9,229! And the gist of their results depends on significance.

Then there is the question of efficiency. How well does the logistic regression fit (predict, account for, explain) wars, if democratization is incomplete? They provide no measure of this. In regular multiple regression, there is the multiple correlation squared (R^2) which tells us the proportion of variation in the dependent variable accounted for by the regression equation. However, such is inappropriate for logistic regression. So, there is a "pseudo R^2" one can calculate, or for the list of wars, one can count the number of nations correctly placed in the no-war, or war category. The authors do neither.

But, there is one thing we can do. The logistic regression comes out with the likelihood -- probability -- that war will occur, given the independent variables, among which is incomplete democratization. But this is usually such a small number in logistic regression that the natural log of the likelihood is given. Now, to get the probability of war from their logistic regression, one takes the anti-log of the log likelihood, which is e^(log likelihood). I did this for their log likelihood of -1339.96, and it is an infinitesimal number. It is so small that the google and my Mac calculators could only give it as zero. Just for e^-13, it is 2.26 x 10^-6; for -130 it is 3.5 x 10^-57. That is, the equations they provide in their book are useless.

Then there is their way of measuring war, which is as yes, or no. And this is a methodological mindset that has led many researchers to mistakenly conclude that democracies are as warlike as other regimes (see my published article on this here). It gives the same weight to a war in which a democracy suffers few killed in combat versus a nondemocracy that has millions killed, e.g., the Boxer Rebellion counted as a war for Britain when it had 34 killed versus 7.5 million for the Soviets in WWII. One war each. This biases results against democracies, which by far have the least killed in wars, as they should by democratic peace theory. Rather, it is the number killed in war that should be counted for each country, and not the number of wars.

In sum, the results about the war likeness of democracy in Electing To Fight do not prove (show, establish, indicate) that incomplete democratization is a danger to peace.