The Debate Continues: Peace by the Numbers

Much recent scholarship, including that of Steven Pinker, Joshua Goldstein and the Human Security Report, has argued that war is in long term decline.  Nassim Nicholas Taleb and Pasquale Cirillo challenged this view and I have participated in some of the back and forth generated by their critique.  (This presentation is particularly germane to today’s blog post.) 

Aaron Clauset brings something new to the table by showing that the Correlates of War (COW) dataset on international war size and timing is compatible with a “no change” model according to which the risks of international wars of all possible sizes have remained stable over the last two centuries.  This is an important new insight that I did not anticipate.  However, I argue below that Aaron’s paper should not precipitate a major shift in our thinking about the decline of war.

That first sentence in the last paragraph is pretty dense so let’s unpack it with a little metaphor.  Imagine that each time humans start an international war the gods of war determine its size through a random draw from an urn full of balls representing possible war sizes.  Assume, further, that they used the same urn and balls from the 19^th century straight through World War II, maintaining a constant probability of wars of all sizes throughout this period.  Aaron then makes the hypothesis that the size of each international war after WWII continued to be generated by a draw from this same pre-WWII urn-and-balls setup.  He tests whether he can reject this “no change” hypothesis and finds he cannot.  Thus, Aaron shows that the historical record is at least consistent with the idea that the gods of war did not switch to a more peaceful war-size generation system after WWII.

Here is some quick background useful for placing Aaron’s finding in context.   Most people now accept that documented war violence has actually declined since WWII.  Aaron does not challenge this perception as we can see from his Table 1, which I augment by two columns on the right that translate frequencies into numbers of international wars by multiplying the frequencies by 38 (the number of international wars post-WWII according to COW).   (Note that severity range 3-4 means 1,000 to 10,000 battle deaths, 4-5 means 10,000 to 100,000 etc.).

This table says that if war size post-WWII has been determined by random draws from the distribution of wars before WWII, i.e., the gods of war did not switch their urns and balls after WWII, then we would expect 14.9 of them to have between 1,000 and 10,000 battle deaths, 14.0 to have between 10,000 and 100,000 etc..

The general thrust of this table is that wars occurring after WWII were weaker.  Specifically, there have not been any truly massive wars and the shift has been toward the smallest range (3-4).  The only contrary evidence against this conclusion is that after WWII there were three wars with 1,000,000 to 10,000,000 battle deaths compared to the expected 1.8.  We could write this off as a random minor glitch although a close examination of the COW data suggests to me that we do not even have to do this.  These three wars must be Korea, Viet Nam and Iran-Iraq for which COW gives battle-death numbers of 910,084, 1,021,442 and 1,250,000, respectively, suggesting that the correct coding would have been “2” rather than “3”.  Moreover, the PRIO Battle Deaths Dataset, widely considered superior to COW, gives “best” estimates of 995,000, 1,625,973 and 644,500 battle deaths, respectively, for these wars, reinforcing my sense that “3” is not good coding here, although I do not think this is a big deal.  More importantly, these three wars are, at best, around the lower edge of the 6-7 range, suggesting that war activity has been less than predicted by the “no change” model in this category as well as in the other high-end ranges. 

In short, the documented evidence on war size in Aaron’s paper shows that wars have been getting weaker since WWII.  Moreover, this conclusion would be strengthened if we normalized the above table by world population which has grown dramatically in recent decades.  The decline-of-war thesis is generally presented in terms of per capita deaths, so the above table actually understates post-WW II changes.

Aaron’s main point is that it is still too soon to rule out the possibility that the same underlying mechanism that generated all the massive wars of the past is still operating but that we have just been lucky enough to go 70 years without any such major war cropping up.  A key part of this argument is the fact that the distribution of war size has a fat tail at the upper end so really big wars are rare but not nearly as rare as they would have been if war size fit a standard Bell-curve-type distribution.  Importantly, the distribution of pre-WWII war sizes has a certain Goldilocks quality.  If huge wars had been more frequent than they actually were pre-WWII then we would already be able to reject the hypothesis that the war generating mechanism had not changed in the post-WWII period.  If, on the other hand, huge wars had been substantially less frequent pre-WWII than they actually were, then the fact that the old war-generating mechanism might still apply would not seem as ominous as it does right now.

Aaron’s point is valid, important and novel.  (Cirillo and Taleb argued something similar but, crucially, they never tested for a break point around World War II as Aaron does).  Yet for at least two broad sets of reasons, I think it should only move the needle a bit in the decline-of-war debate.  First, failure to reject the “no change” thesis does not mean this thesis is actually correct. Indeed, one could make other hypotheses that would be more favourable to a decline-of-war thesis and that also would not get rejected. So we must not overreact to just one non-rejected hypothesis.  Second, Aaron’s analysis sets aside all pieces of evidence that do not pertain to the time series on war size, but this evidence is still relevant and supports the idea that violence really has declined across a broad spectrum of human activity.  Steve Pinker and I offered the following summary in Significance magazine:

“Homicides, slavery, torture, executions, corporal punishment, wife beating, extreme cruelty towards animals, and other forms of aggression are in long-term decline. Unlike wars, these forms of violence are not dominated by a small number of big events, so it would not be sensible to project that they might suddenly display massive surges. Moreover, to the extent that ascendency of the “better angels or our nature” drives this broad-spectrum decline in violence, as argued by one of us (Pinker), these angels could help us lower the probability of future massive wars.”

The preponderance of all the evidence still favors the decline-of-war thesis.  While Aaron’s work is new and important, it is one piece of data in an evolving discussion, and in the context of that discussion I feel that it is premature to reject all the other information we know already that suggests that the decline in war is real.

Michael Spagat Bio

Michael Spagat is a Professor of Economics at Royal Holloway College, University of London.  He gained his Ph.D. at Harvard University and earlier held faculty posts at Brown University and the University of Illinois.  Spagat’s published papers on armed conflict have been published in Nature, Scientific Reports, the New England Journal of Medicine, the Journal of Peace Research, the Journal of Conflict Resolution and PLoS Medicine among other places.  His current research addresses universal patterns in modern war, fabrication in survey research, the Dirty War Index civilian casualties in the Iraq conflict and problems in the measurement of war deaths.   He blogs on war at https://mikespagat.wordpress.com/  and tweets at @Michael_Spagat. For more information see  http://personal.rhul.ac.uk/uhte/014/.