Friday, September 21, 2012

What Long Peace?

There has been a lot of discussion about "the long peace", the "retreat from doomsday", and even how we've discovered the better angels of our nature.  An impressive list of scholars has weighed in on whether this trend is likely to continue.

In this post, I'm going to argue that the trend is somewhat different than many of us realize before offering some thoughts on how to understand the frequency of warfare over time.

First, let's just establish some preliminaries.  The frequency of major power war has declined over time, and, as Josh Goldstein has noted, those wars that do occur are killing fewer people than they have in the past.  From 1950 to 1989, 180k people a year were killed in war.  In the 1990s, that dropped to 100k per year.  Since 2000, it has been 50k a year.  That's a wonderful thing.  And it has to be part of the story.

But it's not at all clear that wars are occurring less frequently.

Below, you will find a simple graph depicting the number of war onsets by year from 1816 to 2007, pooling across interstate wars, intrastate wars, and extrastate wars (data taken from the Correlates of War project).  There's a lot of fluctuation there, but to a first approximation, the number of war onsets is increasing (modestly) over time.  To be fair, this is not inconsistent with any claims made about the likelihood of major power war.  The title of this blog post is deliberately provocative.  But if you were under the impression that organized violence was in secular decline, that the international system was growing steadily more peaceful, you could be forgiven, based on the claims some people have made.  You would, however, be mistaken.

So what's going on here?

In a new project with Scott Wolford and Kyle Joyce1, which I'll be discussing at the upcoming meeting of the Peace Science Society, we argue that the systemic distribution of military capabilities has an impact on the likelihood of war onset at every level.  That is, we argue that we shouldn't just focus on whether bipolar or unipolar systems are more or less prone to producing wars fought between the major powers, but any war.

We analyze a four-player bargaining model in which A and B negotiate over the division of some good in the shadow of war, knowing that any war that does occur may expand to include C, D, or both.  We assume that C is a major power who is sympathetic to B and D a major power aligned with A.  We demonstrate that even if A and B know everything there is to know about one another, and even if they face no commitment problems, war may still occur if B has a better estimate of the likelihood that C will intervene on his behalf than does A.  The mechanism here is the familiar risk-return tradeoff, but note that there's a unique twist here.  In a world where C and D did not exist, A and B would have no reason to fight.  And when they fight, it is entirely possible that neither C nor D will intervene.  Our model suggests that strictly bilateral wars may be caused by uncertainty over the intentions of the major powers.  Thus, it is inappropriate to focus strictly on the incidence of major power war if we are to understand the implications of major power politics.

Our model suggests that the relationship between the actors' capabilities and the likelihood of war is very complex.  We're still working out the full details.  But some simulations I've run suggest that, to a first approximation, wars between A and B (which may or may not ever involve C and D) are more likely to occur when C and D are roughly equal than when either is more powerful than the other.

Nothing in our model tells us that we should think of A and B as both being formal members of the interstate system.  So we assess this by looking at all types of war, not just interstate wars, and certainly not just major power wars.  Using the measure of military capabilities I've discussed previously, we demonstrate that the number of war onsets that is expected in any given year, though generally increasing over time, is also responsive to the distribution of capabilities between the two most powerful states in the system.  According to one model I've estimated (email me for details if you're interested), we expect approximately 1.3 additional wars to begin in any given year as we move from the most imbalanced conditions that have observed over the past 200 years (when the most powerful state in the system has three times the military capabilities, according to my measure, than the second most powerful state in the system) to conditions of perfect parity (that is, when the two most powerful states in the system are evenly matched in terms of their military capabilities, as I have measured them).

This next graph depicts the expected number of new war onsets in any given year based on a model that accounts for the general trend over time (increasing) as well as the level of parity between the two most powerful states in the system.  The parity values are those that are actual observed, while the predicted number of wars is an estimate produced by the statistical model.

As you can see, the relationship is far from perfect.  And I will freely admit that the model is not accounting for a particularly large amount of the variation in the dependent variable.  But the number of wars that is expected to begin in any given year does appear to depend on the distribution of capabilities between the two most powerful countries.  We're not claiming that the systemic distribution of capabilities tells us everything.  It clearly doesn't.  Not by a long shot, in fact.  But it's an important part of the story that's missing from many popular accounts.  It may be that we've seen a change in norms -- though I wonder why there is a modest increase in the frequency of war onset over time if that is the case.  It may well be that the spread of democracy and the deepening of economic globalization and the development of nuclear weapons and the UN's growing commitment to peacekeeping, and many other factors besides, are deeply important -- though, again, it's important to note that wars are not becoming rarer, even if they are becoming less deadly.  All I'm saying for now is that any responsible conversation about aggregate patterns of war and peace needs to have a proper baseline in mind against which to judge the relative explanatory power of those factors.  And, so far, that's been missing.  The fact that fewer people are dying in those wars that do occur is also very important to keep in mind.  I'm not disputing that.  But we need to acknowledge that there are no signs that war is going away.  And the simple fact that the Soviet Union has collapsed itself would have been expected to decrease the frequency of war over the past 20 years.  Any analysis of the impact of norms or democracy or peacekeeping or whatever else has to take into account that the proper counterfactual is not the frequency of war onset during the Cold War, but the frequency of war onset that would be observed in a relatively unipolar system absent those factors.

1. Disclaimer: while this blog post is based on research the three of us are conducting jointly, I am responsible for the content of this blog post.  Especially the objectionable parts.  Anything that strikes you as clever was either Kyle's or Scott's doing. 


  1. Phil:
    1) If you are going to pooh-pooh the norms argument (at least as made by Mueller), you shd state that argument accurately first, which you haven't done. The norms argument as made by Mueller is that major war has been obsolescent since WW1 (and even more since WW2) as a result of widespread revulsion against killing on this scale, facilitated by the presence of an active (albeit small in relative numbers) antiwar mvt in the belligerent countries before WW1, which served to inject antiwar notions into public discourse, which was primed to be more receptive to those antiwar notions as WW1 dragged on (and esp. after it finished). Before WW1 some substantial number of intellectuals and publicists in the 'West' talked about war as glorious, manly, honorable, necessary for the health of the species, etc. After WW1 this kind of discourse becomes much, much less prominent. That is really the key marker of normative shift for Mueller as I understand his argument; it has nothing to do w nuclear weapons and not all that much to do w democracy, globalization and the other factors you list. Goldstein of course emphasizes UN peacekeeping but that is a diff argument than the normative argument of Mueller (and see C Fetteis' extension/modification of the latter). Instead of distinguishing these arguments in this post, you've just basically lumped them all together (and pretty much dismissed them).

    2) The title of the post is not just provocative, it's actively misleading. "The long peace" as I understand it refers specifically to the fact that the US and USSR never directly went to war w each other and perhaps secondarily to the absence of great-power war since 1945 (or '53, depending on definitions). "What long peace?" is a misleading title since the post doesn't cast doubt on those facts.

    3) The post says the intl system is not becoming more peaceful b.c war onsets are increasing. This is a definitional question. It depends on how you measure "more peaceful". The standard def of "war" in 'the literature' and the COW data is, of course, any conflict producing at least 1000 battle deaths, so you can have a bunch of very small wars beginning (1001 deaths will make the cut) and the system still becoming more peaceful if the latter is defined in terms of deadliness and size of wars, not just number of war onsets. You don't bother to mention the 1000 threshold. Presumably most readers will know this but a non-scholar stumbling across this post cd conceivably come away w a wrong impression.
    Note btw that Waltz in his exchange w Colin Kahl in the current Foreign Affairs (wh i have only glanced at) refers to wars causing 1000 deaths as skirmishes. Of course 1000 battle deaths is 1000 too many, but there is a big difference btw the Kargil war betw India and Pakistan (the referent for Waltz's remark) and, e.g., the first day on the Somme (60,0000 British casualties in one day: 20,000 dead, 40,000 wounded).

    4) Are there any, just out of curiosity, real-world exs. of bilateral wars that were set off b.c of diff estimates (or uncertainty) by belligerents of the likelihood of intervention by major powers? (probably, I suppose, but one wd have to look through eg Blainey to find them).

    1. Hey LFC.

      1.) I wasn't suggesting that Mueller is concerned with nuclear weapons or democracy. I was, as you say, lumping together different arguments. Without question, anyone who is interested in assessing the relative merits of different explanations must know the details of those explanations, and I did not provide the details of any of the competing explanations, including Mueller's. After all, a blog post isn't an essay in FA or a journal article. Note that I left out a *great* deal of detail on my own argument as well.

      2.) Fair point about the title. I did, however, state quite clearly in the post that I did not contest the absence of major power war. And many people, when discussing the same topic, do make much grander claims. An alternative title would probably have been more appropriate, but I think anyone who reads the post carefully will understand that I'm taking issue with grander claims that are often made as part of the bigger discussion as opposed to Gaddis' thesis about major power war specifically.

      3.) You are absolutely right that it all depends on what you mean by "peace". Again, I was pretty straightforward about this in the post. I acknowledged, several times, that fewer people are dying in war and that this is very important and needs to be part of the conversation. But there are people out there claiming that we've solved the problem of war, or nearly done so, and that is hard to reconcile with an increased frequency of war onset, even if those wars are lower in intensity.

      4.) I'm uncomfortable with mono-causal accounts of any conflict. There are some conflicts where one could plausibly argue that the decision of the defender to resist the challenger was influenced but a mistaken belief that they would receive assistance from a third party, but arguments like this are focused on identifying systematic patterns that emerge in aggregate because of modest changes in the likelihood of conflict. In the context of the theoretical model, war occurs *only* because of uncertainty about C. But what makes this interesting is that setting aside all the other factors we know to influence the likelihood of war (even those that do so in the same class of models, like commitment problems or uncertainty over the preferences of other actors) allows us to see something that we might otherwise have overlooked, and to analyze clearly the impact of some key factors on that possibility. I don't believe that, in the real world, the sole source of uncertainty concerns the preferences of third parties. Nor that commitments are always credible, or that actors never make mistakes, or that actors are uninfluenced by psychological biases, etc etc etc. I just hold those things aside to gain insight on a very complicated problem. So if I point to a specific war, you are likely going to respond by saying that I've ignored x, y, and z, all of which you think are more important to understanding why that war occurred. And perhaps you'll be right to do so. That will not, in any way, detract from the broader point I'm trying to make. So I'd prefer not to play that game.

  2. spelling correction: Fettweis

  3. Our differences here are partly, I think, but only partly, a question of what one chooses to emphasize. It is worth knowing that 1.3 additional wars may start each yr as the dist. of capabilities betw the 2 most powerful states moves toward parity. Ultimately though it seems one would have to point to some real-world exs., presumably down the road, where the dist. of capabilities was one factor (among many factors) in contributing to a particular conflict. Otherwise you're left w an aggregate pattern with no specific instantiations (examples, whatever).

    I wd also like to look at the war onset figures more closely, but I haven't done that yet. It looks from the 1st graph as if the trend in very recent yrs is actually downward, even tho the trend over the whole pd 1816-2007 is up. You get that big spike in the mid-20th cent. of course with the world wars and one needs to take that into account. Which your statement about "to a first approximation the overall trend is up" perhaps doesn't fully do. You just say there's a lot of fluctuation but sometimes it's helpful to look behind particular fluctuations.

    1. I think you are right that part of our difference is in emphasis, yes.

      I understand what you are trying to get at with the examples, but I think you're still missing my point. Suppose, to take an analogy, that I were to claim that drinking tomato juice reduces the risk of heart attack by 2%. If you looked at large population of people over the course of 200 years, the number of additional years lived as a result of tomato juice consumption might well be significant, from an aggregate view. But if you asked me for an example of someone who got a heart attack *because* they did not drink tomato juice, I honestly wouldn't even know what to say. That's not a perfect analogy, and if we went through the list of every single war that occurred since 1816 and there wasn't a single instance in which I could plausibly claim that my mechanism was at work, that would indeed be a problem. But the tomato juice example is not entirely inapt.

      You are right that in very recent years, the trend is downward. When I said "to a first approximation" while acknowledging that there is a lot of fluctuation, what I meant is that if you were to make a sweeping statement about the overall trend, it would be more appropriate to say that it is positive than negative. Put differently, the decline in deaths suffered in war began several decades before the recent decline in the frequency of war, and thus any discussion of how peaceful the international system is that fails to take note of the fact that the frequency of war was increasing for much of the period during which fatalities were declining would be a little misleading. It would not be correct to say, without any qualifiers, that the frequency of war is strictly increasing over time. Which is, of course, why I did qualify my claim.

      I agree that it's useful to look behind specific fluctuations. For that reason, I would note that we see a decline in parity around the same time as the decline in the frequency of war. It's not a 1 to 1 mapping, and I wouldn't claim that parity alone can perfectly account for this variation, but it is worth noting that there's some overlap there.

    2. Thanks. Particularly your second-to-last paragraph here I find helpful. The combination of the frequency of war increasing for much of the time during which deaths were declining is in itself sort of interesting.

  4. Actually the big spike is not the world wars -- if they are counted as one onset each (which they perhaps shdn't be but i don't know how the graph wprks exactly) -- but prob. rather a lot of 'colonial' and 'peripheral' wars, also mid 20th cent.

    1. The world wars are mostly counted as one war each. A few wars that overlapped with the world wars and can't really be understood in isolation from the world wars, but are still technically separate conflicts, are counted separately. You are right, these spikes have more to do with "colonial" and "peripheral" wars.