No peace in our time?
There’s a new paper arguing, contra Pinker, that the world is not getting more peaceful:
On the tail risk of violent conflict and its underestimation
Pasquale Cirillo and Nassim Nicholas Taleb
Abstract—We examine all possible statistical pictures of violent conflicts over common era history with a focus on dealing with incompleteness and unreliability of data. We apply methods from extreme value theory on log-transformed data to remove compact support, then, owing to the boundedness of maximum casualties, retransform the data and derive expected means. We find the estimated mean likely to be at least three times larger than the sample mean, meaning severe underestimation of the severity of conflicts from naive observation. We check for robustness by sampling between high and low estimates and jackknifing the data. We study inter-arrival times between tail events and find (first-order) memorylessless of events. The statistical pictures obtained are at variance with the claims about “long peace”.
Every claim in the abstract is supported by the data—with the exception of the last claim. Which is the important one, as it’s the only one really contradicting the “long peace” thesis.
Most of the paper is an analysis of trends in peace and war that establish that what we see throughout conflict history is consistent with a memoryless powerlaw process whose mean we underestimate from the sample. That is useful and interesting.
However, the paper does not compare the hypothesis that the world is getting peaceful with the alternative hypothesis that it’s business as usual. Note that it’s not cherry-picking to suggest that the world might be getting more peaceful since 1945 (or 1953). We’ve had the development of nuclear weapons, the creation of the UN, and the complete end of direct great power wars (a rather unprecedented development). It would be good to test this hypothesis; unfortunately this paper, while informative, does not do so.
The only part of the analysis that could be applied here is the claim that:
For an events with more than 10 million victims, if we refer to actual estimates, the average time delay is 101.58 years, with a mean absolute deviation of 144.47 years
This could mean that the peace since the second world war is not unusual, but could be quite typical. But this ignores the “per capita” aspect of violence: the more people, the more deadly events we expect at same per capita violence. Since the current population is so much larger than it’s ever been, the average time delay is certainly lower that 101.58 years. They do have a per capita average time delay—table III. Though this seems to predict events with 10 million casualties (per 7.2 billion people) every 37 years or so. That’s 3.3 million casualties just after WW2, rising to 10 million today. This has never happened so far (unless one accepts the highest death toll estimate of the Korean war; as usual, it is unclear whether 1945 or 1953 was the real transition).
This does not prove that the “long peace” is right, but at least shows the paper has failed to prove it wrong.
http://stevenpinker.com/files/pinker/files/comments_on_taleb_by_s_pinker.pdf Pinker’s response to Taleb
Pinker’s response is shall we say interesting, first he says:
He than proceeds to demonstrate that he doesn’t in fact understand the implications of these statistics by writing stuff like this:
A power law distribution means that at any given time not during a major war it looks like war has fallen to unprecedented levels.
A hypothetical proto-Pinker writing in 1913 could similarly note that the turning point was in 1814, possibly even citing the Franco-Prussian war to show how wars between great powers are now short and limited.
Not disagreeing, but I repeat that the hypothesis that 1945/1953 represent turning points is not unreasonable. and should be tested directly.
What does “turning point” mean in the context of a power law distribution?
The simplest power law , a*x^k has two parameters which govern the overall location and the fatness of tails, I think. You could expect a change in either or both. So you could fit a time-series model in which a and k can change each year by an amount drawn from a distribution, and then see whether the data supports a large net change since 1945? This is what I take Stuart as suggesting.
The problem is that you can’t estimate ‘k’ well enough, at least that’s what Taleb argues at various places.
The available data must support some range of ks, some precision, and if allowing any shift of k over time indicates that k has fallen a lot lately, that’s pretty bad for their theory. If they say you should ignore the data, then they’re doing theology.
The point is range of ks is quite large. That’s what Taleb’s work as a professor of Risk Engineering is about.
If the range of ks is large then the posterior probability of a shift (or to put it another way, the estimated probability that pre-WWII ks differ from post-WWII ks) will be appropriately small and Taleb will have demonstrated what he wants to demonstrate without so much rhetoric and an analysis that largely misses the point.
Note that that looks like Steven Pinker’s 2012 response to Taleb’s earlier criticism of the book. I don’t think it’s a newly written response to Cirillo & Taleb’s paper.
The Second Congo War is estimated to have had killed up to 5.4 million people, although not directly through violence. Do casualties include wounded (so not just deaths?), because in that case a few more wars would fit the bill.
5.4 million is the upper estimate—and my “5-10 million since WW2” is really “10 million per world population in 2015 of 7.2 billion”, which is 10 million (6.7/7.2) = 9.3 million in 2008 and 10 million (5.97/7.2) = 8 million in 1998; both are above 5.4 million.
It’s not stated in the paper what they mean by “victims”, but they count 70 million or so victims of WW2, which has a death toll ranging from 50-80 million, so it seems that “victim” means “dead”.
Well, 70 years of 1⁄37 risk still has 13% chance of showing zero wars. Could happen. (Since we are talking about smaller ones rather than WWIII anthropics doesn’t distort the probabilities measurably.)
One could buy a Pinker improvement scenario and yet be concerned about a heavy tail due to nuclear or bio warfare of existential importance. The median cases might decline and the rate of events go down, yet the tail get nastier.
Indeed. This is not a proof of the “long peace”, just showing the paper doesn’t disprove it.
I am curious why you put the sign of the cross there. None of these people appear to be dead. (?)
Ooops, sorry—because I copy-pasted it from the article and didn’t remove the cross.
Cold approach works and feels good but the actual value given by a random woman, even when you select for the attractive or peacocking ones, tends to be very very low. I think many intellectuals overestimate the value of randomly picked women because those they have come across in male-dominated feels tend to be exceptional—very intelligence, skilled, curious etc. If you’re like me, you’ll find yourself yearning for the physical beauty of the wider community when you’re around rationalist women, and the interest in rationality from the rationalist community when you’re around beautiful women. The secret, I’ve been learning, is simply to see the good in everyone, and select for people who are good at managing interpersonal relationships above all other traits, in one’s social circle.
Your link seems to argue that peace interventions aren’t maximally efficient, not that they’re not useful. As for why peace is a good metric to use, war (for the majority of the population affected) is a sufficiently clear negative that it would take extremely strong arguments to show peace is not an important positive good. Just look at the revealed preferences of refugee flows, if nothing else.
Do I get it right that the concept of violence is reduced to the concept of “official” war between nations? Why? How about murder and violent crime? One is selling death in wholesale, the other in retail, but it can add up. For example Albania was not involved in many official wars but the death rate from clan feuds was so high that it was affecting gender roles.
Pinker argues that all types of violence are going down. Cirillo and Taleb deliberately chose to ignore homicides:
I’d note that they chose to ignore homicides, and it seems that homicides undermine their general point.
Pretty sure I remember reading Taleb on this, stating that homicides follow a normal distribution and he accepts Pinker’s argument there. He de-couples this from the other kinds of violence, saying that the narrative stretches too wide—specifically warfare should be analyzed separately, since that follows Pareto distributions.
That’s fine.
I’m not sure about that. The whole concept that inner cities are dangerous no-go zones is recent. In Victorian London one mugging every couple months was considered a crime wave.
I believe that is incorrect—London used to be a fantastically violent city.
But the reason I suspected it undermined their case is that they said “We may have observed lower rate of homicide in societies but most deaths risks come from violent conflict.” in the paper.
Hey, look—data :-)
Interesting. But there’s a couple of features in there that make me leery of relying on this.
First, the table there is tracking convictions at jury. Clearance rates, especially for property crime, have never to my knowledge been high; if we assume similar figures they’d be undercounting reported crimes (most comparable to the crime statistics we’re familiar with) by a factor of three to five, and undercount committed crimes by more. That isn’t necessarily a good assumption, though, and there’s the rub: we can’t use conviction rates to estimate crime rates unless we know something about how likely cases were to make it through the system.
Second, buried in the bottom of that page there’s a sentence saying that about 17,000 summary convictions (excluding some minor fines) were imposed independently by police magistrates. No information on type, which means the table would further undercount crimes by some unknown proportion depending on how likely summary punishment was. But 17,000 is roughly four times the total jury convictions cited, so it’d probably be large.
I haven’t been able to find another source going back to the 1830s yet, but this data suggests that as many murders were recorded in London in the late Victorian era as in the mid-Sixties, when population was about 20% higher. (Population in 1838 was much lower—the city grew hugely over the 19th century.)
Well, I don’t have an intimate relationship with this data, this was only a Google quickie :-D But I think NRx made a big deal out of the safety (as they claimed) of Victorian London and it’s something they bring up regularly. It might have started with Mencius Moldbug, but I’m too lazy to go looking for the original write-up which, IIRC, included some data analysis.
Even at high homicide rates (e.g. Russia) it would take many many years to add up to even a small war’s casualties.
False, unless one sets an unusually high bar for an armed conflict to qualify as a “war”.
Russia had “over 12,300 homicides” in 2013.* For comparison, about 900 people died in the Falklands War; about 1,400 in the 2006 Lebanon War; and the 1999 Kargil War between India & Pakistan killed 900–5,600 people, depending on whose counts you trust.
The United Nations Office on Drugs and Crime has data for 2008–2012 as well, which allows some more interesting comparisons. In 2008 Russian police counted 16,617 murders, or about 60× as many deaths as in the war between Russia (& South Ossetia & Abkhazia) and Georgia in the same year.
The numbers add up quickly over multiple years. Lumping together the 2008–2013 counts gives about 86,400 recorded murders. This matches International Crisis Group’s estimate of deaths in the Eritrean-Ethiopian War (one of the few unambiguous examples of a recent interstate war arising from a territorial dispute), and beats (for example) the Croatian and Kosovan wars of Yugoslav succession put together (16,000 & 17,000 deaths respectively). Wait another year and the running murder total’ll match the war in Bosnia & Herzegovina.
* Edit to add: the link says “homicides and attempted murders reported” but I’m pretty sure it only counts successful homicides, since the “12,300″ number’s similar to the UNODC numbers for preceding years, and the UNODC counts are of intentional homicide, “defined as unlawful death purposely inflicted on a person by another person”.
Try Mexico. Now that arguably qualifies as a war, but most of the sides involved aren’t official states.
Or how about the threat of violence to keep people under control?
That does not seem to change much throughout history.
What about this one?
Article: http://researchnews.osu.edu/archive/wardecline.htm Paper: http://papers.ssrn.com/sol3/papers.cfm?abstract_id=2317269&download=yes
That paper provides an alternate explanation for the long peace thesis.
However, it rejects “per capita deaths” as a good measure of rate of conflict, which makes it pretty dubious (people are fully aware that per capita is ideal for comparing homicide rates; why suddenly reject it for war deaths?) They write nonsense like:
More people means more soldiers, larger economies (hence more manufacturing of weapons), more and larger groups with reasons to rebel/fight/steal each other’s stuff.
They do have an interesting point with “war as information gathering/negotiations”. But a lot of the rest seems to be a conflation of “the reasons things are getting more peaceful are not nice reasons” with “things are not genuinely getting more peaceful”.
Violence might not be the exact opposite of peace. Intuitively, peace seem to mean a state where people are intentionally not committing violence and not just accidentally. A prison might have lower violence than an certain neighbourhood but it might still not be considered a more peaceful place exactly because the individual proclivity to violence is higher despite the fact violence itself isn’t. Proclivity matters.
I am generally sceptic of Pinker. I have read a ton of papers and Handbooks of Evolutionary Psychology, and it is clear that while he was one of the top researchers in this area in the 90′s this has dramatically changed. The area has shifted towards more empirical precision and fined-grained theories while some of his theories seems to warrant the “just-so story” criticism.
Pinker seems to prevent good evidence for the long peace, but not for his explanations as to why it happened.
“The proliferation of states in the 20th century” was not an exogenous event, but explicit decisions made by the victors of WWI and WWII, specifically to prevent violence, so they should get credit for them.
Pinker tries to provide several complementary explanations for his thesis, including game-theoretic ones (asymmetric growth, comparative advantages and overall economic interdependence) which could be considered “not really nice reasons for measuring our (lack of) willingness to destroy each other”. Like SA said, Braumoeller seems to conflate ‘not very nice reasons to maintain cooperation’ with ‘our willingness to engage in war hasn’t changed’. And this is one of the reasons why Taleb et al. missed the point on Pinker’s thesis. To test if it’s business as usual, if our willingness, what ever that is isomorphic to, is the same, one needs to verify if State actors are more likely to adopt the risk dominant equilibrium than the payoff equilibrium or if there is intransitivity. There is a connection between the benefits of cooperation and the players willingness to coordinate. What if the inability to dominate places our future in a context where we see don’t see fat tails in deaths from deadly conflicts? What if the benefits of cooperation increases over time along with the the willingness to coordinate? What if it’s not business as usual?