Cover up the two outliers with your thumb, and the impression of randomness is total.
FIGURE 5–5. Random and nonrandom patterns
Sources: Displays generated by Ed Purcell; reproduced from Gould, 1991, pp. 266–67.
FIGURE 5–6. Richardson’s data
Source: Graph from Hayes, 2002, based on data in Richardson, 1960.
You can probably guess what the data represent. Each segment is a war. The horizontal axis marks off quarter-centuries from 1800 to 1950. The vertical axis indicates the magnitude of the war, measured as the base-ten logarithm of the number of deaths, from two at the bottom (a hundred deaths) to eight at the top (a hundred million deaths). And the two segments in the upper right correspond to World War I and World War II.
Richardson’s major discovery about the timing of wars is that they begin at random. Every instant Mars, the god of war, rolls his iron dice, and if they turn up snake eyes he sends a pair of nations to war. The next instant he rolls them again, with no memory of what happened the moment before. That would make the distribution of intervals between war onsets exponential, with lots of short intervals and fewer long ones.
The Poisson nature of war undermines historical narratives that see constellations in illusory clusters. It also confounds theories that see grand patterns, cycles, and dialectics in human history. A horrible conflict doesn’t make the world weary of war and give it a respite of peaceable exhaustion. Nor does a pair of belligerents cough on the planet and infect it with a contagious war disease. And a world at peace doesn’t build up a mounting desire for war, like an unignorable itch, that eventually must be discharged in a sudden violent spasm. No, Mars just keeps rolling the dice. Some half-dozen other war datasets have been assembled during and after Richardson’s time; all support the same conclusion.36
Richardson found that not only are the onsets of wars randomly timed; so are their offsets. At every instant Pax, the goddess of peace, rolls her dice, and if they come up boxcars, the warring parties lay down their arms. Richardson found that once a small war (magnitude 3) begins, then every year there is a slightly less than even chance (0.43) that it will terminate. That means that most wars last a bit more than two years, right? If you’re nodding, you haven’t been paying attention! With a constant probability of ending every year, a war is most likely to end after its first year, slightly less likely to end within two years, a bit less likely to stretch on to three, and so on. The same is true for larger wars (magnitude 4 to 7), which have a 0.235 chance of coming to an end before another year is up. War durations are distributed exponentially, with the shortest wars being the most common.37 This tells us that warring nations don’t have to “get the aggression out of their system” before they come to their senses, that wars don’t have a “momentum” that must be allowed to “play itself out.” As soon as a war begins, some combination of antiwar forces—pacifism, fear, rout—puts pressure on it to end.38
If wars start and stop at random, is it pointless even to look for historical trends in war? It isn’t. The “randomness” in a Poisson process pertains to the relationships among successive events, namely that there is none: the event generator, like the dice, has no memory. But nothing says that the probability has to be constant over long stretches of time. Mars could switch from causing a war whenever the dice land in snake eyes to, say, causing a war whenever they add up to 3, or 6, or 7. Any of these shifts would change the probability of war over time without changing its randomness—the fact that the outbreak of one war doesn’t make another war either more or less likely. A Poisson process with a drifting probability is called nonstationary. The possibility that war might decline over some historical period, then, is alive. It would reside in a nonstationary Poisson process with a declining rate parameter.
By the same token, it’s mathematically possible for war both to be a Poisson process and to display cycles. In theory, Mars could oscillate, causing a war on 3 percent of his throws, then shifting to causing a war on 6 percent, and then going back again. In practice, it isn’t easy to distinguish cycles in a nonstationary Poisson process from illusory clusters in a stationary