And Cain talked with Abel his brother: and it came to pass, when they were in the field, that Cain rose up against Abel his brother, and slew him.
~ Genesis 4:8
Ötzi, the Iceman, is our historical Abel. He was murdered. The X-ray image shows the arrowhead that did him in. Life in the Copper Age was no picnic.
In his latest book, The Better Angels of Our Nature, Steven Pinker makes the case for the idea that human-on-human violence has steadily decreased over the last 10,000 years or so. I watched his TED talk and attended a public lecture he gave at Brown University on February 13. I have not read the book, but I would like to share some comments.
The data Pinker has gathered from archival sources are impressive. Graph after graph shows the decline of various indices of violence (e.g., homicide) over historical time. The indices show rates of violence relative to a population of a fixed size (e.g., 100,000) and can thus be understood as measures of risk. Absolute numbers do not accomplish this. There were fewer homicides in the Stone Age because there were fewer people around.
The decline is monotonic. There is no apparent periodicity. The numbers just go down. There are perturbations of the trend and some blips (e.g., during great wars), but no serious disruption of the pattern. Pinker’s case is interesting, in part, because the finding is so hard to believe. He knows the audience’s incredulity and the first line of his talk addresses it. Pinker explains this disbelief as a result of media focus on violence and the large casualty numbers during the world wars of the 20th Century. However, the media and human memory are selective, and what matters are the relative, not the absolute, numbers.
The theme of vanishing violence is a special case of a more general theme that things will become better and better. This idea was common in Western societies until World War I, which shattered it. Since then, many people have experienced dramatic reversals and expect more. In a way, Pinker’s finding is a return to an optimistic view of social evolution. This is not too broad a statement because to explain vanishing violence, Pinker needs to find other positive trends over historical time. Something else must be moving towards more desirable states if vanishing violence is to be understood in causal terms.
Pinker suggests that there are several broad historical trends that can jointly explain falling rates of violence. Among these are the expansion of trade and the consolidation of political power. These trends discourage violence in the individual with threats of punishment and negative reciprocity (no more trade and loss of wealth). At the same time, these trends interact with psychological capacities (the better angels of our nature). Pinker refers to our capacity for empathy, reason, and self-control. But we have always had those. By themselves, these capacities cannot explain falling rates of violence, but improving socio-economic conditions may provide a more suitable context for them to operate.
The broad socio-political trends are interdependent. Increased top-down regulation can bring about trade where none was before. Well-organized states can build infrastructures and enforce contracts; tribal societies cannot. Following Hobbes, Pinker sees social evolution as a move from anarchy to the Leviathan. Anarchy forces people to look out for themselves in whatever way they can. The Leviathan pacifies and civilizes.
Pinker tells a terrific story. Man has become increasingly civilized, and steadily falling rates of violence have been a dividend. Much as I appreciate the story, a few questions about its causal logic must be asked.
First, causes must precede effects. I assume that in Pinker’s historical scenario, this condition is met, and I ignore the possibility that falling rates of violence may in turn support the consolidation of political power or expand trade (i.e., reduce anarchy). Second, for a cause to be sufficient, it needs to be more probable than the effect. I therefore assume that the socio-economic trend away from anarchy has been overall stronger and broader (more probable) than the reduction in violence. Third, I assume that even in the worst of times (anarchy) violence is less probable than the lack thereof.
These basic assumptions enable us to ask what kind of statistical association one might find between cause (anarchy) and effect (violence). Dawes (1993) discovered that it matters how the association is calculated. He noted that most efforts to explain a phenomenon proceed retrospectively. They ask: “Given that the effect occurred, what is the probability of the cause?” and “Given that the effect did not occur, what is the probability of the cause?” The difference between these two conditional probabilities can be transformed into the correlation coefficient such as phi. Conditioning probabilities on the effect ignores the base rate of the effect, which, according to the second and the third assumption, is low. This analysis pretends that the occurrence and the non-occurrence of the effect are equally likely.
An alternative is to take a view oriented to the present. Phi may be computed using the compound probabilities of cause and effect (i.e., the probability that both are present, that both are absent, that only the cause is present, that only the effect is present). Finally, one can take a prospective view and condition the probability of the effect on the presence or absence of the cause. Now the base rate of the cause is ignored.
Dawes found that given the assumptions listed above, a systematic asymmetry emerges. Coefficient phi becomes smaller as we move from the past to the present to the future perspective. This is important because it is so subtle and unforeseen. One does not suspect that a retrospective causal analysis overestimates the predictability of the future. Dawes’s asymmetry adds to the concerns one has about other biases of retrospective reasoning, such as overconfidence or the hindsight bias.
One might ask if the conditions necessary for the Dawes asymmetry are reasonable. Quite so. Dawes’s numerical example is the effect of smoking on lung cancer. Currently, only a minority of individuals smoke, the probability of lung cancer is lower still, and there is a positive association between the two. Playing with numbers, I convinced myself that Dawes was right. When comparing smoking rates between cancer and noncancer cases, a larger statistical association emerges than when comparing cancer rates between smokers and nonsmokers. Not to diminish the evils of smoking, but risk is about the probability of adverse things happening in the future, not about the probability of an antecedent having been present in the past if the evil has already occurred.
Pinker’s analysis is historical. In this lecture, he did not make any forecast about future rates of violence. When I asked him in conversation if he considers himself to be a historicist (like Hegel or Marx), he said no. He does not believe in inexorable forces that work themselves out through history. If it were so, one could predict the future after having understood those forces. Still, from an inductive point of view, one can be optimistic even without referring to causes. Given that rates of violence have declined for so long, a continuation of that trend is more probable than a reversal. A more textured argument would be to first predict the future of social evolution. If there is reason to believe that anarchy will become less probable, and if anarchy is a cause of violence, then we may predict further reductions in the rates of violence. Assuming that Dawes’s conditions are met in this case, we must be ready to expect, however, that Pinker’s story will not continue as neatly as it looks on his graphs with the historical data. Even if the forecast is good, we may have to figure out why it is that anarchy continues to disappear, and not go all Hegelian about it.
Dawes, R. M. (1993). The prediction of the future versus an understanding of the past: A basic asymmetry. American Journal of Psychology, 106, 1–24.
Pinker, S. (2011). The better angels of our nature. New York: Penguin.