An aphorism in the natural sciences states that one should either write the first article on a subject or the last one. The statistical study of war began in the 1930s and 1940s with the work of Lewis Richardson and Quincy Wright, then expanded massively in the 1960s with the Correlates of War project based at the University of Michigan. Those were the first articles. This book is potentially the last important one.

D. Scott Bennett and Allan C. Stam, in a massive synoptic data analysis, have assembled a comprehensive set of variables—packaged in the marvelously user-friendly EUGene software package [http://www.eugenesoftware.org/]— and applied a uniform methodology to test most of the major theories that have suggested empirical predictors for interstate war. This industrial-strength system for the slaughter of sacred cows will doubtlessly be greeted by howls of protest from die-hard advocates of the various theories, but to this observer—who has no stake in any single theory—the authors have gone to great lengths to be fair and impartial. The theories are not straw men, but instead have elicited thousands of journal citations. If they were as good as their advocates claim, they should have stood out in these tests. With three exceptions—strong effects for geographical contiguity, mutual democracy, and a systemic concentration of power—they do not.

It is clear that this is not the outcome that the authors expected or desired. But if their objective had been merely to delegitimate the quantitative study of war, there were far easier ways to do so than with a 12-year effort involving data collection, software development, and statistical analysis. While one can quibble with any of the various analytical choices made in the exercise, the overall package is solid, and it seems quite unlikely that minor modifications of their methods would lead to major modifications in their results. The authors, in fact, try some alternative formulations—including one requiring 60 days of computation—and find that their overall conclusions hold.

One of Bennett and Stam's more intriguing innovations is adopting the standards of medical research to assess whether variables really make a difference. The 1-in-14,000 risk of war per dyad-year is similar to the risk of some diseases that have invoked policy responses, notably lung cancer. Most of the proposed causes of war show nowhere near the level of risk that is associated with policy changes in the domain of public health, and instead are at levels that epidemiologists have learned will usually not survive subsequent tests.

Bennett and Stam's results are also consistent with two other lines of evidence. First, the past 40 years of extensive statistical work has found consistently strong results only for contiguity and the democratic peace (the strong effect of systemic power concentration is the novelty in this study). The other proposed causes have proven to be “oat bran theories” that shine briefly in one set of articles, only to fade upon additional examination. The closest parallel exercise in the qualitative literature—Geoffrey Blainey's (1988) The Causes of War—comes to the same general conclusion: Very few proposed theories about war can survive a broad historical examination.

In view of this rather staggering assessment, where does one go next? Bennett and Stam deal with this question extensively in their first and final chapters. The obvious alternatives would be to revert to either Leopold von Ranke's historicist approach that emphasizes the uniqueness of events or, in the contemporary world, to a postmodern nihlism that denies most notions of objective evidence. Both approaches are rejected by the authors with detailed arguments grounded in the assertions that quantitative statistical methods are the best means of integrating large amounts of empirical evidence, and that most historical and case study approaches by political scientists do, in fact, view history as having repeated patterns instantiated in an objective reality.

Bennett and Stam suggest a number of alternative statistical agendas. One potentially productive avenue—pursued in some initial tests in the book itself—is to subset the cases on time, space, and other characteristics. This is particularly credible since the time period in the comprehensive test—1816 to 1992—encompasses changes in warfare greater than any seen before in human history. Changes in international politics, notably the expansion and demise of global colonial empires, were almost as great. A 177-year period in medieval Europe, imperial China, the Mayan city-state system, or the steppes of Central Asia would most certainly show far greater behavioral regularity.

Unfortunately, good data for these periods is difficult to acquire, and there is an additional implicit assumption in most contemporary international relations research that one should seek to develop theories relevant to the modern system, not for Philip II of France or Genghis Khan. But with a bit of rereading of the original theories (and possibly some interpolation as to what their authors meant to say based on the choice of case studies), and more importantly, giving up the exceedingly ambitious objective of having a single “law” that covers an extraordinarily heterogeneous sample—black powder muskets and smart bombs; nineteenth century Central American dictatorships and the European Union—one may yet find strong statistical regularities.

Subsetting is effectively what research on the democratic peace has done with great success. By restricting the applicability of the theory to democratic dyads, one effectively imposes a set of controls on space, time, and other circumstances implicitly required of the theory. When the cases are conditioned on these restrictions, a very nice empirical result emerges. Subsetting, while not used in most statistical studies of war to date, is consistent with a behavioralist approach, and hardly a retreat to either historical uniqueness or postmodern nihilism.

The Behavioral Origins of War is a tremendously important work, but it brings a message—there is no there here—that many well-established research traditions will not want to hear. Time will tell whether it is treated as the sympathetic wake-up call its authors intend, or die a death by a thousand critical cuts from entrenched interests.