International Security

The Red Queen Goes to War

Aug 2 '11

“Now, here, you see,” said the Red Queen to Alice in Lewis Carroll’s Through the Looking Glass, “it takes all the running you can do, to keep in the same place. If you want to get somewhere else, you must run at least twice as fast as that!”

Apparently the same goes for would-be insurgents and terrorists, a new article in Science declares (“Patterns in Escalation in Insurgent and Terrorist Activity”). Neil Johnson, a physicist by training, and his wonderfully interdisciplinary coauthors–amazingly, five are undergraduates–argue that insurgents (“Red Queens,” in their language) are locked in two-way arms races with counterinsurgents (“Blue Kings”) that are marked by constant innovation and adaptation on both sides. Insurgent adaptation, in their view, approximates a power-law progress curve that, put simply, means that the timing between attacks decreases at a known rate as insurgents become more proficient at their craft. If Red Queen’s adaptation (or “running,” in keeping with Carroll’s imagery) does follow a stable distribution, then it becomes possible to estimate with reasonable confidence the escalation rate, timing, and location of fatal attacks against the Blue King’s men.

This, needless to say, would be a major coup.

Yet should we accept Johnson et al.’s conclusions? I would argue no, for three reasons.

1. Most importantly, the data used in this study are, for lack of a better word, problematic. These authors fit their power curve through four different datasets: IED attacks that killed at least one US soldier in Afghanistan (2001-10) and Iraq (2003-09), as documented by; 3143 attacks by 381 terrorist organizations during 1968-2008, as recorded from newspapers by the Memorial Institute for the Prevention of Terrorism (MIPT); and suicide bombing data for Hezbollah (1982-1985) and Pakistani militants (1995-2008), both from the University of Chicago’s Project on Security and Terrorism.

The majority of the article deals with data from Afghanistan and Iraq, so let’s concentrate our efforts there. The decision to focus solely on a subset of insurgent violence (IEDs) that kill US soldiers (really, a subset of a subset of attacks) introduces enormous selection bias into their analysis. They record 440 fatal IED attacks in Afghanistan, for example, during 2001-10, and yet internal International Security Assistance Force (ISAF) data reveals that there have been at least 21,000 IEDs detonated or discovered over this time period (a coverage rate of about 2%). Similarly, they track about 1200 fatal IED attacks in Iraq (2003-09); the known total of IEDs detonated or discovered is closer to 86,000, a coverage rate of about 1.4%.

Dropped, too, are all other forms of insurgent violence against the Blue King’s men (small arms, indirect fire, and the like), as are IEDs that don’t kill soldiers, IEDs that are discovered, and planned attacks that are not executed. Their analysis is also silent on insurgent violence against Afghan and Iraqi security forces, not to mention civilians, both of which are now bearing the brunt of insurgent attacks in Afghanistan.

In short, this amounts to fitting a power law curve through a soda-straw view of insurgent attacks in Afghanistan and Iraq. No theoretical reason is given for why we should care only about fatal IED attacks against US forces or why these fatal attacks, but not others, would be subject to power law distributions. It isn’t clear, either, why these attacks should be aggregated to the provincial level, as there’s so much within-province heterogeneity in the timing, location, and types of attacks in both Iraq and Afghanistan that they risk obscuring the more important district and village level trends.

2. The proposed model of a Red Queen-Blue King arms race has a decidedly old school flavor. Indeed, it harkens back to Lewis Fry Richardson’s classic work on how fatalities in conventional interstate wars appear to follow a power-law distribution. Yet this two-player game is curiously out-of-step with both current theorizing in political science (and economics) and the US military itself on the nature of a counterinsurgency war, which privileges not the arms race between opponents but the struggle for civilian hearts and minds. If we have a Blue King and a Red Queen, where are all the subjects of the realm? The role of the civilian population is entirely absent from this model yet arguably much of the contest between King and Queen is for control, if not loyalty, of these actors. Yes, all models abstract from reality, but if King’s and Queen’s tactical choices are conditional in part on (expected) civilian attitudes and behaviors, then it becomes difficult to assume that adaptation is random, that Kings and Queens cannot engage in tactical substitution, or that civilians can be safely excluded from the model.

3. Perhaps the most worrisome aspect of the article was its casual disregard for alternative explanations and, in particular, context-specific knowledge.

Our broad-brush theory does not require knowledge of specific adaptation or counteradaptation mechanisms, and hence bypasses issues such as changes in insurgent membership, technology, learning, or skill set, as well as a need to know the hearts and minds of local residents (p.83).

That sound you hear is a host of civil war scholars crying out in anguish.

Why not test the power-law argument against rival explanations directly? Unfortunately, the article’s only attempt to engage the now-voluminous literature in political science on insurgent violence consists of the use of Wikipedia’s page on the Afghan War (I’m not kidding) to identify “key events” that might also account for the power law’s observed trends. A visual inspection of two trend charts then leads the authors to reject at least six alternative explanations because they “represent a poor fit to…the progress curve” (p.82).

This may be a disciplinary divide, but I was expecting a more robust research design that would permit clear (or any) causal identification of how these variables (especially Blue King’s actions) condition insurgent attack propensity. Without such an effort, these scholars are left unable to deal with either plausible alternative explanations or a host of inferential threats (selection bias, endogeneity, etc).

Ultimately, the study leaves a mistaken impression that the complexity of an insurgency can be reduced to a simple power-law. Ironically, this study emerges at a time when scholars and the military alike have come to recognize the importance of modest generalizations derived from hard-won data, area knowledge, and clever research designs. It’d be a shame to trade these gains for the false hopes of a universal (power) law.

h/t to NPR and Ethan