Stats Test

More than a few people have come out with withering critiques of Jonathan Tepperman's Newsweek article on why Obama should learn to love the bomb. Stanford's Martin Hellman, who is running a previously discussed project that applies risk analysis methodologies to nuclear issues, has chimed in with a straightforward statistical analysis of how long we can reasonably expect a global deterrence architecture to hold up sans incident. In a memo published yesterday, Hellman writes the following.

Statistics tells us that, to be 95% confident of our statements, we cannot project the last 64 years of nuclear non-use more than 31 years into the future. Even if one drops the required confidence level to 50%, that only increases the time horizon from 31 to 44 years. And, with the fate of the earth at stake, a higher confidence level would seem appropriate. If we want to be 99% confident about our statements, the 64 years of non-use that we have experienced cannot be used to justify a time horizon of even 14 years. Statistics does not rule out that we might survive significantly longer than these time horizons, but it does say that the data thus far cannot be used to justify such hopes with any degree of confidence.

Leaving aside the bit of hyperbole concerning earth's ultimate fate, it is worth noting that these are only statistical probabilities. It may take significantly longer than 31 years for the next nuclear weapons attack to occur, and it may never occur at all. However, it may also occur a lot sooner, especially in a future with more states acquiring nuclear weapons, which increases the probability that all varieties of potential uses - incidental or accidental, by states or by terrorists - will occur.

Of course, Tepperman's logic runs the other way, as he believes that more nuclear weapons will result in more robust deterrence and therefore less chance of a nuclear attack of any kind. This unfortunately conflates the mere presence of the weapons themselves with their intended effect of deterrence. However, it also true that the same mitigating political factors that undermine Tepperman's ability to draw direct causal links between nuclear weapons as objects and the psychological effect of deterrence may nonetheless, perhaps even serendipitously, serve to reduce the risk indicated by strict statistical analysis. But the "practice of deterrence," so to speak, has always been complicated, imperfect, subject to failure, and has had unintended consequences - for example, the dangerous Cold War legacy of loosely guarded weapons and material in the former Soviet Union and other states' ambitions, some of them realized, to get in on the deterrence business by acquiring nuclear weapons to offset military inferiority in addressing security dilemmas.

Hellman goes on to draw a frightening parallel between Tepperman's "nuclear optimism" - which might be called "deterrence optimism" - and the optimism of NASA scientists leading up to the failed Challenger launch in 1986.

It helps to consider related "space shuttle optimism" arguments that led to the loss of Challenger and her crew. The engineers who had designed the shuttle's booster engine tried to delay Challenger's final launch because the weather that morning was unusually cold, and previous cold weather launches had a higher incidence of partial "burn through" on O-rings designed to seal the booster. But those at NASA responsible for the launch decision suffered from the common misperception that the shuttle's prior 23 successful launches provided ample evidence that it was safe to proceed with launch number 24 . . . [but] even 23 perfect launches would not have provided sufficient evidence to confidently predict success for launch 24.

Of course, what constitutes "sufficient evidence" to "confidently" predict events is largely a political question. And statistical analysis does not tell us anything conclusive about when, why or how a world with a given number of nuclear weapons-wielding states may either avoid or succumb to nuclear attack. Nonetheless, nuclear weapons present high stakes issues and Hellman is correct in reminding us that it is entirely appropriate and prudent to check assertions against what we can and cannot know given statistical constraints on future telling.