Michael KreponExperts Predict 2010 Will Be Terrible

In June 2005, Senator Richard Lugar released the findings of a survey he commissioned of nonproliferation experts. Eighty-five of us responded. Here are some of our pessimistic findings:

Question 1: In your estimate, how many nations that do not currently possess a working nuclear weapon will be added to the nuclear weapons club during the next 5 years?

More than 78% of respondents agreed that one or two new nations would acquire nuclear weapons during the next five years. More than 89% agreed that between one and three new nuclear nations would emerge during this period.

Question 4: In your opinion, what is the probability (expressed as a percentage) of an attack involving a nuclear explosion occurring somewhere in the world in the next 5 years?

Almost 60 percent of respondents judged the risk of a nuclear attack during the next five years to be at least 10%. Almost a third of respondents thought the risk was 20% or more. Nine experts thought the risk was at least 50%.

Question 9: In your opinion, what is the probability (expressed as a percentage) of a major biological terrorist attack that inflicts numerous fatalities in the next 5 years?

The group judged a major biological attack to be slightly more likely than a nuclear attack. More than half of respondents saw the risk of a biological attack in the next five years as between 10% and 30%. Three respondents thought the risk was zero, while three others saw the risk as above 75%.

Question 11: In your opinion, what is the probability of a major chemical weapons terrorist attack that inflicts numerous fatalities in the next 5 years?

The risk of a major chemical attack was judged to be similar to the risk of a biological attack over the same period. As with a biological attack, about half of respondents estimated the risk of a deadly chemical attack over the next five years to be between 10% and 30%.

Question 13: In your opinion, what is the probability of a terrorist attack using a radiological dispersal device (dirty bomb) that affects a major portion of a city during the next 5 years?

In general, respondents judged the probability of a major radiological attack over the next five years to be greater than the probability of a biological, chemical, or nuclear attack over the same time period. The average and median responses (27.1% and 25%) were higher for a radiological attack than for the other three types of WMD attack. Even within the limited time span of five years, 82% said that there was at least a 10% chance of a radiological attack that affects a major portion of a city.

It has been four years since this survey was released, and we have somehow managed to avoid worst cases. I am lousy at statistics, but it would appear that, for the five-year estimates provided to Senator Lugar to have validity, 2010 will be a very gruesome year.

Truly bad news could happen tomorrow, but many yesterdays have passed without major incident. What accounts for the prevalence of proliferation pessimism? Richard Betts has a simple, but persuasive answer:

Those interested in the question [are] those inclined to be worried about it.


  1. AndyJ (History)

    let’s assume that the probabilities of a nuclear/biological/chemical/radiological attack are independent. using figures

    P(nuke) = 0.1
    P(bio) = 0.2
    P(chem) = 0.2
    P(rad) = 0.3

    then we can use the inclusion/exclusion rule to see that

    P(disaster in 5 years) = 0.6168

    and if we make another baseless assumption, that the average rate of attacks is 0.6168 per 5 years, then the distribution of time till the first attack, T ~ Exp(0.6168/5). so:

    P(attack this year | no attack in the last 4) = P(attack this year) = P(T < 1) = 0.1161

    of course the assumptions stated make such figures worthless – i did this more to demonstrate intuition isn’t much use when it comes to probability.

  2. Alex W. (History)

    An aside: One of the many good things that Michael Gordin does in his new book, Red Cloud at Dawn, is to show how the estimate of “five years” until the USSR had a bomb was prevalent through the late 1940s, but nobody advanced the counter. Thus is was “five years” in 1945, and “five years” in 1946, and “five years” in 1947, and so on, until 1949 when suddenly the Russians had the bomb “five years” before anyone expected them to.

    Personally, I’m unclear what “10% chance” really means in this kind of prediction. The use of probabilistic terms doesn’t strike me as the best way to conceptualize this kind of threat, at least not in this kind of vague way.

  3. VS (History)

    Those asked have applied for grants.

  4. MK (History)

    The five year window of maximum danger has been a constant of sorts. As I recall (and stand to be corrected), it was present in the analysis of NSC-68, Gaither, Team A/B, and most recently, the WMD Commission.

  5. John Field (History)

    Is it the case that a single variable – like pessimism – can explain much of the response weight?

    How about an eigenvector decomposition of the data to look for this sort of thing?

    Is the raw anonymized data available?

  6. Josh (History)

    As a famous expert allegedly once remarked, “It’s tough to make predictions, especially about the future.” But this does not explain the consistent lean toward pessimism. There, Betts may be on the money.

    Another explanation, though, is the difficulty of estimating very small probabilities. There is a general tendency to overestimate them. For this reason, there may be better ideas than just asking for the probability of each event.

  7. Stephen Schwartz (History)

    Perhaps we should also pay heed to the recent JASON report Rare Events on the inherent difficulties in forecasting such occurrences:

    “Predicting WMD-T [Weapons of Mass Destruction-Terrorism] rare events is obviously a desirable goal. This has been stated many times by DoD, DHS, the IC, academic researchers, and others. However, it is simply not possible to validate (evaluate) predictive models of rare events that have not occurred, and unvalidated models cannot be relied upon.”