Drew Conway pointed me to this:
The article entitled, “Bayesian Analysis for Intelligence: Some Focus on the Middle East,” was written by Nicholas Schweitzer . . . JIOX provides no information on the essay’s origins, but . . . it appears to be a declassified CIA piece written sometime in the 1970’s (note mentions of Presidents Asad and Sadat, and Prime Minister Rabin on page one). . . . Schweitzer concludes that in general the Bayesian technique was able to more quickly predict “non-events” (i.e., when no hostilities would occur among Middle Eastern nations) than analysts using only their expertise and intuitions. The research design included no baseline for comparison to an actual event; therefore, we are left wondering if the Bayesian technique described here would be able to predict when something will actually happen. Despite this obvious shortcoming, it is very encouraging to observe the level of sophistication being implemented by CIA analysts some thirty-odd years ago.
I actually participated a couple years ago in an (unclassified) meeting on Bayesian analysis for military intelligence, so I know that these ideas are still out there. My only comment, regarding the Bayesian issue per se, is that the key to good statistical methods is typically making use of relevant information; non-Bayesian methods can also be effective if they can be adapted to use the info that goes into a Bayesian procedure.
There was an early Bayesian text by Samuel A. Schmitt. He was coy about who he worked for but it is evident that it was a US intelligence [meaning secret government work] agency in the Washington area. Leibler (of Kullback-Leibler) worked in the same place, as I recall.
Even more obviously, the venerable Bayesian statistician I.J. Good worked in intelligence for many years, both during and after WWII, and latterly has made clear how Alan Turing (the Turing) used Bayesian ideas in codebreaking that he never published, although some were later published by others.
E.H. Simpson was part of the same group in WWII. He is now best known for Simpson's paradox (which was discovered earlier by Yule, and by Karl Pearson) and by ecologists for Simpson's index of diversity, which was not his at all but a rediscovery by Turing of a measure invented by Gini. This is the Gini measure also used in some flavours of CART. Some economists know it from yet another reinvention by Herfindahl.
Interesting article, thanks for posting.
The full reference is:
Schweitzer N. (1976), Bayesian Analysis for Intelligence: Some Focus on the Middle East. Studies in Intelligence 20(2), p.31-44.
Nick,
That's interesting. I recall reading I. J. Good writing about Turing and all that. My impression was always that this was counterproductive for Good's later career, in that everything was a letdown after working with Turing. Good's individual articles and books are interesting, but as a whole I feel he didn't make the larger contributions he could have. It almost seems that he was too sensible to get excited about the statistical theory of the 1950s and onward, but not connected enough to applications to do more in that direction.
I.J. [Jack] Good was born in 1916 and is apparently still active. Rice University Press has just put out a book collecting his columns for a journal. It is prefaced by appreciative pieces by Stephen Fienberg and others. Good has always had a highly individual style, writing hundreds of very short pieces, many of them deliberately facetious or wacky in whole or in part. That more than anything else has perhaps reduced his impact in some quarters. His first book does not seem to me to age very well — the field has mostly just moved in different directions — but his 1965 book on categorical probabilities repays close re-reading from time to time. (It's very short.) With Lindley among British-born Bayesians he bridged the generations between Harold Jeffreys and those currently leading the field. Several others who watch this blog should be able to comment with more authority.
In Intelligence for an Age of Terror: Gregory F. Treverton quoted an senior analyst from the Intelligence Community:"it's all Bayesian these days".
We're quite interested in the application of Bayesian Networks to link analysis in the context of mapping out relationships, as well as for Indications & Warning for intelligence estimates. For amplification refer to the classic: Anticipating Suprsise by Cynthia Grabo http://www.ndic.edu/press/pdf/5671.pdf
Feedback would be greatly appreciated.