# Are there statistical lessons from the “Bible Code” episode

Although this question is somewhat subjective, I hope it qualifies as a good subjective question according to the faq guidelines. It is based on a question that Olle Häggström asked me a year ago and although I have some thoughts about it I do not have a definite answer and I would appreciate some help from others.

## Background:

A paper entitled "Equidistant letter sequences in the book of Genesis," by D. Witztum, E. Rips and Y. Rosenberg made the extraordinary claim that the Hebrew text of the Book of Genesis encodes events which did not occur until millennia after the text was written. The paper was published by "Statistical Science" in 1994 (Vol. 9 429-438), and was offered as a "challenging puzzle" whose solution may contribute to the field of statistics.

In reply, another paper entitled "Solving the Bible code puzzle" by B. McKay, D. Bar-Natan, M. Bar-Hillel and G. Kalai appeared in Statistical science in 1999 (Vol. 14 (1999) 150-173). The new paper argues that Witztum, Rips and Rosenberg's case is fatally defective, indeed that their result merely reflects on the choices made in designing their experiment and collecting the data for it. The paper presents extensive evidence in support of that conclusion.

(My own interests which are summarized in Section 8 of our paper are detailed in another technical report with Bar Hillel and Mckay entitled "The two famous rabbis experiments: how similar is too similar?" See also this site.)

## The questions:

Olle Häggström's specific question was:

"I once suggested that your paper might be useful in a statistics course on advanced undergraduate level, for the purpose of illustrating the pitfalls of data mining and related techniques. Would you agree?"

In addition to Olle's question let me ask a more general question.

Is there something related to statistics that we have learned, (including perhaps some interesting questions to ask) from the Bible Code episode.

Just to make it clear, my question is restricted to insights related to statistics and not to any other aspect of this episode.

-
this is an interesting subject. I am curious why you (McKay et al 1999) would choose 'War and Peace' as a control rather than, for example, random strings of letters (perhaps weighted by their observed frequencies). In other words, is it sufficient for the text to be sufficiently long, or does it have to be sufficiently long and comprehensible (or sufficiently long and of some literary value)? – David Jan 17 '11 at 15:33
David, the Choice of "War and Peace" as a control text (More precisely the beginning of the Hebrew translation of "War and Peace" of the same length as the Book of Genesis) was done by the original researchers. The story according to Aumann is this: When Bob Aumann who carefully followed the experiment told Kenneth Arrow about the marvelous findings in "Genesis", Arrow asked what about "War and Peace". Aumann then started reporting about the war and peace situation in Israel but it turned out that what Arrow asked about was if the same phenomenon cannot be found in "War and Peace". – Gil Kalai Jan 17 '11 at 16:57
The Bible code episode would be a good illustration of the strengths of the Bayesian view of probability. In particular, the Bayes factor $P(D|H)/P(D|not H)$ are insufficiently large given that we would assign a small prior probability to $P(H)$. (H being the hypothesis there exists some mechanism whereby world events are encoded in the Bible.) – charles.y.zheng Mar 9 '11 at 9:58
By the way, you are free to post your own answers. I'd be very interested, as you have presumably weathered a lot of analyses of the whole experience. – Iterator Aug 12 '11 at 1:29