IN THE BEGINNING WAS THE RABBIS EXPERIMENT...

The case for the Bible Code has been misrepresented from its very beginnings, in the famous "Rabbis Experiment" conducted by the research team of Doron Witztum, Eliyahu Rips and Yoav Rosenberg (to be referred to as "WRR" for the remainder of this writeup), who published their results in an article titled "Equidistant Letter Sequences in the Book of Genesis" in Statistical Science in 1994. This article became the foundation and inspiration for Michael Drosnin's research into what he calls the Bible Code.

Every believer in the Bible Code likes to flaunt the fact that the original article describing the Rabbis Experiment was published in a professional journal after peer review, as if that proved the truth of the whole story beyond any reasonable doubt. What these proselytizers never mention is that the authors' original work actually failed its peer review when it was submitted to the Proceedings of the National Academy of Sciences for the USA, a much more prestigious journal. After the authors ignored several requests from Academy members for more thorough statistical controls - the "review" part of peer review - the Academy rejected the paper. A revised and even more selective version of it was then published in Statistical Science. What kind of peer review it underwent there is unknown.

In the Rabbis Experiment, the researchers searched the Book of Genesis for ELS word pairs - specifically, the names and apellations, and birth OR death dates, of 34 famous rabbis (with the dates written in letters, following standard Hebrew convention). Unlike Michael Drosnin, who developed their surprising findings into an even more sensational book, they did not claim that similar sequences could not be found in any other book. Indeed, it should be immediately obvious that any sufficiently large set of letters or numbers would contain as many Equidistant Letter Sequences, even in word pairs, as one could wish for. What the researchers did claim was that there was an abnormally large number of these pairs in unusually close proximity in the text of Genesis. They determined that the "significance level" of these pairs' proximity was 1 in 60,000.

While they were preparing their findings for publication, WRR received a request for additional statistical tests from Persi Diaconis, then Professor of Statistics at Harvard University (remember the peer review?). Diaconis requested that they use a "randomly chosen cyclic shift" to alter the dates accompanying the rabbis' names, and check the resulting distances between names and dates against the researchers' original findings. He also requested that another test be performed, this time matching different people with their birth or death dates, as a statistical significance test (similar to a control group).

Although another colleague requested the same sort of shifted permutation test, the WRR researchers only performed one permutation test, shifting the dates searched for by +1 instead of altering them at random. They completely ignored the request for a "control group" significance test. It was then that their paper was rejected by 'Proceedings', and subsequently revised and published in Statistical Science.


METHODOLOGY FLAWS IN THE RABBIS EXPERIMENT:

In 1999, the work of the WRR team was rather thoroughly disproven by another team of researchers, Brendan McKay, Dror Bar-Natan, Maya Bar-Hillel, and Gil Kalai (henceforth "MBBK"), in an article which was also published in Statistical Science. Presumably, this debunking went through the exact same peer review that made the original Rabbis Experiment bulletproof - with the major difference that this team's finding were nowhere near as sensational, and much more parsimonious.

While I can't claim to understand all of the statistical methods involved in the rebuttal, what I do understand of it seems fairly clearly to show that the findings of the WRR team were greatly exaggerated through flawed (or intentionally skewed) methodology, a significant level of data corruption in the text of the Bible, and an eagerness to find a particular result. I will attempt to explain a few of the key points below.

The first major point of contention is that WRR's findings are significantly sensitive to a very small part of the data. As MBBK put it in their rebuttal, "If the 4 rabbis (out of 32) are removed, the overall 'significance level' jumps from 1 in 60,000 to an uninteresting 1 in 30. Historically speaking, these rabbis are not particularly important compared to the others."

Even worse, a few of the particular appellations (out of 102 that WRR searched for) are hugely influential, with one appellation contributing a factor of 10 to the result all by itself. Again, the appellations in this small group are no more important historically than any of the others.

This demonstrates that the results of the searches are extremely sensitive to seemingly minor elements of the experiment's design. With this in mind, certain aspects of that design deserve to be looked at very closely. The WBBK team then examined the dating system used by WRR, and it turns out that out of 8 possible formats for the dates searched for, WRR chose to use the three that were far more favourable for their test than the other 5. Had they consistently used the date formats traditional in Judaic studies, or the format used by their main source for data on the rabbis they chose, they would have found a "significantly lower" significance level.

Similar selection was used for the list of appellations searched for, with even stronger results. The list of traditional appellations for famous rabbis is quite extensive and follows no particular rules. Rabbis known by one name under a certain tradition are known by several others in different contexts. For example, "Rambam" is the most common Hebrew name - an acronym really - for Rabbi Moshe ben Maimon, better known in English as Maimonides. One rabbi on the list, the Maharit, has 17 different appellations. Of course, WRR did not use all the possible appellations. They selected roughly half of the possibilities, and when questioned about the methods of selection, were unable to supply any consistent reasoning.


CONCERNING "THE ORIGINAL HEBREW" OF THE MASORETIC TEXT:

The other major fault in WRR's and Drosnin's claims is that the text that they used in their searches is definitely not the original text of the Bible. Although much has been made of the "fact" that the amazing predictive ELSes can be found only in the original Hebrew text, the fact is that the original Hebrew text has been lost for thousands of years. Generation after generation of hand copying have caused numerous mutations to creep into the text. With the addition of a letter here, the deletion of a letter there, and occasional substitution of letters (Hebrew is resistant to this, but it does happen on occasion), the text changes gradually, eventually obliterating any secret code that might conceivably have been built into the original text. Thus the Koren Bible that WRR, Gans and Drosnin all used for their research is not the same as any of the numerous Genesis fragments found amongst the Dead Sea Scrolls, nor is it an exact copy of the Yemenite version or the Leningrad Codex (which is generally considered to be the oldest extant copy of the complete Bible).

When MBBK attempted to duplicate the original Rabbis Experiment using these other texts, the results were far less dramatic. In fact, it turns out that the Koren Bible that WRR used gives the best results possible for their experiment. And, although Drosnin and other Bible Code supporters claim that the Koren is as close to the original text as possible, this is somewhat less than a half-truth. The Koren Bible uses the Masoretic Text that is the standard, most widely accepted version of the Bible. However, every Jewish scholar knows that the Masoretic Text is not, letter for letter, the original text of the Bible. The Talmud contains numerous quotations from Biblical passages that are considerably different from the MT. These differences include everything from the addition and deletion of matres lectionis, to the possible deletion or insertion of whole phrases (in some versions, Genesis 1:7-8 contains the phrase "and God saw that it was good". In others, God just goes about his business).

A word about matres lectrionis: this refers to letters that are not phonetically required. Although Hebrew is considered a phonetic language, this is not 100% true. For example, the word Torah is generally written Tav-Vav-Resh-Hay, where the Vav produces an O sound. This is a mater lectionis, an early form of Hebrew vowel whose use has continued to this day. However, it has not always been used. Some instances of "Torah" in the Bible contain a Vav, some don't, and some did in early versions of the text and don't anymore. This is one of the most common routes for mutation in the text of the Bible.

Even worse, one passage in the Babylonian Talmud tells us that the "first scribes" found the middle letter of the Torah in Leviticus 11:42. The Koren and other Masoretic Text Bibles' middle letter is 4830 letters earlier, in Leviticus 8:28. So, either the first scribes and the authors of the Babylonian Talmud - whose interpretations are the very foundation of Jewish theology and practice - were completely mistaken about the length of the Torah, or major alterations were made to the Torah sometime before the compilation of the Masoretic Text.

What all of these mutations add up to is a text that is, although widely accepted, significantly different from the original text of the Bible. It may be close enough for government work. But close isn't nearly close enough when one is making as sensational a claim as Drosnin makes. Either the predictions are encoded into the exact, original text, or they are complete nonsense. To claim that the future mutations in the text were accounted for when designing the Code makes a mockery of the whole concept and violates parsimony.


CONCERNING THE LAW OF LARGE NUMBERS, OR HOW TO SELECT FOR SUCCESS:

It has been said that million to one odds happen eight times a day in New York City. This is something to keep in mind when looking at the coincidences found in the Bible Code. If you have a large enough set of numbers, you're almost surely going to find a large number of striking combinations. The secret is to pick the right words to search for, and disregard all findings that don't suit your purpose. (Start by ignoring all the incidences of "God is dead" and "there is no God", obviously.)

It may be that this sort of selection process enhanced the results of the original Rabbis Experiment. MBBK certainly seemed to think so. It cannot be denied that this is how Michael Drosnin wrote most of his book, combing the Bible for meaningful crossword puzzles and ignoring the ones that didn't work out to anything that he liked.

In fact, there are quite a few ELSes in the Bible that conflict with Drosnin's picture of an omniscient Creator. Using the same ELS technique to look for phrases of his own choosing, Dr. James D. Price has found several statements of heretical nature, including hundreds of incidences of "God is dead" and "Jehovah is dead" - strange statements for the author of the Bible, whoever he was, to intentionally encode in his text. Price also found multiple instances of "There is no God", "There is no Jehovah", "Jehovah is a liar", and other phrases that God would probably not appreciate.


"DICK WRITER PREDICTED DI'S DEATH!!!"

Finally, let's look at Michael Drosnin's claim that only the Torah - not the whole Bible, nor any other book - contains the interesting "crossword puzzles" that predict various events. It should be noted that WRR's claims were not nearly as sensational. They did say that there were more meaningful ELSes in the Bible than in any other book they had searched.

But such pedestrian claims would never have satisfied Michael Drosnin. He wasn't looking for a mere article on interesting statistical anomalies. He wanted a bestseller, and by God (or His ghost writers) he was going to have one. In the first Bible Code book Drosnin announced, "Consistently, the Bible Code brings together interlocking words that reveal related information. With Bill Clinton, President. With the Moon landing, spaceship and Apollo 11. With Hitler, Nazi. With Kennedy, Dallas. In experiment after experiment, the crossword puzzles were found only in the Bible. Not in War and Peace, not in any other book, and not in 10 million computer-generated test cases."

In an interview for Newsweek, printed in the June 9, 1997 issue, he went on to say, "when my critics find a message about the assassination of a prime minister encrypted in Moby Dick, I'll believe them." These words would come back to haunt him in the end, because the same ELS technique that he used has since revealed all sorts of messages in all sorts of books. For starters, Moby-Dick.

It turns out that Moby-Dick contains quite a few "astonishing" retrodictions just like Drosnin's. People whose assassinations Herman Melville predicted include: Indira Gandhi, Rene Moawad, Leon Trotsky, Martin Luther King, Austrian Chancellor Engelbert, Robert F. Kennedy (with the name Sirhan Sirhan and mention of sharpshooting), and John F. Kennedy. Even the accidental death of Diana, Princess of Wales, was foreseen by the whalewise and presumably omniscient Melville.

(Note that English is actually a much harder language to do this in than Hebrew, because of the vowels. Hebrew script normally uses only consonants, which can mean different things depending on the inflections used. For example, "wrote", "writing", "reporter", and the imperative "write!" are all written the same in Hebrew - Caph-Tav-Vav (without matres lectionis, of course). In an English text, we might find an ELS like "The writing will die on October 31," which would have to be discarded as obvious nonsense. The same ELS in Hebrew could be conveniently interpreted as "the reporter will die on October 31." This is no reflection on Drosnin's work, but it does perhaps demonstrate that Moby-Dick is an even more powerful prophetic work than the Bible.)


SOURCES AND INCRIMINATIONS

  • The MBBK analysis and refutation of the Rabbis Experiment can be found at
    http://cs.anu.edu.au/~bdm/dilugim/StatSci/
    At 45 pages, it's a rather hefty article, but well worth reading. My little excerpt of their findings is only the tip of the iceberg. For me, the points that I mentioned pretty much wrapped up the case against the Rabbis Experiment, and hence the Bible Code. But if you want more, there's plenty more.
  • Further refutations of Michael Drosnin's Snake Oil Code can be found all over the Web. There is actually a large group of people who continue to search for ELSes that will disprove Drosnin's ideas or make him look like an idiot. Personally, I find most of this ongoing research as silly as the Bible Code itself.
  • Million to one coincidences can be found eight times a day in New York City.