Tolstoy loves Brendan more than he loves Doron
Brendan McKay

In July 1997, I posted on the internet the first of several "codes experiments" that appear to demonstrate remarkable information hidden in the Hebrew translation of War and Peace. Of course, I don't really believe in such hidden codes, and my page clearly stated that the demonstrated convergences occurred entirely by chance. The purpose of the page was to caution everyone against accepting Doron Witztum's "experiments" in the Book of Genesis.

Now Witztum, the Guru of the Codes, has published a "refutation" of my experiment, proving that all those nice patterns occurred entirely by chance. Since that is exactly what I claimed from the very beginning, it is cause for some hilarity.

Of course, if there is anyone who is qualified to identify problems with my experiment, it is Doron Witztum. After all, every trick that I used was invented by him.

The purpose of the present document is not to defend my experiment, though I might not be able to resist showing some flaws in Witztum's refutation. I am responding because parts of Witztum's article clearly show him to be a fraud. That information is not known as widely as it should be.

We will use Michigan-Clairmont encoding of Hebrew letters. My original paper and Witztum's reply (transliterated) are available if readers wish to see their full text.

Let's go.

Note how Witztum carefully avoids mentioning that my article itself states that the events displayed occurred by chance. He wants to trick his readers into thinking he is showing something contrary to what I wrote, not something that agrees with what I wrote. Let me quote myself:

The lesson to be drawn from this paper is clear enough. Anyone with the skill and the perseverance can make ELS experiments that seem to show remarkable results. In this paper we found a significance level well below 1/1000 from a single name and a single date.   Did it happen by chance? Yes!

In true cult-leader style, Witztum is laying down "the rules". No matter that he doesn't have any relevant qualifications. No matter that his work is riddled with the most elementary errors.

The Method of Measurement

Lest anyone be misled, the "alteration" Witztum mentions here is not in his preprint. So why is he mentioning it? Anyone who follows Witztum's work will know the answer immediately: because it is to his advantage! Since he is going to apply "The Correct Method" to my data and wants a bad result, we can bet our life savings that his "alteration" will make the result worse. Sure enough, it does -- by a factor of 4 or 5.

The pattern is always the same. Witztum's ability to make a priori decisions in his favour is simply mind-boggling. A true prophet in our midst!

Charismatic words

Mr Witztum is deliberately distorting what I wrote. It is not some of the names, it is the set of names as a whole that have many more ELSs than PLSs. The difference is very important, as we shall see in a moment.

Mr Witztum is fully aware that the effects do not cancel out. The set of names is charismatic on average. More on this below.

Aside: The test specified by Persi Diaconis

Is it really possible that six months or more after both Prof Rips and Prof Aumann admitted that Diaconis' test was not used, Doron Witztum still doesn't know? I don't think it is possible. Is it really possible he thinks I won't catch him in this lie? What sort of mind is this we are dealing with?

Witztum is consistent in this deception. For example, in a Dec 1997 letter to the Israeli magazine Galileo he wrote:

After the great success of the second list's measurement, Prof. Diaconis suggested we use a new method of measurement on the second list. We did so, and the surprising results of our experiment are brought at the beginning of our letter.

Let's set the record straight on this. An exchange of letters between Persi Diaconis (representing the journal to which the paper had been submitted) and Robert Aumann (representing Witztum and Rips) took place in 1990 in order to establish the rules for the experiment. The method finally agreed to was described in a letter from Diaconis to Aumann on September 5. We have made a scan of the letter available with the kind permission of Professor Diaconis.

Prof Diaconis's test required calculation of an aggregate distance from each set of names to each set of dates. This is mathematically incompatible with the test actually done. There is no 32 x 32 table of distances in the WRR method. Instead they performed a different test - one that could easily have been predicted to have a great chance of success given the tests they had already done.

Prof Diaconis's 32 x 32 table can be calculated from the WRR distances in a variety of ways. One, which he suggested himself in an earlier letter, is to take the minimum distance. Another, suggested by him later, is to take the average. Both of these options make the result hundreds of times worse. So much worse that, according to my calculations, they would not have achieved the criterion for success indicated in Diaconis' letter.

Another little-known aspect of the history is that Diaconis asked them around the same time to prepare a third list of rabbis. He was concerned by the fact that the second list had already been in existence for three years. With the vigorous support of a senior mathematician who should have known better, they adamantly refused. To his later regret, Diaconis gave in. The rest is history.

One of the reasons given for not preparing a third sample is that less famous rabbis are not so likely to be "encoded". This argument totally ignores the fact that the second list of rabbis gave better results than the first list. In fact, statistical analysis shows that the individual performance of each rabbi has very low correlation with the size of the entry in the encyclopedia. If there is a trend at all, it is towards less famous rabbis performing better.

Back to charismatic names

Dr McKay was right. As mentioned above, the names of the rabbis are charismatic on average. This implies that the more distances that are measured from those names, the better the score will be (on average). It so happens that the permutation that matches each rabbi to his own dates produces more word-pairs and so more distances than over 98% of random permutations. Therefore it has some advantage. This statement has the non-negotiable status of a mathematical theorem.

This property of being charismatic on average is in fact the primary reason why the original (totally invalid) probability P2 gave a result much better than the WRR permutation test gives.

So why didn't you use it?

Here we see two more deceptions.

The first deception is in Witztum's comparison of the two computations. Has he forgotten that he already claimed that the "real" ranking of P4 is 59 in 100 million? He said so himself in his Hebrew preprint cited above. That's better than 1 in a million, not worse. To make the comparison more accurately, I computed both scores using the same 100 million permutations for each. The original method achieved a rank of 55, and the new method achieved a rank of 510. In other words, the change that Witztum claims as an improvement actually makes the result worse by a factor of 9.

The second deception (though to be fair it might be due to Witztum's ignorance of statistics) is the fallacy that one can measure the probability of an event just by looking more closely at that event. Wrong, one must examine the probability space that the event lies in. [Readers who don't know what I'm talking about can take solace from the knowledge that Mr Witztum probably doesn't either.]

Witztum's test permutes the letters of the words on the right. Mine permutes the letters of the words on the left. My weakest students could see the symmetry in that. They are mathematically the same, and if there was a fallacy in my calculation so there is in his. Somehow, this "scientist" failed to notice this obvious fact.

One of the ways of writing my year of birth ($NTHT$YB) is extremely anti-charismatic. So much so that its distance to almost 1/3 of all words is exactly 1 (the largest possible). Obviously, permuting its letters provides him with precisely the same type of advantage that he claims I had from permuting the letters of "Brendan".

Beyond this argument, there is the question of why it is an error to not allow for charisma. That can only be claimed on the basis of a thorough understanding of what the codes phenomenon is. Witztum thinks he knows the mind of the creator of the codes. [Although, since Witztum himself created the codes, I guess he is right. Damn, foiled again.]

More seriously, even if charisma is the entire explanation for the result of my experiment, Witztum has not given any reason why my name should be more charismatic in War and Peace than any permutations of its letters. Why aren't I entitled to claim that Tolstoy loves me on the basis of that?

"Dr McKay" is not very charismatic anyway.

We can hardly wait.

The mathematical expression "more stable" has no evident meaning in this context. Perhaps what it means is "gives a result 351 times better" (which indeed it does for the overall measure used then). Don't expect Doron Witztum to tell you that.

From this statement we can positively conclude that Witztum has found some examples where BEST does better than TOT. It is quite impossible that he would have used it otherwise. However, from the last sentence we can also infer that he is painfully aware that it doesn't work at all well on some earlier experiments he adjusted for TOT. So he has to declare it "an additional or complementary measure" in an attempt to have it both ways at once. The argument is like that over whether swishing the tea-leaves in the clockwise or anti-clockwise direction makes the fortune-telling more accurate.

It is also impossible that he would have mentioned it here unless it did worse on my data.

Selection of the headings

After all that mucking around, Witztum is now finally getting to the real issue. As he knows very well, having done it many times himself, selection of the data is the most important ingredient in a well-cooked experiment. Rather than bother with a line-by-line response, I will quote my own description verbatim from a private letter I wrote to a fellow codes-skeptic on September 10, 1997.

The honest truth is that most of the 1/5400 is just good luck.

Let me consider what choices I had. I could have used a different text, but War and Peace is the only non-Bible Hebrew text I have and was also the one used by WRR. I could have used different dates, such as only day+month like WRR instead of the year as well. Whatever I chose, I had no real choice in the forms of dates. I could have used a different method of analysis, but as I showed it doesn't seem to matter terribly. Probably the only place where I deliberately made a choice in my favour was the spelling MQYY of "McKay". I could have used MKQY or MQQY, but would not have been able to write an excuse like I could for MQYY. Actually MKQY works quite a lot better than chance too, but not as well as MQYY. I could have used multiple spellings, but decided not to. Certainly the freedom I had was very limited.

I think that the real choice might have been to do the experiment at all. I found that BRNDN was good against my date of birth some time ago, while looking for an example for Drosnin. Only much later did I happen to try DRMQYY, after hearing that Maariv had spelled it that way. If it didn't work I would have forgotten about it and some other day found something different that worked.

To understand this, it is necessary to know that at the time I didn't know of any ways to write "McKay" except the three mentioned. If I knew of others I would have tried them too, but in fact I didn't. I also did not know of the WRR method for "header samples", as it was only available in Hebrew at that time. It is interesting (but not really surprising to a mathematician) that the method I invented was nearly the same.

My letter also makes reference to other methods of analysis. Readers may well wonder why Witztum is studiously avoiding the fact that I used two completely different methods of analysis in my paper that both gave good results. He sure would like to "disprove" the other method too, but the poor guy can't. You see, the other method was almost identical to one proposed by Prof Rips. Funny but true.

I knew of that scurrilous pack of filthy lies, yes, but I never thought of asking how it spelt my name. If I had, all I would have had to do was to write "the first time my name appeared in the Israeli press" instead of "the only time". My story would have been saved. Big deal. (Remember that the story only needs to be Witztum-quality, which is laughably easy.)

For people who can't read it, the dedication expresses my disgust at the author of the article, who hid behind a cowardly fake name. A brave and honest Haredi friend was labelled a "traitor" for daring to question Witztum and his codes. It was not an isolated example, and shows very clearly what the codes cult is like.

Recently I heard that Mishpacha had changed ownership. Let's hope that the new owners have the will to drag it out of the gutter.

The Selection of the Expressions in the List

Witztum is quite correct, though he might have mentioned that I put "Canberra" (and "Australia") not in the same list, but in a separate list. It is invalid, all right, so why does Witztum do it? The fact is that the practice of tuning a list by adding extraneous words is an old Witztum stunt. An example is discussed in my paper on his Auschwitz experiment.


Save us.