Sunday, March 04, 2007

Tomb Statistics: Feuerverger Responds

Andrey Feuerverger, who was the statistical consultant on the Jesus Family Tomb programme, has responded to critics on his homepage here:

The Tomb Computation

The latest version of this is dated today, 4 March. Feuerverger's response confirms my concerns, earlier mentioned, about the assumptions that went into the statistical calculations. I was concerned especially about the way that "Mariamne Mara" and "Judas son of Jesus" were being dealt with. It seemed to be that "Mariamne Mara", far from being treated as a neutral or a non-match element, was actually being used as the "Ringo" piece of data, i.e. the really distinctive name that was clinching the identification, something that is highly problematic for reasons stated here and in other places on multiple occasions. And "Judas son of Jesus" was regarded simply as a neutral piece of data, not a non-match of such decisive importance that it should in fact have been radically detracting from the overall probability. In the light of these concerns, now note the following important elements in Feuerverger's response. He rightly makes clear that: "The results of any such computations are highly dependent on the assumptions that enter into it" and lists as one important assumption the following:
We assume that `Marianemou e Mara' is a singularly highly appropriate appellation for Mary Magdalene. Note that this important assumption is contentious and furthermore that statistically this assumption drives the outcome of the computations substantially.
Note too:
We assume that the presence of Matya does not invalidate the find but we assign no evidentiary value to it (other than factoring in its combinatorial role). We also assume that the Yehuda son of Yeshua ossuary does not invalidate the find but we ignore it in the computations. This last assumption is contentious.
Later, he also notes:
The apparent absence of `negatives' in the finding, i.e. of archeological details that would in and of themselves invalidate `the hypothesis' or that would appear to lessen its likelihood.
Well, Judas son of Jesus is just the kind of "negative" which, I have argued, should have seriously detracted from the alleged plausibility. The fact that it was not factored into the assumptions governing the calculations remains highly problematic.

As I see it, the problem with all of this goes back to the difficulty of Jacobovici mediating the discussion between archaeologists, biblical scholars and statisticians. The data that Feuerverger is working with is data that already has a particular spin on it, data that is cherry picked and manipulated. This is an unacceptable basis for a strong statistical analsysis.


JDT said...

Mark, I think you have misunderstood Feuerverger's point about depending on assumptions that Mariamenon=Mary Magdalene, etc. I just talked to him on the phone for over an hour on this and what I posted on my Blog yesterday and sent around to you and others is correct. This also came up on the Koppel interview. The numbers do not change one bit whether Mariamenon is Mary Magdalene or not. The name is given a rarity value based on its frequency. An entirely separate question is then possible, namely, is this an appropriate name for Mary Magdalene? That is really not a question for the statistician and the answer to that question does not effect the statistics. What Andre is saying is that if he, as a statistician, then applies his final numbers to a final question--what are the probabilities that the names in this tomb belong to the Jesus family, assuming the identifications are correct, he comes up with 1/600. But note, he does not need to ask that final question, but since he did, he of course based his conclusion about the identifications on what he was told. Hence all the confusion about that quote. It is the difference between onomastics and prosopography. He has expertise in the first, but not necessarily in the second, so to do the second, which he did not have to do, he relied on others. I hope this is clear.


james Tabor

Eric Rowe said...

This debate is getting somewhat frustrating, as the matter of pure statistics on its own is objective and shouldn't permit disagreements like what we're seeing. First, I don't really see Dr. Feuerverger directly respond to the qualifications that were demanded by the letter Dr. Goodacre made available here by D'Mello. And what Dr. Tabor has now posted only troubles me more.

Dr. Tabor, are you really still standing by the language that "the probabilities that the names in this tomb [I think you meant to say "do not"] belong to the Jesus family, assuming the identifications are correct, [are] 1/600"? This differs from the seemingly more accurate claim that the probability of this not being Jesus's family's tomb is 1/600 only if it is first assumed 100% certain that one of the known tombs from the time period was that of Jesus's family (per D'Mello's explanation). The difference between these claims is huge. And unless D'Mello is wrong in his letter, then your repetition of this claim is indefensible.

Randy Ingermanson said...

I have a Ph.D. in theoretical physics from UC Berkeley and have worked as a computational physicist for 18 years. I also wrote one of the books debunking the alleged "Bible code" years ago. So I have substantial experience in evaluating the probabilities of alleged "rare and amazing word combinations."

The same problem that came up in the flawed "Bible code" calculations years ago is coming up here: people are insisting on using low-probability spellings of names as if those were the only names that would have supported the hypothesis that this tomb is related to Jesus.

The fact is that ANY variant form of "Mary" would have been accepted as evidence for a possible member of the inner circle of Jesus. It is irrelevant that "Mariamenou" is very rare, because Simcha Jacobovici would have been JUST AS EXCITED to find a plain old "Mariam" or a second copy of "Maria". There are a number of variant forms for Mary, and there is no reason to assume that any of them is more likely to be one of the Marys of the Jesus circle than any other.

Therefore, no matter which form you find, whether it's "Maria" or "Mariam" or Mariamne" or "Mariamenou", you still can only give that a probability factor that accounts for the whole class of variant forms. In Rachel Hachlili's article, this is assigned a value of 21.4%. In The Jesus Family Tomb, this number is given as a round 25%. Either number is fine. What is NOT valid is to use a value of 1/160! Doing so is a failure to count the many other possibilities that you would have been equally happy to find.

Similar comments apply to the probability assigned to the "Yose" inscription. We can't prove that any of the variant forms of "Yoseph" are more or less likely to be related to the Jesus circle, so if we find ANY of these forms, we have to use the probability for the entire class. Rachel Hachlili assigns this a value of 14%, and Feuerverger correctly uses this in computing the probability for "Jesus son of Joseph." But it is incorrect to lower this probability to 5% for the "Yose" inscription. You have to again use 14%.