The percentage was very surprising: the Amsterdam notary Arnold van den Bergh would have betrayed Anne Frank’s hiding place to the Nazis with a probability of 85 percent. A ‘forensic statistician’ had calculated that, the international cold case team (CCT) reported on Sunday 16 January, which had been investigating the possible traitors of Anne Frank for five years.
Historians with knowledge of the persecution of the Jews rejected Van den Bergh’s designation because of speculative reasoning and missing sources. The claim that Van den Bergh had lists of addresses of people in hiding, collected by the Jewish Council, in particular met with criticism: there was no evidence whatsoever for the existence of such lists, according to Leiden historian Bart van der Boom, for example.
What kept ringing in newspapers and on social media was that 85 percent, a figure that, incidentally, is missing from the CCT book written by Rosemary Sullivan, Anne Frank’s Betrayal. It does contain the name of forensic analyst Frans Alkemade, who drew up the report from which the CCT had derived the 85 percent.
Alkemade appears to be unhappy with the way in which the percentage has been publicized. That number was marketed by the CCT and the media as an “absolute probability,” while he reported it as a “conditional probability.” “The difference between the two is not an insignificant subtlety,” says Alkemade. “The conditional statement”as this witness speaks the truth, and is the 85 percent chance that the suspect is guilty”, has a fundamentally different meaning than the absolute statement ‘The probability is 85 percent that the suspect is guilty’.”
Alkemade’s analysis in the Anne Frank case concerned the conditional probability that Van den Bergh was the informant, as the facts and insights provided to him by the CCT would be correct. Moreover, Alkemade had no way of independently checking all those facts and insights.
He says that the cold case team only approached him when their investigation was almost finished, on the advice of an employee of the cold case team of the The Hague police. Over the past ten years, Alkemade has regularly drawn up reports on the evaluation of evidence in criminal cases on behalf of courts, the public prosecutor and lawyers for suspects, using so-called Bayesian analyses. In addition to lectures at universities, he also teaches officers, lawyers and judges.
“The Anne Frank team asked me to also conduct such analyzes on the different scenarios they had drawn up about the possible traitor, in order to support their own conclusions,” says Alkemade, who is a physicist by training. “Historical investigations are different from criminal investigations, but the Bayesian methodology is useful for comparing the probabilities of the different traitor scenarios in this case. However, I have repeatedly informed the CCT that the results of my analyzes should always be presented in the right context and with great care. In a historical case like this you have less facts and information at your disposal than in a criminal case. And any additional evidence, incriminating or exculpatory, is usually lost, making the outcomes more uncertain.”
Weight of evidence
The most likely scenarios emerged from his analysis: Ans van Dijk, who was executed in 1948 for the betrayal of two hundred Jews; and the notary Arnold van den Bergh, who died in 1950, who had already come into the picture in the 1940s through an anonymous letter to Otto Frank, Anne’s father, in which this notary was identified as the traitor without proof. The ‘Van den Bergh hypothesis’ turned out to be the most probable after Alkemade’s calculation, but on the condition that the assumptions made by the CCT about the possibility, motive and means of passing on Anne Frank’s hiding place to the Nazis were indeed correct. would be. Depending on the evidential value he assigned to these and various additional assumptions, he arrived at a probability of between 50 and 95 percent that Van den Bergh had been the Nazi informant, with 85 percent being the best guess, he stated in his report.
That contains several warnings to the CCT about the scope of its analysis. Alkemade writes that he deliberately uses the term ‘informant’ and not ‘traitor’, because he also considers Nazi informants as victims instead of perpetrators. He also writes in the report that his calculation does not show that Van den Berg was actually the informant. “I don’t come to that conclusion, and neither should others. First, 85 percent don’t come close to beyond reasonable doubt. In addition, you have to realize that we are investigating events from 75 years ago, which means that the collection of data we have to work with is limited.”
Alkemade also points out in the report that even a 95 percent probability would be insufficient to identify a perpetrator. “If we were to sentence suspects from 95 percent probability, that could mean that 1 in 20 prisoners is innocent in the cell.” An unacceptably high number.
Despite Alkemade’s warnings in the report, ’85 percent’ took on a life of their own because of the oversimplified presentation of the CCT. Alkemade is now consulting with the CCT about a joint statement, in which that number is still nuanced and which makes it clear that archives that are inaccessible or unknown to the researchers may contain documents “that completely turn the matter over”. Pieter van Twisk, head of research at the CCT, said when asked that the CCT will respond to the criticism next week. It will also explain how to deal with Alkemade’s report.
Ronald Meester, professor of probability theory at the Free University, was very curious about the calculation of the 85 percent, but did not find that percentage in the book that he bought immediately after publication. He did read that the cold case team had consulted Alkemade, with whom he has regularly argued in recent years about his irresponsible wide use of Bayesian models in criminal cases. Meester agrees with Alkemade that a Bayesian model can be used well for analyzing scenarios in historical research. “That is not fundamentally different from criminal cases, which by definition are about something that happened in the past. And there too there can sometimes be a lot of time between the events and the criminal case, think of Nicky Verstappen, who died in 1998.”
Meester’s objection is that Alkemade assigns a number to a very complex situation, not only in the Anne Frank case, but also in his reports for criminal cases. “His analyzes concern the criminal case as a whole, to which he then attaches a number. That goes way too far and gives the wrong impression that you can do magic with Bayesian models.”
Meester points out that no judge has yet accepted Alkemade’s reports on full criminal cases, while the use of Bayesian models for the interpretation of evidence has been fully accepted for elements of criminal cases in the form of ‘plausibility quotients’ [zie kader].
Meester is supported by Marjan Sjerps, statistician at the Netherlands Forensic Institute (NFI) and professor by special appointment of forensic statistics at the University of Amsterdam. „What Alkemade does is described in literature’linear Bayes‘ because it oversimplifies reality. Better application of the Bayesian model makes clear how complex that reality is. You cannot reduce a complex lawsuit to a simple calculation.”
According to Sjerps, the Bayesian model is very suitable for inferring how strong evidence and especially combinations of different evidence are. Bayesian models have also proved very useful in visualizing common fallacies. These models also highlight the importance of formulating an alternative hypothesis to the prosecution’s hypothesis: to prevent tunnel vision.”
Frans Alkemade is not impressed by the criticism of his work. “In principle, I use the same mathematical methodology as the NFI in my analyses, but I also include ‘initial probabilities’ and other than purely forensic evidence. The NFI does not dare to do this because they believe that they will then be sitting in the judge’s seat too much. But I think my reports do help judges get a better sense of the weight of the different evidence in a criminal case. I also often see elements of my analyzes reflected in the statements. And of course I always present my calculations with the same perspectives as I did with the traitor scenarios.”