[Probabilities cannot be calculated retrospectively--not even in the courtroom].

Chance events are part of everyday life, but coincidence of diseases often raises suspicions about hidden causes, for example when power lines are blamed for the geographical clustering of cancer. Recently, criminal procedures in the Netherlands have revolved around the question of whether statistical 'predictions' are a valid reason to hold a hospital nurse accountable for the occurrence of excess deaths during her duty hours, or a kindergarten employee for unexplained respiratory problems in several infants. In both cases, the appeals court judges did not accept the statistical 'argument' in the absence of other evidence. In the UK, however, Sally Clark's initial life sentence for the double murder of her 2 babies was largely based on 'probabilities in retrospect', put forward by the paediatrician Sir Roy Meadow as an expert witness. 4 years later she was acquitted, whereas Meadow was struck off the medical register on a charge of professional misconduct. There is no Bayesian or other mathematical solution to the problem of chance events. Only the detection of causal factors that are plausible and supported by new evidence can help to reinterpret coincidences as relationships. Scrupulous reasoning about probabilities is required, not only of physicians but also of judges and politicians.