How Did A Figure Of One In Seventy-Three Million Reach A Murder Jury?
How Did A Figure Of One In Seventy-Three Million Reach A Murder Jury?
Sally Clark's two sons died just over a year apart. Christopher was eleven weeks old when he died in December 1996. Harry was eight weeks old when he died in January 1998. At their mother's trial, jurors heard a calculation that appeared to give those losses a mathematical meaning. For a family with particular characteristics, the chance of two sudden infant deaths was put at roughly one in seventy-three million. The number became more famous than either child's name.
Clark was wrongly convicted of both murders in November 1999. The Court of Appeal quashed both convictions in January 2003. The route between those decisions involved two quite different failures. The statistic rested on an assumption for which adequate support had not been supplied. Meanwhile, laboratory results from Harry's post-mortem examination had failed to reach the defence. Understanding the case requires looking at what the jury heard and then at the evidence nobody gave it the chance to consider.
Two Deaths Became Evidence About Each Other
Christopher's death was initially understood as natural. After Harry died, the earlier death was re-examined. A second unexpected death in one household obviously creates reasons for clinicians to ask more questions. A family might share an exposure or inherited susceptibility. There could be a risk to another child. A criminal investigation might also be necessary. Deciding to investigate is a much smaller step than concluding that examination has identified a murderer.
By the time the prosecution opened its case, the two deaths could influence how every finding was read. An observation in Christopher's records that had seemed compatible with natural death might acquire a different significance once a doctor knew how Harry had died. A disputed finding in Harry's case might seem more persuasive because Christopher had also died. Neither change in interpretation necessarily involved the discovery of new physical evidence from the first child.
Consider how readily the same evidence can travel in both directions. If one death had independently been established as a homicide, an investigator might have reason to revisit the other. If both remained in dispute, however, allowing the first to establish the second and the second to establish the first would settle neither. The way out of that circle was to examine the medical evidence for each boy, including every test capable of supporting an alternative account.
Where The Number Came From
Professor Sir Roy Meadow gave evidence drawing on an estimate for a sudden infant death in a family with characteristics broadly resembling the Clarks'. The figure used was one in 8,543. This was a population estimate derived from a study. Nobody had measured a personal risk for Christopher or Harry that could be described with such precision. Selecting a subgroup was intended to take account of factors known to influence the frequency of sudden infant death.
The celebrated calculation was then remarkably short. Multiply 8,543 by itself and the result is 72,982,849. Rounded, that gives about seventy-three million. The multiplication is accurate. The argument depends on whether the same estimated probability could properly be applied twice in that way. A calculator can confirm the product to the last digit without telling its user anything about the validity of the assumption.
Multiplying two probabilities in this manner is familiar with independent events. If a particular outcome has a one-in-six chance on a fair die, the chance of obtaining it twice in two independent throws is one in thirty-six. The die has no family history. One throw does not reveal that the second throw comes from a household with an unusual but previously undetected characteristic. The relationship between siblings is what makes the apparently simple analogy unsafe.
To estimate the chance of both events, the relevant second probability is the chance of a second death given that a first death has occurred in that family. Replacing it with the risk for a family chosen from the population requires evidence that knowing about the first death supplies no relevant information. That was precisely what the calculation assumed. It was not something demonstrated by the existence of the original one-in-8,543 estimate.
How Family Differences Alter The Arithmetic
Imagine two households that fall in the same broad research category. They may share certain visible characteristics and still differ in ways the study has not captured. If an unmeasured characteristic raises the risk for more than one child, a family that has experienced one death may be more likely to possess it. Observing the first loss changes what an investigator reasonably believes about the risk surrounding the second.
It is unnecessary to prove that the Clark family carried a particular genetic variant to make that objection. The statistical question comes before any such diagnosis. Was there enough evidence to assume independence? Without it, the multiplication lacked the scientific foundation needed for presentation as a probability applicable to these deaths. The Royal Statistical Society raised this objection publicly. It was an objection to the model being used, rather than a dispute over the arithmetic.
There is another difficulty with the appearance of precision. A study can estimate the rate of an event in a group of families only from the people observed, the way cases were defined and the uncertainty in those observations. A rounded starting estimate cannot become a highly precise individual prediction by being multiplied by itself. Printing an eight-digit result in a calculation risks suggesting levels of accuracy the underlying observations could never support.
Even an excellent study of sudden infant death would have had a limited remit. Its purpose might be to identify risk factors, guide prevention and direct further research. Its researchers would not have observed a sample of mothers charged in circumstances exactly matching Clark's and counted which were innocent. Once a population estimate is taken into a different setting, the person using it must say what connects the study's question to the one the jury must answer.
The More Serious Error A Listener Could Make
Suppose, for the sake of argument, that researchers had established an entirely reliable probability for two natural deaths in the same household. It would still be wrong to read that number as the probability that a mother accused in a particular case was innocent. The first statement begins by specifying a situation and asks how often a series of events arises. The second begins after those events have been observed and asks which explanation accounts for them.
A simple test is to notice what is absent from the multiplication. There is no estimate of the frequency of two murders by a parent, no allowance for other possible combinations of causes and no evaluation of the medical evidence from either child. A number calculated without any of those ingredients cannot weigh up two alleged homicides against all reasonable alternatives. Yet heard in a murder trial, an immense denominator invites exactly that interpretation.
A jury does not need a mathematical formula for reasonable doubt. It does need to know when a witness has supplied a number that answers a quite different question. The practical danger was especially acute here because the alleged events had already selected the family for attention. Begin with a vast population and an unusual combination of events will eventually lead investigators to an unusual household. Its appearance in court is not a second independent occurrence of the same coincidence.
This is sometimes called the prosecutor's fallacy. The phrase describes faulty reasoning, wherever it occurs. It cannot establish what individual jurors thought in retirement. No reliable reconstruction is available in which a historian removes the number, asks the same jury to deliberate again and discovers how the votes would have changed. What can be assessed is the risk of misunderstanding when evidence is presented in a form so readily confused with the odds of innocence.
What The Statisticians Asked The Lord Chancellor To Consider
In January 2002 the president of the Royal Statistical Society wrote to the Lord Chancellor. The date is revealing. The first appeal had already failed and the second had yet to succeed. Statisticians were asking the justice system to examine the way numerical evidence had been presented while the legal outcome remained in dispute. Their argument survives in a document that can be read separately from the later appeal judgment.
One way to see the difference between the two objections is to imagine making only one correction. Supply excellent data about recurrence within families and a listener might still mistake the risk of two deaths for the probability of innocence. Give the jury a perfect explanation of that logical error and the original multiplication might still be scientifically unsupported. Repairing how a number is communicated does not validate how it was obtained. Improving how it is obtained does not guarantee a jury understands its limited purpose.
For a lawyer preparing to question an expert, the distinction produces two manageable questions. What evidence justifies the probability used for the second child? What proposition should jurors draw from the result if they accept it? A witness who can answer one has not necessarily answered the other. Both could have been asked without requiring any juror to make a fresh estimate of the frequency of murder in British households.
The Medical Results Outside The Courtroom
Harry's post-mortem examination produced microbiological results. Samples had been sent for testing, and results indicating the presence of Staphylococcus aureus were available. Of particular concern in the eventual appeal was the evidence relating to cerebrospinal fluid. The defence did not have the material at the trial. Experts called to consider the cause of death had therefore worked without information that might have directed them towards an infection.
Laboratory findings require interpretation. Bacteria found after death do not automatically establish that infection killed a child. Experts may have to consider where samples came from, how they were taken, the possibility of contamination and whether results from different sites fit a coherent clinical account. Those questions explain why disclosure mattered. The qualified people who could ask them had to know the tests existed before they could offer an opinion on what the results meant.
The Criminal Cases Review Commission subsequently obtained expert evidence supporting an overwhelming staphylococcal infection at the time of Harry's death. Its published account identifies this as fresh evidence unavailable to the trial. It regarded both the medical and statistical evidence originally presented as misleading and the failure to disclose the bacteriological evidence as serious. The commission's role was to consider whether the case should be sent back to the Court of Appeal.
The distinction between a new test and a new interpretation is particularly important. The crucial opportunity had been available while the case was being prepared. Time had not suddenly produced a sample from Harry that could never have been examined in 1999. The later experts had access to results and information that the first jury's experts should have been able to consider. In that sense the decisive discovery was partly a discovery about the contents of the case itself.
Why A Finding In Harry's Case Affected Christopher's
A bacterium identified in a sample from one baby cannot establish what caused another baby's death more than a year earlier. The review commission made a narrower and more consequential point. Medical opinion about Christopher had been reconsidered in the light of Harry's death. Evidence that materially altered the possible explanation for Harry's death required a re-evaluation of the opinions formed about Christopher under the earlier assumption.
Think of the two counts as two files with references running between them. Correcting a page in Harry's file does not physically change a photograph or slide in Christopher's. It can invalidate an inference an expert has written beside that photograph or slide. To judge whether the opinion remains sound, somebody has to follow the reference back to its origin. It is insufficient to say that the newly discovered result carries Harry's name and so belongs in only one file.
The commission received Clark's application in November 2001. Her first appeal had failed. In March 2002 the commission referred the case back, having concluded that there was a real possibility the convictions would be found unsafe. Referral itself did not quash a conviction. It placed the evidence before a court with the power to do so. In January 2003 the Court of Appeal allowed the appeal and Clark was released.
The reasons delivered in April explain the consequences of the non-disclosure and the judges' concern about the statistical evidence. Reports that describe a conviction overturned by a better statistic miss the particular route to the decision. The successful appeal did not replace seventy-three million with a smaller figure and then draw a different numerical conclusion about guilt. The court examined whether verdicts reached without important medical evidence and amid seriously problematic expert testimony could stand.
What Disclosure Would Have Allowed
Had the results been available to the defence before trial, the first step might have been to ask a microbiologist to assess their significance. If that expert identified infection as a realistic possibility, the pathologists could have been asked to reconsider their views. Counsel could have tested prosecution opinions against it. Jurors could have heard the competing accounts from people answering questions about the same body of information.
None of those steps required an expert to promise that infection was the only conceivable explanation. A criminal defendant is entitled to challenge the prosecution's account even when medicine cannot supply a final diagnosis. The distinction mattered profoundly to a mother accused of killing two children. Requiring her to prove the exact natural mechanism of each death would have transferred a burden that belonged to the prosecution.
Disclosure enables disagreement as well as agreement. One expert might regard a bacterial culture as important, another might offer reasons for treating it cautiously, and a third might ask to inspect the original records. That is what testing evidence looks like. When the existence of a result is unknown to one side, apparent consensus may be the product of a shared omission. More witnesses repeating an assessment based on incomplete information cannot repair it.
There is an enduring practical question here for anyone reading a court report that describes a panel of distinguished experts. Did they each consider the relevant material, or did they inherit somebody else's selection of what mattered? In Clark's case the contents of a laboratory record changed the terms on which experts could sensibly disagree. The qualifications of the original witnesses had never guaranteed that the evidence placed before them was complete.
Why The Case Is Still Misremembered
The number is easy to print in a headline. The medical finding takes a sentence to identify and several more to explain. That difference in convenience has helped one error eclipse another. Yet a reader who remembers only the probability may come away believing that a faulty piece of mathematics was the sole obstacle between the jury and a clear medical account. In reality, the courtroom heard disputed medical opinions without crucial information for testing them.
Clark was a solicitor before the proceedings made her name synonymous with a miscarriage of justice. Christopher and Harry were infants with different histories and different medical records. Treating them as two interchangeable occurrences of an event was useful for producing a spectacular multiplication. It was a particularly poor way to discover what happened to either child.
The Court of Appeal could remove wrongful convictions. It could not reconvene the original trial with the missing results on the table, give the family back the years of imprisonment or answer every medical question about two sudden losses. The most important correction it made was nevertheless exact. Sally Clark was entitled to leave court with both murder convictions quashed. A statistic about families similar to hers had never been a verdict on her own.
