In 1999, a British jury heard that the odds of two babies dying of natural causes in the same family were 1 in 73 million, and a mother went to prison for murder.
That number sounded like a mathematical verdict. It was not. Statisticians later called it false, and the reasoning behind it already had a name: the prosecutor’s fallacy.
The myth: a tiny chance of coincidence proves guilt
The intuition goes like this. If an innocent explanation is incredibly unlikely, then the guilty explanation must be almost certain.
William C. Thompson and Edward Schumann put a name on this trap in 1987. The prosecutor’s fallacy treats the probability of a chance match as if it were the probability of innocence. Those are two completely different questions.
A simple example shows the gap. Imagine a blood type shared by 10 percent of people, and a suspect who has it. A prosecutor might claim a 90 percent chance of guilt. But in a town of 1,000 people, 100 share that blood type. Without any other evidence, the suspect is just one of 100, roughly a 1 percent chance.
The case of Sally Clark
Sally Clark was a solicitor living in Wilmslow, Cheshire. Her first son, Christopher, died in December 1996 at 11 weeks old. Her second son, Harry, was found dead on January 26, 1998, aged 8 weeks. She was charged with two counts of murder.
At her trial at Chester Crown Court in October 1999, the paediatrician Sir Roy Meadow told the jury that the chance of two sudden infant deaths in an affluent family like hers was 1 in 73 million. He compared it to backing an 80 to 1 outsider in the Grand National and winning four years in a row.
On November 9, 1999, the jury convicted her by a 10 to 2 majority. She received a mandatory life sentence.
Two mistakes hidden in one number
Meadow started from an estimate of 1 in 8,543 for a single cot death in a family like the Clarks, then simply multiplied it by itself. That only works if the two deaths are completely independent, like two separate coin flips.
They are not. Unknown genetic or environmental factors can make a second death in the same family more likely, not less. Both babies were also boys, which raises the risk, a detail Meadow ignored. The professor Ray Hill later estimated that once these links are accounted for, a double natural death was 4.5 to 9 times more likely than a double murder.
The second mistake was the fallacy itself. Even a correct small number would not mean much on its own, because double murder by a mother is also extraordinarily rare. The real question is which of two rare events is more likely, and the 1 in 73 million figure says nothing about that. Press reports still treated it as the chance she was innocent.
On October 23, 2001, the Royal Statistical Society publicly warned about the “misuse of statistics in the courts”. In January 2002 it wrote to the Lord Chancellor that “the calculation leading to 1 in 73 million is false.”
How the conviction finally collapsed
A first appeal failed in October 2000. What broke the case was something else. The pathologist Alan Williams had withheld test results showing the bacterium Staphylococcus aureus in Harry’s body, including in the fluid around his brain.
The conviction was overturned in January 2003, after Clark had spent more than three years in prison. The appeal court said Meadow’s figure “should never have been before the jury” in the form it was presented, and that the Grand National comparison may have strongly influenced them.
Hundreds of other cases were reviewed. Angela Cannings and Donna Anthony had their convictions overturned, and Trupti Patel was acquitted, all cases in which Meadow had testified. Sally Clark died in 2007, aged 42. Her family said she never recovered.
The same trap is everywhere
The error is not limited to courtrooms. Picture a breathalyzer that never misses a drunk driver but wrongly flags 5 percent of sober ones, with 1 driver in 1,000 actually drunk. Most people guess a positive result means a 95 percent chance the driver is drunk. The real figure is about 2 percent. It is the same blind spot behind the 99 percent accurate medical test that is wrong most of the time.
Researchers have found one reliable fix: talk in counts, not percentages. Saying “1 driver out of 1,000” instead of “0.1 percent” helps people reason correctly, which is why groups like the Cochrane Collaboration recommend it. When one number stands between a person and a prison cell, asking “out of how many?” is not pedantry. It is the whole question.

