Stat Fails
Documented cases where a statistical error had real consequences, with the reasoning worked through and a source behind every number.
The counsellors who were 100% sure
In 1998 a researcher posing as a low-risk client asked 20 German AIDS counselling centres whether a positive screening test could be a false alarm. Most said no, several quoted 100% certainty; the honest answer was about 50%. Base-rate neglect, documented on the professionals who ran the test every day.
Read the full storyThe pill scare that caused 13,000 abortions
In 1995 UK doctors were warned that third-generation pills doubled the risk of a blood clot. The absolute increase was one clot per 7,000 women, smaller than the clot risk of the pregnancies that followed. The scare is estimated to have caused about 13,000 extra abortions in England and Wales.
Read the full storyThe sepsis AI that missed two thirds of cases
A sepsis model built into Epic ran in hundreds of US hospitals, sold with an AUC of 0.76 to 0.83. An independent validation on 38,000 hospitalizations measured 0.63: two thirds of sepsis cases missed, alerts fired on 18% of all patients, most of them false.
Read the full storyThe 82% vs 44% that meant nothing
In a 2007 campaign ad, Rudy Giuliani claimed US prostate-cancer survival was 82% against Britain's 44%, proof, he said, that American care was better. But the death rates were nearly identical - about 26 versus 27 per 100,000 men. The survival gap was lead-time bias and overdiagnosis, not medicine.
Read the full storyStat Exam Pro drills the statistics behind these stories as exam-style questions with worked answers, in English and French.
Get it on iPhone