Stat Fails collects documented cases where a statistical error had real consequences, with the reasoning worked through. References at the end.
In October 1995, the UK Committee on Safety of Medicines sent an urgent warning to doctors and the press: the newer, third-generation contraceptive pills doubled the risk of a blood clot compared with the older ones. The message that reached the public was three words long.
Twice the risk.
Women stopped taking the pill in droves. But here is the number the warning left out — the one that changes everything:
1 in 7,000clots per year, old pill
2 in 7,000clots per year, new pill
+1 in 7,000the entire real increase
That is the doubling: one extra clot for every 7,000 women. And clots are usually treatable, so the extra deaths were a tiny fraction of even that.
The panic had a body count of its own. Women who stopped the pill had unplanned pregnancies — and pregnancy carries a higher clot risk than any pill. In the year that followed, pill use among under-16s fell from 40% to 27%, and the scare is estimated to have caused around 13,000 extra abortions in England and Wales — thousands of women pushed into a hard decision to dodge a risk that was barely there.
The mechanism in one line: a large relative increase on top of a tiny baseline is still a tiny increase. Doubling almost nothing gives almost nothing.
This is the relative-versus-absolute trap in its most costly form. We built an interactive version where you can pull the two risks apart and watch the relative and absolute numbers diverge:
When you hear that something doubles a risk, ask the next question before you react: doubles it from what to what? A relative change means nothing until you know the absolute numbers it sits on.
The full breakdown, in plain English
The quick version above is enough to change how you read a health headline. This is the same story, slowly, with the numbers drawn out.
Two ways to say the same thing
A relative risk compares two risks as a ratio: twice as likely, 50% more, half as much. It hides the starting point. An absolute risk is the plain number: so many per hundred, or per thousand, or per hundred thousand. The 1995 warning was true in relative terms and almost meaningless in absolute ones, and only the relative half was printed.
The same fact, drawn twice. On the left, the doubling that made headlines. On the right, the actual rates: 15 and 30 clots per 100,000 women per year.
The numbers in people
Percentages slide past the eye; people do not. So put it in natural frequencies.
Take 7,000 women on the old pill for a year: about 1 has a blood clot. Take 7,000 women on the new pill: about 2 do. The famous doubling is 1 extra clot per 7,000 women — roughly 15 versus 30 per 100,000 per year. Most clots are treated and survived, so the extra risk of dying was smaller still.
The part that makes it tragic
A risk this small only matters next to what you compare it with. Set the pills beside the alternative that frightened women adopted — pregnancy — and the picture inverts.
Blood-clot risk per 100,000 women per year. The gap between the two pills is real but small; pregnancy carries a higher risk than either. Many women fled toward the larger risk.
The documented cost
The consequences were counted. Furedi found pill use among under-16s dropping from 40% to 27% in a single year, a sharp rise in unintended pregnancies among younger women, and a bill to the NHS of about £21 million in maternity care and at least £46 million in abortion provision; widely-cited estimates put the extra abortions at around 13,000 in England and Wales. Thousands of lives disrupted to avoid an extra clot in a few. The relative figure travelled the world in a headline; the absolute figure that would have calmed everyone was one sentence long and did not.
The twist: the doubling may not even have been real
Here is the sharpest part. If the third-generation pill truly doubled clot risk, then when women abandoned it — use fell from 53% to 14% — clot rates should have dropped. Farmer and colleagues looked in the UK GP records and found no change at all: the same rate of clots before and after the scare. The regulator itself later revised the risk down from twofold to about 1.7. So women fled a risk that was tiny in absolute terms and, on this evidence, may have been smaller than advertised in relative terms too.
What to carry out of this
A relative risk with no absolute number attached is not information, it is a mood. Before you change what you do, ask for the two plain rates it was built from — and, when a treatment or a test is involved, run them through the numbers yourself.
Sources
Furedi, A. (1999). Social consequences. The public health implications of the 1995 ‘pill scare’. Human Reproduction Update, 5(6), 621–626. https://doi.org/10.1093/humupd/5.6.621
Gigerenzer, G., et al. (2007). Helping Doctors and Patients Make Sense of Health Statistics. Psychological science in the public interest, 8(2), 53–96. https://doi.org/10.1111/j.1539-6053.2008.00033.x
Farmer, R.D., Williams, T.J., Simpson, E.L., & Nightingale, A.L. (2000). Effect of the 1995 pill scare on rates of venous thromboembolism among women taking combined oral contraceptives: analysis of the General Practice Research Database. BMJ, 321(7259), 477–479. https://doi.org/10.1136/bmj.321.7259.477
Stat Exam Pro drills relative versus absolute risk, and the rest of medical statistics, as exam questions with worked answers, in English and French.
