Steelman
Texas sharpshooter
Also called clustering illusion (related), data dredging (related).
Painting the target around the bullet holes: finding a pattern in random data after the fact.
What it is
Picking out a cluster or similarity in data after seeing it and treating it as meaningful, while ignoring the differences and the randomness that produced it, like firing at a barn and then painting a target around the tightest group of holes.
The pattern
Look at lots of data. Find a cluster. Draw a boundary around it. Claim the cluster reveals a cause.
An example
Residents notice six cancer cases on one street and conclude that the nearby phone mast must be causing them.
Across thousands of streets, some will have clusters by chance. Epidemiologists check whether the rate is really higher than expected and whether exposure explains it. The cluster alone can’t.
Why it fails
Random processes produce clusters all the time. If you choose the pattern after seeing the data, you can always find one, but the “hit” was never predicted. A cluster can suggest a hypothesis, but that hypothesis must be tested on new data.
When it’s not a fallacy
Most lists skip this part. Calling a fair point a fallacy is a mistake too.
It can be fair when
- The cluster is used to form a hypothesis that’s then tested independently. In 19th-century London, John Snow noticed cholera cases clustered around a water pump, then tested the idea that cholera spread through water.
Try one
A question from the app. The wrong answers are its usual look-alikes.
Priya, on the fan forum: “The new album goes platinum. Calling it now. I logged everything about the band’s 14 albums: keys, tempos, track counts, cover colors, producers, release months. The four that went platinum all came out in October, and the new one drops in October.”
What’s the flaw?
Priya checked dozens of features across 14 albums, so one of them was bound to line up with the four hits by luck. A pattern picked after looking at the results isn’t evidence until it predicts something it wasn’t fitted to.
A fair reply
“Log enough things and one will match the hits by luck, so October has to predict something before it proves anything.”
Often confused with
- Cherry-picking
Cherry-picking drops inconvenient data. The Texas sharpshooter redraws the boundaries around whatever data happen to cluster.
- Correlation is not causation
The sharpshooter usually ends in a causal claim, but its specific error is choosing the pattern after seeing the data.
Where the name comes from
A sharpshooter is an expert marksman. The Texan in the joke fires at a barn, then paints a target around the tightest bunch of holes. The “clustering illusion” is a related trap: seeing meaningful clumps in random data. “Data dredging” scoops through piles of data, like dredging a riverbed, until something turns up.
A version of the story about a Texas marksman appeared in print in 1977, but the mathematician John Venn made the same point with a similar anecdote in 1866. The name was popularized by the surgeon and writer Atul Gawande in 1999.
Sources
- The Texas Sharpshooter Fallacy, The Fallacy Files
- Texas sharpshooter fallacy, Wikipedia
In the app
Practice spotting it on arguments that sound real.
Steelman has more examples of Texas sharpshooter, how to answer it in a real conversation, and practice rounds that mix it with its look-alikes, so you learn the difference, not just the name. Coming soon to iPhone and Android.
You’re on the list.
We’ll email you the day Steelman is out.
Thanks for offering to test. We’ll write before the beta with how to join.