Finding a pattern in data you’ve already seen isn’t prediction
The story goes like this.
A Texas sharpshooter fires a bunch of rounds at the side of a barn, then walks over and paints a target around wherever the bullet holes clustered thickest. “Look how accurate I am.”
That’s the fallacy.
Finding a pattern after the fact and pretending it was predicted. Treating whatever cluster shows up in existing data as if it confirmed something somebody was already looking for.
The Texas Sharpshooter fallacy is cherry picking’s cousin from 15.18, but with a specific twist.
Instead of selecting favorable data while ignoring unfavorable data, somebody finds clusters or patterns in random data and treats them as meaningful. Then acts as if those patterns were what they were looking for all along. They’re not picking data. They’re drawing significance around noise after they already know what the noise did.
This happens constantly in health scares.
Somebody notices that five people on the same street got cancer. That’s frightening. It feels like it can’t be a coincidence. There must be a cause – something in the water, a power line, a nearby facility. The logic feels airtight. The cluster is right there.
But here’s the part that trips up almost everybody.
If you have a large enough dataset, random clusters will appear. That’s what randomness actually looks like. It doesn’t distribute itself evenly. It clumps. Random distribution produces streaks, gaps, and dense patches that the human brain immediately reads as meaningful. The cluster might be entirely meaningful, and sometimes there really is contaminated groundwater. The fallacy is assuming the cluster proves causation without doing the investigation to rule out randomness first.
Investment advisors are particularly prone to this.
They can always find a pattern in past market data that would have predicted the outcome – after the outcome is known. Tested prospectively on new data, those patterns don’t hold. The chart-reading and the market-calling confidence that goes with it is mostly a story somebody painted around old bullet holes. Which is why most professional active managers underperform index funds over long stretches. They’re good at painting targets. They’re not actually better at shooting.
Political analysts do the same move with elections.
After a candidate wins, commentators identify the decisive factors with great confidence. But before the election, the same commentators often pointed to entirely different decisive factors. The explanation gets constructed around the result, which means it explains nothing. It’s just a story that fits the outcome now that the outcome is known.
This fallacy shows up in conspiracy thinking a lot too.
People find coincidences in dates, numbers, or names, and treat the clustering as proof of a hidden plan. Given enough events, coincidences will always appear. Given enough numbers, somebody will always be able to find a pattern. The pattern isn’t evidence of design. It’s evidence that the human brain is relentlessly good at finding patterns even when no pattern exists.
The antidote is prospective thinking.
What did somebody predict before the data came in? If the pattern only appeared after the outcome was known, it needs to be tested on new data before it’s trusted. A theory that perfectly explains the past but fails to predict the future isn’t a theory. It’s a description with good posture.
The useful question – did I find this pattern before I knew the result, or did I find it because I was looking for something that explained the result I already knew?
Those are very different discoveries.
Only one of them is actually a discovery.