Texas Sharpshooter Fallacy: Drawing Bullseyes After the Bullets Have Landed

A Texan cowboy fires a dozen rounds at a barn wall, walks up to examine the scattered bullet holes, and draws a bullseye around the cluster where the most shots happened to land. "Perfect shot group," he announces. The joke captures something profound about a failure mode that distorts scientific research, drives medical panics, powers financial fraud, and saturates our daily reasoning. This is the Texas sharpshooter fallacy.

What the Fallacy Describes

The Texas sharpshooter fallacy occurs when someone identifies a pattern in data after examining the data, and then presents that pattern as if it had been predicted or as if it represents a meaningful signal rather than chance variation. The defining feature is the reversal of the proper order: the hypothesis is formed after the results are known, then retrofitted to explain them — giving the illusion of predictive accuracy.

In standard scientific reasoning, you form a hypothesis first, then collect data to test it. The hypothesis is at risk — the data could falsify it. In Texas sharpshooter reasoning, you look at the data first, identify whatever pattern stands out, and then declare that pattern significant. The "hypothesis" is never at risk, because it was designed to fit what you already found.

The fallacy belongs to a cluster of related errors involving selective attention to data: confirmation bias (noticing evidence that supports your beliefs), data dredging (exhaustively testing many hypotheses until one appears significant), and apophenia (perceiving meaningful patterns in random noise). What distinguishes the Texas sharpshooter is the specific temporal reversal — results first, hypothesis second — combined with the presentation of the outcome as meaningful discovery.

Cancer Clusters and the Fear of Patterns

In the 1980s, researchers and public health officials began noticing elevated rates of certain cancers in specific Texas counties near petrochemical facilities. Maps were drawn. Clusters were circled. Local newspapers ran alarming stories. Residents were frightened.

But the statistical analysis told a different story. Texas has 254 counties. When you track hundreds of different types of cancer across hundreds of geographic units over multiple years, you generate thousands of data points. By pure chance, some county-disease combinations will appear elevated above the expected rate. The question is not "did we find a cluster?" — in large enough datasets you will always find clusters — but "is this cluster more than you'd expect from chance?"

This is precisely the problem the Texas sharpshooter fallacy identifies. When you examine a map of disease rates and circle the areas that look elevated, you have drawn your bullseye after the fact. The circled area was selected precisely because it looked elevated — which guarantees it will look elevated. You have not predicted anything; you have described the data you used to define your target.

Subsequent systematic studies of many such claimed cancer clusters found that the overwhelming majority could not be confirmed by rigorous prospective epidemiological research. The clusters existed in the data; the question was whether they were signals or noise, and post-hoc pattern selection cannot answer that question.

Power Lines and the Multiple Comparisons Problem

A landmark example emerged from research into childhood leukaemia and proximity to high-voltage power lines. A widely cited 1979 study by Wertheimer and Leeper found a suggestive association. The finding generated enormous public concern and considerable follow-up research through the 1980s and 1990s.

The problem, which later analyses identified, was the scale of the data exploration involved. When researchers tested associations between power line proximity and hundreds of different health outcomes, the sheer number of comparisons created a statistical near-certainty that at least one association would appear significant purely by chance. With a standard significance threshold of p<0.05, you expect one false positive in every twenty tests even when there is no real effect. Run eight hundred tests and you expect forty false positives from chance alone.

When the "significant" finding is then selected from among hundreds of tests and presented as a discovery, the multiple comparisons problem makes it a classic Texas sharpshooter result. Subsequent rigorously designed prospective studies found no convincing evidence of a causal link between power lines and childhood leukaemia.

Stock Market Gurus and Backtested Strategies

Financial markets are a fertile habitat for the Texas sharpshooter fallacy, for a structural reason: there is an enormous quantity of historical price data, the incentives to find patterns are enormous, and the human mind is exquisitely well-suited to finding patterns even in random noise.

Backtesting — the practice of testing a trading strategy against historical data — is standard in quantitative finance. But naive backtesting is almost always a Texas sharpshooter exercise. If you test ten thousand different combinations of technical indicators against historical stock prices, some combination will appear to have worked brilliantly. The "strategy" was constructed to fit the data it was tested on. It carries no predictive validity for future performance.

This is why academic finance has long documented that almost no actively managed funds consistently outperform passive index funds after fees over the long run. The track records used to market these funds are typically selected from among many strategies that were tried; the successful-looking ones survive, are marketed, and attract capital. The selection process guarantees that some funds will have impressive historical records from pure chance — survivorship bias plus Texas sharpshooter.

The same pattern drives the "stock picking guru" industry. Among thousands of people who make public predictions about stock markets, some will appear prescient purely by chance. A single successful prediction sequence, selected post-hoc from a sea of failures, creates the impression of sharpshooter accuracy.

Nostradamus and Prophetic Texts

The Wikipedia article on the Texas sharpshooter fallacy notes that it "is often found in modern-day interpretations of the quatrains of Nostradamus." This is a perfect non-scientific example. The 16th-century poet's verses are intentionally vague, allusive, and metaphorical. Across centuries of human history, interpreters locate historical events that seem to match various quatrains — and present the matches as evidence of prophetic accuracy.

But the methodology is pure Texas sharpshooter: search through hundreds of cryptic verses and centuries of history for combinations that appear to match, then present those combinations as predictions. The verses were not specific predictions with defined criteria for success or failure; they were verbal inkblots that can be shaped to fit whatever events you already know occurred. Any bullseye found this way was always drawn around a bullet hole.

Why Our Minds Are Vulnerable

The Texas sharpshooter fallacy exploits cognitive tendencies that are adaptive in most contexts. Humans evolved as pattern-recognizers. In the ancestral environment, erring on the side of perceiving patterns — the rustle in the grass might be a predator — was less costly than missing real patterns. The cost of a false positive (unnecessary vigilance) was lower than the cost of a false negative (eaten by a lion).

This legacy means we are systematically prone to apophenia — perceiving meaningful patterns in random stimuli. The Texas sharpshooter fallacy is apophenia weaponized: not just perceiving the pattern, but retrofitting a narrative of prediction and significance onto it.

The problem is compounded by confirmation bias. Once a pattern has been identified and framed as meaningful, we selectively attend to further data that confirms it. The bullseye has been drawn; we now notice all the shots within it and ignore the shots outside.

The Statistical Solution: Pre-Registration

The scientific community has developed a direct institutional response to the Texas sharpshooter problem: pre-registration. Researchers publicly register their hypotheses, methods, and planned analyses before data collection begins. The registered protocol creates a verifiable record of what was predicted before the results were seen.

A pre-registered study that finds a significant result is far more credible than an unregistered study with the same result, because the hypothesis was on record before the data was drawn. The bullseye was painted before the shots were fired.

Pre-registration has become standard in many areas of medicine and psychology, partly in response to the "replication crisis" — the discovery that a large proportion of published findings in these fields failed to replicate in independent studies. Many of those failures are attributable to exactly the Texas sharpshooter dynamic: patterns found by extensive post-hoc exploration of data, dressed up as confirmatory results.

Recognizing the Fallacy in the Wild

Warning signs that you may be encountering a Texas sharpshooter argument:

• The pattern or correlation was reported as "discovered" rather than tested against a pre-specified hypothesis.
• The study examined many variables, outcomes, or subgroups — creating many opportunities for chance findings.
• The "finding" was highlighted because it was the most interesting result in a data set, not because it was the primary test.
• Historical data is being used to validate a strategy or prediction that was never made before the data was available.
• The cluster, pattern, or anomaly was defined only after examining where the data clustered.

See Also

• Apophenia — the tendency to perceive patterns in random noise
• Confirmation Bias — selectively attending to confirming evidence
• Data Dredging — exhaustively mining data for any apparent correlation
• P-Hacking — manipulating analysis to achieve statistical significance
• Publication Bias — the selective survival of positive results
• False Cause — inferring causation from co-occurrence or correlation

Sources & Further Reading

• Gould, Stephen Jay. The Mismeasure of Man. Norton, 1981. — On the dangers of data-mining and post-hoc pattern finding.
• Gigerenzer, Gerd. Calculated Risks. Simon & Schuster, 2002. — Accessible treatment of statistical reasoning failures.
• Ioannidis, John P.A. "Why Most Published Research Findings Are False." PLOS Medicine 2(8), 2005. — Landmark paper on the replication crisis.
• Thompson, William C. "Painting the Target Around the Matching Profile: The Texas Sharpshooter Fallacy in Forensic DNA Interpretation." Law, Probability and Risk, 2009.
• Wikipedia: Texas Sharpshooter Fallacy
• Open Science Collaboration. "Estimating the Reproducibility of Psychological Science." Science 349(6251), 2015.