Texas Sharpshooter Distortion

Picking out the facts that match a story, while ignoring the rest, makes a pattern look real even when it happened by pure chance. The story feels validated purely by the selection.

This bias selectively attends to confirmatory evidence, creating the illusion of a meaningful pattern out of random or heterogeneous data. Analysts overfit their interpretation by retroactively clustering outcomes to support a preferred hypothesis, undermining objective inference.

A manager notices three employees in the northwest corner of the office were all promoted last year, and concludes that sitting in that area must boost performance — ignoring the dozens of other employees in that same area who weren't promoted, and the many promotions that happened everywhere else.

A litigator defending a pharmaceutical company conducts a post hoc subgroup analysis of a clinical trial, identifying a narrow slice of patients in whom the drug shows a statistically significant benefit. Without a hypothesis specified in advance, that cluster was effectively constructed around the data after the fact — the target painted around the bullet holes. Presented to a court as a supporting pattern, judges anchoring on that manufactured cluster give it undue weight in the evidentiary record, when proper correction for the number of comparisons made would reveal it as a chance artifact.

Matching facts get noticed, and attention then narrows to only those facts. That narrowed focus is exactly what makes the chosen facts feel more significant than they actually are.

Observers impose a target region onto a dataset after the fact, weighting the internally coherent observations more heavily than the rest. That selective sampling skews the resulting conclusion toward the constructed cluster, rather than toward what the full data actually shows.

Looking at all the data before deciding whether a pattern is real is the direct fix. Testing whether the pattern would plausibly appear by chance alone settles the question.

Defining the hypothesis and the clustering criteria before inspecting the outcomes prevents the retroactive construction entirely. Out-of-sample tests or randomization checks assess whether the pattern actually holds up against fresh data.

False pattern identification; Overconfident conclusions; Ignoring contradictory evidence

An adversarial actor can deliberately mine a large dataset for any subset of cases that superficially support a desired legal precedent or policy conclusion, then present that cherry-picked cluster as representative evidence while suppressing disconfirming records. In litigation or regulatory proceedings, this can manufacture the appearance of a robust evidentiary pattern—e.g., selectively citing only favorable case outcomes to establish a false doctrinal trend. Combined with citation concentration and precedent application asymmetries, this weaponized post hoc clustering can anchor judicial or regulatory reasoning around a manufactured target region that was never prespecified.

Preregistration of hypotheses and clustering criteria before data inspection is the primary structural defense, preventing retroactive target construction. Analysts should demand out-of-sample validation or randomization tests to assess whether an apparent cluster survives contact with held-out or independently drawn data. In legal and evaluative contexts, requiring adversarial parties to disclose the full dataset from which cited precedents were drawn and applying systematic doctrinal congruence audits materially reduce the exploitable surface.