Confirmation bias is a pattern in which evidence search or evaluation favours an existing hypothesis instead of testing what could disconfirm it. Wason's original hypothesis-testing experiment examined a preference for confirmatory tests in a conceptual task. Applying the idea to betting requires an observable search record, not an assumption about a person's motives.
A hypothesis is not the bias
Research needs hypotheses. The problem appears when supportive and adverse evidence receive different treatment.
| Research choice | Symmetric version | Possible confirmatory version |
|---|---|---|
| Form window | Same declared window for both teams | Change the window until one team looks strong |
| Availability | Record absences and returns for both sides | Count only the opponent's absences |
| Source quality | Apply one source hierarchy | Accept weak sources only when supportive |
| Metric threshold | Freeze before comparison | Move the threshold after seeing values |
| Missing data | Use one declared rule | Treat missing adverse evidence as irrelevant |
The original confirmation-testing research used a particular reasoning task. It supports the concept but does not prove that a football research choice has the same cause.
Worked falsifier sheet
Illustrative hypothesis: "Team A's recent shot profile supports a higher probability of winning than the current model gives."
Before collecting more evidence, write:
- Target: regulation-time Team A win in a named fixture.
- Current probability: 0.48 from model version 3.2 at 10:00 UTC.
- Support test: same-provider shot quality remains higher after opponent and venue adjustment.
- Falsifier: the effect disappears under a predeclared longer window or after lineup availability is updated; the purpose is to attempt elimination rather than collect only support, consistent with the original hypothesis-testing experiment.
- Exclusion rule: no source without a timestamp and provider definition.
- Decision rule: no action if either team's required data is missing.
Now run the same queries and thresholds for Team A and its opponent. Store rejected evidence with the reason. A conclusion that survives the test may still be wrong; the audit only makes the evidence treatment visible.
Prevent hindsight from rewriting the search
Outcome knowledge can change how a prior decision is evaluated. The original outcome-bias experiments are a reason to freeze search terms, source cutoffs, and exclusions before the result. After the event, add a new review record rather than editing the old one.
Practical checks
- State the proposition in falsifiable terms.
- Write at least one observation that would reduce confidence.
- Apply the same time window, metric definitions, and source standard to alternatives.
- Log evidence that was considered and excluded.
- Let missing or contradictory evidence produce no action.
- Review the search process before revealing the outcome.
What the audit cannot do
A symmetric search does not guarantee complete data, causal inference, a calibrated probability, or a favourable price. Search engines and feeds can themselves rank or omit information. Two analysts can make different defensible modelling choices. The record should expose those choices instead of claiming objectivity.
Next step
Use Cognitive Biases Betting for the next part of this topic.
Continue learning
- Next guide: Dunning-Kruger Effect in Betting
- Related guide: FOMO in Betting
Assumptions and limitations
The example is hypothetical and does not recommend a selection. Confirmation bias is a research concept, not a diagnosis or a synonym for being wrong. Similar one-sided records can arise from poor data access, time pressure, or undeclared methods. Bias checks do not remove uncertainty or gambling harm.
GambleAware
GamCare
Gambling Therapy
GAMSTOP