Confirmation bias is easy to recognise in somebody else's argument and surprisingly difficult to notice in your own investigation. You may be gathering real evidence, asking sincere questions, and still choosing tests that protect your favourite explanation. This guide develops a practical alternative: write a competing account, predict where it differs, and look for that difference. Modic's reasoning games let you practise the process with immediate feedback.
What is confirmation bias?
Confirmation bias is a tendency to seek, interpret, or remember evidence in ways that favour an existing belief. It is broader than deliberately ignoring facts. A person can run a test honestly yet choose examples that would fit several possible explanations. The test returns true information, but that information may not identify the explanation the person thinks it confirms.
Nickerson's review of confirmation bias surveys this broader pattern. Use the idea as a question about your process rather than an insult or a diagnosis. Disagreeing with you does not establish that someone is biased, and having a strong belief does not make that belief false. The issue is how evidence is selected and weighed.
Start with a rule you can actually test
Open the confirmation bias experiment before reading the next section if you want to try its hidden rule unaided. The opening sequence is 2, 4, 6. You can submit triples and receive truthful feedback about whether they fit. Your first explanation might focus on even numbers, equal spacing, ascending order, or some combination. The exercise asks what each test can actually tell you.
A triple such as 8, 10, 12 can fit both equal-spaced even numbers and a broader increasing-number rule. Acceptance therefore does not distinguish those hypotheses. Repeating similar accepted examples can increase confidence without narrowing the explanation. This is a small, noiseless task; real decisions are messier, but the difference between confirming and discriminating evidence still transfers.
A worked example: design a test that separates two rules
This walkthrough reveals the task's rule, so play first if you want the surprise. Suppose your candidate explanation requires equally spaced even numbers, while a rival explanation accepts any strictly increasing triple. Both fit the opening example. The useful next test is something the rival accepts and your first rule rejects, rather than another obvious fit for both.
- Write the two hypotheses explicitly. For the narrow rule, 1, 2, 9 should be rejected; for the increasing-number rule, it should be accepted.
- Submit 1, 2, 9. The game's acceptance rules out the narrow hypothesis as the complete rule, although it does not by itself prove every detail of the broader one.
- Test a descending or repeated-number triple, then compare the pattern of feedback. Reveal the rule only after recording what would make each explanation fail.
Examples in work, health information, and investing
A team believes a discount caused a sales increase. It lists successful discounted weeks but does not compare similar weeks without discounts or check whether advertising changed simultaneously. An investor expects a company to recover and reads only commentary agreeing with that forecast. A reader researching a symptom searches for one feared explanation rather than a range of possibilities. In each case, the search can become organised around preserving a conclusion.
The corrective move is not to give every explanation equal weight. Some have weak support, and expertise matters. Ask what relevant observation separates credible alternatives. For a business decision, that might be a comparison group or a pilot. For a consequential health question, it may mean using qualified medical evaluation rather than trying to diagnose yourself through selective search results. The concept improves the evidence process; it does not replace domain expertise.
Ask what evidence would be expected under the alternative
Wason's original hypothesis-testing paper supplies the historical reasoning-task context. The game is a teaching implementation, not a personality test. Its central lesson is to compare predictions from rival rules. Before gathering more evidence, write the result each explanation predicts. An observation shared by both can be real and useful, yet weak for choosing between them.
The Wason selection task offers another chance to inspect whether a selected case could expose a rule violation. The Bayesian updating experiment then shows why evidence should change confidence in relation to its likelihood, not merely its emotional vividness. These tools address different parts of reasoning: choosing the observation and interpreting what it means after it arrives.
A practical confirmation-bias checklist
Before making a consequential choice, write one sentence for the favoured explanation and one for a plausible rival. Specify a result that would reduce confidence in the first. Record the prediction before you observe the outcome, because after-the-fact explanations can make almost any result seem expected. You do not need an elaborate research programme to make a small decision more falsifiable.
Ask a colleague to challenge the evidence selection, not simply argue for the opposite conclusion. A productive review might ask whether the sample excludes failures, whether a changed denominator explains the result, or whether your proposed test would come out the same under the rival. The value of information experiment helps distinguish useful investigation from collecting more material that cannot change the action.
- What exactly do I believe, and how confident am I?
- What plausible alternative fits what I already observed?
- Where do the alternatives make different predictions?
- What result would make me revise my view?
- Could the information change my decision enough to justify collecting it?
Avoid two common overcorrections
First, searching for disconfirming evidence does not require abandoning a well-supported belief after one anomaly. Measurements can be noisy and observations can be mistaken. Inspect reliability and the total body of evidence. A good process makes revision possible without making confidence infinitely fragile. The noiseless yes/no game deliberately omits this important real-world complication.
Second, balance is not achieved by giving unsupported claims the same credibility as established evidence. A serious rival needs a reason to be considered. Your aim is to find informative differences between plausible explanations, not to manufacture uncertainty around every conclusion. Clear thinking includes both openness to correction and proportionate weighting of the available evidence.
Try a small decision log
Choose a question you will revisit within a month. Record the hypothesis, rival, predicted result, planned observation, and action you would take under each outcome. When the review date arrives, compare the original predictions with the evidence. Notice whether you are changing the success definition after seeing the result or explaining away the result that previously would have mattered.
The benefit is not a badge proving you have eliminated bias. It is an inspectable record that makes your reasoning easier to revise. Keep the scope small enough to complete. One carefully chosen observation can teach more than a folder full of agreeable examples, especially when you can explain why that observation distinguished the alternatives.
Try the ideas for yourself.
These are teaching models. Follow the assumptions in each experiment; the results are not real-world forecasts.
Sources & further reading
Current-event context was checked on October 14, 2026. Follow the original source for newer updates. Worked scenarios are illustrative unless explicitly identified as reported data.
- Nickerson: Confirmation bias review ↗
Research review hosted by a university; the checklist here is an original practical application.
- Wason: failure to eliminate hypotheses ↗
Original hypothesis-testing research that provides context for the interactive rule task.