Confirmation bias
myside bias · confirmatory search
The tendency to look for, notice and weight evidence that fits what you already think. It operates on search and on interpretation, which is why more information often produces more confidence without producing more accuracy.
In practice
You decide the problem is the marketing, then find three pieces of evidence that it is the marketing. The evidence is real, and you would have found three for any hypothesis you started with.
The common mistake
Thinking it is corrected by being open-minded or by gathering more data. Both make it worse on their own, because the search itself is what is biased and a larger biased sample is more convincing.
The failure is not believing what you want to be true. It is subtler and much harder to catch: the evidence you encounter is selected before you evaluate it, and the selection is done by the hypothesis you are already holding.
Two mechanisms
Biased search. You look where the confirming evidence would be. Suspecting a client is unhappy, you reread their terse emails. You do not reread the warm ones, because they are not relevant to the question you asked — and that is exactly the problem, since they are.
Biased interpretation. Ambiguous evidence gets read as supporting whatever you already hold. The same flat reply reads as annoyance if you expect annoyance and as busyness if you do not, and in both cases the reading feels like observation.
The second mechanism is why more information can increase confidence without increasing accuracy. Two people with opposite views given the same mixed evidence both come away more sure, and both are reading honestly.
The experiment worth knowing
Peter Wason set people a simple task in 1960: here is a sequence, 2-4-6, produced by a rule. Propose other sequences and you will be told whether each fits. Name the rule when you are confident.
Almost everyone forms a hypothesis — ascending even numbers — and then proposes 8-10-12, 20-22-24, 100-102-104. Every one comes back yes. Confidence rises with each. Then they name the rule and are wrong.
The rule was any three ascending numbers. The confirming tests could never have revealed that. Only proposing something the hypothesis forbids — 1-2-3, or 5-4-3 — carries information, and hardly anyone does it.
A test that cannot fail tells you nothing. That is the whole lesson and it applies far outside the puzzle.
What it costs in practice
Hiring. An impression forms in the first minutes and the rest of the interview gathers support for it. Structured questions asked identically of every candidate exist entirely to interrupt this.
Diagnosis. The first hypothesis about why revenue fell determines which numbers get pulled. Somebody who starts with "the market softened" and somebody who starts with "our follow-up got slower" will both find their evidence in the same dataset.
Strategy. The most expensive version. A plan is committed to, and every subsequent piece of information is sorted into support and noise. This is performance theater's quieter cousin — nobody is pretending, and the reporting has been shaped by what somebody expected to see.
What actually helps
Open-mindedness does not, on its own. Nor does more data.
Ask what would change your mind, before looking. Name the specific observation that would falsify the hypothesis. If you cannot name one, you do not have a hypothesis, you have a commitment — which is falsifiability applied to your own week.
Look for the disconfirming case deliberately. Not "is there evidence for this" but "what would I expect to see if this were false, and is it there?"
Give the opposite view to someone with an incentive. Assigning a person to argue against the plan works because it makes disconfirmation somebody's job, which is the only reliable way to get it done.
Write the prediction down first. Same fix as cognitive dissonance, for the same reason: a record you cannot revise removes the cheapest escape.
Concept web
Open the full webQuestions
What is confirmation bias?
The tendency to search for and interpret evidence in ways that support what you already believe. It operates on which evidence you look for and on how ambiguous evidence is read, so more information can raise confidence without raising accuracy.
What is the Wason 2-4-6 task?
A puzzle where people must find the rule behind a number sequence. Most propose only sequences that fit their hypothesis, receive confirmation each time, and guess wrong, because a test that cannot fail carries no information.
How do you counter confirmation bias?
Name in advance the observation that would change your mind, deliberately look for the disconfirming case rather than supporting evidence, assign someone to argue the opposite, and record predictions before outcomes are known.