What is confirmation bias?

Confirmation bias describes a family of related tendencies: searching preferentially for evidence that supports an existing belief, interpreting ambiguous evidence as supportive, applying much harsher scrutiny to disconfirming information than to confirming information, and recalling supportive evidence more readily. Peter Wason demonstrated the search component in the 1960s with a task in which people testing a rule about number sequences overwhelmingly proposed examples that would confirm their hypothesis rather than examples that could falsify it, which is the less informative strategy.

The uncomfortable finding is that education and intelligence do not protect against it and can amplify it. Studies of politically charged topics have found that people with greater numeracy and subject knowledge show larger partisan divergence in interpreting the same data, because greater ability provides better tools for constructing arguments against unwelcome conclusions. That is motivated reasoning operating through, rather than despite, competence.

Countermeasures with some evidence behind them are procedural rather than attitudinal, since deciding to be open-minded does not work. Consider the opposite, meaning deliberately generating reasons a belief might be wrong, reduces the effect in experiments. Specifying in advance what evidence would change your mind, and writing it down, makes later goalpost-moving visible. Pre-registration in science serves this function institutionally by fixing the hypothesis and analysis before the data arrive. Actively seeking out the strongest version of an opposing argument rather than the weakest helps, as does exposure to people who disagree in good faith. Structurally, adversarial processes, peer review, red teams, and devil's advocacy exist because individual self-correction is unreliable and distributed disagreement works better.