Confirmation bias
Claim assessed: People seek out and interpret evidence in favour of what they already believe.
Published · how we work
In short
Confirmation bias is, in Nickerson’s (1998) definition, “the seeking or interpreting of evidence in ways that are partial to existing beliefs, expectations, or a hypothesis in hand.” Nickerson calls it “a ubiquitous phenomenon in many guises”: it shows up in what you choose to check, which sources you read and how you judge an argument. The studies support the basic tendency well. But different things are gathered under the same name, and the most dramatic of them, the idea that contrary evidence makes people believe even more strongly what they believed before, has turned out to be rare.
What it says
The 2-4-6 task. Wason (1960) gave participants the number sequence 2-4-6 and told them it followed a rule they had to discover. They could propose other sequences of three numbers and were told each time whether it followed the rule. The real rule was very broad: any three numbers in increasing order.
Of 29 participants, only 6 found the rule without first announcing a wrong one. Nickerson (1998) describes the pattern: people assumed something like “even numbers, going up by 2” and then proposed only sequences that fitted their hypothesis. Each got a “yes”, so the hypothesis seemed confirmed. But since the real rule covered all of them, these tests could never show that the hypothesis was too narrow. For that, they would have had to try a sequence that, by their hypothesis, should have got a “no”.
Not every positive test is an error. Klayman and Ha (1987) showed that many phenomena labelled “confirmation bias” are better explained by a positive test strategy: you check the cases where you expect what you are looking for to appear. The strategy is not biased in itself. In many real situations it is a good method, but it leads to systematic errors in situations like 2-4-6, where your hypothesis is a special case of the real rule.
How we judge evidence. Lord, Ross and Lepper (1979) picked students who supported or opposed the death penalty and had them read two made-up studies: one that seemed to show the death penalty deters crime and one that seemed to show the opposite. Each side found the study that confirmed its view more convincing, and participants reported that their views grew stronger along the way. The result became the classic example of polarization: the same evidence pushes people apart instead of bringing them together.
Which sources we choose. Hart and colleagues (2009) pooled studies where people could choose between information that supported their views and information that challenged them. On average they preferred the supportive information, with a moderate effect (d = 0.36). The preference was weaker when the view was not tied to their values, when they were more confident in it, and when the contrary information was useful for a concrete goal. In the last case, people actually preferred the contrary information.
Example
You have formed the view that a colleague is unreliable. From now on you notice every time they are late and every missed deadline, but the days they stayed late don’t stick in your mind. Each late arrival confirms your view. As in the 2-4-6 task, the useful question isn’t “what fits my view?” but “what should I see if I were wrong, and have I looked for it?”. The example is an illustration, not a study cited here.
How to apply it
The steps below are a practical way to apply the idea, proposed by us based on the studies cited.
- Look for the test that would prove you wrong. The lesson from 2-4-6: ask which result would show you were wrong, and check exactly that case, not one more that fits.
- Think of alternatives. Nickerson (1998) cites studies where the tendency dropped when people were asked to consider alternative hypotheses, but notes that they don’t do it spontaneously.
- Find reasons against your own judgement. Nickerson suggests looking for reasons for and especially against a conclusion, because overconfidence comes partly from not thinking about why you might be wrong.
- Separate making a case from weighing evidence. When you build an argument, you collect evidence for it. When you decide what is true, you also weigh the evidence against. Nickerson observes that many people mix the two up.
- Read sources you disagree with, too. Hart et al. (2009) found that people choose contrary information when it is useful to them. Make it useful: look for the other side’s best argument, not its weakest.
Limits and nuances
- One name for several things. Nickerson (1998) and Klayman and Ha (1987) show that the term covers different phenomena: how you choose tests, how you choose sources, how you judge evidence. They don’t necessarily share the same cause or size.
- Polarization is rare. Guess and Coppock (2020) ran three large experiments, on gun control, the minimum wage and the death penalty, one of them with the original materials of Lord et al. (1979). They found no evidence of backlash: where the evidence changed anything, it changed views in its direction, in the same way on both sides. This doesn’t contradict that we judge evidence in a biased way, but it shows that contrary evidence rarely pushes people the other way.
- Positive testing is often reasonable. Klayman and Ha (1987) show that the positive test strategy works well in many real situations. The problem arises in specific situations, not every time.
- The effect varies. The preference for supportive information is moderate on average, but it depends on how important the view is, how confident you are in it and how useful the contrary information is (Hart et al., 2009).
- What we read. We read Nickerson (1998) and Guess and Coppock (2020) in full. For Wason (1960), Lord et al. (1979), Klayman and Ha (1987) and Hart et al. (2009) we read the abstracts. The description of the 2-4-6 task comes from Nickerson.
Sources
- Peter C. Wason (1960). On the failure to eliminate hypotheses in a conceptual task
- Charles G. Lord, Lee Ross, Mark R. Lepper (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence
- Joshua Klayman, Young-Won Ha (1987). Confirmation, disconfirmation, and information in hypothesis testing
- Raymond S. Nickerson (1998). Confirmation bias: A ubiquitous phenomenon in many guises
- William Hart, Dolores Albarracín, Alice H. Eagly, Inge Brechan, Matthew J. Lindberg, Lisa Merrill (2009). Feeling validated versus being correct: A meta-analysis of selective exposure to information
- Andrew Guess, Alexander Coppock (2020). Does counter-attitudinal information cause backlash? Results from three large survey experiments
See also: Anchoring, Availability heuristic, Cognitive dissonance, Consider the opposite, Fundamental attribution error, Halo effect, Pygmalion effect