Understanding Confirmation Bias

Three behaviours, one label

Confirmation bias is usually described as the tendency to look for evidence that supports what you already think. That is one of three distinct behaviours the term covers, and the three do not respond to the same remedies.

Biased search. You look in places likely to produce supporting evidence. You choose search terms that presuppose the answer. You ask people who agree with you.

Biased interpretation. You apply different standards to evidence depending on whether it fits. Supporting evidence is accepted at face value; contrary evidence is examined for flaws. Both examinations may be competent. The asymmetry is the problem.

Biased recall. You remember the confirming instances more readily. The prediction that came true is available; the three that did not are not.

Separating these matters because “be aware of your biases” is advice that does almost nothing, while specific countermeasures aimed at specific behaviours do somewhat better.

Why it is not stupidity

Confirmation bias is not a failure of intelligence, and there is no evidence that being well informed protects against it. In some settings, greater knowledge provides more material with which to construct a defence of a prior position.

It is also not always a defect. Treating your established beliefs as a default and requiring evidence to overturn them is a reasonable policy. A person who abandoned every belief at the first contrary datum would be unable to function. The problem is not having a prior. The problem is applying a standard of evidence that varies with whether you like the conclusion.

That reframing is useful because it gives you something concrete to watch for. The question is not “am I biased?” It is “am I using the same standard in both directions?”

Countermeasures that do something

For biased search: specify the search before you run it

Decide in advance what you would search for if you wanted to find out you were wrong. Write the query down. Then run it.

This is more effective than trying to search neutrally, because neutral searching is difficult to verify from the inside. A pre-written contrary query is checkable.

For biased interpretation: apply the reversal test

When you find yourself dismissing a study, an article, or an argument, ask: if this had reached the opposite conclusion, would I be scrutinizing it this closely?

If the answer is no, the objection may still be correct. But you now know your standard is asymmetric and can decide deliberately whether to keep it.

For biased interpretation: state the criteria first

Before looking at the evidence on a contested question, write down what result would count as support and what would count as disconfirmation. Doing this in advance makes it much harder to redefine the standard after the fact.

For biased recall: keep a record

Memory will not do this fairly. Written predictions with dates are the only reliable remedy. This applies to organizations as much as individuals: a team that records forecasts and reviews them later learns things a team relying on recollection does not.

For all three: talk to someone who disagrees and is well informed

The important qualifier is “well informed”. Arguing with a weak version of the opposing position tends to increase confidence rather than reduce it. The productive conversation is with someone who holds the opposing view for good reasons and can state your position back to you accurately.

What does not work well

Simply knowing about the bias. Awareness of a bias does not reliably reduce it, and can produce a new problem: confident identification of the bias in others while remaining unable to see it in yourself.

Consuming opposing content passively. Reading material from a source you distrust, without a structure for engaging with it, often reinforces the original position rather than challenging it. Structure matters more than exposure.

Trying harder to be objective. Effort is not the limiting factor. Procedure is.

Worked example

You believe a proposed policy will reduce a particular harm, and you are looking for evidence.

Biased search would be querying “policy X reduces harm Y”. The pre-committed contrary search is “policy X evaluation null result” and “policy X unintended consequences”, run before you look at anything supportive.

Biased interpretation shows up when you find a supportive study with a small sample and accept it, then find a contrary study with a small sample and note that the sample is small. The reversal test catches this.

Biased recall shows up six months later, when you remember having seen “lots of evidence” for the policy. A written note listing what you actually found, with dates, prevents the number from growing.

Practical exercise

Choose a belief you hold with moderate confidence, about something that matters but is not central to your identity. Write:

1. The belief, stated precisely enough to be wrong.
2. The three strongest pieces of evidence you have for it.
3. The search query you would run if you wanted to find out you were mistaken.
4. What result from that query would actually move you, specified before you run it.

Then run the query.

The exercise is harder than it looks, and the difficulty is the useful part.

Summary checklist

• I can distinguish biased search, biased interpretation, and biased recall.
• I wrote a contrary search query before running it.
• I applied the reversal test to at least one objection I raised.
• I stated my criteria for support and disconfirmation before looking.
• I keep a written record rather than relying on recall.

No external references were required for this page. Constructed examples are used throughout, and the countermeasures described are procedural rather than empirical claims about effect sizes.

Next step

Practice stating what would change your mind

Topics covered: Critical ThinkingInformation Resilience