Confirmation bias is the tendency to search for, interpret and remember information in a way that supports what we already believe. It is one of the most robust findings in judgement research, and it is also one of the most overused labels. The useful question is not whether people show it. It is when, how strongly, and which countermeasures actually change the outcome.
What confirmation bias is
People rarely look for evidence that could prove them wrong. They ask questions that invite a yes, they weigh a supporting study more heavily than an opposing one of equal quality, and they recall the hits and forget the misses. The bias works at three stages: the search for information, the evaluation of what is found, and the later recall of it.
The term covers at least three operations that research often separates. Selective search means choosing where to look. Biased interpretation means reading identical evidence differently depending on the conclusion it supports. Selective memory means remembering the confirming cases better. A single label hides three different failure points, and each one calls for a different fix.
The original evidence
The earliest well-known demonstration is Peter Wason’s 2-4-6 task (Wason, 1960). Participants saw three numbers, 2, 4, 6, and had to discover the rule that generated the sequence. They could propose triples and receive feedback on whether each one fit. Most participants formed a narrow hypothesis, such as “even numbers increasing by two”, and tested it only with triples that confirmed it. Only a minority found the actual rule, which was simply “any three numbers in ascending order”. Those who found it tended to test triples designed to fail.
Wason’s 1968 selection task showed the same pattern in logic (Wason, 1968). Participants saw four cards and had to test a conditional rule. Most chose cards that could confirm the rule and few chose the card that could falsify it. The result is often quoted as a reasoning failure. It is better read as evidence that people default to positive tests, which is useful for some problems and misleading for others.
Lord, Ross and Lepper (1979) moved the question from abstract reasoning to attitudes. They recruited 48 undergraduates who held strong, opposing views on capital punishment. Participants read two studies, one apparently supporting and one apparently opposing the deterrent effect of the death penalty. Both groups rated the study that matched their prior view as better conducted and more convincing. Exposure to the same mixed evidence pushed both groups further apart. This is the origin of the phrase “biased assimilation”.
Nickerson’s 1998 review in Review of General Psychology is the reference that gave the field its map (Nickerson, 1998). He documented the bias across laboratory work, reasoning, scientific practice, medicine and legal decisions, and argued that it is a pervasive tendency rather than a laboratory curiosity. Kunda’s 1990 paper set the theoretical frame that many later studies use: people want accurate conclusions, but they also want the conclusions they prefer, and the second goal shapes how they search and weigh evidence (Kunda, 1990).
How strong is the evidence today
The basic pattern is robust. Positive testing shows up in many tasks, and selective search is easy to reproduce. The weak points lie elsewhere.
First, the Lord et al. effect rests on a small sample, a single pair of studies and a specific topic. I could not find a large direct replication in my search, so the study should be read as a clear demonstration that the effect can occur, not as a precise estimate of its size. Second, the strength of the bias depends on how much the person cares about the topic. Taber and Lodge (2006) found that participants scrutinised arguments against their political views more harshly than arguments for them, with more politically sophisticated participants showing the pattern at least as strongly (Taber & Lodge, 2006). This is a motivated-reasoning finding, and it shows that the bias is not limited to people who know little.
Third, a newer and more precise line of work asks when people seek opposing views at all. Kappes and colleagues (2020) report that people weigh others’ opinions in a way that favours confirmation, with the strength of the opposing opinion mattering less than it should (Kappes et al., 2020). The effect is real, but the sizes reported across studies vary with the design, and I would not quote a single percentage from one experiment as a general rule.
A common overstatement needs correcting. Confirmation bias is often presented as the reason people hold false beliefs. That claim goes further than the evidence. The bias changes how strongly people hold views and how they react to new information, which matters a great deal, but the mechanism is one of several that produce error.
Everyday and business examples
The bias shows up wherever a person has a stake in a conclusion. The table below lists common cases, the typical confirming behaviour and a countermeasure that fits.
| Setting | Typical confirming behaviour | Countermeasure that fits |
|---|---|---|
| Product launch | Team tests the feature with users who already like it | Recruit users who are likely to reject the idea |
| Hiring | Interviewer’s first impression guides every later question | Use the same structured questions for every candidate |
| Investing | Investor reads analysts who agree with their thesis | Write down the disconfirming signal before buying |
| Marketing | Campaign is judged by clicks from the target audience only | Set a holdout group and compare outcomes |
| Medicine | Clinician keeps the first diagnosis and filters later symptoms | Ask what finding would make the diagnosis wrong |
| Relationships | Partner recalls the evidence that the other person is careless | Each side names one counterexample before arguing |
The pattern in business is consistent. A manager who has committed to a strategy tends to collect dashboards that show progress and to discount the churn data that suggests otherwise. The problem is rarely that the data is missing. It is that the question was framed to find support.
The science case is similar. A researcher who expects an effect tends to test with the design most likely to show it, and to treat a null result as a flaw in the study. The replication crisis made this visible across psychology. For the broader context, see behavioural science replication crisis.
How to reduce it
Five practical moves help, each aimed at a different stage of the bias.
- Ask for the disconfirming test before the confirming one. Write down the result that would change your mind before you look at data. This targets selective search directly.
- Assign a red team with a brief to defeat the conclusion. The brief must be real. A colleague who is asked to “think about risks” usually reports the risks they already accept.
- Blind the evaluation where possible. Reviewers who do not know which condition or author produced a result judge the evidence more evenly.
- Set a fixed rule for weighing contrary evidence, such as requiring the same level of scrutiny for a result that you like and one that you dislike.
- Use structured decision tools, such as a pre-mortem or a decision log, that record the reasoning before the outcome is known. A log makes later selective memory visible.
These moves do not remove the bias. They shift the default so that the first search is less one-sided. Structure also helps more than good intentions. Telling people to be objective has little measurable effect in most studies, while changing the procedure changes what people see.
Whether a given tactic works depends on the setting. Choice architecture can help when the goal is to change what a person sees by default. For the wider context of fast intuitive judgement and slower checking, see System 1 and System 2.
In my view, the most useful reading of confirmation bias is operational. Ask where in the process the one-sided evidence enters: the search, the weighting, or the memory. Then change that step, because a general warning to “be more objective” changes almost nothing. Research still has to show which of the five moves hold up in field settings with real stakes, and how much each one reduces error compared with a control.
If your team makes repeated decisions with real money or people at stake, start with a pre-mortem and a decision log. For a structured approach to changing decision routines, see applied behavioural science for business.
Frequently Asked Questions
What is confirmation bias in simple terms?
It is the habit of looking for, reading and remembering evidence that supports a belief you already hold, and giving less weight to evidence that cuts against it. It can happen without any intent to deceive.
Is confirmation bias the same as motivated reasoning?
They overlap but are not identical. Confirmation bias describes the pattern of one-sided search and interpretation. Motivated reasoning adds the claim that the pattern is driven by a goal, such as protecting self-image or a group identity. Kunda’s 1990 framework connects the two.
Can you eliminate confirmation bias completely?
No, and claims of complete elimination should be treated with caution. Structured procedures, such as red teams, blinding and decision logs, reduce its influence on specific decisions. The evidence for large, lasting reductions across all settings is still limited.
References
Kappes, A., Harvey, A. H., Lohrenz, T., Montague, P. R., & Sharot, T. (2020). Confirmation bias in the utilization of others’ opinion strength. Nature Neuroscience, 23(1), 130-137. https://doi.org/10.1038/s41593-019-0549-2
Kunda, Z. (1990). The case for motivated reasoning. Psychological Bulletin, 108(3), 480-498. https://doi.org/10.1037/0033-2909.108.3.480
Lord, C. G., Ross, L., & Lepper, M. R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37(11), 2098-2109. https://doi.org/10.1037/0022-3514.37.11.2098
Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175-220. https://doi.org/10.1037/1089-2680.2.2.175
Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American Journal of Political Science, 50(3), 755-769. https://doi.org/10.1111/j.1540-5907.2006.00214.x
Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129-140. https://doi.org/10.1080/17470216008416717
Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20(3), 273-281. https://doi.org/10.1080/14640746808400161