Trap/Cognitive Bias/No. 0191
Confirmation Bias
You decide a job offer is right, then spend the evening reading only the reviews that make it sound good.
- Evidence
- Well established
- Read
- 6 min
- Links
- 20 connections
01You've seen this when…
- in life
A friend sends a short reply. You already think they’re upset, so you treat the brevity as proof and skim past their invitation to lunch.
- at work
An engineer suspects the database is causing a slowdown. She studies database logs for an hour without checking whether the delay starts before any database call.
- out in the world
You support a proposed housing development. A report predicting lower rents looks credible; one predicting traffic problems sends you straight to the authors’ funding disclosures.
02The idea
Once you have an explanation, evidence stops arriving on neutral ground. A matching fact feels like another piece clicking into place. A mismatch feels like something that needs explaining away.
Confirmation bias is favoring information that supports an existing belief or hypothesis. The bias can enter while you look for information or judge what you find, and it can shape what you remember afterward. You can collect accurate facts and still build a misleading picture if your selection consistently favors one answer.
The belief doesn’t have to be important to you. A tentative guess about a software bug can guide your search just as a political commitment can guide your reading. You don’t need a conscious wish to defend it.
The useful test is whether you gave alternatives a fair chance, even after finding supporting evidence. If the result pointed the other way, would you seek the same information and accept the same methods—and would you notice the same weaknesses?
03Why it happens
- A guess tells you where to look. Once you suspect the database, its logs become the obvious place to investigate. That narrows the search before you’ve established where the problem lives.
- Matching evidence is easy to recognize. A prediction comes true, and you count it as support. Several competing explanations may predict that result, so evidence matters most when it helps you tell them apart.
- Scrutiny can become uneven. An unwelcome result gets a close inspection; a welcome result passes with fewer questions. Motivated reasoning can intensify this when your identity, reputation, or wishes depend on the answer, but those stakes aren’t required.
- Memory can preserve the pattern. Episodes that fit your account may be easier to bring to mind. Later, the belief seems supported by experience, even though your recollection leaves out contrary episodes.
These processes can reinforce each other. A narrow search produces a tilted collection of facts, which makes the starting belief feel stronger, which narrows the next search.
04A worked example
In a 1979 experiment, psychologist Charles Lord worked with fellow psychologists Lee Ross and Mark Lepper. They recruited students who favored or opposed capital punishment. Participants evaluated descriptions of purported studies about whether the death penalty deterred murder. The researchers constructed the materials to provide evidence pointing in both directions.
What it looks like Careful evaluation. Participants notice methodological strengths and weaknesses rather than simply accepting everything they read.
What’s actually going on Those evaluations track prior beliefs. Participants tend to find the research supporting their position more convincing and criticize the research opposing it more heavily. People on opposing sides find support in a deliberately mixed body of evidence. The experiment concerns how people evaluate evidence. Whether capital punishment actually deters crime is a separate question.
What would have helped Setting methodological standards before seeing the conclusions, then using those standards equally to evaluate study methods with the findings hidden. The experiment left these proposed safeguards untested. The aim is to make agreement with the conclusion less able to decide whether the method looks sound.
05How to spot it
None of these proves bias on its own. Look for a repeated imbalance in how you treat supporting and opposing evidence.
06What to do instead
- Write down what would lower your confidence. Do this before searching. Name an observable result and use a standard that evidence could realistically meet. Use a falsification test to give your belief a chance to fail by looking for evidence that would challenge it.
- Keep a rival explanation in play. Beside your preferred account, write one plausible alternative. Then look for an observation on which their predictions differ. For a slowdown, compare time spent before the database call with time spent inside it.
- Apply one quality checklist. Decide what matters: sample selection, measurement, comparison groups, or relevant expertise. Use the same checklist whether you like the conclusion or not.
- Seek a strong opposing case. Use the consider-the-opposite strategy: explain how your current view could be wrong. Start with a knowledgeable critic who can make a strong opposing case.
- Make disagreement collaborative. Ask someone with a different view to help choose what evidence would distinguish your accounts. Adversarial collaboration turns competing convictions into a shared test.
The goal is to make your confidence responsive to evidence, so any change of opinion follows from what you learn. Bayesian updating captures that principle: new information should move your belief according to how strongly it favors one explanation over another.
07When it isn’t confirmation bias
Looking for a predicted result isn’t automatically biased. Psychologists Joshua Klayman and Young-Won Ha showed why positive testing—checking cases where your hypothesis predicts something will happen—can be informative. Whether it’s a good strategy depends on the task and the alternatives. The problem is treating agreement as decisive when rival explanations agree too.
You can give claims different weights. Rejecting a badly designed study can be sound judgment. The question is whether you’d reject an equally bad study that supported your view.
Two nearby ideas differ. Belief perseverance is continuing to believe after the original support has been discredited. Confirmation bias concerns the handling of evidence. Motivated reasoning concerns how desired conclusions shape reasoning; confirmation bias can occur even when you’re indifferent to which answer wins.
Finally, disagreement alone leaves bias unproven. Calling someone biased leaves two questions open: whether your evidence is better and whether you’ve treated it fairly.
08Roots
In 1960, British psychologist Peter Wason gave people three numbers: 2, 4, 6. Their task was to discover the rule behind the sequence. They could propose other triples to learn whether each followed the rule and announce their answer when ready.
A natural guess was numbers increasing by two. Triples such as 8, 10, 12 would receive approval and appear to confirm it. But Wason’s actual rule was broader: any three numbers in ascending order. A triple such as 1, 2, 9 could expose the gap. Participants often tested examples that fit their guesses without trying cases that could distinguish those guesses from the broader rule.
The task made an old reasoning problem visible in a small, repeatable experiment. As later research moved beyond testing rules to interpreting studies and choosing information, it also examined how people handled personal beliefs. Raymond Nickerson’s 1998 review brought these strands together under confirmation bias.
One possible adaptive explanation is that searching where we expect to find something often saves effort. Klayman and Ha’s analysis helps explain when that strategy works. A claim about evolutionary origins would require evidence beyond a strategy’s efficiency in some settings. Any evolutionary origin story remains speculative; the demonstrated point is that a useful search habit can become misleading when it fails to distinguish alternatives.
09How solid is this?
Favoring belief-consistent evidence is well documented in hypothesis testing, information selection, and evaluation. Its strength varies with the task and incentives; confirming tests can also be informative, so not every search for support demonstrates bias.
10Connections
- Often confused with Halo Effect, Hindsight Bias
- Countered by A/B Testing, Bayesian Updating, Consider-the-Opposite Strategy, Falsification Test, Adversarial Collaboration, Red Teaming, Steelmanning
- Can lead toBelief Perseverance, Overconfidence Effect, Texas Sharpshooter Fallacy
- Part of Motivated Reasoning
- Includes Cherry Picking, Myside Bias, Congruence Bias
- See also Anchoring Bias, Bias Blind Spot, Illusory Truth Effect, Negativity Bias
+ 10 more in the list
11Origin and sources
Peter C. Wason’s hypothesis-testing experiments (1960) provided the classic demonstration. Raymond S. Nickerson’s review (1998) synthesized the broader research on confirmation bias.
- [1]Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129–140.
- [2]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.
- [3]Klayman, J., & Ha, Y.-W. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94(2), 211–228.
- [4]Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175–220.
Suggest an edit· Updated 2026-10-02