Cognitive
Confirmation Bias
The cognitive tendency to seek, interpret, and remember information that confirms what we already believe
Confirmation bias is the systematic tendency to favor information that confirms preexisting beliefs while discounting contradicting evidence. Peter Wason coined the term in 1960 with his "2-4-6" rule-discovery task, in which participants tested only hypotheses that fit their initial guess. The bias operates through three channels: biased search (Googling supportive sources), biased interpretation (reading ambiguous data favorably), and biased memory (recalling confirming examples). Effect sizes are large and robust — meta-analyses (Hart et al. 2009, k=91) show people prefer congenial information about 2:1. It underlies stereotype persistence, scientific stagnation, jury verdicts, and online echo chambers. Debiasing requires actively considering the opposite — passive awareness rarely helps.
- Coined byPeter Wason (1960)
- Core mechanismsBiased search, interpretation, memory
- Meta-analytic ratio~2:1 preference for congenial info (Hart 2009)
- Famous taskWason's 2-4-6 rule-discovery
- Real-world domainsPolitics, medicine, science, juries, hiring
- Best-known fix"Consider the opposite" (Lord, Lepper, Preston 1984)
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
Why confirmation bias matters
- Science. Drives p-hacking, selective reporting, and replication failures unless preregistration is enforced.
- Medicine. Misdiagnosis often begins with anchoring on a first guess and confirming it through selective testing.
- Politics. Voters seek out and remember congenial coverage, deepening polarization across parties.
- Law. Police and jurors who form an early suspect theory weigh later evidence asymmetrically.
- Hiring. Interviewers form impressions in seconds and spend the rest of the interview gathering supporting evidence.
- Investing. Investors over-weight news supporting current positions, holding losers and missing reversals.
- Personal relationships. Stereotyped expectations of partners and colleagues persist because contradicting moments are forgotten.
Common misconceptions
- Only the unintelligent fall for it. High-IQ and highly educated people show equal or larger bias when topics are identity-relevant (Kahan 2017).
- Awareness fixes it. Knowing about the bias does not reduce it; only structured prompts like "consider the opposite" reliably help.
- It is the same as stubbornness. Stubbornness is conscious refusal; confirmation bias operates outside awareness during search and recall.
- It only happens to the other side. "Bias blind spot" research (Pronin 2002) shows people see bias in others while denying it in themselves.
- The internet caused it. Documented since the 1960s; the internet amplifies but did not create the underlying tendency.
- It is always bad. A mild positive-test strategy is computationally efficient when hypotheses are roughly true; the failure mode is refusing to update under disconfirmation.
Frequently asked questions
What is confirmation bias?
A systematic tendency to favor information that supports existing beliefs and to discount information that contradicts them. It works through three routes — what we look for, how we interpret what we find, and what we later remember. Wason (1960) demonstrated it with a number-rule task; subsequent decades of research showed the bias is pervasive, robust across cultures, and operates even when people are explicitly motivated to be accurate.
How is it different from motivated reasoning?
Motivated reasoning is the broader phenomenon of letting goals shape conclusions; confirmation bias is one specific mechanism. Confirmation bias can occur even without strong motivation — people simply find it cognitively easier to test hypotheses by looking for confirmations. Motivated reasoning adds an emotional or identity stake. Kunda's (1990) review treats confirmation bias as a tool that motivated reasoning recruits.
What was Wason's 2-4-6 task?
Participants are told 2-4-6 fits a rule and must discover the rule by proposing new triplets. The actual rule is "any ascending sequence." Most people guess "even numbers increasing by 2" and then test only sequences that fit (8-10-12, 100-102-104). They never propose 1-2-3 or 5-10-15 — tests that could falsify their hypothesis. About 80% report a wrong rule with high confidence, illustrating positive-test strategy.
How big is the effect?
Hart, Albarracin, Eagly et al.'s (2009) meta-analysis of 91 studies on selective exposure found people choose attitude-consistent information about 67% of the time vs 33% for inconsistent — roughly a 2:1 ratio. The effect is larger when beliefs are tied to identity, when commitment has been made publicly, and when the inconsistent information is hard to refute. Effect sizes (d ~ 0.36) are moderate but extremely consistent.
Does it cause echo chambers online?
Partially. Algorithmic curation amplifies what users already engage with, and users self-select into congenial feeds. But research by Bakshy et al. (2015, Facebook) and Guess et al. (2021) suggests cross-cutting exposure online is actually higher than offline; the bias lies more in interpretation than exposure. Still, identity-protective cognition (Kahan) means even balanced exposure can deepen polarization when issues become tribal markers.
How can it be reduced?
The most reliable fix is "consider the opposite" — Lord, Lepper, and Preston (1984) showed that explicitly asking participants to argue the contrary position reduced biased assimilation. Other tools: pre-registering predictions, adversarial collaboration, structured analytic techniques (used by intelligence analysts), red-team review, and seeking disconfirming rather than confirming tests. Mere awareness of the bias, without these structured prompts, does little.
Where does it cause real harm?
Medical misdiagnosis (anchoring on a first hypothesis and ignoring contrary labs), wrongful convictions (police and jurors locking onto an early suspect), failed scientific replications (analysts steering toward expected results), bad hiring (interviewers deciding in 30 seconds and confirming over the rest of the interview), and investment losses (holding losers because confirming news feels weightier). Costs run into billions annually across these domains.