In short
Congruence bias is the habit of testing a hypothesis by checking what it predicts, rather than by running the test that would separate it from the alternatives. It is narrower than confirmation bias: the evidence is read honestly when it arrives, but the test chosen was one that every live alternative predicted the same way, so the expected result, when it comes, leaves all of them standing.
How congruence bias works
A hypothesis suggests its own test. If you believe red triangles open the lock, the test that comes to mind is a red triangle, and when it opens, the belief is confirmed and nothing has been learned. This is the narrow habit the page is about: not reading evidence partially, which is confirmation bias, but choosing a test that every live hypothesis predicted the same way. The evidence that arrives is read honestly. When it comes out as all of them expected, which is the usual case, it separates none of them. A surprise would have carried information, and would have put every one of those hypotheses in question at once, but a surprise is not what the test was chosen to find.
The reason it feels like testing is that the result is real. A card opened a lock; a prediction came true. What is missing is the comparison. A test distinguishes the hypotheses on the table only when they predict different results from it, and the test a hypothesis suggests is usually one where the rivals agree. Jonathan Baron, who named the bias, describes it as a failure to consider alternative hypotheses rather than a failure of honesty, which is why it shows up in people with no stake in the answer.
Baron, Beattie and Hershey found the same habit in diagnostic reasoning in 1988. Given a hypothesis about what a patient might have, people chose the test that would be positive if the hypothesis were true, even when a different test would have separated the candidate diagnoses better. They called the shortcut the congruence heuristic, and they noted that it often works: in many real problems, the test a hypothesis suggests is also the most informative one available.
That caveat matters, and the page would oversell without it. Klayman and Ha argued in 1987 that the positive test strategy, examining cases where you expect the property to hold, is a reasonable default across many problems and gives the wrong answer in a particular shape of problem: when the true rule is broader than the hypothesis, every positive case fits both, and only a case the hypothesis excludes can tell them apart. Wason's task was built with exactly that shape, which is why it is so hard.
The original study
Peter Wason published "On the failure to eliminate hypotheses in a conceptual task" in the Quarterly Journal of Experimental Psychology in 1960. Twenty-nine participants were told that the experimenter had a rule in mind that generated triples of numbers, and that 2, 4, 6 was an example of it. They could propose as many triples as they liked, were told each time whether the triple fitted the rule, and were asked to announce the rule only when they felt certain.
Six participants announced the correct rule at their first attempt. Twenty-two announced at least one incorrect rule along the way and one reached no conclusion. The rule was any three numbers in ascending order, which is much broader than the hypotheses people brought to it: ascending even numbers, numbers increasing by two, the third number being the sum of the first two. Those hypotheses are all consistent with 2, 4, 6 and all much narrower than the rule, so any triple generated to fit them was told that it fitted, and the agreement was read as support. Wason reported that participants who announced two or more wrong rules had been unable or unwilling to test the hypothesis they were working with.
Read the diagram and evidence as text
An accented testing loop remains inside a narrow hypothesis enclosed by a broader true-rule region. A dashed possible test lies inside the true rule but outside the hypothesis, beyond an untouched boundary. The regions are schematic, not set sizes or probabilities. The reported result, 6 of 29 correct without an earlier incorrect conclusion, is separate from the illustrative tests.
Testing the rule behind 2, 4, 6
Conceptual sequence; counts from Wason (1960)
Hypothesis: “Numbers increasing by two.” It fits 2, 4, 6.
Congruent test: Propose another triple the hypothesis predicts
Told that it fits: The narrow hypothesis and the broader true rule both predict this answer. Only 6 of 29 participants named the rule without a previous incorrect conclusion.
Disconfirming test: A triple the hypothesis excludes: the only kind that can separate the rivals
Hypothesis → suggests its own test → Congruent test (error transition)
Congruent test → confirmed; nothing separated → Told that it fits
| Condition | Value and unit | Evidence quote |
|---|---|---|
| Announced the correct rule at their first attempt | 6 participants | Six out of 29 subjects reached the correct conclusion without previous incorrect ones [Display conversion: 6 is the numeric transcription of Six.] |
| Participants in Wason's 1960 rule-discovery task | 29 participants | Six out of 29 subjects reached the correct conclusion without previous incorrect ones |
Wason returned to the structure in 1968 with the selection task, in which four cards show a vowel, a consonant, an even number and an odd number, and the rule is that a card with a vowel on one side has an even number on the other. Most people turn the vowel, and many turn the even number, which cannot break the rule whatever is on its back. The card that can break it, the odd number, is rarely chosen. The error is the same: the chosen test is the one the rule predicts rather than the one that could refute it.
The term congruence bias comes from Jonathan Baron's work. With Beattie and Hershey he showed in 1988 that people asked to choose a diagnostic test picked the one congruent with the hypothesis under consideration rather than the one that best discriminated between the candidates, and in his writing on thinking he treats the habit as a failure to search for alternatives rather than as motivated reasoning.
Everyday examples
The pattern appears wherever a check is designed by the person who already has an answer in mind. The rows below are illustrations, not findings from the studies cited on this page.
| Situation | What the bias does | What helps |
|---|---|---|
| Debugging code you believe is broken in one function | You add logging inside that function and find the values you expected, which rules nothing out | Write the test that would pass if the fault were somewhere else, and run it first |
| A clinician who suspects one condition | Orders the test that would be positive under that suspicion, rather than the one that separates it from the next candidate | Name the two most likely explanations and ask which test tells them apart |
| Interviewing to check whether a candidate is a strong writer | Asking for examples of good writing returns examples; the question cannot come back negative | Set the same writing task for everyone and score it blind |
| Testing whether a new process is faster | Measuring only after the change confirms the change looks fine, with nothing to compare it against | Measure the old process over the same period, or alternate between the two |
| Checking a rumor by asking the person who told you | The source agrees with itself, and the agreement feels like a second data point | Look for a source that would know if the rumor were false |
How to counter it
The fix is not to distrust your hypothesis but to design the test against a rival.
- Write down a second hypothesis before choosing a test. With only one in the room, every test is congruent with it.
- Ask what result would make you drop the hypothesis. If no available result would, the test is decoration.
- Prefer the test where the candidates disagree. Informativeness lives in the gap between predictions, not in the strength of the prediction.
- Try a case your hypothesis excludes but a rival allows. In Wason's task the breakthrough triple is one you expect to be rejected; in the demo above, after two triangles have opened the lock, it is the square.
- Have someone else design the check, especially someone who holds the other hypothesis. They will pick a different test, which is the point.
- Remember that positive tests are often fine. The time to be careful is when your hypothesis might be a special case of something broader, because then every confirming case fits both.
What it is not
Three nearby patterns all concern evidence and belief. They differ in where the failure sits.
- It is not confirmation bias, which is the broader habit of seeking, reading and recalling evidence in ways that favor what you already believe. Congruence bias is one part of that: the choice of test, made before any evidence arrives, and it occurs with no preference for the outcome at all.
- It is not information bias, the appetite for information that cannot change what you will do. When a congruent test returns the result every current candidate predicted, it distinguishes none of those candidates; an information-bias test is information that cannot change the decision.
- It is not belief bias, judging an argument to be valid because its conclusion sounds right. That is about evaluating reasoning that is already in front of you, not about which evidence gets collected.
Tests and tools that use this
Related biases
Congruence Bias FAQ
What is congruence bias?
It is the habit of testing a hypothesis by checking what that hypothesis predicts, instead of running the test that would separate it from the alternatives. A congruent test is one that every rival still in play predicts the same way, so the expected result does not distinguish them. A result none of them predicted would call all of them into question, but that is not what the test was chosen to find. Jonathan Baron named the pattern, and Wason's 1960 rule-discovery task is the standard demonstration of it.
How is congruence bias different from confirmation bias?
Confirmation bias covers the whole cycle: which evidence gets sought, how it is interpreted and what is remembered. Congruence bias is one step inside it, the choice of test, and it does not require any wish for a particular answer. A person who would happily accept a refutation still tends to pick the test their hypothesis predicts, because that is the test the hypothesis suggests.
What was Wason's 2, 4, 6 task?
Participants were told the experimenter had a rule that generated number triples and that 2, 4, 6 fitted it. They proposed triples, were told each time whether the triple fitted, and announced the rule when certain. The rule was any three ascending numbers. Of 29 participants, 6 announced it correctly at the first attempt; most generated triples that fitted their own narrower hypothesis and read the agreement as support.
How does the demo on this page work?
You get one fact, that a red triangle opens the lock, and two tests. A card that every rule still in play predicts the same way separates none of them when it does what they all expect, whether that is opening or staying shut, so spending a test on one counts as the congruent choice. A card that two live rules predict differently can rule one of them out. The reveal gives the real rule and names the tests that would have found it.
Is testing your own hypothesis always a mistake?
No. Klayman and Ha showed in 1987 that examining cases where you expect a property to hold is a reasonable strategy in many problems and gives useful information. It fails in a specific shape of problem: when the true rule is broader than your hypothesis, every positive case fits both, so only a case your hypothesis excludes can separate them. Wason's task has that shape by design.
Where does congruence bias show up outside the laboratory?
Anywhere the person who holds a hypothesis also designs the check: a developer adding logging where they already suspect the fault, a clinician ordering the test their first guess predicts, a manager asking a question that can only produce agreement. Baron, Beattie and Hershey reported the pattern in diagnostic test selection in 1988 and called it the congruence heuristic.
References
- Wason, P. C. (1960). On the failure to eliminate hypotheses in a conceptual task. Quarterly Journal of Experimental Psychology, 12(3), 129-140. doi.org
- Baron, J., Beattie, J., & Hershey, J. C. (1988). Heuristics and biases in diagnostic reasoning: II. Congruence, information, and certainty. Organizational Behavior and Human Decision Processes, 42(1), 88-110. doi.org
- Klayman, J., & Ha, Y.-W. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94(2), 211-228. doi.org
- Wason, P. C. (1968). Reasoning about a rule. Quarterly Journal of Experimental Psychology, 20(3), 273-281. doi.org
- Nickerson, R. S. (1998). Confirmation bias: A ubiquitous phenomenon in many guises. Review of General Psychology, 2(2), 175-220. doi.org
Cite this source
Psychology.com. (2026, October 23). Congruence Bias. Psychology.com. https://psychology.com/biases/congruence-bias