Self-serving bias

The fair exam and the unfair one

A student gets an A on the first statistics midterm: “I really understood the material this time.” She gets a C on the second: “Half the questions were on things we barely covered, and they were worded to trick you.”

Both exams were written by the same instructor for the same course. She may be right about the second one, but nothing in her account suggests she checked: she didn’t compare it with the syllabus or ask how the rest of the class did. The explanation moved from herself to the test when the grade did.

Claiming the team’s win

Three coworkers prepare a pitch together. When it wins the contract, one of them tells a friend, “Honestly, my section is what sold it.” A month later a similar pitch by the same three loses, and she says, “The other two sections were too long. I said so at the time.”

Here the outside cause is other people rather than luck. Her share of the credit rises when the result is good and falls when it’s bad, although her part in the work was about the same both times. In a preregistered study of three-person groups, people rated their own responsibility for a shared decision higher when it paid off than when it didn’t, even though the outcomes had been fixed in advance and nobody’s choice affected them (see Evidence).

Skill when it goes in, luck when it doesn’t

Four friends run a month-long stock-picking contest with pretend money. One of them finishes first and says it shows he has a feel for the market. The next month he finishes last: “Nobody could have predicted that news week. It was a coin flip.”

Over a single month, rankings in a contest like this are dominated by chance, and he says so himself about the losing month. If a month’s results are mostly luck, then so was the winning month. The less obvious move here isn’t blaming someone else. It is treating the success as stable (a lasting talent) and the failure as a one-off, which is one of the dimensions researchers measure.

Explaining a success and a failure differently is often correct:

• When you have evidence about the cause. If the instructor admits two questions were flawed, or you know you studied twice as long for the exam you passed, the different explanations follow the facts.
• When the outcome is unusual for you. Someone with a long record of success is right to treat one bad result as more likely a fluke than a new pattern. The base rate is evidence.
• When the situations really differed. A race run in extreme heat, a project with half the usual budget: different conditions call for different explanations.
• When you apply the same standard in both directions. Crediting yourself is fine if you’d also have accepted the blame had the result gone the other way, on the same kind of evidence.

The test: if the outcome had gone the other way, would you have explained it by the same kind of cause?

The slow race on a hot day

A runner who has finished her local 10K in under 50 minutes every year for six years runs it in 56 minutes on a day that reached 95°F. She says the heat did it, not a loss of fitness. The results show that nearly every returning runner was several minutes slower than the year before.

She blames circumstances for a bad result, which is the pattern the bias describes. But she has two kinds of evidence that the circumstances mattered: a long record that makes one slow race unusual for her, and a whole field that slowed down by a similar amount. That’s the unusual for you and situations really differed conditions above.

Credit that was checked

A bakery owner redesigns her online ordering page. Orders over the next three months rise 30%. Before crediting the change, she compares the same months in the two previous years, when orders were flat, and checks with two other bakeries in town, whose orders didn’t rise. Then she says, “The redesign worked.”

She attributes a good outcome to her own decision, but only after ruling out the obvious outside causes (the season, a town-wide rise in demand). The internal explanation survived a comparison that could have undermined it. That is evidence about the cause, not a flattering default.

Two families of explanation were long treated as rivals.

• Motivation. Taking credit for success and distancing yourself from failure protects self-esteem. A 1999 meta-analysis organized many of the conditions that change the bias’s size (how important the task is, self-esteem, self-focused attention and others) under a single idea: the more an outcome threatens how you see yourself, the stronger the bias.
• Information and expectation. Dale Miller and Michael Ross argued in 1975 that people expect to succeed and intend to, so a success matches what they tried to do and a failure doesn’t. An explanation that points to yourself for the expected outcome and to something else for the unexpected one can follow from those expectations without any wish to feel good.

A 2008 review concluded that the two often work together rather than competing. The information account also marks where the pattern shades into reasonable inference: if you usually succeed, a failure really is more likely to have an unusual cause.

The self-serving bias is a relative of Unrealistic optimism, which concerns expectations about what will happen; the self-serving bias concerns explanations of what already did.

These are practical checks; none has been tested as a remedy for this bias in particular.

• Run the flip. Before settling on why something went well or badly, ask what you’d say if it had gone the other way. If the answer is “the opposite kind of cause,” look for evidence that tells the two apart.
• Look for a comparison. How did others do under the same conditions? How did you do the last few times? Those are what separate a real outside cause from a convenient one.
• Ask someone else to assign the credit. In the preregistered study below, the bias was stronger in people’s ratings of their own responsibility than in their ratings of other group members’.

Status: replicates robustly. Several independent meta-analyses, covering hundreds of studies, lab tasks and real sporting outcomes, find the pattern, and a preregistered Registered Report found it where outcomes were fixed in advance. None of the meta-analyses described here settles the question on its own (the largest doesn’t report a correction for publication bias, and the sport meta-analysis found some), so the preregistered study is what meets this site’s bar. The large average effect in the biggest meta-analysis comes with big differences between groups of people, and how much of the pattern is error rather than reasonable inference is still discussed.

• Miller and Ross (1975) reviewed the early evidence and found it weaker than the idea’s popularity suggested: some support for people taking credit after success, but only minimal evidence that they deflect blame after failure. They proposed the non-motivational explanation above.
• Campbell and Sedikides (1999) meta-analyzed experiments that tested the bias with 14 different moderators and concluded that the bias is widespread and that conditions threatening the self make it larger.
• Mezulis, Abramson, Hyde and Hankin (2004) meta-analyzed 266 studies (503 independent effect sizes) that compared how internal, stable and global people’s explanations were for good events versus bad ones. The average effect was large (d = 0.96) and present in nearly all samples. It was largest in children and older adults, much smaller in Asian samples (d = 0.30) than in US (1.05) or other Western (0.70) samples, and smallest in samples with depression (0.21).
• Allen, Robson, Martin and Laborde (2020) meta-analyzed 69 studies of about 10,500 athletes explaining real competitive results. Athletes credited their own wins to themselves and their losses to outside factors (standardized mean difference 0.62), did the same for their teams’ results (0.63), and in a handful of studies claimed more personal responsibility for team wins than team losses (0.28). The authors report some publication bias and heterogeneity.
• Jaquiery and El Zein (2022), a Registered Report with a preregistered sample of 500 online participants, had people vote in three-person groups on which of two gambles to take. Unknown to them, every outcome was fixed in advance, so no choice affected anything. People still rated players more responsible for rewards than for missed rewards, especially the player who received the outcome, and the effect was stronger when they rated themselves than when they rated others. The average difference was small (d = 0.24 for ratings of the self), but it appeared where nobody’s choice could have made a difference.
• Okamoto and colleagues (2025) used a movement task in which rewards depended either on skill or on chance. Participants credited successes to their ability and blamed failures on randomness, and these attributions shaped their later decisions.

The link to the actor–observer asymmetry. Malle’s (2006) meta-analysis of the Actor–observer asymmetry found no general difference between how people explain their own and others’ behavior, but did find a valence pattern. For bad events, actors cited internal causes less than observers did (d ≈ 0.24); for good events the gap reversed (about −0.15). Malle read this as a self-serving pattern rather than an actor–observer one. It was small, it shrank after a correction for possible missing studies (to about 0.11 for bad events and 0.01 for good), and the bad-event gap held for internal attributions only. Three studies with East Asian participants showed the reverse of the Western pattern. So the self-serving bias measured within a person (good versus bad events) is large, while the version measured as a difference between explaining yourself and explaining someone else is much smaller.

Culture. Mezulis and colleagues found the attributional bias reduced, but still present, in Asian samples. Heine and Hamamura (2007), meta-analyzing self-enhancement more broadly across 30 methods, found a clear self-serving pattern among Westerners (d = 0.87) but none among East Asians (d = −0.01), and concluded that East Asians do not self-enhance. That broader question is disputed: Constantine Sedikides and colleagues argue that when the comparison is limited to traits that matter to a given culture, East Asians self-enhance too, and Heine’s side replies that this leaves out the methods that disagree. It is also not the same measure as attributions for success and failure; how far the attributional bias generalizes across cultures is not settled.

Not the better-than-average effect. The self-serving bias is about explaining outcomes (why did this go well or badly?). The better-than-average effect is about comparing yourself with others (am I more careful, kinder or more capable than the average person?). A 2020 meta-analysis of that effect (291 samples, more than 950,000 participants) found it robust (dz = 0.78) with little evidence of publication bias, and also smaller among East Asian than European American samples. The two often appear together, but someone can rate themselves average and still explain their wins and losses in a self-serving way.

What remains uncertain. How much of the pattern is error. Malle argued that when an experiment gives someone a failure that is out of line with their usual performance, explaining it as a fluke can be normatively defensible, and an observer without that history has less reason to. The clearest evidence of error comes from designs like the Registered Report above, where the outcomes couldn’t have depended on anyone.

Popular version. The bias is often summarized as “people take credit and blame others.” In the research, the outside cause is often the task or luck rather than another person, the shift can be about how lasting a cause seems rather than whose it is, and the size varies widely: it is small among people with depression.