Robust, Firmly Established, Pervasive

You know this one. People excuse their own behaviour by pointing at circumstances and explain everyone else's by pointing at character. It is in every management course. A meta-analysis of the entire published literature found average effect sizes running from slightly negative to trivially small.

Our Grades For These Claims

Applying the scheme from the first article in this series.

That the general actor-observer asymmetry does not exist is Grade A. A meta-analysis of the full published literature, by a specialist in the field, finding effects bracketing zero, is about as clean a negative result as this series has encountered.

That narrow conditional versions survive is Grade B, being the moderator findings from the same analysis.

That the correspondence bias is a separate and better-supported phenomenon is Grade C on our sourcing, which is a paper title and an encyclopedia entry we did not obtain.

Our position: this is the strongest negative finding in the series, and the effect it overturns remains in general circulation twenty years later.

A Note On Method

Everything here is verified to August 2026.

We obtained the published abstract in full from PubMed and from the author's own institutional copy of the paper, together with three further reproductions, all identical[1][2][3][4].

We did not obtain the full text, and report no analysis beyond what the abstract states.

We read portions of the paper's reference list, which is itself informative and which we use for two specific observations, each flagged.

One observation in this article concerns how a third party cited this paper. We found that text on a publisher page, we do not name its authors, and we treat it as an illustration rather than as a criticism of anyone.

All arithmetic converting effect sizes is ours.

This article reviews psychological research. It is not management or human resources advice.

Three Things That Get Conflated

A distinction that has to come first, because the rest of the article depends on it. The three are routinely treated as one idea in business writing.

The actor-observer hypothesis, from Jones and Nisbett (1971): people explain their own behaviour situationally and other people's dispositionally[1]. This is a claim about a difference between two perspectives.

The fundamental attribution error, a term attributed to Ross (1977)[5]: observers over-attribute behaviour to disposition and under-attribute to situation. This is a claim about one perspective being wrong.

The correspondence bias, associated with Gilbert and Malone (1995)[6]: the tendency to infer that behaviour corresponds to an enduring disposition. This is a claim about an inference.

Two observations, ours.

Only the first is the subject of the meta-analysis below. A negative result on the actor-observer asymmetry does not automatically dispose of the other two.

But they are constantly merged, and the business version, being roughly "we judge others by their actions and ourselves by our intentions", is closest to the one that did not survive.

The Hypothesis

The claim, stated precisely.

"The actor-observer hypothesis (E. E. Jones & R. E. Nisbett, 1971) states that people tend to explain their own behavior with situation causes and other people's behavior with person causes."[1]

Two observations, ours.

It is symmetrical and simple, which is part of its appeal. One sentence, two clauses, immediately recognisable from ordinary life.

And it is testable, which is what made 173 studies possible. Ask people to explain their own behaviour and someone else's, code the explanations as internal or external, and compare.

How It Was Described

The standing it enjoyed, in the meta-analysis's own words.

"Widely known in psychology, this asymmetry has been described as robust, firmly established, and pervasive."[1]

Three observations, ours.

Those three adjectives are the strongest available vocabulary for an empirical claim in psychology. Robust means it survives varied conditions. Firmly established means the question is closed. Pervasive means it turns up everywhere.

The author is quoting the field's characterisation, not his own, and the placement is deliberate: the next sentence of the abstract is the result.

And note the word "surprising" in the paper's own title, in parentheses[2]. That is a specialist signalling that he did not expect what he found, which is a mark in favour of the analysis rather than against it.

The Result

The finding.

"However, a meta-analysis on 173 published studies revealed average effect sizes from d = -0.016 to d = 0.095."[1]

Three observations, ours.

173 published studies. This is not a single failed replication. It is the accumulated literature, and it was large enough that its size was part of the effect's reputation.

The lower bound is negative. The range does not merely approach zero from above; it crosses it. Depending on the measure used, the average effect ran from very slightly backwards to very slightly forwards.

And the analysis found the effects were not there having been designed to find them. A funnel plot appears among the paper's figures[3], indicating the author checked for publication bias, which is what you do when you expect an effect and want to know whether the literature is trustworthy.

What Those Numbers Mean

Translating, with our own arithmetic.

At the upper bound of d = 0.095: an actor's explanation is more situational than an observer's about 53.8 percent of the time, against 50 percent by chance, with roughly 96 percent overlap between the two distributions.

At the lower bound of d = -0.016: about 49.4 percent, which is to say slightly the wrong way, with roughly 99 percent overlap.

Both conversions are ours, computed under standard normal assumptions, and appear in no paper.

Set against what this series has reported: brainstorming penalty at 1.39 on our own conversion, feedback interventions at 0.41, the disputed nudge headline at 0.43, growth mindset interventions at 0.080, choice overload's meta-analytic mean at 0.02.

The actor-observer asymmetry's upper bound sits between growth mindset and choice overload, and its lower bound sits below zero. It is the smallest effect this series has reported, and it carried the largest description.

The Four Conditions

Where it did appear.

"A moderator analysis showed that the asymmetry held only when the actor was portrayed as highly idiosyncratic, when hypothetical events were explained, when actor and observer were intimates, or when free-response explanations were coded."[1]

Note the word only. Four conditions, and outside them, nothing.

Taking them in turn, with our own gloss on what each involves.

Highly idiosyncratic actor. The person being explained was presented as unusual, which is an invitation to reach for a dispositional account.

Hypothetical events. Nothing happened. Participants explained imagined behaviour.

Intimates. Actor and observer knew each other well, which is the opposite of the stranger-judging-stranger case the effect is usually invoked for.

Free-response coding. Participants explained in their own words and a researcher classified the answers as internal or external.

What Those Four Have In Common

Our own analysis, offered as reasoning rather than as a finding.

Three of the four move away from the situation the effect is popularly used to describe.

The popular version concerns an ordinary observer judging an ordinary person about something that actually happened. That is the manager assessing an employee, which is why the idea travelled into management writing.

But the effect appeared when the actor was unusual rather than ordinary, when the event was hypothetical rather than real, and when the parties were intimates rather than a manager and a report.

Two consequences.

The surviving conditions are close to the complement of the applied case. Whatever holds under them is not obviously the thing a management course is describing.

And the fourth condition is methodological rather than substantive. That the asymmetry appears when a researcher codes free-text answers, and not when participants choose from fixed options, is a fact about measurement. It is compatible with a real effect that only free response can detect, and equally compatible with coder expectation, and the abstract does not settle which.

The Valence Reversal

The finding that reframes the whole thing.

"In addition, the asymmetry held for negative events, but a reverse asymmetry held for positive events. This valence effect may indicate a self-serving pattern in attribution, but across valence, no actor-observer asymmetry exists."[1]

Three observations, ours.

Read what a reversal means. For bad outcomes, people explained their own situationally and others' dispositionally. For good outcomes, the pattern flipped: their own dispositionally, others' situationally.

That is not an actor-observer effect at all. It is a self-serving one, and the author says so: it may indicate a self-serving pattern in attribution[1].

And the two halves cancel. Because the pattern reverses with valence, averaging across positive and negative events leaves nothing, which is exactly what the headline numbers show.

Our own summary: the data are consistent with people arranging explanations to their own credit, which is a different and older idea than the one being tested, and which the perspective framing had obscured for thirty-five years.

The Author's Own Sentence

The conclusion, quoted so that nobody has to take our word for the strength of it.

"Across valence, no actor-observer asymmetry exists."[1]

Two observations, ours.

There is no hedge in that sentence. Not "small", not "weaker than believed", not "conditional". The claim is that it does not exist.

Contrast the sixteenth article in this series, where Tetlock's own careful "modestly" was inflated into "crushed" in transmission. Here the author's own wording is stronger than what circulates, and the transmission failure runs the other way: a categorical negative that has largely not been transmitted at all.

A Line In The Reference List

Something we noticed while reading the paper's bibliography.

Among the studies included is Garrido and de la Torre Garcia (1981), in Revista de Psicologia General y Aplicada, titled Actor-observador. Una hipotesis no confirmada en la theoria de la atribucion, glossed in the reference list as "Actor-observer: An unconfirmed hypothesis in attribution theory"[4].

Two observations, ours.

That is 1981. A paper announcing in its title that the hypothesis was unconfirmed sat in the literature for a quarter of a century before the meta-analysis, while the effect was being described as firmly established.

It was published in Spanish, in a Spanish-language journal. We are not in a position to say whether that is why it was overlooked, and we are not asserting it was. We record it because a reader assessing how a field concluded something was settled should know the disconfirming title was on the shelf.

The Citation That Proves The Point

An illustration we found while sourcing this article. We report it carefully.

On a publisher page hosting the meta-analysis, third-party text citing it reads: "One well-documented bias is the actor-observer asymmetry (Jones and Nisbett, 1971), where individuals tend to attribute their own actions to external circumstances but perceive others' actions as resulting from internal traits (Malle, 2006)."[3]

Four things about how we are treating this, all ours.

The citation attached to that sentence is the meta-analysis that concluded no such asymmetry exists. It is being cited as authority for the claim it refuted.

We do not name the authors and are not criticising them. This is an extremely easy mistake to make, and the point is structural rather than personal.

We found the text on a publisher page alongside the paper and did not obtain the citing work itself, so we cannot see its full context.

And it is the cleanest available demonstration of why this series checks primary sources. A paper is cited for the opposite of its finding, in a passage sitting on the same page as the paper. The first article in this series argued that a reputation and a finding are different objects; here they have separated so far that one is being used to support the other's negation.

Dead, Long Live The Other One

What the field did next, reported from a title.

A reference list identifies Gawronski, B. (2004), Theory-based bias correction in dispositional inference: The fundamental attribution error is dead, long live the correspondence bias, in the European Review of Social Psychology, 15(1), 183–217[5].

We did not obtain this paper and report only its title and citation.

Two observations, ours.

The title is an argument in itself: it proposes retiring the fundamental attribution error as a construct while retaining the correspondence bias, which is precisely the distinction the third section of this article drew.

And it is dated 2004, two years before the actor-observer meta-analysis. So by the middle of the last decade, serious people were publicly arguing that two of the three constructs in this territory should be retired or renamed, while all three continued to circulate together in applied writing as one idea.

What Survives

Being fair to a literature this article has spent most of its length dismantling.

Four things, and this assessment is ours.

The self-serving pattern. The valence reversal is a finding, not a null, and the author names it.

The four conditional asymmetries, which are real within their conditions and are reported by the same analysis that found the general effect absent.

The correspondence bias, which is a different construct that this meta-analysis did not test and which at least one paper argues should be the surviving one[5].

And a related finding from the same author's other work, recorded in an indexed description: "There was no evidence for the traditional actor-observer hypothesis, but systematic support was found for the actor-observer asymmetries hypothesized by the folk-conceptual theory."[4] We did not obtain that paper, and note only that the author proposes a replacement rather than leaving a hole.

The Version A Manager Should Keep

Our own reasoning about what to do with this, given that the popular version is gone.

The management use of this idea is usually: before concluding an employee is careless, consider their circumstances.

Three observations.

That advice is not damaged by this meta-analysis, because it does not depend on the asymmetry being real. It depends only on situations sometimes explaining behaviour, which is not in dispute.

But the justification has to change. You cannot support it by saying research shows managers systematically over-attribute to character, because the study of that perspective difference found nothing across valence.

And the self-serving finding gives a better-supported version: on the valence result, the pattern that appeared was people explaining bad outcomes in the way that flattered them. In a performance review, that predicts something specific and checkable. Both parties will tend to construct the account that reflects better on themselves, and both will do it in good faith.

That is a more useful warning than the original, because it applies to the manager and the employee equally, and it does not require anybody to be systematically biased against anybody.

A Theme Running Through This Series

Placing this alongside its neighbours. Ours.

Three articles here have now landed on the same distinction between a property of a person and a property of a situation.

The burnout article found the WHO definition locating the condition in chronic workplace stress that has not been successfully managed, which describes an environment.

The financial worry article found the strongest result comparing the same farmer to himself, which cannot support a claim about kinds of people at all.

This one finds that the general tendency to make that mistake, at least in the perspective-difference form, was not there.

Two observations.

Those first two are cases where the situational reading is better supported than the dispositional one, on their own evidence.

And this article is a caution against the obvious next move, which is to adopt a general rule that everyone always over-attributes to character. That rule is exactly what 173 studies failed to find, and a series arguing for evidence would be careless to smuggle it back in as a theme.

The Asymmetry That Does Matter

A closing observation, ours, about what is genuinely asymmetric in a workplace.

Set aside attribution style entirely. Two asymmetries remain, and neither needs this literature.

Information. You know why you were late. You do not know why they were. That is not a bias, it is a difference in access, and no amount of good faith closes it.

Consequence. When a manager forms a view about an employee's character, it affects assignments, promotion and pay. When an employee forms a view about a manager's character, it usually does not.

Our own conclusion: the practically important asymmetry in a workplace is in what each party knows and what each party's judgment can do, and both of those are structural facts you can address by asking and by process. Neither requires a contested effect to justify it.

What To Do

Stop citing the actor-observer asymmetry as established. A meta-analysis of 173 studies reports average effect sizes from minus 0.016 to 0.095, and its author concludes that across valence no such asymmetry exists.

Keep the three constructs separate. The actor-observer hypothesis, the fundamental attribution error and the correspondence bias are distinct claims, and the negative result applies to the first.

Use the self-serving version instead. The valence reversal is a positive finding, applies to both parties equally, and requires nobody to be biased against anybody.

Keep the practical advice, change its justification. Considering someone's circumstances before judging their character is sound, and it does not need this research to support it.

Expect both sides of a review to flatter themselves. That is what the surviving finding predicts, in good faith, on both sides of the table.

Address the information gap directly. You know your own reasons and not theirs, which is a difference in access rather than a bias, and it is closed by asking.

Notice whose judgment carries consequences. A manager's view of an employee's character moves assignments and pay; the reverse usually does not, and that asymmetry is structural.

Check what a citation actually says. This article's clearest example is a passage citing the refuting meta-analysis as authority for the refuted claim.

The Limits Of This Analysis

Several caveats matter. This article reviews psychological research and is not management or human resources advice. Everything is verified to August 2026. We obtained the published abstract in full but did not obtain the full text, and report no analysis beyond what the abstract states; we have no detail of the coding scheme, the inclusion criteria, or the moderator analyses beyond their summary. We did not obtain any of the 173 studies, the Ross (1977) chapter, the Gilbert and Malone (1995) paper, the Gawronski (2004) paper, or the folk-conceptual theory work, and report several of these as citations and titles only, with no findings. Our conversions of the reported effect sizes into superiority and overlap percentages are our own, computed under standard normal assumptions, and appear in no paper. The observation about a third party citing the meta-analysis for the claim it refuted rests on text found on a publisher page; we did not obtain the citing work, cannot see its full context, and do not name its authors. Our note on the 1981 Spanish-language paper records only its title as glossed in a reference list; we did not obtain it and make no claim about why it was overlooked. The analysis of what the four moderating conditions have in common, the reading of the valence result, the recommended managerial substitution, and the section on information and consequence asymmetries are our own reasoning, not findings. Roughly twenty years of subsequent literature was not reviewed, and later work may have revised this picture.

Related Reading

Behavioural Finance The Same Farmer Where a within-person design settled that a finding was about a situation, not a person. No. 20 A Conflict Resolved One found happiness plateaus above seventy-five thousand dollars. The other found it never stops rising. Rather than argue, they pooled their data,... Behavioural Finance Point Zero Eight The other effect whose size sits near the threshold of mattering at all.

Frequently Asked Questions

Do people really excuse themselves and blame others?

Not in the way the classic hypothesis claims. A meta-analysis of 173 published studies found average effect sizes from minus 0.016 to 0.095, and concluded that across valence no actor-observer asymmetry exists. The pattern that did appear was self-serving rather than perspective-based.

What is the self-serving pattern?

The asymmetry held for negative events but reversed for positive ones. For bad outcomes people explained their own situationally and others' dispositionally; for good outcomes the reverse. The author notes this may indicate a self-serving pattern, and because the halves cancel, the overall average is nothing.

Is the fundamental attribution error also dead?

That is a separate construct and this meta-analysis did not test it. We note a 2004 paper whose title argues the fundamental attribution error is dead and the correspondence bias should survive, but we did not obtain it and report only the title.

Should I stop considering people's circumstances?

No. That advice never depended on the asymmetry being real, only on situations sometimes explaining behaviour, which nobody disputes. What has to change is the justification: you cannot support it by claiming research shows managers systematically over-attribute to character.

What should a manager use instead?

The self-serving finding, which is better supported and more useful because it applies to both parties equally. In a review, expect each side to construct the account that reflects better on them, in good faith, and design the conversation accordingly.

How is a refuted effect still taught?

We found the clearest illustration while researching this article: a passage citing the meta-analysis itself as authority for the asymmetry it refuted. We do not name the authors, because the point is structural. A finding and its reputation are different objects, and here they have fully separated.

References

1. Malle, B. F. (2006). The actor-observer asymmetry in attribution: a (surprising) meta-analysis. Psychological Bulletin, 132(6), 895–919. DOI 10.1037/0033-2909.132.6.895, published abstract via PubMed, on the actor-observer hypothesis of Jones and Nisbett (1971) stating that people tend to explain their own behavior with situation causes and other people's behavior with person causes; on the asymmetry being widely known in psychology and described as robust, firmly established, and pervasive; on a meta-analysis of 173 published studies revealing average effect sizes from d = -0.016 to d = 0.095; on a moderator analysis showing the asymmetry held only when the actor was portrayed as highly idiosyncratic, when hypothetical events were explained, when actor and observer were intimates, or when free-response explanations were coded; on the asymmetry holding for negative events but a reverse asymmetry holding for positive events; and on the conclusion that this valence effect may indicate a self-serving pattern in attribution, but that across valence, no actor-observer asymmetry exists. Note: we obtained the published abstract in full; we did not obtain the full text and report no analysis beyond what the abstract states. pubmed.ncbi.nlm.nih.gov
2. Author's institutional copy of Malle (2006), hosted by the author's university laboratory, reproducing the abstract identically and confirming the paper's title including the parenthetical "surprising". Note: the author's own institutional copy; we obtained the first page. research.clps.brown.edu
3. Publisher record for Malle (2006), reproducing the abstract, listing among the paper's figures a funnel plot between each raw effect size and its standard error across the 173 studies and a plot of the asymmetry as a function of time of hypothesis test across seven roughly equal-sized subsets of studies; and separately carrying third-party text describing the actor-observer asymmetry as a well-documented bias where individuals attribute their own actions to external circumstances but perceive others' actions as resulting from internal traits, with Malle (2006) given as the supporting citation. Note: a publisher record. The third-party passage is reported in this article as an illustration; we did not obtain the citing work, cannot see its full context, and do not name its authors. researchgate.net
4. Hosted copies of Malle (2006) including portions of its reference list, confirming the abstract; recording Garrido, M. E., & de la Torre Garcia, A. (1981), Actor-observador. Una hipotesis no confirmada en la theoria de la atribucion, glossed as "Actor-observer: An unconfirmed hypothesis in attribution theory", Revista de Psicologia General y Aplicada, 36, 697–718; recording Funder, D. C. (1987), Errors and mistakes: Evaluating the accuracy of social judgment, Psychological Bulletin, 101, 75–90; and carrying an indexed description of related work by Malle, Knobe and Nelson stating there was no evidence for the traditional actor-observer hypothesis but systematic support for the actor-observer asymmetries hypothesized by the folk-conceptual theory. Note: hosted copies and indexed descriptions. We did not obtain the 1981 paper, the Funder paper, or the folk-conceptual theory work, and report titles and citations only. academia.edu
5. Reference list on a decision-science reference site, identifying Ross, L. (1977), The intuitive psychologist and his shortcomings: Distortions in the attribution process, in Berkowitz (ed.), Advances in Experimental Social Psychology, 10, 173–220; Gawronski, B. (2004), Theory-based bias correction in dispositional inference: The fundamental attribution error is dead, long live the correspondence bias, European Review of Social Psychology, 15(1), 183–217; and Gawronski, B. (2007), Correspondence bias, in Baumeister and Vohs (eds.), Encyclopedia of Social Psychology, 194–195. Note: a commercial reference website's citation list. We obtained none of these works and report titles and citations only, with no findings. thedecisionlab.com
6. Semantic Scholar record for Malle (2006), reproducing the abstract and moderator findings, and carrying indexed descriptions of related literature including that correspondence bias, citing Gilbert and Jones (1986), and Jones and Nisbett's (1971) actor-observer hypothesis are treated as distinct constructs in research on everyday explanation; and recording Gilbert, D. T., & Malone, P. S. (1995), The correspondence bias, Psychological Bulletin, 117, 21–38. Note: a bibliographic record with third-party indexed summaries; we did not obtain the correspondence bias literature. semanticscholar.org

This article reviews psychological research and is not management or human resources advice. The meta-analysis was not obtained in full; nothing is reported beyond its published abstract. None of the 173 studies, nor the works on the fundamental attribution error and correspondence bias, were obtained, and several appear as titles and citations only. The conversions of effect sizes into superiority and overlap percentages are the authors' own. The observation about a third party citing the meta-analysis for the claim it refuted rests on text found on a publisher page; the citing work was not obtained and its authors are not named.