Skip to main content
Category: Effects
Type: A difference between the cause given by the person who acted and the cause given by a person who watched, for the same act. The classic prediction gives the situation to the actor and a stable disposition to the watcher
Origin: Edward E. Jones and Richard E. Nisbett, The Actor and the Observer: Divergent Perceptions of the Causes of Behavior, General Learning Press, 1971. The pamphlet was not opened. The sentence below is the quotation in Bertram F. Malle, Psychological Bulletin 132(6), November 2006, pages 895–919, DOI 10.1037/0033-2909.132.6.895, which cites it as page 93. Malle’s footnote says the paper is often cited as the chapter in Jones, Kanouse, Kelley, Nisbett, Valins, and Weiner, eds., Attribution: Perceiving the Causes of Behavior, 1972, pages 79–94. That chapter was not opened. The counted case is Michael D. Storms, “Videotape and the attribution process: Reversing actors’ and observers’ points of view,” Journal of Personality and Social Psychology 27(2), 1973, pages 165–175, DOI 10.1037/h0034782. That PDF was read
Also known as: Actor–observer asymmetry, the name in the 2006 meta-analysis. It is not the Fundamental Attribution Error as one score, and it is not the Self-Serving Bias as one score
Quick Answer — Actor-Observer Bias is the claim that the person who acted and a person who watched explain one act with different causes. Jones and Nisbett stated it in 1971: the actor names the situation, and the watcher names a stable disposition. Storms (1973) moved those scores by turning a camera. Malle (2006) pooled 173 studies and 14,686 people and found an average classic gap of d = 0.095. After good and bad events are balanced, the adjusted averages sit at 0.042 and −0.021.

​
What is Actor-Observer Bias?

Actor-Observer Bias is a difference in the causes two people give for one act: the person who did it, and a person who watched.
actors tend to attribute the causes of their behavior to stimuli inherent in the situation, while observers tend to attribute behavior to stable dispositions of the actor
That sentence is Jones and Nisbett (1971), page 93, as Malle (2006) prints it. The 1971 pamphlet was not opened. In Malle’s scoring, a positive d means the watcher had a higher internal-minus-external score than the actor. That direction is the classic claim. A negative d is the reverse. A useful picture is a late arrival. The person who arrived names the train. The colleague who watched names a habit. The picture fails in two places that the 2006 pool actually measured. On successes, the average goes the other way. Across all events, the average classic gap is small. Fundamental Attribution Error is a watcher’s inference that a constrained act reveals a trait. Actor-Observer Bias is the gap between two explanations. Self-Serving Bias is a person’s claim on credit and refusal of blame. Malle reads the good-event and bad-event split as a relative of that pattern, and he also says the split is not, by itself, proof of a motive.

​
Actor-Observer Bias in 3 Depths

  • Beginner: When you explain your own delay with the train and a colleague’s delay with their character, write both sentences down. That pair is the claim. It is not yet a rate.
  • Practitioner: Before a review of a miss, ask the person who acted and a person who only saw the result each to give one cause in words. In the 2006 pool, a gap showed up more often in coded words than on rating scales, and more often for bad events than for good ones.
  • Advanced: The camera in one Yale room moved a situational rating. The average of 173 studies did not. A tactic that copies the camera, or copies d = 0.095, into a hiring meeting has left both samples.

​
Origin

The claim is the 1971 paper. Malle quotes Jones and Nisbett as offering two cognitive accounts, alongside self-justification, which they did not treat as the whole story. One account is where the eyes point. The watcher sees the actor. The actor looks out at the situation, and, as Malle quotes them on page 85, is “directed outward” because seeing one’s own behavior is hard. The other account is information. The actor has a history across situations. The watcher often has one slice. Malle’s pool did not support the prediction that strangers, who lack that history, would show the larger gap. Intimates showed the larger one. Storms (1973) tested the visual account with a videotape. Nisbett, Caputo, Legant, and Marecek (1973) are in Malle’s discussion as the suggestion that actors avoid internal causes in order to keep a sense of freedom. That 1973 PDF was not opened, so its sample sizes are not on this page. Malle says very few studies measured that sense of freedom. Among them, he reports, Miller and Norman (1975) found the opposite of the predicted pattern. Their paper was not opened.
I must conclude that after 35 years of research and more than 170 studies, the classic actor–observer hypothesis is—counter to what textbook descriptions and commonly held beliefs suggest—neither firmly established nor robust and general and that evidence for it is neither pervasive nor plentiful
That sentence is Malle’s discussion. He attaches it to Aronson (2002), Fiske and Taylor (1991), Jones (1976), and Watson (1982). Those four books and papers were not opened. The phrases he quotes from them — “pervasive,” “plentiful,” “robust and quite general,” “firmly established” — stay inside his citations.

​
Key Points

Four distinctions keep one conversation, one average, and one bad-event split from collapsing into a single slogan.
1

The score is a difference between two people

Malle’s H1 asks whether the watcher shows a greater preponderance of internal over external causes than the actor does. Across 113 articles, 173 studies, and 14,686 participants, published from 1971 through 2004, the random-effects estimate for that internal-minus-external score is d = 0.095, with a 95 percent interval from 0.032 to 0.159. The fixed-effects estimate is d = 0.032. The internal score alone is d = 0.062, interval −0.019 to 0.143. The external score alone is d = 0.023, interval −0.064 to 0.109. Neither component differs from zero. In correlation units the gap runs from r = −0.01 to r = .05. Sixty-eight studies sat at or below d = 0. Ninety-seven did not reach d = 0.20. The abstract’s range, from d = −0.016 to d = 0.095, is the span of those estimates, not a second sample.
2

A bad event and a good event pull opposite ways

Where the event’s valence was clear, bad events gave the classic direction: d = 0.241, interval 0.135 to 0.347, 61 studies, fixed-effects d = 0.106. Good events gave the reverse: d = −0.158, interval −0.315 to −0.002, 45 studies, fixed-effects d = −0.148. On bad events the internal score is d = 0.311, interval 0.169 to 0.469. The external score is d = −0.020, interval −0.164 to 0.123, which includes zero. Malle’s valence-adjusted averages, correcting for the mix of good and bad events, are d = 0.042 in the random-effects model and d = −0.021 in the fixed-effects model. Self-Serving Bias is the neighbor he names. He also says the pattern may not be a motive: an actor can be right to refuse a trait for a single miss.
3

Words, imagined events, and intimates are where a gap appeared

Content coding of spoken explanations, 20 studies, gave d = 0.318, interval 0.137 to 0.500. Rating scales, 152 studies, gave d = 0.062, interval −0.004 to 0.129. The contrast is Q(1, 170) = 6.7, p < .01. Hypothetical events, 24 studies, gave d = 0.280, interval 0.122 to 0.438. Real events, 149 studies, gave d = 0.059, interval −0.097 to 0.129. Q(171) = 6.3, p < .01. Intimates, defined as a parent and child, close friends, or partners, N = 29 studies, gave d = 0.245, interval 0.098 to 0.392. Strangers and acquaintances, N = 143, gave d = 0.072, interval −0.014 to 0.128. Q(1, 169) = 5.1, p < .05. Jones and Nisbett had predicted the smaller gap among intimates. The pool went the other way.
4

The camera is one room, then five later studies

Storms (1973) is the first perspective test in Malle’s set of six studies across five articles. Malle computes Storms’s standard effect as d = 0.482 and the perspective-switched reversal as d = −0.636. Five later studies, 1975 to 1984, averaged a reverse gap in the control condition, raw d = −0.229, and a similar value after the switch, raw d = −0.289. Pooled, six control conditions sit at d = −0.077, interval −0.385 to 0.065, and six switched conditions at d = −0.280, interval −0.509 to −0.051. With no classic gap in the controls, Malle says it is unclear what the switch achieved. One study he cites, Martin and Huang (1984), found no difference between a real switch and an irrelevant music video when the delay matched. That paper was not opened.

​
Applications

These four moves use splits the 1973 experiment and the 2006 pool actually make. They are not a claim that a camera, a written cause, or a reminder about bad events changes a court’s verdict, a classroom grade, or a household argument by a measured amount.

Write both causes before you pick one

Put the actor’s cause and the watcher’s cause on the same line. The claim is the difference. A meeting that records only the watcher’s trait word has not yet run the comparison Jones and Nisbett stated.

Split a miss from a success

The classic direction in this pool is the bad-event cell, d = 0.241. The good-event cell goes the other way, d = −0.158. A rule that always trusts the actor’s situation, or always trusts the watcher’s trait, erases that split.

Ask for a sentence, not only a slider

The coded explanations showed a gap. The rating scales, on average, did not. If the only record is a 1-to-9 mark for “personality” and a 1-to-9 mark for “situation,” you are in the cell whose average was d = 0.062.

Treat a replay as a question

In Storms’s room, actors who saw themselves lowered the situational sum from the mid-20s to 20.70, and their difference score rose to 6.80. Five later studies did not hand Malle a classic control gap to reverse. A replay can change what people mark. It is not a verdict that the situation was absent.

​
Case Study

One hundred and twenty Yale undergraduate men, strangers to one another, sat in 30 groups of four. Two were actors, called participants in the script, and they talked for five minutes to get acquainted. Two were observers. Each observer watched one matched actor. Everyone was told the session would be videotaped so the tape could help with the questionnaire. A random draw then made the session either a no-tape control or a tape session. In a tape session the experimenter said only one camera had worked, and every experimental group saw the tape of Actor 1. That lie assigned the four roles: the actor who saw himself, the actor who saw the partner he had already been looking at, the observer who saw the same actor again, and the observer who saw the other participant. No one, at debriefing, said they had suspected the tape was the manipulation. Twenty men sat in each of the six cells. The questionnaire asked, for friendliness, talkativeness, nervousness, and dominance, how important personal characteristics were and how important the situation was. Each cause was a 9-point scale from extremely unimportant to extremely important. The totals below sum the four scales, so a cause score runs from 4 to 36. The index is the dispositional sum minus the situational sum. A higher index is relatively more dispositional. With the original view repeated on tape, actors marked dispositional 26.10 and situational 25.95, index 0.15. Observers marked 27.10 and 22.20, index 4.90. With no tape, actors marked 27.35 and 25.10, index 2.25. Observers marked 27.30 and 22.50, index 4.80. The no-tape contrast is p < .12. The same-orientation contrast is p < .05. A tape that only repeated the live view did not differ from no tape, q = 1.79 for actors and q = 1 for observers. When the view was new, actors who saw themselves marked dispositional 27.50 and situational 20.70, index 6.80. Observers who saw the other participant marked 25.75 and 24.15, index 1.60. The role-by-orientation interaction is F = 9.72, df = 2/114, p < .001. The dispositional sums alone did not interact, F = 1.38, df = 2/114. The situational sums did, F = 5.78, df = 2/114, p < .005. The reversed contrast itself is p < .05. Dominance did not carry the effect, F < 1. Friendliness, talkativeness, and nervousness did. The lesson inside this room is that the visual orientation lined up with the situational rating, and that a self-view made actors relatively more dispositional than the matched observers. The boundary is the same table plus Malle’s later pool. The no-tape cell did not clear .05. The men, the five minutes, and the fake broken camera are the sample. Malle’s five later perspective studies did not reproduce a classic control gap for a switch to reverse.

​
Boundaries and Failure Modes

Actor-Observer Bias does not say that watchers name traits and actors name situations as a general rate. Malle’s conclusion is that the classic hypothesis is not firmly established. The component scores for internal causes and for external causes do not differ from zero. The valence-adjusted averages are 0.042 and −0.021. Two uses overshoot the table. One exports Storms’s 0.15, 4.90, 6.80, or 20.70 onto a class, a shop, a household, or another year. Those sums belong to 120 Yale men in a five-minute stranger conversation. The other exports d = 0.095, d = 0.241, or d = −0.158 onto a performance review as if the review were one of the 173 studies. A within-subjects split of valence, which Malle also tested, did not show the good-event and bad-event opposition. That test was read and its coefficient is not printed here, so this page does not pretend the opposition appears inside one person. The perspective boundary is sharper than a single famous study. Storms found the reversal he predicted. The five later studies in Malle’s set averaged a reverse gap already in the control condition. Martin and Huang (1984), as Malle summarizes them, matched a perspective switch against a music video and found no difference. Their paper was not opened. Malle, Heim, and Knorek (2006), cited in the same discussion as a later laboratory disconfirmation on person-versus-situation ratings, was not opened either. Jones and Nisbett’s intimacy prediction is a separate miss. Knowing the actor did not shrink the gap in this pool. It enlarged it. A practice that “corrects” only for strangers, and trusts intimates to be accurate, follows the prediction the studies did not keep.

​
Common Misconceptions

Three readings merge this claim with a different score, or promote one room to a law.
Fundamental Attribution Error is a judgment of someone else’s act, often an act the situation constrained. Actor-Observer Bias is a comparison of two explanations of one act. A study can find a trait inference in watchers and still find no actor-minus-watcher gap. Malle keeps the two claims apart. The existing page on the fundamental attribution error calls the asymmetry the same phenomenon. That page was not edited. The comparison on this page is the one Jones and Nisbett stated and the one Malle tested.
Textbooks said the evidence was plentiful. Malle’s pool is the count: d = 0.095 on the difference score, components indistinguishable from zero, and 97 studies below d = 0.20. A bad event, a coded sentence, an imagined event, or an intimate pair is where a larger gap appeared. A single late train is none of those counts. Confirmation Bias would keep the anecdote and drop the pool. The pool is the check.
Actors who saw themselves in Storms’s study cut the situational sum to 20.70 and posted the highest dispositional-minus-situational index in the table, 6.80. Storms himself warns that self-viewing may leave people “undersituational,” discounting a real situation that is still there. The later perspective studies did not give Malle a clean classic gap to erase. Naïve Realism is the further claim that one’s own view is the view of the facts. A camera changes a rating. It does not certify the rating.
These six pages sit beside Actor-Observer Bias. Each one asks a different question.

Fundamental Attribution Error

A watcher’s inference from another person’s act to a trait. Actor-Observer Bias asks whether that watcher and the actor explain the same act differently.

Self-Serving Bias

Taking credit for success and handing failure to the situation. Malle’s good-event and bad-event split points this way. His adjusted average across valence does not leave a general actor-watcher gap.

Naïve Realism

Treating one’s own perception as a direct reading of the facts. A camera can move an attribution score. It does not make that score the situation as it was.

False Consensus Effect

Expecting other people to share one’s own choice or view. Actor-Observer Bias is about the cause of an act, not about how common the act is thought to be.

Confirmation Bias

Keeping the evidence that fits a judgment already made. The textbook claim was repeated for decades. The 173-study average is the evidence that claim had not collected.

Hindsight Bias

Reading a finished outcome as having been predictable. A disposition named after the act is easy to treat as what one knew at the time. Predictability is a different question from whose cause was marked.

​
One-Line Takeaway

Write the actor’s cause and the watcher’s cause for the same act, and do not treat one five-minute conversation, or d = 0.095, as the rate for every miss and every success.