Fundamental attribution error

The quiet new coworker

On her first day, a new team member joins the group for lunch but says almost nothing. Afterward, someone remarks, “She seems pretty cold.”

A first day among strangers makes most people quiet and careful. Her behavior is what the situation would produce in nearly anyone, so it carries almost no information about whether she is warm or cold. The judgment skips that step and treats the behavior as a view of her personality.

The assigned side

A book club decides to try a debate. The organizer draws names to pick who will argue for and against the month’s novel, and Dana is drawn to defend it. She makes a strong case. A few weeks later, when someone suggests another book by the same author, a member says, “Dana will be thrilled. She loved the last one.”

Everyone in the room watched Dana’s side get picked from a hat, and her argument was still taken as her real opinion. Knowing about the situation isn’t enough on its own: people tend to discount it, but not by enough. This is the form the research has studied most closely (see Evidence).

The people who wrote the questions

At a team trivia night, each round is hosted by a different coworker, who writes questions about subjects they know well. By the end of the evening, the consensus is that the hosts are the smartest people in the office.

Here the situation is easy to miss. Each host picked questions from their own store of knowledge, while the players had to answer questions on topics they didn’t choose. Any host would look brilliant and any player would look slow, whoever they were. In a 1977 experiment built on this setup, with the roles assigned at random, both the contestants and the onlookers rated the question-writers as more knowledgeable.

Inferring something about a person from their behavior is often the right thing to do. It’s legitimate:

• When the behavior repeats across many different situations. If the settings vary and the behavior doesn’t, the person is the common factor.
• When the person freely chose it, and other choices were easy. Chosen behavior says more than required behavior.
• When the behavior goes against what the situation was pushing for. Acting against the pressure of a situation is stronger evidence about a person, not weaker.
• When other people in the same situation acted differently. Then the situation alone can’t explain it.
• When you only need to predict behavior in the same situation. If you’ll deal with someone in the same role again, how they acted in that role is a fair guide, even if it isn’t a guide to their character.

Spotting the error doesn’t show that the person lacks the trait. The quiet coworker may really be reserved. The error is treating behavior the situation explains as evidence for the trait.

The test: would most people have done the same thing in that position?

Late to everything

Over five years, a friend has arrived late to casual dinners, to a wedding he was in, to his own birthday party and to two job interviews, in different cities and at different times of day. You suggest to a new friend that they tell him to arrive half an hour before the real start time.

This is a judgment about his character from his behavior, but the behavior has been observed across many situations that have little in common except him. No single situation explains the whole pattern, so inferring a lasting habit is sound. This is the repeated across situations condition above.

Speaking against the room

In a meeting, a manager makes it clear he wants to approve a vendor contract, and everyone who speaks agrees. One analyst says she thinks it’s a bad deal and explains why. Afterward, a colleague says, “She’s willing to push back when she thinks something’s wrong.”

The situation was pushing toward agreement, and she went the other way. Behavior that runs against the pressure of a situation is more informative about a person, not less. The colleague also draws a narrow conclusion that fits what was observed. This is the against the situation condition above.

Researchers don’t agree on a single cause. Daniel Gilbert and Patrick Malone argued in a 1995 review that several different failures produce the same result:

• The situation is invisible. Many of the forces that shape behavior (a rule, a request made earlier, the advantage of writing the questions) aren’t visible in the moment you see the behavior.
• Expectations are off. You may know about the situation but underestimate how strongly it pushes people, so ordinary behavior in it looks unusual.
• The situation colors how the behavior looks. Knowing someone is under pressure can make their behavior look more extreme than it was, which then cancels out the allowance you made for the pressure.
• The correction comes too late. Evidence suggests people form an impression of the person quickly and automatically, then adjust for the situation as a slower, effortful second step. When attention is taken up with something else, the adjustment tends not to happen.

Culture plays a part. Explaining behavior by personal traits is common everywhere, but a 1999 review of cross-cultural research found that people in East Asian cultures give more weight to context, and are more likely to avoid the error when the situational pressures are obvious.

A few remedies have been tested in the lab:

• Expect to explain your judgment, before you form it. In a 1985 study, people who knew in advance that they would have to justify their impressions to someone else were more cautious about inferring the true opinions of writers who had been assigned their position, while still drawing strong conclusions about writers who had chosen freely. Being told about the need to justify only after reading the essay did not have this effect.
• Don’t judge people while your attention is elsewhere. In experiments where observers were given a distracting task, they failed to adjust for the situation even when they had been told about it, and in one study even when they had been made to memorize it.
• Training can help, at least on tests. In a 2015 study, a single session of an interactive training game reduced people’s scores on a fundamental attribution error measure immediately and about two months later; an instructional video helped less. The measures were designed by the researchers, so it’s less clear how far the improvement carries over to everyday judgments.

The practical question behind all three is the test above: what would most people have done in that position?

Status: replicates robustly. The basic pattern, inferring a disposition from behavior that the situation largely explains, has been found for more than 50 years and in a large preregistered replication run in many countries. What is still debated is why it happens and how large it is in different populations.

• Jones and Harris (1967) asked Duke University students to estimate the true attitude of a student who had written a short essay about Fidel Castro’s Cuba. Participants were told the writer had either chosen a side freely or been assigned it by a course instructor as an exam question. Even in the assigned condition, writers of pro-Castro essays were rated as considerably more pro-Castro than writers of anti-Castro essays (about 44 versus 23 on a scale from 10 to 70). The authors had set out to test a different hypothesis, which wasn’t confirmed; they described the correspondence in the assigned condition as the most striking result.
• Ross (1977) coined the term fundamental attribution error for the tendency to underestimate situational influences and overestimate dispositional ones, and argued that it explains findings from many different experiments.
• Ross, Amabile and Steinmetz (1977) randomly assigned people to write difficult general-knowledge questions or answer them. Contestants and observers rated the questioners as more knowledgeable, failing to allow for the advantage the role gave them.
• Gilbert and Malone (1995) reviewed the literature and argued that the effect, which they called the correspondence bias, is several phenomena with separate causes (see “Why it happens”). Gilbert, Pelham and Krull (1988) had shown that distracted observers failed to correct their impressions for situational information.
• Many Labs 2 (Klein et al., 2018), a preregistered replication project run by dozens of labs in countries around the world, tested this with 7,197 participants. It the essay paradigm with essays about capital punishment whose positions participants knew had been assigned. Writers assigned the “for” side were judged far more in favor than writers assigned the “against” side, a large effect (d = 1.82). The effect was stronger when participants found the essay more persuasive.
• Bauman and Skitka (2010) tested the essay paradigm on a nationally representative sample of American adults, not only students, and found that about half of them (53%) showed the bias.

Culture. Choi, Nisbett and Norenzayan (1999) reviewed evidence that East Asians show the bias less, mainly because they give more weight to situations when those are salient. The Many Labs 2 team found no difference between broadly “WEIRD” (Western, educated, industrialized, rich, democratic) and less-WEIRD samples in an exploratory comparison. The authors of the original cultural study, Miyamoto and Kitayama (2018), reanalyzed the same data and did find a stronger bias in the American samples than in the Japanese one, while noting that other countries’ results didn’t follow a simple East–West pattern. The effect exists across cultures; how much it varies is still open.

Names. “Fundamental attribution error” and “correspondence bias” are usually used as synonyms, and this entry treats them that way. Some researchers separate them: Gawronski (2004) uses correspondence bias for the observed pattern and argues against Ross’s explanation of it (that people simply underestimate how much situations matter), concluding that the bias comes from several different ways people apply, misapply or set aside their beliefs about situations. The pattern is robust; the explanation built into the name is not settled.

What’s still uncertain. Besides the mechanism, it’s hard to say exactly how much of the measured effect is a mistake. Jones and Harris ran a second experiment partly to rule out participants doubting that the essays were really assigned. And an assigned essay can still carry a little real information about its writer: in Many Labs 2, the effect grew when participants found the essay more persuasive.

Popular version. The effect is often described as “people blame character, never circumstances”, as if everyone did it all the time. The research claim is narrower: on average, people under-adjust for situations when inferring what someone is like. Individuals and cultures vary.

Not the actor–observer asymmetry. A different claim, that people explain their own behavior by situations and other people’s by traits, is often presented alongside this one. A meta-analysis of 173 studies (Malle, 2006) found average effects close to zero for that asymmetry, holding only under particular conditions. That result doesn’t bear on the fundamental attribution error, which is about how observers adjust for situations, not about a difference between self and others.