Defensive Attribution Hypothesis: When Worse Damage Means More Blame

The hypothesis starts from a discomfort. An accident that nobody could have prevented is a reminder that the same thing could happen to the person hearing about it, and the worse the accident, the less welcome that reminder is. Assigning responsibility to someone removes it: if Lennie should have had the brakes checked, then a person who does have their brakes checked is safe. On that account blame rises with the severity of the outcome, even when the behavior being judged has not changed at all.

Walster's own data say something precise about how the blame is produced. Her subjects did not describe Lennie as more careless when the accident was severe; the ratings of what he had probably done about the handbrake, the wheels and the brake check barely moved. What moved were the standards they applied. Asked how often a person is morally obliged to have the brakes checked, subjects in the severe conditions named shorter intervals. The behavior was perceived the same way and judged against a stricter rule.

Kelly Shaver added the other half in 1970. If blame is defensive, it should depend not only on how bad the outcome was but on how much the observer resembles the person who caused it, because blaming someone like you offers no protection at all. Shaver's experiments varied personal similarity and found that observers who saw themselves in the actor assigned less responsibility, not more, which is the version of the hypothesis most of the later literature tests.

The evidence has not been kind to the simple form. Walster herself did not reproduce the severity effect in a 1967 follow-up with a different scenario. Jerry Burger's 1981 meta-analysis pooled 22 studies and found the severity main effect statistically significant but weak, while the relationship Shaver predicted between similarity and lower responsibility held up well. The careful statement is that outcome severity moves blame a little and unreliably, and that who the observer identifies with moves it more.

Elaine Walster published "Assignment of responsibility for an accident" in the Journal of Personality and Social Psychology in 1966. Her 88 subjects, 44 women and 44 men from an introductory psychology course at the University of Minnesota, each heard one of four taped accounts of a young man called Lennie B. The tapes were identical except for the ending: his mother described him, a craft instructor and a history teacher spoke about him, and a neighbor described the car rolling away from the top of a hill after the rusted handbrake cable broke.

The endings crossed two things: whether the damage was trivial or serious, and whether anyone other than Lennie could have been hurt. On two tapes only his own car was at stake, and it either picked up a dent against a stump or was destroyed against a tree. On the other two there was a store at the bottom of the hill, and the car either stopped at the stump or went through the window and injured the grocer and a child. Subjects then rated Lennie's responsibility on a four-point scale running from not at all responsible to completely responsible.

Responsibility tracked the damage. The means were 2.5 and 3.0 for the pair where only Lennie's car was involved, and 2.6 and 3.2 for the pair with the store, and the overall difference between mild and severe outcomes was statistically significant. The carelessness Lennie was credited with did not rise with severity; the moral standards subjects reported did. Walster also found that the pattern in the store conditions came from the men in her sample, with the women rating Lennie about equally responsible whether the car reached the store or not, and she wrote that she could not explain it.

What followed is the part most summaries leave out. The severity effect proved hard to reproduce, starting with Walster's own 1967 study. Shaver's 1970 experiments reframed the hypothesis around personal relevance, and Burger's 1981 meta-analysis of 22 studies found weak support for severity and good support for the similarity prediction. The defensive attribution hypothesis survives as a description of when blame moves rather than as a law that worse outcomes always draw more of it.

The rows below are places where the same conduct is judged against different outcomes, not findings from the studies cited here.

The useful move is to separate the conduct from the outcome, because the bias works by letting the outcome speak for both.

• State what the person knew and could have done before you state what happened. Once the outcome is in the room, the standards shift without anyone noticing.
• Write the rule down before the case. If the question is how often brakes should be checked, answer it in general terms first, then apply it.
• Run the counterfactual in both directions. Ask how you would judge the same conduct if the car had stopped at the stump, and how you would judge it if the outcome had been worse still.
• Treat near misses as cases. Organizations that only investigate harm are training themselves on outcomes rather than on conduct.
• Check the identification. Blame can fall as well as rise: if the person in the story is a lot like you, the pressure runs toward excusing rather than blaming.
• In anything formal, separate the finding of fact from the measure of harm, and have different people do them where you can.

Three nearby entries are about judging a decision by how it turned out, and the differences are in the motive and the target.

• It is not outcome bias, which is the general tendency to rate a decision by its result rather than by what was known when it was made. Defensive attribution is a specific explanation for one case of that, built on the observer's need to feel safe.
• It is not the just-world hypothesis, the belief that people get what they deserve. The just-world account explains blaming the victim; defensive attribution concerns anyone who might have prevented the harm, and it predicts less blame as the observer comes to resemble them.
• It is not the fundamental attribution error, the habit of explaining behavior by character and overlooking the situation. That happens regardless of how things turned out, while defensive attribution is driven by the severity of the outcome.