Group Attribution Error: Why a Verdict Is Not Every Juror's View

A group's decision is a product of its members' views and a rule for combining them, and the group attribution error is the habit of reading the product as if it were the views alone. A jury that convicts 7 to 5 is reported as a jury that convicted; the five who voted the other way disappear into the verdict. A country whose parliament passes a law is described as wanting it; the margin and the turnout vanish into the outcome. Scott Allison and David Messick named the error in 1985 after showing that people infer a typical member's attitude from a group's decision even when the decision was made by a single leader and the members were never consulted.

The error is a cousin of the better-known fundamental attribution error, in which a person's behavior is explained by their disposition and the situation that constrained it is overlooked. Here the "behavior" is a group's outcome and the overlooked situation is the decision rule. The rule can be anything from unanimity to a single decider, and the same outcome under different rules says different things about the members: a unanimous verdict does tell you about each juror, a 7 to 5 verdict tells you less, and a chair's decision tells you nothing at all about the eight people who did not decide. People treat all three as though they were the first.

Allison and colleagues went on to show that the inference runs in several directions. Diane Mackie and Allison found in 1987 that when a group's decision changed between two occasions, observers inferred that the members' attitudes had changed, even when the outcome had shifted because the rule or the turnout had, and they called this the illusion of group attitude change. Leila Worth, Allison and Messick found in 1987 that the error reaches people's own groups too, though more weakly: participants inferred a closer match between a decision and members' attitudes for another group than for their own.

The shortcut survives because outcomes are public and simple; the counts and rules behind them are often unreported and always more effort to use. And the inference is often right enough, since most groups do lean the way they decide. The error is in forgetting how far the outcome can diverge from the members, and in how little attention a stated rule receives even when it is in plain sight, as it was in the three sliders above.

Allison and Messick published "The group attribution error" in the Journal of Experimental Social Psychology in 1985. The paper reported a series of experiments in which participants read about a decision reached by a group at a national, state or local level, and then rated the attitude of a typical member of that group toward the matter decided.

The decision rule was the manipulation. In some versions the matter was settled by a single leader with no popular vote; in others by a popular vote of more than 90 percent; in others by a vote of approximately half the population. If participants had been using the rule, their inferences should have differed sharply: a 90 percent vote justifies a confident inference about the typical member, a bare majority a weak one, and a leader's decree none. Instead, participants inferred in all three cases that the typical member's attitude matched the group's outcome. The decision rule, which was stated in the materials, barely moved the ratings.

The 1985 paper reported the pattern as a finding about inference rather than as a measured percentage, and this page does not quote a rating from it. Later work filled in the boundaries. Mackie and Allison showed in 1987 that observers who saw a group's decision change between two votes inferred a change in the members' attitudes even when the change in outcome was produced by a change in the rule. A 1996 review by Allison, Mackie and Messick placed the error within a broader family of outcome biases, in which what happened is used as the evidence for what people wanted or believed.

The demo on this page follows the logic of the 1985 materials with three decisions whose rules are stated, a 7 to 5 jury, a chair deciding alone against a known head count, and a referendum passed on low turnout. It uses arithmetic as the benchmark, because the share of members who actually voted for an outcome is the only thing a stated count can support. A visitor who places a random juror at 58 percent has used the count; a visitor who places the juror at 85 percent has used the verdict.

The pattern appears wherever a collective result is reported without the count or the rule behind it. The rows are illustrations, not findings from the cited studies.

The error lives in the gap between an outcome and the numbers that produced it, so the counter is to ask for the numbers.

• Ask for the count. Before attributing a view to a member, find out the vote share. Seven of twelve is 58 percent, not a consensus.
• Ask for the rule. A decision by a leader, a committee or a bare majority says less about each member than a unanimous one, and nothing at all in the first case.
• Ask for the turnout. A 90 percent result on a 40 percent turnout means 36 percent of the eligible group voted yes; the rest did not say.
• Describe results as counts. "Passed 52 to 48" preserves the information; "the town decided" discards it.
• Check the direction of your inference. When you learn what a group did, notice the moment you start to picture its members agreeing, and ask what in the count supports the picture.

The group attribution error sits among several biases about groups and their members, and it is worth separating from its nearest neighbors.

• It is not the fundamental attribution error, which explains an individual's behavior by disposition while overlooking the situation. The group attribution error applies the same move to a group outcome, with the decision rule as the overlooked situation, and Allison and Messick introduced it as the group-level parallel.
• It is not the ultimate attribution error, which is explaining an out-group member's bad behavior by their group's character and their good behavior by luck. That concerns the direction of explanations for individuals; the group attribution error concerns inferring individuals' attitudes from a collective result, in-group or out-group alike.
• It is not out-group homogeneity bias, which is seeing less variety in another group than in your own. That bias assumes agreement among members in general; the group attribution error assumes agreement with one specific outcome, and Worth, Allison and Messick found it operating, in weaker form, within people's own groups as well.