Debiasing Errors: Insufficient, Unnecessary, and Overcorrection

WHEN PEOPLE TRY TO CORRECT, HOW SUCCESSFUL ARE THEY?

Even in the rare instances in which people believe that their judgments are biased, they may not successfully debias these judgments. In fact, their corrected judgments might be worse than their uncorrected ones.

The dilemma people face is similar to the predicament of a hiker who arrives at the intersection of several trails. There is a sign pointing toward one of the paths indicating that it leads to the hiker’s destination. However, the hiker has been told that a mischievous boy often moves the sign so that it points in the wrong direction, and she has just seen a boy running away from the sign, laughing gleefully. She is pretty sure that the sign is pointing to the wrong path (i.e., that it is “contaminated”), but how should she correct for this fact? Given that the sign is pointing to the trail on the far left, should she take the one on the far right? Or should she assume that the boy only had time to move the sign a little, and therefore take the middle trail? Or that he is a clever boy who tried to convince her that he moved the sign, when in fact he did not? Just as people are at the mercy of their theories when deciding whether a response is biased, so are they at the mercy of their theories when deciding how to correct for this bias.

Unfortunately, people’s theories about debiasing do not appear to be particularly accurate. People who are exposed to contaminating information and engage in debiasing rarely end up with judgments similar to people who were not exposed to the contaminant. Three kinds of errors have been found: insufficient correction (debiasing in the direction of accuracy that does not go far enough), unnecessary correction (debiasing when there was no bias to start with), and overcorrection (too much debiasing, such that judgments end up biased in the opposite direction).

A number of studies found evidence for insufficient correction, which is the best of the three types of errors (because the corrected judgment is more accurate than the uncorrected one). Petty and Wegener (1993, Study 4), for example, found that rating the desirability of vacations in very appealing locations, such as Hawaii, lowered people’s ratings of vacations in American cities such as Indianapolis and Kansas City. Giving people a subtle cue that their initial ratings might bias their later ones led to some, but not complete, debiasing. People in the subtle cue condition formed judgments that were still biased, but not as biased as people who did not receive the subtle cue. These results suggest that some debiasing is better than no debiasing.

However, this same study found evidence for unnecessary correction in another condition. Some participants received a more blatant cue that their ratings of the American cities might be biased. People in the control condition, who did not first rate vacations in Hawaii but received the blatant warning, engaged in unnecessary correction. That is, they seem to have assumed that their evaluations of the American cities were biased when in fact they were not. They became more negatively disposed toward vacations in these cities, even though their evaluations had not been biased in the first place.

Stapel, Martin, and Schwarz (1998) found evidence for overcorrection. Similar to the Petty and Wegener (1993) study, people first rated the desirability of the weather in tropical locations and then rated the desirability of the weather in midwestern cities. In the absence of any warnings, this led to a contrast effect: Compared to a control group who did not first rate the tropical locations, people rated the weather in the midwestern cities as less desirable. When people were warned to avoid unwanted influences on their ratings, they overcorrected such that they showed an assimilation effect: Compared to the control group, they rated the weather in the midwestern cities as more desirable.

In sum, just because people attempt to correct a judgment they perceive to be biased is no guarantee that their result will be a more accurate judgment. Either unnecessary correction or overcorrection might occur, resulting in judgments that are more biased than they were before they were “corrected.”