Prediction Biases: An Integrative Review. (June 2021)

• Record Type: Journal Article
• Authors: Yang, Yang; Hsee, Christopher K.; Li, Xilin
• Is Part Of: Current directions in psychological science. Volume 30:Number 3(2021)
• Page Start: 195
• Page End: 201
• Publication Date: 2021-06
• Subjects: prediction bias -- forecasting errors -- overgeneralization -- regression -- relevance insensitivity; Psychology -- Periodicals; 150
• DOI: 10.1177/0963721421990341
• Languages: English
• ISSNs: 0963-7214

Abstract

Research in psychology and related fields has documented a myriad of prediction biases, such as the underprediction of hedonic adaptation and the overprediction of other people's concern for fairness. These prediction biases are ostensibly independent, each with its own cause. We argue, however, that many of these seemingly disparate biases are specific instances of a general bias—situation insensitivity: People are insensitive to variations in the situational variable that underlies the target variable (the variable to be predicted). Consequently, when encountering a condition in which the situational variable is at one of its ends such that the value of the target variable is low, people overpredict the value; conversely, when encountering a condition in which the situational variable is at its other end such that the value of the target variable is high, people underpredict it. Most prior research documenting prediction biases has focused on only one end of the situational variable and therefore has shown either only an overprediction bias or only an underprediction bias. We argue that at the other end of the situational variable, the originally documented bias can disappear or even reverse. Our framework not only explains known biases but also predicts new biases.