Learning versus Unlearning: An Experiment on Retractions

Widely discredited ideas nevertheless persist. Why do we fail to “unlearn”? We study the effectiveness of retractions—the revoking of earlier information—in correcting beliefs. Our experimental design identifies belief updating from retractions— unlearning—and compares it to updating from equivalent new information—learning. Subjects do not fully unlearn from retractions and update approximately one-third less from retractions versus new information. Although we document several well-known biases in belief updating, our results require an explanation that treats retractions as intrinsically different. We find evidence for one such mechanism, while ruling out several others: retractions are more complex than direct information.