Warning: Warnings Can Backfire Even When They Provide New and Important Information

We clarify the conditions under which warnings that provide useful information backfire. Our analysis is based on three observations: (1) warnings can increase the attention given to the warned-against behavior, (2) in many settings, counterproductive warned-against behaviors (like texting while driving) are typically rewarding, and (3) decisions from experience reflect insufficient sensitivity to rare outcomes. A simple model that abstracts the implications of these observations predicts that a backfiring effect initially emerges when the warned-against behavior is unlikely to be considered without the warning. In addition, the model predicts an increase in the choice rate of the warned-against behavior, with experience, when the probability of losing from the warned-against behavior is low. Experiment 1 tests and supports these predictions in the context of warnings concerning risky clicks. Experiments 2 and 3 demonstrate the generality of these predictions in more complex settings that involve warnings concerning danger zones and misinformation.