I show that a simple intervention, asking people to forecast what others do, can reduce selfish behavior in morally ambiguous settings. Across two preregistered experiments, a moral wiggle room game and a dice-rolling game, forecasting others' behavior before choice reduced information avoidance by 18 percentage points and reduced aggregate excess payoffs associated with misreporting by 19%, although the latter estimate is only marginally significant. When beliefs were elicited after choice, stated forecasts were more closely aligned with participants' own selfish actions, a pattern consistent with ex post rationalization or self-projection. Regardless of elicitation timing, evaluations of the appropriateness of behavior showed no detectable treatment differences. In a third experiment, a binary-choice dictator game with transparent payoffs and a high baseline rate of choosing the 50-50 split, belief elicitation had no significant effect. The evidence is consistent with a theory of latent norms in which internally held standards may be insufficiently salient and inactive at the moment of choice. The results also caution that belief elicitation is not a neutral measurement tool, with implications for experimental design and for comparisons between laboratory and field behavior.

This paper develops a belief-based Markov model in which behavioral change occurs through two sequential gates: the burden of the incumbent behavior governs reconsideration, and the burden of the alternative governs acceptance. At zero habit bias, the model has the same long-run steady states as the standard logit model of discrete choice with social interactions, but absolute cost levels determine the speed of adjustment. Populations with the same relative incentives can therefore converge to the same long-run behavior at sharply different rates. With belief-dependent costs, the familiar critical-mass and hysteresis results carry over. I interpret conformity feedback as salience-scaled: perceived prevalence has greater behavioral force when the relevant norm is focal. Temporary material support combined with stronger conformity feedback can move a population into a self-sustaining high-compliance basin, while heterogeneity in effective salience can sustain behavioral differences even under shared, accurate beliefs. The framework thus provides a process-based account of slow, uneven, and path-dependent norm change.

People often avoid information about the harm their choices impose on others. Because such choices occur against beliefs about what others do, descriptive norms may license or restrain this strategic ignorance. In a preregistered 2 × 4 moral wiggle-room experiment (N = 1,020), participants predicted how often oth- ers acquired information, received randomly assigned truthful feedback about the ignorance rate observed in a comparable study, and then chose whether to reveal a recipient’s payoffs. Feedback moved behavior where it contradicted priors: news that ignorance was common raised avoidance by 10 percentage points, concentrated among participants who expected widespread revealing, while news that ignorance was rare reduced avoidance only among those who expected widespread ignorance. Because most participants held optimistic priors, licensing dominated in aggregate. Stated beliefs showed no self-serving distortion: participants who knew they would face the decision reported the same expectations as those who did not (equivalence within ±5 points). Descriptive norms shape strategic ignorance by licensing and cor- recting behavior—not through motivated belief formation—providing a belief-level account of when norm information helps and when it boomerangs.

While people often avoid learning about inconvenient or undesirable consequences of their actions in order to behave selfishly, many social situations involve another person who is in a position to impose this information. How does the presence of a potential informer affect information, behavior, and welfare in social decisions with moral wiggle-room? We introduce a third-party informer into the moral wiggle-room game. Almost half of the dictators tried to avoid information only to have it imposed upon them by the informer. These dictators frequently revised their behavior to benefit the recipient, even at their own expense. Given an opportunity to reward informers, most dictators chose to do so, but those who had bad news thrust upon them by the informer were more likely to withhold it. Interestingly, a subtle change in the choice interface - separating the dictator’s ignorance and allocation choices in two separate screens - caused a substantial reduction in the share of dictators choosing ignorance.

Although humans exhibit many prosocial behaviors, when the social benefits of their options are uncertain, surprisingly many avoid learning them before choosing, using ignorance as an excuse to dodge moral obligations and revert to selfish behavior. This kind of willful ignorance is robust in the sense that researchers have documented it using a wide array of methods, across diverse settings, and a time period spanning nearly two decades. At the same time, however, the degree to which it manifests is inconsistent across and within studies. Some of these inconsistencies stem from obvious factors, while the moderators driving others have yet to be identified or are poorly understood. This study synthesizes and organizes these contextual factors, providing recommendations for future research.