The Social Norm Mismatch: Why We Desire Norms Precisely When They Are Not in Place

Many of the world’s societal and environmental challenges are characterized by two sources of uncertainty: ecological uncertainty of a potential disaster and social uncertainty about others' behavior. Minimizing disasters is costly, creating individual incentives to delay or reduce investments, especially if it is uncertain how severe the disaster would be (in terms of probability and/or impact). Passing the safe bandwidth between sustainable and exploitative behavior may have catastrophic results. Social norms are often proposed as solutions to these challenges. Providing simple behavioral instructions, people might look for norms particularly under environmental uncertainty. However, since norms are formed endogenously with collective behavior, they are stronger when uncertainty is lower, because beliefs about the right action are more strongly embedded. This creates a paradox where norms are malfunctioning when they are needed most.

We test this using a collective risk social dilemma with endogenous and continuous impact reduction. Participants contribute not for a collective gain, but to prevent a collective loss. The size of the loss depends on the amount of points contributed to the collective fund. We show theoretically that social uncertainty is highest for intermediate risk and experimentally test for three disaster levels (low: 10%, middle: 50%, high: 90%) the hypotheses that for intermediate risk behavior is more heterogeneous, participants earn less, and more people desire social information. N=900 subjects (N=300 per treatment) participate in the main decision-making experiment. They play the endogenous CRSD, provide their personal beliefs, and decide whether to pay to receive information about the behavior of others. N=600 subjects (N=200 per treatment) guess the distribution of behavior and beliefs of study 1 participants. All hypotheses, the experimental design, and the analyses are preregistered using AsPredicted. The experiment is programmed in oTree and will be conducted with a representative sample of Prolific users in November 2024.