Risk Creep: Learning Under Uncertainty Leads to a Creeping Tolerance for Risk

Abstract

How do people learn appropriate behavior in highly uncertain environments? Decades of work on reinforcement learning (coupling behaviors with outcomes) and social learning (inferring cues from others’ behavior) requires assumptions that do not hold under high uncertainty. We measure two new learning mechanisms that might operate under high uncertainty. Experiential learning occurs when people use perceptions of risk from their own past behaviors to guide future behaviors—even in the absence of definitive outcomes from those behaviors. Minimal signal learning occurs when sparse signals from the infrequently observed behavior of others guides future behavior—particularly when norms are clearly absent. We track behavioral patterns at the height of uncertainty during the COVID-19 pandemic, post-lockdown and pre-vaccine. Our results suggest that within a naturalistic setting, people are capable of learning in highly uncertain environments, absent any external reinforcement of behavior and prior to norm emergence via these new learning mechanisms. Specifically, we find that people learn based on assessing whether or not the utility from a past behavior was “worth” taking part in the risk behavior (experiential learning) and that they learn based on observations of weak and noisy signals from others’ behavior (minimal signal learning). Moreover, results show a gradual increase in precarious activity under these two learning mechanisms, what we call risk creep, and suspect this may be a common by-product of learning in highly uncertain contexts.  

Keywords: COVID-19 Norms, Risk Perception, Decision-Making, Social Learning, Experiential Learning, Decision-Making Under Uncertainty

JEL Classification: D8, D80