2026/07/09 by Paul K. Bergmann, Dominik Deffner · 1 voice
Economics, Econometrics and Finance · Social Sciences · Physics and Astronomy · #Complex Systems and Time Series Analysis #Experimental Behavioral Economics Studies #Opinion Dynamics and Social Influence
paper · doi:10.31234/osf.io/kqpdu_v2
Individuals often estimate the risks and benefits of decision alternatives from their own experiences and the behaviours and outcomes of others. In both simulations and experiments, social learning has largely been found to improve decision-making compared to individual learning alone. While previous work treated decision outcomes over repeated choices as independent, in many real-life contexts, including financial investments, life-course developments, or social networks, decision outcomes depend on previous outcomes, producing self-reinforcing ``the-rich-get-richer" dynamics. The adaptive consequences of social learning under such dependent, multiplicative conditions remain unknown. Using social reinforcement learning models in an agent-based simulation framework, we identify diverging effects of several well-established social learning strategies. In contrast to independent, additive environments, nearly all forms of social learning in multiplicative environments undermined performance and increased inequality compared to individual learning. This occurred because an overly risk-seeking but fortunate minority provided systematically misleading cues. Payoff-biased learning strategies, usually considered a safeguard against herding, further deteriorated decision-quality, whereas conformity-biased copying slightly improved performance. These findings were observed even under time-optimal utility transformations of private and social payoff information. Together, our findings demonstrate disadvantageous collective learning under stochastic multiplicative growth, emphasizing that nearly all forms of social influence exacerbate inequality in these systems.