Contradictory Deviations from Maximization: Environment-Specific Biases, or Reflections of Basic Properties of Human Learning?
12 May 2022 @ 3:00 pm - 4:30 pm
Online Research Seminar
3:00 pm, Thursday, 12 May 2022
Analyses of human reaction to economic incentives reveal contradictory deviations from maximization. For example, underinvestment in the stock market suggests risk aversion, but insufficient diversification of financial assets suggests risk-seeking. The leading explanations for these contradictions assume that different choice environments (e.g., different framings) trigger different biases. Our analysis shows that variation in the choice environment is often not a necessary condition. It demonstrates how certain changes in the incentive structure are sufficient to trigger six pairs of contradictory deviations from maximization even when the choice environment does not change. Moreover, our analysis shows that the direction of these deviations can be captured with simple “partially attentive sampler” models. These models differ from the popular reinforcement learning models in two ways: They assume that choice propensities reflect reliance on small samples (rather than temporal difference learning), and a partially attentive choice rule (rather than softmax or epsilon-greedy rules). A three-parameter abstraction of these assumptions, estimated to fit the contradictory deviations we studied, provides useful prediction of choice behavior in a preregistered study with 40 new randomly selected choice tasks. These results suggest that ignoring the predictable impact of the incentive structure can lead to exaggeration of the importance of environment-specific decision biases. We propose that the descriptive value of the reliance on small samples assumption reflects the fact that, in static settings, this assumption approximates the output of complex cognitive processes that resemble machine learning classification algorithms, and that features of the choice environment can impact the likely classification rules.
About the speaker
Ido Erev, PhD University of North Carolina, in Cognitive/Quantitative Psychology, is Professor at the Faculty of Industrial Engineering and Management of Technion – Israel Institute of Technology. He is an ex-President of the European Association for Decision Making, and the Vice Dean in charge of the startup MBA program in the Faculty of Industrial Engineering and Management in the Technion. His research clarifies the conditions under which wise incentive systems can solve behavioral and social problems. Among the contributions of this research is the discovery of a robust experience-description gap: People exhibit oversensitivity to rare events when they decide based on a description of the incentive structure, but experience reverses this bias and leads to underweighting of rare events. Comparison of alternative models favors the assumption that people tend to select the option that led to the best outcome in a small sample of similar past experiences. These observations imply that incentives are most effective when they ensure that the socially desirable behavior maximizes payoff, and minimizes the probability of regret.