Frequency effects in decision-making involving loss minimization

Recent work provides evidence for frequency effects during decision-making, where less-rewarding options that are presented more frequently are selected more often than more-rewarding options presented less frequently. This is predicted by the Decay but not the Delta reinforcement-learning (RL) model. The Decay model assumes that higher-frequency options are preferred because their past outcomes are more available in memory than those of lower-frequency options. However, most of this research has involved decision-making with gains, rather than losses. In loss-minimization scenarios, the Decay model predicts a reversed frequency effect because it assumes greater memory for losses, for the more frequently encountered alternatives. We tested this prediction in three experiments and found that the Decay model provides a very poor fit to data in loss-minimization scenarios. In Experiment 2, where participants tried to minimize their expenditures in a hypothetical shopping scenario, we observed a modest frequency effect. In Experiments 1 and 3, where participants were asked to minimize losses as points, without the hypothetical shopping scenario context, frequency effects were attenuated, but not reversed. These effects were best-accounted for by two novel models, the Prospect-Valence Prediction-Error Decay model (PVPE-Decay), which assumes relative rather than absolute processing of rewards, and the Delta-Uncertainty model which assumes aversiveness to less frequent options that are higher in uncertainty. These results dovetail with recent work showing that people process reward outcomes in a context-dependent manner, and they suggest smaller losses can be perceived as relative gains if framed in familiar scenarios involving cost-minimization.