Similarity-based learning and the wavy recency effect of rare events

Ori Plonsky

Abstract

Many behavioral phenomena can be the product of a tendency to rely on small samples of past experiences. Previous studies suggest that
this can be a product of a cognitively efficient tendency to rely on the most recent outcomes. Congruently, the most popular models of learning assume that people mostly rely on a few recent outcomes. In this talk, I will review research suggesting a very different explanation: People rely on a
small set of the most similar past experiences. My investigation explores settings of repeated binary choice with feedback. I will present a model,
designed for these settings, that judges similarity as a function of sequential patterns of outcomes. A computational analysis shows this
model can be extremely effective across wide classes of dynamic decision settings (more effective than basic reinforcement learning models).It
further shows that in static settings with rare events the model predicts a unique wavy recency pattern. Empirical analysis of multiple datasets (including
decisions from experience with full or with partial feedback, probability learning tasks, and repeated decisions under risk with feedback) support this
wavy recency prediction, a pattern that was ignored by prior research and violates the basic assumption of recency in popular learning models. a
pattern that was ignored by prior research and violates the basic assumption of recency in popular learning models. a pattern that was ignored by prior
research and violates the basic assumption of recency in popular learning models.

Paper

Plonsky, O., Teodorescu, K., & Erev, I. (2015). Reliance on small samples, the wavy recency effect, and similarity-based learning. Psychological review, 122(4), 621.

Export event as iCal