On the impact of experience on probability weighting in decisions under risk.

Abstract

Previous research demonstrates that feedback in decisions under risk leads people to behave as if they give less weight to rare events. We clarify the boundaries of this phenomenon and shed light on the underlying mechanisms. In a preregistered experiment, participants faced 60 different decisions-under-risk choice tasks. Each task was a choice between a safe prospect (e.g., "59 with certainty") and a "rare disaster" gamble ("60 with p ϭ .98; 10 otherwise"). Additionally, each option also incurred a small cost (a draw from the set {Ϫ8, Ϫ6, Ϫ4, Ϫ2, 0} was added to the payoff). After each choice, participants received full feedback concerning the (total) realized payoffs of each option. The experiment compared 2 conditions that differed in the dependency between the 2 added costs. The results reveal high sensitivity to this dependency. Underweighting of rare events (preference for the rare disaster gamble) emerged with experience only when this dependency implied that in most cases, the rare disaster alternative provides a higher outcome than the safe alternative. In contrast, when in most cases the final outcomes from the safer option were higher, feedback appeared to increase the weighting of rare events (i.e., increased preference for the safe option). Common decisions-under-risk models (e.g., prospect theory) that assume the value of each prospect is judged only as a function of its own payoff distribution cannot account for this difference. Yet, the results can be explained with the hypothesis that choice reflects reliance on small samples of past experiences with similar decision tasks.

On the effect of perceived patterns in decisions from sampling.

“…One possible explanation for the similarities across the four conditions, despite differences in instructions and incentives, is that participants did not read, understand, or believe the instructions of the task (but see Cohen et al, 2020). Thus, it is possible that participants in Conditions Sam-pleEV and SampleNat.Mean did not realize that the sequence is irrelevant.…”

Underweighting of rare events in social interactions and its implications to the design of voluntary health applications

“…Research on individual decisions from experience documented a robust tendency to behave as if rare (low probability) events are underweighted (and common experiences are overweighted). 13 Research documented such bias in repeated decisions with partial (Barron & Erev, 2003), complete (Camilleri & Newell, 2011), and biased feedback (Plonsky & Teodorescu, 2020), one-shot decisions from description (Cohen et al, 2020) and from sampling (Hertwig et al, 2004), two-stage decisions (Y. Roth et al, 2016), investment decisions (Taleb, 2007), market entry games (Erev et al, 2010), and animal choice (Shafir et al, 2008).…”

Over and under commitment to a course of action in decisions from experience.

“…This hypothesis predicts that frequent losses (i.e., condition Cost) will lead to earlier stopping compared to frequent gains (i.e., condition RareLoss). This assertion relies on the fact that reliance on small samples implies preference for the drawing policy that gave the best outcome in most previous experiences (see Cohen et al, 2019; Erev et al, 2017). 3 Note this is the opposite of the prescription of the sunk cost effect, that is, that higher sunk costs would imply later stopping.…”