Decisions from valuations of unknown payoff distributions.

Abstract

Four experiments are presented that clarify the impact of experience on the way people use valuations. In each of the 100 trials of Study 1, participants were asked to choose between the status quo and an unknown binary lottery based on valuations by two expert systems: a well-calibrated “expert” reporting the expected values, and an expert that ignores the low probability outcome and reports the medians (that equaled the modes). The results suggest that experience decreased the inclination to follow the recommendation of the well-calibrated expert. This deviation from maximization appears to reflect two biases: the tendency to follow the expert that has recommended the action that provided higher payoff in most cases, and the tendency to follow the more extreme valuation. Studies 2 and 3 suggest similar reliance on experts that recommend the best choice in most trials, in choices between two payoff distributions, and even when these experts do not provide the median or the mode of the distributions. Study 4 shows that the impact of experts that direct people toward the optimal choice can be increased by exaggeration (inflating the estimated advantage of the payoff maximizing option). However, the long-term impact of exaggeration depends on the proportion of cases in which the optimal choice leads to the best payoff; a lasting positive effect of exaggeration was observed only when this proportion was high. These results can be captured by assuming several “expert weighting rules” and the selection between the rules based on small samples of past experiences.

Failures in the communication of risk: Decisions and numeracy

“…Bolton and Katok (2018) study the efficacy of probability forecasts versus recommendation forecasts. They found that forecasts in the form of a recommendation are prone to a “cry wolf effect”—a tendency to ignore the forecast when the forecast has been wrong in the past (also found in previous studies; e.g., Bliss et al., 1995; Erev et al., 2020; Meyer & Bitan, 2002). Overall, probability forecasts performed better.…”

“…One obvious alternative, natural language descriptions of uncertainty, has been shown to be ambiguous: Interpretation of verbal descriptions of uncertainty are sensitive to context (e.g., Harris & Corner, 2011; Weber & Hilton, 1990) and vary widely by individual (e.g., Karelitz & Budescu, 2004; Wallsten et al., 1986), limiting their ability to deliver a clear message about risk. Recommendations based on expert assessments of the forecast offer a way around both user numeracy and verbal description shortcomings (Erev et al., 2020). Non‐numeric in nature, recommendations can, in theory, convey the optimal action to the decision maker as effectively as quantitative measures.…”

The common origin of both oversimplified and overly complex decision rules

“…Table 2 shows that in 6 of the 16 problems, the difference between the risk‐rates in the two description conditions was significant. In all six cases, the direction of the significant difference can be explained with the assertion that the description of the cues as experts' recommendations increased the tendency to follow the rule “take the risk if the cue includes at least two T's.” In Condition Experts, this rule implies following the modal recommendation and taking the risk when most experts recommend it (see a similar pattern in Erev et al, 2022). Importantly, in five of the six significant cases, following the modal recommendation impaired expected return.…”

“…Table 2 shows that in 6 of the 16 problems, the difference between the risk-rates in the two description conditions was significant. In all six cases, the direction of the significant difference can be explained with the assertion that the description of the cues as experts' recom- most experts recommend it (see a similar pattern in Erev et al, 2022).…”

On the descriptive value of the reliance on small-samples assumption

“…While this set of situations has clear boundaries, it contains many important members. Examples include settings in which safety devices increase accidents (Cohen & Erev, 2018), taxation backfires (YCNE), people over and under-commit to a course of action (Cohen & Erev, 2021), experience reduces the tendency to trust well-calibrated experts (Erev et al, 2022) and it is necessary to enforce rules (Plonsky et al, 2021). Yet, more insight into how people respond to different dimensions of similarity, and how these similarities interact, is necessary to predict behavior when dynamic regularities are easily detectable.…”