Variance, Skewness and Multiple Outcomes in Described and Experienced Prospects: Can One Descriptive Model Capture It All?

We determined the scope of five decision models of choices across four environmental niches defined by whether outcome probabilities are described (risk) or experienced by sampling (uncertainty) and whether lotteries are simple (one or two outcomes per prospect) or complex (three or four). The majority of participants chose in accordance with cumulative prospect theory only in simple environments involving decisions from description (75%). In complex environments involving decisions from description and experience, however, skewness-preference models were more prevalent (57% and 68%, respectively). Consequently, in niches outside of simple lotteries under risk, rank dependence and nonlinear probability weighting failed to accurately describe the majority of choices. Exploiting elicited subjective beliefs in decisions from experience, we found that experienced (sampled) outcome likelihoods outperformed elicited beliefs in predicting choices and found scant evidence for two-stage models of decisions under uncertainty. Finally, we found statistically significant evidence that 90% of participants chose as if they relied on different models across environments; nonetheless, assuming as if participants used a single model across all environments to predict out-of-sample choice only minimally reduced prediction accuracy. We discuss the implications of model mimicry and task diagnosticity in light of these results in terms of both economic and statistical significance, both for model comparisons and inference.

Automated summary

The study determined the scope and prevalence of decision models in different environments and evaluated the accuracy of predictions made by these models.