We congratulate Sara Arts, Qiyan Ong and Jianying Qiu on the publication of their research in Experimental Economics.
Read the findings of the study below.
There are many decisions in life that people may not be able to make with full confidence. It has been shown that a lack of decision confidence has the potential to explain a wide range of behavioral anomalies, such as the willingness-to-accept and willingness-to-pay gap, preference reversals, stochastic choices, and hyperbolic discounting in intertemporal choices. Yet, a major difficulty in these studies remains: how to measure decision confidence reliably?
Past studies have mostly elicited decision confidence by directly asking whether people were unsure of their choices. Due to the lack of incentive and the explicit reference to decision confidence, these measures potentially face behavioral biases such as priming effects and experimenter demand effects.
Building on earlier studies, we examine whether the way people randomize between options can be used as an incentivized behavioral measure for decision confidence. We demonstrate theoretically how randomization emerges from the optimization behavior of an individual who faces uncertainty regarding her preference between two options. In two experiments where subjects faced pairs of options (a lottery and a varying sure payment), we allowed subjects to choose randomization probabilities according to which they would receive each option. Separately, we obtained two measures of self-reported confidence for choosing between the two options. We further exogenously manipulated decision confidence by a) presenting a simple lottery with two outcomes and a complex lottery with the same expected value but more payoff outcomes over a wider range of possible values, and b) increasing subjects' experience with the lottery by allowing them to either observe the outcome draws of the lottery or to make hypothetical choices and observe the payoffs of their choices and the counterfactuals.
Consistent with the theoretical predictions, we find that subjects' randomization probabilities were strongly and positively correlated with the two self-reported confidence measures (median Spearman correlation between 0.86 and 0.89). Further, increasing the complexity of the lottery led to a decrease in self-reported decision confidence, while increasing experience with lotteries led to an increase in self-reported decision confidence. These exogenous changes in self-reported decision confidence were met with corresponding changes in randomization probabilities. As a result, the correlations between randomization probabilities and self-reported decision confidence measures were robust to the exogenous manipulations of decision confidence.
Overall, our study provides direct evidence of the connections between several important concepts in the literature, such as decision confidence, cognitive uncertainty, preference uncertainty/imprecision, incomplete preference, preference for randomization, and stochastic choice. Our finding of a systematic relationship between randomization, alternative measures of decision confidence, and stochastic choice suggests that there may exist a common psychological underpinning for these various concepts.