Research
Research
Abstract: This paper studies how outcome bias distorts incentives in delegated decision-making. In a theory-guided experiment, agents choose between two lotteries on behalf of a principal: one is first-order stochastic dominant, but the dominated lottery is more likely to yield a higher payoff state-by-state. Principals, despite perfectly observing the agent's choice, tend to reward agents for choosing the lottery that realizes a higher payoff -- thereby creating incentives for agents to choose the dominated lottery. These perverse incentives are robust: they arise in a lab and in two online experiments with a broad U.S. population, both in one-shot settings and when principals can directly influence agent behavior through their rewards. An awareness and revision intervention highlighting the incentive consequences of reward decisions fails to eliminate them. Structural estimation reveals stark heterogeneity in outcome bias. Agents' choices reflect perverse incentives, with aggregate choice patterns driven by agents high in cognitive reflection. The findings suggest that outcome-based evaluations can systematically distort incentives in domains such as politics, finance, and corporate governance.
Abstract: A growing literature documents that many individuals deliberately delegate decisions to random devices such as coin flips, even when one option clearly dominates. We propose a novel explanation: randomization allows for hedging against risk of regret. In our model, a decision-maker engages in outcome biased ex-post evaluations of their choices, projecting realized outcomes onto past decisions. Randomization ameliorates ex-post regret – if a bad outcome occurs, one can blame it on the coin flip. We conduct online experiments where participants choose mixtures over two lotteries, one of which first-order stochastically dominates the other. Holding marginal distributions fixed, we systematically vary the correlation of payoffs. As predicted by the model, participants randomize most under perfect negative correlation, less under independence, and least when one lottery dominates state-wise. In a clustering exercise, regret-hedgers emerge as the most prominent behavioral type. We further find that withholding outcome feedback on the non-chosen lottery decreases rates of randomization. Our model can rationalize puzzling findings in the literature, and we document the first direct evidence that randomization is a manifestation of deliberate regret-hedging.
Abstract: Allowing risk preferences to be sensitive to the correlation between lottery outcomes can resolve classical deviations from expected utility theory and provides a plausible explanation for phenomena in various real-world settings. However, evidence on correlation sensitivity is limited and mixed. In this paper, we first show that correlation-sensitive preferences in the general framework of Lanzani (2022) can be classified into three categories. We propose a novel experimental task that allows to classify experimental subjects according to this categorization. In a series of experiments, we find that aggregate choices display correlation sensitivity but in the opposite direction as often assumed in regret and salience theory. Individual level analysis suggests that the aggregate findings are driven by a minority who consistently exhibit this behavior even when it violates first-order stochastic dominance. Finally, we disentangle between correlation sensitivity due to deliberate within-state comparisons and incidental payoff comparisons due to the framing of decision problems, and find that both channels produce correlation sensitivity, with deliberate comparisons being somewhat more important.
Abstract: Salience theory relies on the assumption that not only the marginal distribution of lotteries, but also the correlation of payoffs across states impacts choices. Recent experimental studies on salience theory seem to provide evidence in favor of such correlation effects. However, these studies fail to control for event-splitting effects (ESE). In this paper, we seek to disentangle the role of correlation and event-splitting in two settings: 1) the common consequence Allais paradox as studied by Bordalo et al. (2012), Frydman and Mormann (2018), and Bruhin et al. (2022); 2) choices between Mao pairs as studied by Dertwinkel-Kalt and Köster (2020). In both settings, we find evidence suggesting that recent findings supporting correlation effects are largely driven by ESE. Once controlling for ESE, we find no consistent evidence for correlation effects. Our results thus shed doubt on the validity of salience theory in describing risky behavior.