When engaging in strategic decision-making, a valuable skill is the ability to assess the risk and time preferences of other people. These assessments may vary greatly in accuracy based on a predictor’s experience and the information they have about an opponent. These factors may be based entirely on immutable characteristics such as presented race or gender. Using a two-phase experiment, we can test how well individuals can predict the risk and time preferences of a sample of people with only gender or other demographic information to make their predictions. Specifically, we explored whether there was an information advantage when participants were making in-group predictions about someone who shared the same gender or political identity as them. In phase 1, participants take part in a modified double multiple price list (DMPL) to elicit their respective preferences. In phase 2, subjects make predictions about decisions made in the DMPL and adjust those predictions to specific groups.
We study the formation of risk perceptions— subjective probability beliefs— of three adverse events—COVID-19 contagion, influenza contagion, and food poisoning—at the onset (2020) and outset (2023) of the COVID-19 pandemic in the United States. We show that perceived risk levels for COVID-19 are similar to those for influenza, are not significantly influenced by proximity to infection, and are shaped instead by an individual’s gender, education, and employment status. Using an instrumental variable strategy, we assess whether these perceptions influence a number of protective behaviors. We find that although risk perceptions are associated with various protective behaviors, they only causally increase the likelihood of phone or online medical consultations by about a percentage point.
We investigate how interpersonal preferences influence decision making under uncertainty. Classical models assume agents maximize expected utility based solely on their own payoffs, yet experimental evidence often shows apparent instability in risk preferences across institutions. We propose and test a utility framework that incorporates altruism, malice, fairness, and competitiveness alongside risk attitudes. Using five choice tasks that hold own-payoff risk constant while varying effects on another participant’s payoffs, we isolate interpersonal factors from the institutional setting and strategic uncertainty. Results show that while average choices align with risk-neutral predictions, individual choices exhibit three systematic patterns: subjects tend to be altruistic and fairness-seeking, malicious and distinction-seeking, or neutral on both dimensions. These findings explain apparent shifts in risk preferences across institutions even when underlying preferences are stable. This highlights the importance of interpersonal considerations in models of choice under uncertainty.
Prior research has documented a tendency for insurance demand to rise abruptly in the aftermath of natural disasters. This may be attributable in part to the information shock that a disaster represents. Understanding how beliefs about risk evolve is of first-order importance for projecting climate adaptation in coming decades. Using a three-wave survey of 815 residents of flood-prone coastal counties in the southeastern United States, including 436 for whom we have longitudinal data, we study how individuals’ beliefs evolve after two very different types of informational shocks about future local flooding risk: viewing flood maps, and experiencing a local flooding disaster. We find that, on average, exposure to maps causes respondents to update their beliefs of their homes’ ten-year flood risk downward substantially (38
We argue that the established and varied research on highway safety suffers from fundamental methodological shortcomings. Researchers face the dilemma of either identifying causal influences on automobile accidents or asking limited policy-oriented questions about auto safety. We develop a theoretical framework to clarify the limitations of the three main approaches taken in automobile safety research: the use of controlled environments, disaggregated data, and aggregated data. We illustrate these limitations in the context of the vast empirical literature that has sought to assess the effectiveness of seatbelt use in reducing fatal accidents. We briefly discuss promising advances in computation and data that may help improve the credibility of automobile safety research as we enter an era of vehicle autonomy.
This paper examines behavior in contests where the prize value is ambiguous. We develop a theoretical model of bidding in a Tullock contest with an ambiguous prize where contestants account for the ambiguity attitude of their rival. Ambiguity affects optimal behavior via two countervailing channels - a direct effect arising from contestants’ ambiguity about the value of the prize and an indirect effect corresponding to the effect of ambiguity on the opponent’s behavior. Using a controlled laboratory experiment, we elicit individual risk and ambiguity attitudes and compare predicted and observed behavior in contests with an ambiguous prize, a risky prize and certain prizes. A comparison between contests with ambiguous and risky prizes, shows that participants invest significantly less under ambiguity. Additionally, we decompose the effect of changing from a certain prize to an ambiguous prize into two components - the first is the effect of introducing risk and the second is the effect of introducing ambiguity. Empirically, we find that both effects are significant, but work in opposite directions.
We examine the relationship between choice inconsistency and household wealth by combining experimental data with administrative records. In a representative Dutch sample, we reveal choice inconsistency through an experiment and quantify its severity using the Money Pump Index (MPI), which reflects potential monetary losses from inconsistent choices. Linking MPI to household wealth from Statistics Netherlands, we find a significant association: a one standard deviation increase in MPI corresponds to a 14.8
We examine how decision-makers’ (DMs’) ambiguity attitudes shape trust for two different sources of financial forecasting: human or machine learning (ML). In an incentivized laboratory experiment, we measure participants’ ambiguity attitudes and optimism regarding forecast accuracy for both sources. Our results reveal that DMs are similarly ambiguity-seeking and ambiguity-generated insensitive (“a-insensitive”; i.e., they insufficiently discriminate between changes in the likelihood of prediction accuracy), regardless of the analyst type. DMs hold more optimistic beliefs about the accuracy of ML analysts, which predicts higher trust in ML analysts over human analysts. However, DMs who are more a-insensitive are less likely to incorporate their beliefs into their trust. DMs’ a-insensitivity increases with financial literacy, suggesting that financially literate DMs perceive greater ambiguity in prediction accuracy. Our findings demonstrate that a-insensitivity acts as a cognitive barrier between beliefs and trust.
Experimental evidence suggests that gender differences in tournament entry may be driven by differences in competitiveness, overconfidence, and risk or ambiguity attitudes. We examine the influence of risk, ambiguity, and complexity on tournament entry using a novel experimental design that mitigates the role of overconfidence. Our experiment uses a rank-order tournament with calibrated ability, ensuring that variation in subjects’ performances is driven by the random noise component. We vary the nature of uncertainty by manipulating whether the distribution of noise is known (risk) or unknown (ambiguity), as well as the complexity of the situation by varying the number of draws used to determine the noise component. We do not find any significant effects of complexity (the number of draws) on tournament entry for either kind of uncertainty. Moreover, in our risk treatments, we observe no gender gap in tournament entry. However, there is a significant gender gap in our ambiguity treatments, where the distribution of the noise component is unknown, driven by the negative influence of ambiguity on entry by women. A separate treatment-based measure of participants’ preference for competition indicates that both men and women exhibit substantial competitiveness in the risk treatments, with no difference across gender. Yet, in stark contrast, ambiguity suppresses the competitiveness of women, without impacting the competitiveness of men. Altogether, our results suggest that greater transparency to minimize ambiguity may help to mitigate the gender gap in tournament entry.
It is well-documented that individuals’ behavior is influenced by reference points. However, while there is plentiful research on the role of loss aversion in financial decision-making, less is known about how individuals adopt, and react to, reference points for their own job performance. Is the performance reference value for a given task based on a rational evaluation of the task’s difficulty level or can individuals be induced to adopt other more salient but less rationally grounded reference points? The answer could have important implications for, e.g., managerial practices. I contribute with evidence from Stableford golf competitions, in which an individual-specific expected result, or reference point, is set for each hole based on the player’s ability (i.e. handicap) and the hole’s difficulty level. Exploiting the fact that the reference point varies across holes for a given ability and across ability for a given hole, I estimate the effect of a stricter reference point on performance in a model with ability and hole fixed effects. In line with adoption of reference points that are salient but not always informative of the hole’s true difficulty level, I find that a stricter exogenously set reference point causally improves players’ performance.
This paper explores the concurrent roles of prospect theory and regret theory in decision making under risk through a series of carefully designed experiments. We demonstrate that both models can be important for explaining real human choices in uncertain environments. Our experimental paradigm uses pairs of binary lotteries designed to induce both loss aversion and regret, with one gamble always involving a loss and the other changing from mixed, with both a gain and a loss, to gain only. Changes in behavior around these switching points elucidate loss aversion. Three feedback treatments are employed—no feedback, partial feedback (where only the chosen gamble’s outcome is revealed), and complete feedback (where outcomes of both gambles are revealed)—to assess how different levels of information influence decision making. In the no feedback treatment, we find strong support for prospect theory. When feedback is introduced, participants’ behavior reflects both prospect theory and regret theory although loss aversion has a greater impact than regret. Our findings underscore the importance of considering such hybrid models in evaluating decision making under risk, as well as the need to account for the information available to decision makers, as behavior varies with different feedback conditions.
Hurricanes and tropical cyclones have become more severe over the past 40 years and are expected to intensify in the future due to climate change. In this paper, we seek to understand how natural disaster experiences shape precautionary economic behaviors in response to future risks. We combine daily, household-level consumer goods purchase data from 2008-2018 with hurricane hit and warning data and use propensity score trimming to obtain a sample of households with a similar probability of receiving a hurricane warning to mitigate the problem of endogenous residential sorting. Using a triple difference-in-differences model, we find that past hurricane experience significantly influences household preparedness for impending storms. While households without prior-year hurricane exposure tend to stock up on emergency supplies only after receiving warnings, experienced households prepare earlier, increasing their purchases up to a week before warnings and prioritizing essential items. The degree of preparedness varies based on the severity of past disasters, household income, and the disaster risk level of the area.
Consider a decision maker who prefers to receive a reward when event A happens rather than—when event B happens, and who also prefers to receive the same reward when event A does not happen rather than—when event B does not happen. We define such decision maker as ambiguity averse (seeking) if event B (A) is unambiguous, i.e. measurable with objective probability. Under transitivity of preferences and a weak form of the first-order stochastic dominance, the proposed definition of ambiguity aversion (seeking) implies subadditivity (superadditivity) of matching probabilities of an ambiguous event and its complement. Implications for biseparable preferences (special cases of which include subjective expected utility theory, Choquet expected utility, multiple priors or maxmin expected utility, and α-maxmin theory) are provided. Ambiguity averse decision makers are willing to purchase full insurance for an uncertain event at a range of prices that are not necessarily actuarily fair. Ambiguity seeking decision makers never insure in full against an uncertain event. Our proposed definitions ofambiguity aversion (seeking) are unrelated to the concepts of smooth ambiguity aversion (seeking) undersmooth ambiguity model. An extension to fully uncertain situation (Anscombe-Aumann acts) is provided.
When faced with a risky prospect, people often make decisions which are inconsistent with an objective assessment of probability. Viscusi’s Prospective Reference Theory (PRT, 1989) model offers a simple and intuitive framework to model such behavior. We conduct an experiment to test the validity of PRT, focusing on the unique pattern of dominance violations predicted by the model. Subjects’ dominated choices are more consistent with PRT than with other models, providing novel empirical support for PRT. In addition we also test the possible source of the probability distortions. Subjects’ tendency to anchor correlates positively with likelihood insensitivity, implying that for some subjects, their decisions under risk may be caused by heuristic processing. To further illustrate PRT’s potential, we discuss how it applies to several new choice anomalies and extend the model to capture other commonly-observed behaviors, such as choices implying an inverse S-shaped probability weighting function.
It is rational to care about proximity; closer is often better. Proximity bias can be found when people overweight proximity and are willing (for example) to suffer serious welfare losses in terms of health or wealth in return for modest welfare gains as a result of proximity. In extreme cases, proximity bias leads people to stay where they are, at significant cost to their own welfare. Proximity bias is paralleled by proximity neglect, which can be found when people underweight the welfare benefits of proximity. Proximity bias can be seen as a product of present bias, though it often has additional or distinctive characteristics (including overestimation of the welfare costs of getting from one place to another). Extreme forms of proximity bias can be counted as pathological (“hodophobia”). There is clear evidence of the importance of proximity, and suggestive evidence of proximity bias, in diverse contexts, including medical care; vaccination; eviction; voting; and public assistance. Proximity bias has significant implications for policy and law. It suggests that there may be large effects from increasing (or reducing) proximity or making proximity less (or more) salient, perhaps through the use of debiasing, online alternatives, or various forms of choicearchitecture.
There is strong evidence indicating that relative risk aversion increases with outcome magnitude whereas impatience decreases with outcome magnitude. This finding seems paradoxical because it cannot be captured by the same utility function: Increasing relative risk aversion requires decreasing elasticity of the utility function, whereas decreasing impatience requires increasing elasticity with respect to outcome magnitudes. We develop a model that organically links the domains of risk taking and time discounting. The resulting two-speeds model generates a magnitude-dependent discount function such that increasing relative risk aversion is not only compatible with the magnitude effect in discounting but actually predicts magnitude-dependent discount weights. Moreover, we conduct a high-stakes laboratory experiment that reproduces the magnitude paradox and enables us to structurally estimate competing models of magnitude-dependent discounting. Both the Bayesian Information Criterion and the Akaike Information Criterion favor our two-speeds model. The results suggest that participants behave as if they apply a heuristic procedure: the value of a future reward consists of a certain percentage of the reward, irrespective of the length of delay, plus a delay-dependent component.
This research considers the role of income for subjective longevity assessments using the 2022 Survey of Consumer Finances. Treating subjective health and longevity assessments as jointly determined is important to understanding the role of income for both judgments. Income has a positive effect on health judgments and does not have a statistically significant effect on subjective longevity when health is rated as excellent or in excellent or good composite. Using analysis from the Health and Retirement Survey together with the SCF indicated that surveys framing the longevity question in direct or probabilistic terms can be complementary in estimating subjective longevity.
Loss aversion is one of the most robust findings in behavioral economics, with individuals typically weighing losses about twice as heavily as equivalent gains, and some even weighing losses many times more than equivalent gains. What drives these differences across individuals? Could it be that frequent exposure to the prospect of loss intensifies this bias? We examine this question in a competitive industry where decision-makers routinely face the prospect of losses that could threaten business survival. Using two distinct approaches, we find evidence of strong to extreme loss aversion. First, via thousands of real-time labor demand decisions from a retail chain and a discrete choice stopping model, we find a loss aversion coefficient of λ =4.2 , rising to λ =9.5 on slow days with smaller management teams, while disappearing on busy days. Second, through structured interviews with business owners and managers, we document a mean loss aversion coefficient of λ =10.1 and median of λ =1.6 , with 74
This paper tests, experimentally, three different perspectives on the relationship between preferences for randomization and the timing of resolution of uncertainty in ambiguous environments. We consider the canonical Expected Utility model, where there is no preference for randomization, Raiffa’s (Quarterly Journal of Economics, 75(4), 690–694, 1961) model, where preference for randomization is independent of the timing of resolution of uncertainty, and modern theories from Ke and Zhang (Econometrica, 88(3), 1159–1195, 2020) and Saito (American Economic Review, 105(3), 1246–1271, 2015), where preference for randomization depends on the timing of resolution of uncertainty. Our experimental results show a strong preference for randomization, including substantial violations of first order stochastic dominance, but are not consistent with any of the theories considered.
The theory of rational inattention posits that individuals maximize expected utility net of information costs, which are often modeled using mutual information. A key prediction of this framework is that posterior beliefs are invariant to prior probabilities. We test this prediction using a novel auditory-information experiment. Our results reveal a clear violation of this prediction: under low-stakes conditions, we observe a simple behavioral asymmetry that is inconsistent with invariance. We then introduce a reduced-form hybrid specification, defined as a convex combination of Bayesian updating–based mutual-information costs and hypothesis testing–based log-likelihood-ratio costs, to quantify the relative empirical contributions of the two components. Within this hybrid specification, the estimated marginal cost of Bayesian learning is positively related to the marginal cost of Type I decision errors, and the estimated weight on the mutual-information component is approximately 0.60 for the modal group of subjects.
