Session 5: Experimental Economics

A rich literature finds that individuals avoid information and suggests that avoidance is driven by image concerns. This paper provides the first direct test of whether individuals avoid information because of image concerns. We build off of a classic paradigm, introducing control conditions that make minimal changes to eliminate the role of image concerns while keeping other key features of the environment unchanged. Data from 6,421 experimental
subjects shows that image concerns play a role in driving information avoidance, but a role that is substantially smaller than one might have expected.

A growing number of studies suggest that minimum wages have limited disemployment effects while at the same time increasing output prices. This finding contradicts the ”law of demand”, which states that output demand, and therefore employment, should fall whenever prices increase. We propose a simple framework to explain this fact and to highlight some aspects of ethical consumption more generally. Consumers derive extra utility when engaging in transactions that can be associated with positive moral attributes. In the context of the minimum wage, consumers derive a higher marginal utility when they know that the good they are consuming is produced by a worker earning a higher wage. Combined with firms’ inability to credibly commit to higher wages, a mandated minimum wage policy can lead to higher output and positive employment effects simultaneously. We implement an online survey experiment in the U.S. to test for the proposed mechanism. We use our findings to reassess the welfare implications of the policy.

Many models of decision-making under uncertainty are motivated by two prominent deviations from expected utility (EU): the common consequence effect (CCE) and the common ratio effect (CRE). Both decision problems were originally proposed as thought experiments by Allais (1953), and later popularized by Kahneman & Tversky (1979). The apparent deviations from EU predictions in each problem have motivated a wide body of decision theories in risky choice.

Although the CRE and CCE both represent violations of the EU axiom of independence, they have been studied mostly independently, and using quite different experimental parameters. In fact, however, the two decision problems are closely related: If conducted at a common set of experimental parameters, the two problems would share three out of four possible options. Moreover, the connections between the two problems are relevant for assessing various non-EU models—i.e., different models predict specific patterns.

In this paper, we extend existing empirical tests by (i) explicitly recognizing the connection between the two decision problems; (ii) conducting a large number of experiments covering connected CRE and CCE problems at different experimental parameters; and (iii) implementing experiments using both paired choice tasks (for comparison to the prior literature) and paired valuation tasks (our preferred approach given the inferential challenges outlined in McGranaghan et al (2022)).

Our results provide important insights on the shape of risk preferences. We find small but significant CR preferences, but systematic reverse CC preferences. Through their connection, this pattern implies that individuals violate betweenness by preferring mixtures. These results are inconsistent with leading non-EU models, and we propose a model to rationalize these findings.

It has been well-documented that people tend to be overconfident. We investigate whether biased beliefs in performance persist in a high-stakes environment. Specifically, we ask whether students are overconfident when estimating their high school entrance exam performance. Students in our environment have strong incentives to accurately assess their exam scores because they need to submit their rank order list under an immediate acceptance mechanism before knowing their exam performance. Combining administrative and survey data on estimated performance and actual performance, we find no evidence for overconfidence in estimation in this high-stakes environment. However, when we remove the high stakes by eliciting students’ recall of their performance in a previous mock exam, they show a strong pattern of overconfidence. Consistent with Benabou and Tirole  ́ (2002)’s theory of the supply of biased beliefs through biased memory, we find suggestive evidence that students rely on their potentially biased memory of past performance to construct their high-stakes estimation.

Individuals often change their decision-making in response to complexity, as has been discussed for decades in psychology and economics, but existing literature provides little evidence on the general characteristics of these processes. We introduce an experimental methodology to show that in the face of complexity, individuals resort to “procedural” decision-making, which we categorize as choice processes that are more describable. We elicit accuracy in replicating decision-makers’ choices to experimentally measure and incentivize the choice process’ describability. We show that procedural decision-making increases as we exogenously vary the complexity of the environment, defined by the choice set’s cardinality. This allows for procedural reinterpretations of existing findings in decision-making under complexity, such as in the use of heuristics.

Recent theoretical work implicates decision confidence as a central component of decision-making under uncertainty, attributing failures of Expected Utility (EU) to a lack of confidence. We design an experiment testing EU’ central independence axiom and contemporaneously eliciting measures of decision confidence. We find that choices characterized by high self-reported levels of decision confidence and low response times are more likely to com-ply with the independence axiom. Contrary to the common certainty effect rationale for independence violations, we show that subjects predominantly violate EU by choosing risky lotteries over certain amounts when they are unconfident in their choices.

The use of Artificial Intelligence (AI) in recruitment is rapidly increasing and drastically changing how people apply to jobs and how applications are reviewed. In this paper, we use two field experiments to study how AI in recruitment impacts gender diversity in the male-dominated technology sector, both overall and separately for labor supply and demand. We find that the use of AI in recruitment changes the gender distribution of potential hires, in some cases more than doubling the fraction of top applicants that are women.This change is generated by better outcomes for women in both supply and demand. On the supply side, we observe that the use of AI reduces the gender gap in application completion rates. Complementary survey evidence suggests that this is driven by female jobseekers believing that there is less bias in recruitment when assessed by AI instead of human evaluators. On the demand side, we find that providing evaluators with applicants’ AI scores closes the gender gap in assessments that otherwise disadvantage female applicants. Finally, we show that the AI tool would have to be substantially biased against women to result in a lower level of gender diversity than found without AI.

We conduct a lab experiment that shows current rules regulating transplant centers (TCs) and organ procurement organizations (OPOs) create perverse incentives that inefficiently reduce both organ recovery and beneficial transplantations. We model the decision environment with a 2-player multi period game between an OPO and a TC. In the condition that simulates current rules, OPOs recover only highest-quality kidneys and forgo valuable recovery opportunities, and TCs decline some beneficial transplants and perform some unnecessary transplants. Alternative regulations that reward TCs and OPOs together for health outcomes in their entire patient pool lead to behaviors that increase organ recovery and appropriate transplants.

Standard models of discrimination assume that decision-makers use all the available information about candidates when making their decisions. Based on research in psychology, I test the hypothesis that when decision-makers have a lot of other information in their mind, they are less likely to remember how particular individuals performed and fall back on stereotypes, which disadvantages women in predominantly male-dominated fields. First, in years when teachers need to evaluate a larger number of students – amplifying memory constraints – girls are considerably less likely to be recommended for top-tier scientific high school tracks. On the contrary, the gender gap in students’ objective math ability does not expand during these years. Second, I conduct an experiment in which teachers assign track recommendations for hypothetical student profiles. Teachers are less likely to recommend girls to scientific tracks if they freely recall the information about them than when they can reference the information (proxying a perfect memory benchmark). Third, when asked to remember a candidate’s past performance on a series of trivia questions in sports and pop-culture, participants tend to remember a higher share of correct sports questions when they are answered by a boy than by an identical girl. The opposite is true for pop-culture questions. The results suggest that a significant portion of gender discrimination is driven by imperfect and selective memory of previously observed information, opening up the scope for policy interventions in the form of structured reminders.

I show experimentally that information persuades not only by shifting beliefs but also by redirecting attention. Participants in my experiment decide whether to purchase a multi-attribute good. At baseline, selective attention generates large distortions in how responsive demand is to the values of these attributes. Randomly providing information about the value of one attribute, even when it is already known and transparently redundant, starkly increases responsiveness to that attribute and distracts attention from others. These forces can produce paradoxical responses to correcting beliefs: reducing overoptimism about an attribute can nonetheless boost demand for its associated good. 

Poverty assistance is often administered in-kind even though cash transfers might raise recipients’ welfare more effectively. We characterize the political economy constraint that paternalistic motives impose on the welfare system. In our experiment, a representative sample of U.S. citizens reveals their motives by deciding whether to constrain real U.S. food stamp recipients’ choices between in-kind donations and cash equivalents we disburse. The modal respondent (40%) imposes the strictest possible constraints, while 30% impose no constraints. Hence, the majority’s behavior is consistent with deontological motives rather than trade-off thinking. Yet, because of biased beliefs about recipient preferences, respondents underestimate the restrictiveness of their interventions, suggesting that they are partly misguided. Overall, respondents’ goal is not to ensure sufficient healthy nutrition, but to prevent consumption of items deemed inappropriate. While respondents reveal racial and gender stereotypes in various survey questions, neither donor nor recipient demographics have substantial effects on restriction decisions, though restrictions increase with respondents’ political conservatism. In-experiment behavior correlates strongly with views about government policy.

In a field experiment among undergraduates, we test the impact of interventions to increase sleep on sleep habits and academic achievement. Offering incentives contingent on sleeping at least 7 hours per night increases sleep during both the four-week treatment period and the one to five-week post-treatment period. The intervention also significantly increases GPA at the end of the semester. Our estimates suggest that causally increasing sleep by an average of 6 - 16 minutes per night improves GPA by 0.12 - 0.14 standard deviations. We additionally examine the role of timing of rewards and reminders and feedback for improving sleep habits. We find that immediate incentives combined with reminders and feedback have the largest impact during treatment, but do not outperform delayed incentives or reminders and feedback alone during the post-intervention period. Our results suggest that interventions targeting sleep are a cost-effective tool for improving educational outcomes.

Designed markets often rely on carefully crafted descriptions of mechanisms. By and large, these descriptions implicitly attempt to convey as directly as possible how outcomes are calculated. Are there principled, alternative theories of how to construct descriptions to expose different properties of mechanisms? Recently-proposed menu descriptions aim to provide such a theory towards exposing the strategyproofness of real-world mechanisms such as Deferred Acceptance. We design an incentivized experiment to test the ability of a menu description (compared to a traditional description) to affect participants' understanding of strategyproofness and behavior. We also design treatments conveying the definition of strategyproofness itself rather than the full details of the mechanism, with one treatment following the menu approach, and one using a common mechanism-design definition of strategyproofness.

Decision theorists usually take a normative view on stochastic dominance: a decision maker who chooses a lottery that puts more weight on options he likes less must be making a mistake. In this paper, I argue that stochastic dominance violations may naturally occur in situations where anticipatory utility is high, such as going on a holiday trip. In such a situation, the decision maker may trade the certainty of going to his favorite destination for the excitement of not knowing where he will go. To document this phenomenon, I conduct an experiment in which participants make a series of binary choices between a sure destination and a lottery over holiday trips. The outcome of the lottery is revealed close to the date of travel. I vary the characteristics of the lotteries to understand when violations of stochastic dominance are most likely to occur and analyze their properties. I discuss the implications for the modelling of anticipatory utility.

We experimentally study the transmission of subjective expectations into actions. Subjects in our experiment report valuations that are far too insensitive to their expectations, relative to the prediction from a frictionless model. We propose that the insensitivity is driven by a noisy cognitive process that prevents subjects from precisely computing asset valuations. The empirical link between subjective expectations and actions becomes stronger as subjective expectations approach rational expectations. Our results highlight the importance of incorporating weak transmission into belief-based asset pricing models. Finally, we discuss how cognitive noise can provide a microfoundation for inelastic demand in the stock market.

We examine the experimenter’s preferences over different populations using statistical power under a fixed budget as the stand-in for the researcher’s utility. We consider five populations commonly used in experiments by economists: undergraduate students at a physical location, undergraduate students in a virtual setting, Amazon MTurk "workers", a filtered MTurk subset from CloudResearch, and Prolific. Focusing on noise due to inattention, observation costs dominate the comparisons, with the larger online population samples superior to the smaller lab samples. However, once we factor in responsiveness to treatment, the lab samples have greater power than either MTurk or Prolific.

We develop indices of the objective and subjective complexity of lottery choice
problems that can be computed for any standard dataset. These indices reflect
which choice set features increase error rates and cognitive uncertainty in gauging expected values. Using these measures, we study behavioral responses to complexity across one million experimental decisions. In line with a model of heteroscedastic cognitive noise, complexity (i) makes choices more inconsistent and regressive to people’s prior; (ii) predicts when subjects accept unattractive gambles; and (iii) spuriously generates complexity aversion and small-stakes risk aversion. In structural estimations, complexity-dependent heteroscedasticity improves model fit considerably more than prospect theory does.

A central factor when choosing an action is its causal e↵ect on the outcome of interest. Yet, causal information is often lacking. People instead observe correlational or historical data, along with causal interpretations and action recommendations provided by experts who frequently disagree with each other. We use a laboratory experiment to study human choice in such settings. Roughly half of our subjects attempt to determine the fit of the causal interpretations to past data, as the literature on model persuasion assumes, and we outline the limits to their ability to do so. Half the subjects’ choices are co-determined by the interpretations’ promises of future payouts, as the literature on narrative competition assumes, or by the downside these choices entail if they are mistaken. Additionally, subjects commonly employ heuristics such as Occam’s razor, but they usually prefer more complex interpretations to more parsimonious ones. We also study the extent to which behavior is robust to framing and has out-of-sample predictive power, as well as the relation between subjects’ choices and their political attitudes and psychological characteristics. Finally, we will characterize the contexts in which subjects’ behavioral tendencies expose them to the greatest losses and render them most receptive to misleading interpretations.

We report results from an experiment designed to study how people extract
patterns from their observations. The novel experimental design asks subjects to organize different sets of observations (data) with the goal of making pre-
dictions in similar situations. We study whether the predictions subjects make in each environment are consistent with them using some “model” that posits specific statistical relationships between different variables. We find that the
predictions of most subjects can be rationalized by some model. Importantly,
we find the most commonly used model is the optimal one in that it maximizes
prediction accuracy. Deviations from the optimal model often involve use of
simpler models that fail to account for statistically relevant correlations in the data. Variation in the set of observations presented to subjects across environments allows us to test whether the way subjects learn from data display a key aspect of causal reasoning: identification of conditional independence between variables. While we find strong evidence for this, we also observe that failures of this increase with the noise in the data. Complemented with ancillary non-choice data that emerges as a by-product of our design, our results provide insights into how people form models of the world by studying data and how they use these models to make predictions.