The Economic Science Association is organizing a seminar series featuring junior job-market candidates in economics who use experimental methods. The virtual seminar series will run from October to December 2023 with sessions occuring at different times to accomodate audiences around the world. Each sessions consists of two to three speakers, each speaking for 20 minutes, followed by a discussant.

Sessions are held via Zoom and announced via the ESA announcements group.

The scheduled speakers and discussants are listed below.

Many decisions rest on the collective judgment of small groups like committees, boards, or teams. However, some group members may have hidden agendas and manipulate this judgment to induce a consequent decision in their interest. Utilizing an incentivized experiment, I analyze how manipulation affects the accuracy and trustworthiness of such group judgment depending on the format of group interaction. I compare group judgments from unstructured face-to-face interaction, ubiquitous in real-world institutions, to group judgments from the scientifically endorsed, structured Delphi technique. To identify mechanisms underlying the accuracy differences, I use structural estimations and analyze emergent communication patterns. Without manipulation, Delphi is more accurate than face-to-face interaction and indistinguishable from the Bayesian benchmark. Manipulation decreases accuracy for Delphi but not for face-to-face interaction. Thus, with manipulation, Delphi is less accurate than face-to-face interaction. Manipulation likely decreases the accuracy of Delphi judgments through more bias and less utilization of valuable information. Trustworthiness does not always match accuracy. Judgments from face-to-face interaction - unjustifiably - enjoy higher levels of trust without hidden agendas. Trustworthiness correctly decreases with hidden agendas for Delphi groups but - unjustifiably - also for face-to-face groups. With hidden agendas, face-to-face groups are simultaneously more accurate and trusted.

Can central banks benefit from communicating the performance of their past inflation projections with the public? If the public underestimates the central bank's projection ability, such communication can improve the anchoring of inflation expectations to inflation projections and build trust. To test these predictions, I incorporate an information provision experiment (N=5527) into the Bundesbank's Survey on Consumer Expectations. The experiment informs a random subset of respondents about the average inaccuracy of the European Central Bank's (ECB) past inflation projections after eliciting beliefs about its magnitude. The results show that the majority of households underestimate the ECB's ability to project inflation. The intervention strengthens the anchoring of expectations, reduces inflation uncertainty, and discourages consumption of durable goods. An analysis of the causal mechanisms reveals that trust in the ECB plays a mediating role in the treatment effects. Thus, central banks can build trust with the public through transparency when they are misperceived.

In the US, Democrats and Republicans increasingly disagree not just on ideologies, but even on facts. Given the economic consequences of political polarization, it is important to understand how information affects belief divergence. Using an online sample of 1100 Democrats and 1100 Republicans, I conduct an information experiment to examine beliefs about hiring discrimination against Black workers in the US. Democrats believe there is more hiring discrimination than Republicans do. I find that wage gap information fails to close this belief gap because Democrats use wage gap information to update their beliefs about racial discrimination in hiring, while Republicans do not. Furthermore, even after both groups agree about the extent of hiring discrimination, participants change their opinions about the observed discrimination depending on their political affiliation. Together, these findings highlight key drawbacks in decreasing political polarization through information.

The agent chooses between binary quality objects, she receives a sample of signals for each object. A common application is the choice of products from online platforms. The agent knows information accuracy if she knows the correlation between signals and quality. I show that, irrespective of behavioral channels such as non-Bayesian updating, an additivity axiom characterizes the agent's choice. Additivity implies an additional sample's impact on belief is independent of a current, already observed sample. However, we often believe an additional sample is more relevant when the current sample is small. This intuition reflects that we believe there is more left to learn about accuracy for smaller samples - and hence accuracy is not known. To test the axiom's validity and understand the actual decision process, I conduct a lab experiment where subjects choose objects along with a sample of signals. In this lab experiment, additivity implies preferences with parallel and linear indifference curves. Instead, almost all subjects display non-parallel indifference curves that "fan out", consistent with a heuristic that approximates optimal choice under accuracy uncertainty. A novel elicitation mechanism for belief confidence reveals further that behavior consistent with heuristic use is associated with higher confidence. The paper shows that agents do not behave as if they know information accuracy, but they do know how to choose and do so with confidence.

First collect some information, then choose an action. In such contexts, apparent mistakes may in fact be optimal if information acquisition is costly. But when can an observer conclude that a mistake was not optimal? I investigate the causes of mistakes in a novel online experiment with an intent-to-collect information stage. I test the axioms of the costly information acquisition model of Caplin and Dean (2015) and estimate within-subject bounds on attention costs. Optimal inattention explains most observed mistakes, but same-day dynamic inconsistency also appears to be a cause of mistakes.

One hypothesis for observed differences in how racial groups prioritize economic issues is that these groups have a social identity based on differing social norms. It has been theorized that the introduction of ethnically charged environmental primes may influence individuals to conform more strongly to these norms. In a highly cited paper, Benjamin et al. (2010) tested this through an experiment that assessed the effect of ethnically charged survey questions on the time and risk preferences of respondents. They found that primed Asian respondents were more patient than those in the control group with no effects observed for primed White respondents. Additionally, primed non-immigrant Black respondents displayed more patient preferences, with no effect on their immigrant Black counterparts. However, many recent replications of priming experiments have failed to obtain the results of original studies. We attempt to replicate the results of Benjamin et al. (2010) using a sample of the general US population that is over twice as large as the original study. We fail to replicate any of the original results and, in some cases, find the opposite results. Priming does not change any groups' time or risk preferences, though black men become more risk seeking. These results suggest that it may be improper to generalize the results found in the original paper or attribute them to the American public at large.

We study the effect of reserving seats for specific groups of students under the Deferred Acceptance (DA) Mechanism and the Top Trading Cycles (TTC) Mechanism. By extending the traditional mechanisms, we allow schools to process seats based on a defined precedence order that determines whether schools fill reserved seats or open seats first. Our theoretical predictions suggest that changing precedence order can lead to large differences in assignment outcomes. The data from a laboratory experiment strongly support the theoretical predictions. The results of interest are students' incentives to report preferences truthfully and assignment outcomes along important dimensions such as efficiency and stability. Our findings suggest that processing open seats first is more efficient and is a more effective way to reserve seats in school choice.

Do biased beliefs always lead to sub-optimal actions in equilibrium? Heidhues et al. (2018) demonstrate that optimal action can be achieved with misspecified beliefs in equilibrium when output depends not on each of the inputs independently but solely on their aggregate. This study provides an experimental test of this proposition. Supporting the theory, Experiment 1 highlights the exacerbated inefficiency that arises when decision-makers allocate tasks to individuals separately, guided by their potentially incorrect beliefs about the relative productivity of each person. However, this harm can be mitigated when decision-makers allocate tasks to a group of individuals, focusing solely on the average productivity of the group. Experiment 2 further establishes a causal link by introducing exogenous belief biases. Nevertheless, we find little evidence of the subjects playing in equilibrium. This study holds significant implications for how to address the negative impacts of belief biases, especially when belief biases are challenging to rectify.

When evaluated in retrospect, decision-makers are often judged as if they could have foreseen a random outcome, even when the outcomes carry no relevant information. Although such outcome-biased retrospection is typically viewed as detrimental, it remains unclear how it impacts the incentives of decision-makers when choosing between different actions, the extent to which decision-makers can anticipate the outcome bias, how it influences their choices, and ultimately, welfare. I address these questions through a theory-guided experiment set in a delegated risk-taking context. In this scenario, an agent, acting on behalf of their principal, must choose between a first-order-stochastic-dominant and a dominated lottery. The principal observes the choice and outcome and decides whether to grant a bonus to the agent. I propose a simple model of outcome bias in which bonus decisions depend on an ex-post counterfactual comparison between the obtained outcome and the forgone one. This mechanism can create a new type of agency problem. When the dominant lottery is likely to yield a lower outcome than the alternative, the outcome bias can effectively incentivize the principal to encourage the agent to choose the dominated lottery. I provide experimental evidence confirming this prediction. Stated beliefs suggest that, on average, agents anticipate the outcome-dependent nature of bonus payments, with a majority believing they have strong incentives to choose the dominated lottery. However, despite their stated beliefs, most agents predominantly choose the dominant lottery. Experimental subjects with high scores in cognitive reflection and those who have studied economics exhibit a greater propensity to choose the dominated lottery when they believe they have an incentive to do so. These exploratory results imply that outcome bias may indeed have a detrimental impact on actions and welfare, particularly in settings where decision-makers excel in strategic reasoning. Estimates of finite mixture models suggest that heterogeneity in the principals’ outcome bias can be parsimoniously characterized by two types: fully outcome biased and nearly unbiased, with roughly 62% falling into the fully outcome biased category. High outcome bias correlates with low cognitive reflection and fast response times, supporting its interpretation as a cognitive heuristic. Agents tend to overestimate the degree of outcome bias among principals, with 50% believing that 100% of the principals are fully outcome biased.

I estimate management opposition to unions in terms of hiring discrimination with a large-scale field experiment in the German labor market. By sending 13000 fictitious job applications, revealing union membership in the CV and pro-union sentiment via social media accounts, I provide evidence for hiring discrimination. Callback rates for union members decrease on average by 151%. Discrimination is strongest in the presence of a high sectoral share of union members and large firm size. I further explore variation in regional and sectoral strike intensity and find weak evidence that discrimination increases if a sector is exposed to an intense strike. Yet, strike activities account only for a minor extent of hiring discrimination. My results indicate that hiring discrimination can be explained by union threat effects. Sectors with low collective bargaining coverage have lower hiring discrimination and in the absence of collective agreements sectors are less likely to follow collective agreement wage setting. Taken together, these results provide the first large-scale experimental evidence of management opposition to labor unions.

Do safety interventions create a false sense of security? We investigated whether people adjust their behavior optimally when there are exogenous changes to risk. In a laboratory experiment, subjects played an insurance-buying game under various risks of losing endowment. We found strong evidence of overcompensation - subjects purchased more (less) insurance than optimum in response to an increase (decrease) in risk. The degree of overcompensation was larger when risk increased than when it decreased. We exposed the same subjects to changes in safety conditions in real life and found little evidence that lab behaviors predict behaviors in the field.

Many economic decisions are made by teams, committees and boards. Individuals may come to these groups with different beliefs that must, ultimately, coalesce to a consensus belief. Yet, relatively little is known about how the beliefs that inform decision making in groups are formed and how these differ from individual beliefs. We conduct an experiment to examine the role of communication in the processing of new information. Overall, neither prior beliefs nor transfers differ between individuals and groups. Groups exhibit asymmetric updating but are not more biased than individuals. Based on text analyses, we identify risk preferences as an important topic in group communication and observe a self-serving bias in updating by more risk-averse groups -- but not by risk-averse individuals. While the group environment does not necessarily lead to more motivated beliefs, communication can amplify individual preferences in a way that leads to more biased belief updating by groups.

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 recruitment tools can impact 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.

In many relevant decision contexts, we are affected by a wide array of behavioral biases. Also, we often have the chance to observe others’ decisions and, possibly, change ours. This paper investigates the impact of social learning on a broad selection of behavioral biases, reflecting economically relevant decision-making frameworks. Through an online experiment, I document how social learning can amplify errors stemming from behavioral biases, leading to worse group outcomes. This detrimental effect of social learning on group outcomes is driven by a misalignment between performance and confidence. For some tasks, high-performing participants tend to be less confident in their answers and more confident in another participant’s incorrect answers, while the opposite holds for low-performing ones. These results shed light on settings where cognitive biases affect decision-making in the presence of social learning, such as the interpretation of statistical information or investment decisions.

In many contests, players are not aware of how many competitors they face. While a rich literature examines how disclosing this information affects contestants’ productive effort, it does not account for the impact of disclosure and concealment policies on destructive behavior such as sabotage. In this paper, I theoretically and experimentally study how disclosing the number of contestants affects effort and sabotage decisions. Relying on a model based on a Tullock contest with exogenous enter probabilities, I show no differences in effort and sabotage decisions between the disclosure policies, resulting in no differences in the expected payoffs. Overall performance, however, is higher under concealment when the probability of being the only contestant is not too low. The experimental results largely confirm the theoretical predictions. Yet, if the probability of not facing any opponent is large, I find slightly larger sabotage decisions. Additionally, under group size disclosure, subjects exert above equilibrium sabotage choices for larger group sizes, resulting in substantially more destroyed value. The results contribute to the study of sabotage in contests and contests with group size uncertainty and have substantial implications for contests’ design. By concealing the number of contestants, the designer can increase the produced value, when the probability of only one active contestant is not too low and competitors can sabotage each other.

Imposing monetary penalties may lead to a decrease in people's willingness to engage in prosocial behaviors, a phenomenon known as crowding-out effects. In other cases, people show rule-following tendencies and, therefore, penalties can encourage prosocial behavior, possibly resulting in crowding-in effects. This article aims to reconcile these opposing theories by examining how monetary penalties influence prosocial behavior. We use a dictator game in which participants were subject to two subtly different penalty conditions – a `fine’ imposed after the dictator take money, and a `fee' paid before taking money. The study finds that while some participants are more likely to take money when facing a penalty (crowding-out effect), others refrain from taking money, even when they are taking large amounts without the penalty (crowding-in effect). Additionally, in the fee condition, fewer people take any money than in the fine condition. The study also shows that monetary penalties can impact social norms and perceived entitlement, which partially explains the crowding-out(in) effects.

Previous studies have shown that our past pro-social actions influence our future ones. So, which pro-social deeds inspire us to do more? Can nudges promoting these deeds crowd out this behavioural spillover? We study a social norm nudge promoting vegetarianism in an online experiment (n=2775). Our findings suggest that those who chose vegetarian options were more likely to donate to environmental charities. However, the social norm nudge crowds out donations among a population segment that initially responded positively to the nudge. Despite their appeal, social norm nudges may, therefore, be counterproductive. Our empirical strategy stems from a model describing how acting pro-socially affects individuals' willingness to do extra pro-social actions and how nudges can crowd out or crowd in this willingness. We identify these effects with an instrumental variable and explore heterogeneity with machine learning.

Behavioral interventions are popular tools to decrease the carbon footprint of consumers' choices, but not much is known about the channels through which they impact behavior. This project investigates one popular intervention in detail: carbon labels on canteen meals. In a series of lab-in-the-field and field experiments, I provide evidence that the labels mainly impact consumers by directing their attention toward carbon emissions. Increasing consumers' knowledge about carbon impact plays a secondary role. Using data from my experiments, I structurally estimate a model in which consumers prefer to avoid carbon emissions, but are influenced by attentional biases and misperceptions about carbon impact when making their purchase decisions. Combining this estimation with an experimental elicitation of consumers' willingness to pay to see or avoid carbon labels, I find that correcting these two frictions has an overall positive effect on consumer welfare. I directly experimentally estimate the effectiveness of the carbon labels in the student canteen as similar to that of a carbon tax of EUR 120/ton.

High-ability employees can use costly signals to distinguish themselves. However, competitors are seldom in the picture. This study proposes that pro-social behaviors toward competitors can serve as a strategic signal, leading to the following research questions: Do individuals help their competitors signal their ability or competence? Do observers recognize the signal component of generous behaviors? And do receivers of help treat it differently when it signals abilities and influences their future welfare? To answer these questions, a real-effort task experiment was conducted in a workplace frame. The results show that individuals are willing to help their competitors at a cost to signal themselves. Employers are able to recognize this signal, and it does not diminish the positive reciprocity of the competitors.

We study the problem of a planner making social choices for groups of varying sizes and compositions. We propose a preference consistency criterion that relates social preferences across different domains; this criterion requires preferences are identical on domains differing only by adding agents with choice-independent payoffs. We then derive a representation theorem for a familiar functional form---an additively-separable social utility function. We then test adherence to the criterion in an online laboratory experiment, with three major findings: choices are more consistent within a domain size than between domain sizes, choices are more consistent between domains differing by addition of agents with choice independent payoffs than duplication of agents with choice dependent payoffs, and the degree of the planner's inclusion in the domain has a substantial effect on consistency. We also document the direction of choice reversals when preferences are inconsistent, finding that the planner favors increased inefficiency aversion/decreased inequality aversion as domain size increases.

Worker sorting into tasks and occupations has long been recognized as an important feature of labor markets. This sorting may be inefficient if jobseekers have inaccurate beliefs about their skills and therefore apply to jobs that don’t match their skills, a behavior we show is common for young South African jobseekers. We run two field experiments that give jobseekers their results from standardized assessments of job-relevant skills. This redirects their search toward jobs that value skills where they score relatively highly without raising their search effort, using measures from administrative, incentivized task, and survey data. This also raises earnings and job quality, consistent with inefficient sorting due to limited information.

It has become common for people to post to social media in support of a cause, which has raised concerns that vital, higher-cost actions of support may be crowded out by these visible, low-effort options. The media voiced this ‘slacktivism’ concern after a surge in low- effort online support in the wake of George Floyd’s murder. I develop a model to formalize slacktivism. Using a between-subject laboratory experiment, I show that subjects who send - or post - a digital message to peers stating ‘I support racial justice’ are less likely to donate to racial justice charities than those who did not have the option to publicize their support. Importantly, posting does not encourage others to give, so that the social media environment not only crowds out an individual’s own donations, it also crowds out their own net contribution. Even subjects who believe posts have a zero or negative impact on others’ decisions continue to post and donate less often. Motivated reasoning and reputational moral licensing are the underlying mechanisms. In a follow-on experiment, I attempt to mitigate slacktivism by informing people of my previous results and encouraging them to donate as well as post support.

Parents make many choices for their children that are often consequential for child’s future outcomes (e.g., choice of school), but in certain situations older children may also hold their own (potentially different) preferences. Traditional models of paternalistic decision-making assume that parents already internalize to the extent they deem optimal their children’s preferences when deciding for them so that providing information about those preferences which the parent could acquire on their own should not make parents more responsive. Yet, anecdotal evidence from the centralized high school application process in Bulgaria where parents submit a ranked list of up to 200 school options for their 14-year-olds suggests that parents may not be fully familiar with how their child would rank their preferred set of schools on their own, which could result in parents not internalizing their child’s preferences to the extent they would like. I conduct a field experiment in the few weeks the parent’s final ranking is due to (1) measure at baseline the extent to which parents have wrong beliefs about their child’s school ranking, and to (2) test whether sharing with parents the school ranking their child has autonomously prepared makes them more reactive to the child’s true preferences. I find that parents guess about 36% of their child’s ranking wrong at baseline. Once I ask parents to submit the ranking they plan to send to the Ministry of Education, I find that treated parents give a ranking that is 20% closer to the child’s true ranking relative to control parents. Next, I estimate an ordinal preference aggregation model where parents weigh their ideal ranking for the child with child’s own ranking (which is the parent’s guess in control, and the true ranking in treatment) to construct the submitted ranking. I estimate that parents assign a weight of 0.46 to their child’s ranking in treatment and 0.37 in control, which remains robust to how confident parents feel about their guess in control. The difference in weights suggests that the intervention not only corrects parents’ beliefs about child’s preferences but also induces parents to respect those preferences more.

Salience and regret theory are two influential models in economics that describe how the manipulation of relative differences between potential outcomes of risky options affects individuals’ decisions. In this project, I present axiomatizations for both theories. The axioms for salience theory formulate hypotheses concerning preferences for a larger difference in outcomes compared to the summation of smaller differences that partition it. In contrast, axioms for regret theory ensure that comparisons between outcome pairs maintain the ordinal measure of their utility differences. To validate these axioms, I conduct an online experiment with 800 participants. To reduce the mental complexity of contingent reasoning, this experiment introduces a new presentation format in which potential outcomes from all options are listed jointly under each state. The current experiment finds supportive evidence for both theories at the aggregate level. However, when focusing on the subjects without stochastic choice patterns, the results indicate substantial heterogeneity. More than 60% of the subjects violate regret theory in different directions, and their deviation patterns are correlated across the tasks. The violation rates of the axioms for salience theory range between 37% and 45%. Nevertheless, the behaviors of these violators are closer to expected utility maximizers when compared to the remaining participants.

Teams play a crucial role in decision-making across organizations and institutions, bringing together a range of perspectives and skills. However, teams also face challenges such as coordination issues and potential delays in decision-making processes due to poor cooperation. This study investigates the influence of team racial and gender diversity on individual behavior in economic decision-making within a dynamic context. The study uses a series of experiments to examine the cooperation and coordination of teams with diverse compositions in a dynamic context. The results show that incumbents contribute higher than newcomers in all three gender groups. However, while incumbents decrease their cooperation and coordination in the public goods game and minimum effort games after team composition changes, newcomers increase their levels of cooperation and coordination after joining new teams.

In a laboratory experiment, participants first choose between two purely redistributive tax regimes and then generate income through a real effort task. High taxes insure against unfortunate gross incomes, but increase effort-reducing moral hazard. The results suggest that subjects’ initial tax choices behind a veil of ignorance are quite heterogeneous. Overconfident subjects are more likely to choose low taxes. When familiar with the task and their relative gross incomes, subjects choose the tax rate opportunistically: High-income subjects tend to choose the low rate, while low-income subjects tend to choose the high rate myopically. 'Voting by feet' eventually leads to a high proportion of subjects choosing the low tax. A control treatment in which the tax rate is exogenous given helps to identify the moral hazard effect because it controls for selection. A further treatment where income is randomly given helps to quantify subjects’ redistributive preferences arising from income uncertainty.

Public goods are key for well-functioning societies, yet individual support for their provision is insufficiently understood. Guided by a theoretical framework, we study how perceived relative income shapes these preferences. In two survey experiments with Indian respondents, we shift relative income perceptions using both a novel method that induces exogenous variation, and a standard information treatment. We then elicit preferences for air quality, including actual contributions to environmental initiatives. When perceived relative income increases, right-wing supporters withdraw contributions. The effect coincides with reduced health concerns and lower intentions to use private protection against pollution. Contributions are unchanged among center-left supporters.

Received wisdom holds that income rank matters for life satisfaction. In much of the literature, however, income comparisons are limited to the national population and evidence is correlational. In this paper, we investigate differences in the causal effects of rank information across reference groups. In a representative sample of mid-career Finns, we randomize individuals to receive personal rank information about educational, municipal, occupational, or age reference groups, and compare the effects, for a set of alternative welfare measures, to the standard national reference group and to a control group that receives no information. We also characterize the accuracy of rank beliefs across groups. Our data, which integrates experimental and register data, finds that rank information causes differences in satisfaction with disposable income, perceived fairness of own income, and wage satisfaction, but not life satisfaction. We also find substantial variation in the effects across reference groups, with those for the national reference group both weak and insignicant.

Decades of evidence on decision-making suggest that complexity is an essential factor influencing choices. Evidence is consistent with individuals using different choice processes in the face of complexity, but choice processes are inherently difficult to observe. We introduce an experimental methodology to test the hypothesis that individuals resort to more 'procedural' decision processes as decisions get more complex. We characterize procedural decision-making as choice processes that are more describable, and we measure this by incentivizing other participants to guess what decision-makers chose based on decision-makers' own descriptions of their choice process. In two different domains and with two different notions of complexity, we show that procedural decision-making increases as we exogenously vary the complexity of the environment. We show evidence that procedural decision-making is a choice simplification that reduces the cognitive costs of decision-making and that using procedural choice processes can lead decision-makers to make different choices and of different quality.

We conduct a field experiment in Rio de Janeiro to understand how perceptions of discrimination impact jobseekers residing in favelas (urban slums). In this context, jobseekers expect substantial discrimination associated with their place of residence. We set up an HR firm and partner with private firms with real job openings to estimate how jobseekers’ anticipated discrimination and stigma visibility affect their job application rates, effort, and interview performance. In a first experiment, removing the need to declare a home address from the application process does not increase job application rates. In a second experiment, we randomly tell jobseekers the results of an audit study where we found no discrimination in callback rates against favela residents. The information successfully decreases expected discrimination, but does not increase interview show-up rates. Finally, we randomize whether job interviewees believe their interviewer knows their address or not, and we find that interviewees perceived their performance decreases when their addresses are known. Performance gauged by the interviewer also worsens, but to a lesser extent. Our evidence suggests that perceived discrimination is usually not marginal for job application or effort decisions --because individuals strategically craft applications, obfuscating their addresses, or ignore anticipated discrimination-- but stigma visibility can hinder performance in a stressful situation, when it is harder to be strategic.

As the cost of rising obesity levels becomes of greater public policy concern, more attention has been given to policies that enforce menu labeling and health warnings. Thus, it is important to uncover the efficacy of such interventions that intend to nudge consumer behavior towards healthier eating. This paper investigates if increased salience of information for nutritional characteristics garners a higher willingness to pay for healthier options using a discrete choice experiment to elicit consumer preferences. While other research has focused on generic or “fad” nutritional claims, this study focuses on how advertisement and labeling information of actual nutritional attributes impacts consumer choice, and how consumer risk preferences affect their willingness to pay for nutritional attributes, based on the “framing” of the label presented. This study serves to provide evidence-based solutions for the rising obesity epidemic impacting both developed and developing countries and serves to form policy conclusions regarding food and nutrition. I find that risk attitudes indeed affect nutrition choices, and the type of messaging received impacts people with higher risk tolerance differently from those with lower tolerance.

I design and implement a field experiment to test whether stigma concerns interfere with social learning, in the context of mental health counseling for Syrian refugees in Jordan. I recruit central individuals in the Syrian refugee population and first measure their beliefs about who in their social network should be targeted with information. Second, I test their willingness to distribute information about a phone counseling service to their network when the social costliness of the information varies. Namely, I vary whether participants are nudged to reveal that they are being paid to share information, holding constant their actual financial incentives. This might provide them with social 'cover' to talk about stigmatized topics, or conversely might reduce the signaling value they earn by appearing prosocial. I also vary whether the message recipient is told she is being targeted as someone who might especially benefit from the information. I find first that individuals have accurate knowledge of who needs mental health services more. Second, I show that senders are sensitive to social image concerns that are activated by different framings of the same information, with rates of sharing and patterns of targeting both varying across messages framings. The effects appear to be driven by increased willingness to share when the sender’s excuse of financial compensation is revealed. These findings imply that community-based targeting may reach different people depending on the associated stigma, and that, surprisingly, making promoters’ financial compensation public may provide them an excuse to promote stigmatized programs more widely.

Women in Bangladesh struggle to access the labor market in general and male-dominated occupations in particular, despite recent progress in education and training. We add other-regarding employer preferences to a standard labor market model and use a two-sided field experiment to identify paternalistic discrimination: the preferential hiring of male workers to protect female workers from jobs perceived as harmful or difficult. We observe real hiring and application decisions for a night-shift job in Bangladesh and experimentally vary employers' and applicants' perceptions of the welfare of the job through informing them of a safe transport worker amenity of the job. Not informing employers about the safe transport home for workers decreases demand for female labor by 25%, more than twice the decrease in female labor supply of 11% from not informing applicants about the transport amenity. Combining the results of the two experiments in an equilibrium model, we demonstrate that eliminating paternalistic discrimination in our experimental setting reduces the gender employment and wage gaps. Finally, our counterfactual exercises suggest that providing safe transport is a cost-effective way to increase employer profits and worker welfare.

Empirical evidence shows that patients’ noncompliance with doctors' recommendations is a prevalent issue in the patient-doctor relationship, yet the impact of such noncompliance on doctors' performance remains unclear. In this paper, I develop a model within the framework of client-expert interaction to address this question. The model predicts that clients choose not to comply with their experts’ recommendations if they perceive the diagnosis to be imprecise. More importantly, the model predicts that this noncompliance will reduce experts’ investment in improving diagnostic precision, particularly among experts who prioritize clients' well-being. A laboratory experiment validates the theoretical predictions and investigates the effectiveness of two interventions, namely communication and reputation, in improving clients' welfare. This study illustrates the complexity of the dynamics of patient-doctor interactions and highlights the unintended adverse consequence of patients’ active participation in their own diagnosis on doctors’ performance which further reduces patients’ welfare. Furthermore, this study proposes potential interventions to improve patient-doctor relationships and promote patient welfare.

We investigate the external validity of commonly used trust measures across different contexts. In a preregistered online study, we measure trust in the US and Japan in three different ways: using the trust game, two measures commonly used in surveys, and 14 real-life scenarios (asked hypothetically). We find that, in both countries, the survey measures are better than the trust game at predicting how people behave in real-life scenarios. Furthermore, according to most (but not all) of our measures, the Americans seem to be more trusting than the Japanese. Our results show that selecting the wrong trust measure(s) for a specific research question can easily lead to erroneous conclusions.

Why do borrowing individuals fail to repay their debt optimally? We design a diagnostic laboratory experiment where we rule out selection into debt and other confounds present in the field as potential explanations. We document that allocations are predominantly suboptimal even when individuals are attentive to interest rates, equipped with optimization ability, and face unrealistically high interest rate differences. We use revealed preference methods and identify mental accounting as the main mechanism through a novel parametric test of fungibility. We further investigate if optimization failures extend to an algebraically identical investment problem. Our structural estimates reveal debt frame substantially increases fungibility violations compared to an algebraically identical investment frame. Choice process data document the debt frame increases subjects’ focus on the irrelevant balance information. These results have implications on boundedly rational models of decision-making, the design of consumer protection policies, and the evolution of wealth inequality.

HIGHLIGHTS: 1. Human attitudes and preferences are susceptible to social influence, which contradicts the axioms of orthodox economics. 2. Reaction time provides useful information about the individual's utilities, often neglected in mainstream economics. 3. Contagion impacted the participants' preferences, leading to a change in their choices. 4. Observing others' behavior will increase the amplitude of the P300 component in the midline and right posterior regions. 5. The absence of late positive potential (LPP) in the time window of 500-650 ms suggests that the presence of P300 may indicate difficulty in making decisions.

In developing countries, despite large number of skill and entrepreneurship development programs, the gender gap in entrepreneurial activities is high. This paper focuses to bridge this gap by studying the role of social networks to facilitate entrepreneurial growth for women. Through an RCT, we vary if women attend a three day training program with a randomly matched peer in the network vs alone. We find that pairing matters only when paired with a close friend that is central.

In a two-period model featuring durable goods, we explore the implications of anticipated price distortions in the context of available storage. Rationing serves as our primary mechanism to rationalize hoarding behavior. We empirically validate our model using three experimental setups: i) a free market without hoarding incentives, ii) a price-capped market where hoarding is anticipated due to potential rationing in the second period, and iii) a market with an impending supply shock, where hoarding is expected but not rationing. Our experimental results reveal a prevalent trend of excessive hoarding among participants, leading to market inefficiencies greater than our theoretical predictions.

Income inequality may be attributed to differences in effort or luck. Some people start with high opportunities, while others are unlucky and grow up in disadvantaged environments. This paper investigates how inequality of opportunity affect effort choice and fairness preferences. In our model, we show that when agents have to reach a target output, those who succeed on average are luckier and exert less effort. We test the model in a real effort experiment in which the outcome is binary: success or failure. In the experiment, effort choice varies within subject depending on the opportunities received. Lucky workers exert less effort yet succeed more. Many unlucky workers are demotivated and make less effort. Subjects make the right inference about the role that luck plays in determining success or failure. They also attribute success to effort. We observe more redistribution of earnings when the level of opportunity that the worker start with is known. The higher level of redistribution is suggestive of meritocratic behavior.

Does the self-selection of information influence how people update their beliefs? We conduct an online experiment in Taiwan where we provide information about the effectiveness of COVID-19 vaccines to participants. We first ask the subject to do three things: i) report their beliefs about vaccine effectiveness, ii) rank vaccines based on willingness to read, and iii) choose how many top-ranked ones to learn about. We then assign the subjects the selected and non-selected information randomly. After the subjects read the information, they report their beliefs about vaccine effectiveness. We have three main findings: i) the subjects prefer the information about the vaccines that they believe are the best, ii) the subjects are persuaded more by the requested vaccine information, and iii) when there is a positive update in beliefs of vaccine effectiveness, subjects' preferences of the vaccine increase. We also find that when the subjects receive the information about the vaccine they selected, their preference for it increases. We propose a model of information acquisition with Bayesian agents, and we find the agent is more willing to consume the information when the information is more likely to alter the agent's decision (either when the prior belief is less accurate or when the belief is closer to the status-quo value). Using this fact, we conclude that the update will be larger when the agent receives the selected information. Our empirical findings match the model predictions.

Economic models estimated from experimental data can guide decisions and influence policy if they can predict real-world behaviors. For instance, financial advisors might use surveys involving binary choices between lotteries to determine a portfolio composition that best aligns with their clients' attitudes toward risk. In this paper, we contribute to the analysis of how well economic models predict behaviors under risk, both within and outside experimental settings, in two distinct ways. First, we introduce a novel structural approach that generates predictions about behavior under risk without relying on specific decision models. Second, using both experimental and field data, we show that our novel approach outperforms popular economic models and black box algorithms in terms of out-of-sample predictions.

While recommendation algorithms have been increasingly used in daily life,little has been done to investigate their effect on decision making in terms of decision quality and preferences. Here we examine this question in an experimental setting whereby subjects from a representative US sample are randomly assigned to five conditions and make sets of binary choices between two lotteries. The two control conditions provide either no recommendation or recommendations based on a randomization device. The three treatment conditions provide recommendations developed by algorithms: one is based on the choice of the majority, and the other two use AI-based recommenders including content-based filtering and user-based collaborative filtering. We find that subjects tend to follow recommended choices and are willing to pay a small fee to receive recommendations for their subsequent decisions. Compared to control conditions, recommendations help subjects make better and faster decisions and behave in accordance with the independence axiom and the betweenness axiom. These results can be explained by some classes of stochastic choice models. Our work adds to the growing literature on the behavioral underpinnings of algorithms including AI and sheds light on the design of choice architecture for decision making under risk.

This study presents the first field experiment on marital status-based hiring discrimination in India and Asia. It introduces a novel prototype experimental design for resume audits and conducts the first online field experiment on Naukri.com, India's largest job portal. The research aims to assess how marital status affects hiring decisions at initial contact stage for male and female candidates in gender-neutral HR domain jobs. In six prominent Indian metro cities, key hubs in the service sector, the study conducts a comprehensive online field experiment. It involves 1964 automated job applications for female applicants and 1868 for male applicants. The experiment employs a resume audit methodology and matched designs for same-gender teams to control for gender effects. Marital status is systematically randomised within experimental blocks, automation techniques ensuring randomisation an counterbalancing of job applications. The experiment was pre-registered to reduce biases. The findings reveal a notable relationship between marital status and discrimination in hiring practices. Married males tend to receive more interview invitations than their single counterparts, though this difference is statistically insignificant. Conversely, single females are significantly more likely to secure an interview compared to their married counterparts. The study advocates a significant gender bias related to marital status in hiring practices. These results offer insights into the declining female labor force participation in India, emphasising the urgency for inclusive policies. The study underscores the pressing need for evidence-based strategies to combat discrimination in hiring practices.

Approximately one-third of global youth encounter bullying, but we're still grappling with how to effectively prevent it. Empathy, a crucial social-emotional skill, is believed to be negatively related to aggressive behaviors, yet the underlying mechanisms remain understudied. This paper fills these gaps by leveraging unique data obtained from conducting a randomized control trial of a parental involvement program on empathy development in middle schools in China. We find that the treated students improved their empathy and reduced bullying. Furthermore, offering this program also changed the social network structure: it increased friendship links, decreased isolation, and reduced the number of bully friends. To understand the mechanisms, we build and estimate a model composed of the empathy production function, network formation, and students' bullying decisions. The model reveals the individual human capital effect of empathy on bullying through private utility and the social effect through changes in the friendship network and associated peer effects. Decomposing the empathy effects, we find that the social effect is non-negligible, accounting for half of the empathy’s human capital effect on reducing bullying. Ultimately, this study aims to foster the innovation of public policy to prevent bullying and violent behaviors more broadly.

When governing entities levy financial penalties for rule violation, they may aim to maximize compliance or revenues. Agents may be uncertain of these objectives; further they may also not know enforcers' detection ability for rule violation. Utilizing a framework of verifiable disclosure game, we investigate how rule enforcers leverage the options to hide or reveal their privately-informed detection ability and how agents respond. Our model derives multiple equilibria. To examine the selection among those equilibria, we conduct laboratory experiments where the enforcer’s objective is known to the agent in transparent treatments, but unknown to the agent in the opaque treatment. In transparent treatments, unraveling occurs. However, under the opaque treatment, only compliance-maximizing enforcers with strong detection ability reveal their detection ability, and agents violate the rule when enforcers hide. Our results outline that when the enforcement objective is opaque to agents, strategic withholding information related to the detection ability benefits revenue-maximizing enforcers.

The recent decision of the Supreme Court raises questions regarding race discriminations for different minority groups. In this paper, we investigate race discrimination with three race groups: White, Black, and Asian. We employ the controlled in-lab audit study. We found that Asians with high signals and Black people with low signals are discriminated against. This observation indicates that stereotype confirming minority groups get discriminated.

Feedback on performance provides individuals with information that can inform their beliefs and help shape their decision-making. We use an online experiment to study how effective qualitative feedback is in updating beliefs and how it is used to inform decision-making. Furthermore, we study whether there are gender differences with respect to how feedback is given and interpreted and whether these differences lead to gender differences in the benefit from feedback. We run a panel study where participants complete a writing task, which is graded by other participants who also provide written feedback on the task performance. In one treatment, participants who receive feedback face a choice to compete. In another, they face a choice to edit. We study the effectiveness of feedback by looking at the estimation of participants’ grade beliefs. We find that qualitative feedback improves the estimates of beliefs and that individuals update their beliefs in the appropriate direction. We document gender differences in the benefit of feedback conditional on performance. We show that there are no gender differences in how feedback is given, but there are gender differences in how positive feedback is interpreted. With respect to the choices, we find evidence of two channels through which feedback impacts the choice to compete: a belief channel and an encouragement channel. We don’t find strong evidence of gender differences for these channels. For the choice to edit, we find that more concrete feedback to low performers increases the likelihood of editing. With no strong gender difference.

The gender pay gap is driven to a considerable extent by hierarchical job segregation between men and women. One potential source of this segregation is supply-side behavioral differences in promotion applications. In this study, we use a controlled laboratory experiment to disentangle the roles of gender, field of study, task difficulty, and preferences in promotion application decisions. Our study provides three key findings. Gender differences in promotion application behavior are present among STEM students and are more prominent when they are exposed to a stereotypically masculine task. Specifically, when an easier alternative is available, women are less willing to apply for a promotion than men. Second, there is no significant difference between men's and women's willingness to apply for promotion in non-STEM fields. Third, we find that previously reported gender differences in self-confidence are present only among STEM majors. The results also suggest that inefficient sorting into jobs leads to a decrease in total welfare. Finally, we propose an easy-to-implement policy intervention to close the gender gap among STEM students when applying for a challenging job.

The risk-incentives tradeoff (RIT) is a fundamental result of principal-agent theory. Yet, empirical evidence has been elusive. This could be due to a lack of robustness of the theory outside of the standard expected utility framework (EUT) or to confounding factors in the empirical tests. First, we theoretically study the existence of RIT under alternative theories: Rank-Dependent Utility (RDU) and Mean-Variance-Skewness (MVS). We show that RIT is remarkably robust under RDU, but not under MVS. Second, we use a novel experimental design that eliminates confounding factors and find evidence for RIT even in the case of risk-seeking agents, which is a distinct prediction of RDU. Our results provide support for the risk-incentives tradeoff and suggest that it applies to a broad range of situations including cases in which agents are riskseeking (e.g., executive compensation).

Calling out more behaviors as ``immoral'' to induce better behaviors can be futile. Broadening a negative moral category normalizes falling into this category. In reaction to the diluted negative moral category, people may behave less morally - willing to bear the decreased reputational cost of being ``immoral'' (diluting effect). This paper conceptualizes the diluting effect and provides direct experimental evidence for it. The objective function of language designers (i.e., policymakers, activists) who shape moral categories governs the relative importance of the diluting effect. However, data from a second experiment suggest that language designers underestimate the diluting effect. The results have implications for designing moral categories and shed light on the origins of misperceived social norms.

People are known to egocentrically discount instrumentally valuable information acquired through social interactions. In this paper, I explore whether this is driven by preferences over the identity of information sources using online experiments where the identity of information sources and the value of information are exogenously assigned. I focus on two types of identity : a. Naturally occurring religious/caste identity in an Indian sample, and b. Experimentally induced minimal group identity in a US/EU sample. I find that while learning increases in the accuracy of information sources, there is no evidence that participants favour information from in-group over out-group sources in either the religious or the minimal context. However, heterogeneity analyses reveal that caste identity drives differences in how individuals choose to learn from in- or out-group sources. Participants display a lower or higher propensity to learn from others, depending on whether they belong to a high- or low-status caste group. The results are not driven by differences in beliefs about the abilities of different identity groups. Taken together, the findings have implications for contexts where social identity plays a large role in daily life, and for how we think about expert advice, policy communication, and team decisions in organisations.

Beliefs about one’s own relative skill matter for many economic decisions. Yet, little is known about how these beliefs affect communication, especially the decision to talk - instead of letting other people do the talking. I use a laboratory experiment to investigate whether overconfidence leads to less successful conversations. In a communication game with aligned incentives, two senders try to inform a receiver. The accuracy of each sender’s information depends on her relative skill, such that the more skillful sender should do the talking. Senders who overestimate their skill may fail to be informative. A treatment variation creates an exogenous shock to the senders’ confidence level. The results confirm that increased confidence translates into a larger frequency of talking. The conversation, however, does not become less informative, as the participants solve their coordination problem with higher precision: in the treatment with an upward shift in confidence, they coordinate better who should talk and who should stay silent.

This paper tests a model of social learning with endogenous timing and variable signal precision in a laboratory setting. Participants in a two-person market decide simultaneously whether to make an irreversible investment. Each participant receives a private signal, either strong or weak, correlated with the investment return. Waiting is costly in case one decides to invest, however, it reveals the investment decision of the other participant. Overall, I find some departures from the equilibrium point predictions. Most of these departures are in the form of either overinvestment or a mistrust of public information. Tolerance for risk and randomization could explain some of this behavior.

The inclusion of response time indicators has become a common feature in the contemporary landscape of social media sites. Does the response time carry private information, and do people use it to improve their welfare? We design a modified cheap-talk game to study the intricate interplay between response time, private information, and its influence on users' well-being, tailored to situations where truth discovery demands an investment of time. Our investigation uncovers a noteworthy sender inclination towards truth-seeking behaviors, even in cases where deceptive intentions ultimately prevail. This preference makes the difference between different types of decisions nuanced. Consequently, the receivers are unable to extract substantial welfare gains through the interpretation of response times. In addition, when the senders are aware of the availability of their response time, we find that they are able to manipulate it successfully.

The emergent abilities of Large Language Models (LLMs), which power tools like ChatGPT and Bard, have led to both excitement and concerns about the impact that AI could have on academic writing. In this paper, we investigate how the use of AI in manuscript preparation is viewed by academics and assessed by detection algorithms. We first conduct a survey of academics in which we elicit perceptions on the ethics and reporting requirements for assistance in manuscript preparation, and we find important differences across academic roles, English language background, and forms of assistance. Next, we use GPT-3.5 to revise abstracts from academic manuscripts, and we find that abstracts are often flagged by detection algorithms as having been generated by AI, even when we just prompt ChatGPT to fix the grammar in the abstract.

As society becomes increasingly diverse, an often overlooked question arises: how does a change in our social context---defined by the individuals that surround us---influence our interactions with each other? This paper delves into this query, emphasizing prosocial behavior. I study redistributive allocations in an experiment where decision-makers are randomly exposed to either a homogeneous context, consisting of peers with same group membership, or a diverse context, consisting of peers from contrasting groups. I show that diverse contexts amplify ingroup favoritism, where decision-makers allocate more to ingroup members and less to outgroup members relative to a homogeneous context. This shift in allocations is driven by changes in perceptions of social proximity to ingroup members. Finally, exposure to a diverse context influence political views, aligning with heightened ingroup favoritism.

This paper introduces a novel menu effect that violates the transitivity principle of rational choice theory. Subscription providers often use pricing strategies in which long-term subscriptions are presented favorably compared to cheaper short-term alternatives. Given a low probability of repeated use, our experiments demonstrate that the willingness to pay for the shorter subscription increases after its reduction to a single-use option (i.e., utility decrease). Participants’ comments from a series of experiments (670 subjects) suggest that a single-use option favors rational evaluation based on the actual re-use likelihood. Instead, when both alternatives represent time spans, irrelevant comparative criteria (e.g., cost/time ratio) gain salience. Two-dimensional models, underlying standard behavioral theories, fall short of encompassing this context effect. We offer an explanation based on salience theory.

How does fact checking influence a person’s ability to persuade another? We present a lab experiment based on a Bayesian persuasion framework in which false messages are flagged with a positive probability. That is, if a Sender sends a false message, there is a positive probability that the Receiver will be fully informed about the state of the world. We find that not all subjects behave as predicted. In particular, Senders do not respond to increases in the probability of message flagging one way or another, i.e., they neither increase nor decrease their frequency of lying. Receivers, however, respond to increases in the message flagging probability as predicted, becoming increasingly more trusting of non-flagged messages and using Bayes rule to update their beliefs. Lastly, we show that these results provide important insights for litigation, lobbying, and regulating misinformation.

Many individuals exhibit inadequate savings behaviors, prompting inquiry into the contributing factors and the efficacy of interventions aimed at promoting increased savings. This study employs a survey experiment targeting College Savings Account holders, matching their responses with corresponding administrative records. I show that financial constraints and misbeliefs surrounding effective saving strategies are both contributing factors to lower savings rates. To address this issue, respondents are randomly presented with one of three messages: (i) a behaviorally-informed message that matches one of their (mis)beliefs, (ii) a behaviorally-informed message that mismatches it, or (iii) financial literacy resources. Results show that belief-matched messages lead to higher savings balances over a six-month period, particularly at the intensive margins and among individuals with lower income, and financial literacy outperforms belief-mismatched messages. Notably, the enhanced savings are attributed to augmented regular contributions, rather than short-term, one-off increments. Additionally, I find no evidence of crowding out effects. I interpret these results through the lenses of a signaling model where relevance influences behavioral response. A simple cost-benefit analysis highlights the need for strategic minimization of data collection costs for personalized behavioral interventions to be cost-effective.

It is well known that individuals avoid information in many important circumstances. By eliciting menu preferences for information ahead of its availability, we reveal that around half of individuals are nonetheless tempted by information they want to avoid. Some individuals are offered the information in a later session, regardless of their menu preferences, and a small share make dynamically inconsistent choices. We show theoretically that deliberation time increases in self-control costs for a resistible temptation. Using this measurement, we find that preferences for commitment reflect sophistication about experienced self-control costs, in turn indicating that avoiding information becomes harder at the time of its availability. To provide intuition for our results, we show that temptation arises in a dual-self model of information preferences with imperfect recall. Since the availability of temptations can harm welfare, our findings provide insight into the implications of recent growth in the supply of information.

People disagree about what is fair. But how important are disagreements about fairness for understanding people's support for welfare policies? In this paper we study this question in a large and representative sample of US Americans in the context of the coronavirus pandemic. Fist, we introduce fairness preferences into a Meltzer-Richard model and test its theoretical predictions. We find that fairness preferences - identified using a novel experimental design - are a stronger predictor of people's support for welfare policies than their income. Fairness preferences also prove to be fundamental because they predict how much weight people attach to beliefs about the causes of inequality. Second, we study the stability of fairness views and policy support in times of crises using individual-level panel data and a complementary experimental manipulation. Our data show that changes in support for welfare policies during the pandemic are rather caused by changes in beliefs about the causes of inequality or by self-interest than by changes in fairness preferences. Our results have implications for prominent models in political economy and for understanding the mechanisms shaping people's demand for fair policies.

Many firms engage in Corporate Social Responsibility (CSR) activities, advertising them among consumers, workers, and job candidates. CSR increases the firm attractiveness to potential employees as reflected in the number of job applications. The economic literature attributes this effect to CSR boosting workers' sense of purpose. We develop an additional explanation for why workers prefer to work for firms that engage in CSR---even when the activity does not affect them directly. We suggest that firms use CSR to signal their future treatment of employees. We show experimentally that managers use CSR in a risky environment to signal to potential workers they will help them if they suffer from a personal negative shock. Workers respond to this signal by accepting more job offers in a risky environment from managers that engage in CSR. However, the result holds only for male employees and managers.

Probability matching, as a classical violation of expected utility maximization in the repeated choice, describes individuals’ tendency to mix or match their choice frequency with the outcome probabilities, when repeatedly asked to predict two payoff-relevant outcomes that only differ in their probabilities of occurrence. We use the correlation between outcomes and frame in the payoff structures as tools to decompose into three broad classes of theories: models of Correlation-Irrelevant Stochastic Choices, models of Correlation-Sensitive Stochastic Choices, and Framing Effect. Our experimental design uses a diagnostic approach, using three treatments to decompose the three classes. We find that the underlying correlation between options accounts for a substantial amount of mixing behaviors: (1) mixing behaviors decrease when the correlation between outcomes increases; (2) learning amplifies the responsiveness to correlation change between options; and (3) once the correlation is fixed, mixing behaviors do not vary with the frame change. Our findings emphasize the importance of developing models that capture the responsiveness of stochastic choices to the correlation between options, a factor widely overlooked by most existing preference-based models including correlation-sensitive preference.

I explore the dynamic cognitive hierarchy (DCH) solution proposed by Lin and Palfrey (2022) in the context of multi-stage games of incomplete information. In this environment, as the game progresses, players simultaneously update their beliefs about other players' payoff-relevant types and levels of sophistication. Applying the DCH solution to a class of two-person dirty-faces games, I find that lower-level players tend to figure out their face types in later periods than higher-level players. This finding sharply contrasts with the standard equilibrium theory. Moreover, DCH predicts a specific violation of strategic equivalence---players might exhibit different behavior in two different games that share the same reduced normal form. To test this prediction, I conduct a laboratory experiment using two strategically equivalent versions of the dirty-faces game. The results indicate significant differences in players' behavior between the two versions. Furthermore, the direction and magnitude of the observed difference align with the prediction of DCH.

Cryptocurrencies are a part of new technologies that allow people to hold, trade, buy and sell digital tokens. While the number of cryptocurrency owners is increasing, there is some reluctance towards ownership. It is generally understood that a barrier to adopting new technology is having some knowledge of it and how it works, which could account for some of the reluctance toward cryptocurrency adoption. To understand the types of knowledge required to encourage cryptocurrency use, we provide subjects with varying knowledge-based messages about cryptocurrencies and evaluate their willingness to own cryptocurrencies after reading the message. Our online survey experiment consists of 1400 Americans and contains a mix of cryptocurrency owners and non-owners. With this survey, we identify characteristics of owners and nonowers of cryptocurrencies and measure the willingness to own cryptocurrencies after short, informational messages. We find that cryptocurrency owners are typically young males with a high tolerance for risk. In addition, our results indicate that brief messages concerning the ease of use of cryptocurrencies and the security features of cryptocurrencies are effective at increasing nonowners’ willingness to own cryptocurrencies in the short run. Our research is an introductory step in understanding the knowledge of and openness toward cryptocurrencies.

Games with multiple equilibria introduce the potential for populations to get stuck in inefficient outcomes. In theory, the introduction of additional equilibria, "stepping stones", could pave the way for a smoother and less risky transition. I run a lab experiment to test if the introduction of these “stepping stones”, can facilitate transitions from an inefficient but safe equilibrium to a risky, payoff dominant equilibrium. I employ different payoffs for the transition strategy and examine the effects that different degrees of information about the game have on group's play. I find evidence that adding these "stepping stones" does help populations transition to the efficient equilibrium. I also find that when groups have more information about each other's payoffs they are able to transition to the efficient equilibrium faster and are less prone to cyclical behavior.

We consider a sender who observes the state and chooses between disclosing the state or spending time and cost to utilize an imperfect reporting device. The sender's payoff is maximized if the receiver guesses the state is good, while the receiver's payoff is maximized by correctly identifying the state. The sender can utilize the device in the bad state to persuade and extract surplus from the receiver. Extraction is contingent on the receiver's inability to differentiate between sender disclosures and device reports. Due to the time requirement and state-dependent utilization of the test, a sophisticated receiver can infer the state if the sender does not strategically choose their disclosure time. We model interaction, solve for all Bayesian Nash equilibria, and conduct an experiment varying the visibility of disclosure time. We explore the impact of this variation on subject behavior, and compare the observed behavior to our theoretical predictions.