Session 12: Psychology and Economics

People make correct inferences from observed events but make mistakes when reasoning about hypothetical events. I study the implications in competitive markets. I define Dynamic Cursed Expectations (DCE), where agents treat events occurring at different future periods as independent. I build on it to define Dynamic Cursed Expectations Equilibrium (DCEE) and prove its existence for a general financial economy. I apply DCEE to an asset pricing model and show that in DCEE agents underestimate the variance of risky assets, which under risk aversion leads to overvaluation relative to the Rational Expectations Equilibrium benchmark.

We study the role of informational complexity on the higher-order rationality of agents in Bayesian games of incomplete information. In a class of dominance solvable Bayesian chain games, we find evidence that informational environments which rely on more private information, and are therefore more complex in terms of contingent thinking requirements, are associated with lower orders of rationality. The results have implications for the optimal design of information structures, by highlighting a potential source of discrepancy between observed behavior and theoretical predictions.

When voters underappreciate the equilibrium effects of policies, they may rank them incorrectly, making socially optimal policies politically infeasible. I study the policy implications of equilibrium neglect (EN) and propose a potential remedy. I first develop a general framework of EN, and then use this framework to construct a portable and generally applicable potential remedy: using off-path contingent rebates a social planner can generate a policy that (i) implements the social optimum, (ii) does not increase the deficit—on and off the equilibrium path—and (iii) is politically feasible even in the presence of equilibrium-neglectful voters. This solution can be applied in a large class of problems. Empirically, I administer an opinion poll on congestion pricing with a large representative sample from six US metros. Respondents are overly pessimistic about traffic-reduction effects of congestion pricing, which in turn translates to strong opposition to the policy. Adding contingent compensation that pays if traffic remains high significantly increases public support, especially among potential recipients.

We study decisions by agents whose information depends on their actions and is based on a random subset of their past experiences. If the empirical distribution of actions converges, the limit must be a stochastic memory equilibrium, and that stochastic memory equilibrium generates the stochastic choices of random utility models. We illustrate how our model can be used to study the effect of reminders on behavior and underreaction to large samples. Extending the model to allow for recency and rehearsal effects enables us to explain correlated prediction errors in forecasts of returns and the equity premium puzzle.

I design an experiment to document and understand misspecified mental models with an illusion of control, i.e., false beliefs in causal impacts of controllable factors on payoff-relevant outcomes. Participants learn from observational data and then manipulate a randomly assigned variable — their choice variable — to potentially affect a payoff-relevant outcome. I directly elicit participants’ mental models about the outcome-generating process and find that their models tend to involve the choice variable, implying an illusion of control when this variable is completely irrelevant. I show that the experience of making choices directly distorts mental models, and provide suggestive evidence that memory plays a mediating role: The choice process causes individuals to selectively retrieve the memory information on the association between their choice variable and the outcome.

Why is in-kind aid a prominent feature of welfare systems? We present a lab-in-the-field experiment involving members of the general U.S. population and SNAP recipients. After documenting a widespread desire to limit recipients’ choices, we quantify the relative importance of (i) welfarist motives, (ii) utility or disutility derived from curtailing another’s autonomy, and (iii) absolutist attitudes concerning the appropriate form of aid. Choices primarily reflect the two non-welfarist motives. Because people systematically misperceive recipient preferences, their interventions are more restrictive than they intend. Interventionist preferences and non-welfarist motives are more pronounced among the political right, particularly when recipients are black.

We examine whether autonomy increases adherence to expert recommendations in technology adoption. In a context where farmers overuse fertilizer, we ran a field experiment that combined recommendations with either a restrictive subsidy tied to expert-recommended inputs or a flexible subsidy preserving farmer autonomy over input choice. In the short run, farmers adopted expert recommendations at similar rates regardless of subsidy autonomy and reduced fertilizer over-use by two-thirds. In the longer run, after the intervention ended, farmers with autonomy were significantly more likely to persist with the expert recommendations. We replicate these findings in a complementary laboratory experiment and find suggestive evidence that autonomy increases persistence by improving recommendation recall. Our results suggest that preserving choice can enhance the longterm effectiveness of expert advice and targeted subsidies.

I develop a structural model to examine how short-form content length interacts with user temptation duration, exacerbating self-control problems on social media. Microdata from a U.S. short drama series indicate viewers consume 23 episodes (82%) more than intended and overspend by $5.51 (23%). With temptation averaging 6.6 minutes per decision, minute-long videos repeatedly trigger present bias. Extending the analysis to TikTok reveals short videos triple efficiency losses compared to YouTube’s 12-minute videos, reducing U.S. monthly consumer surplus by $10.2 billion. Implementing a default time limit of 12 minutes per day could recover $6.8 billion of this loss, underscoring the value of targeted policy interventions.

Simple choices between positively-valued options are common in our daily lives and are susceptible to robust attentional choice biases. However, we also encounter forced choices between negatively-valued options (“aversive choice”). Based on previous studies and the predictions of attentional choice process models, we hypothesized that attentional choice biases would reverse as people switch from choices between gains to aversive choices. However, in this paper, we find not only that the reversal does not occur, but that the direction of attentional choice biases is the same in aversive choices as it is in appetitive choices. To explain this behavior, we use a variation of theAttentional Drift-Diffusion-Model that incorporates reference-dependent value signals. These results suggest that even in aversive choices, decision-makers are still susceptible to the same attentional manipulations at work in gains.

Pakistan ranks in the lowest decile in female labor force participation, and even in sectors where women are more prevalent, such as teaching, they earn 70 cents for each dollar men earn. While we have extensive evidence on the prevalence of gender bias in hiring, promotions and wages, we know less about the mechanisms underlying this bias and the extent to which certain personnel policies may mitigate or exacerbate these biases. To test this, I conduct a large scale field experiment with 3,600 employees in 250 schools and randomly vary i). how often managers observe a given employee and ii). whether manager evaluations affect employee’s pay or are just used for feedback. First, I find when there are no financial stakes associated with performance evaluations, there is minimal difference in scores between men and women. This holds even when controlling for a rich set of controls of teacher productivity, such as value-added, clock in and out time, time use, and pedagogy measured via classroom observations. In contrast, when principals’ evaluations determine teachers’ end of year raise, we see that female teachers receive 20% lower raises, controlling for productivity. However, when principals are randomly assigned to conduct more frequent classroom observations of the teacher, this increases the evaluation of female teachers and closes two-thirds of the gender gap under financial stakes. To understand mechanisms, I conduct a follow up vignette survey to test whether our results are due to differential manager expectations about employee reactions to low raises (e.g. higher turnover by men) or differential perceived deservedness of scarce financial resources. The results favor this second explanation as managers favoring single-earner (lower income) households over dual-earner (higher income) households, which is highly correlated with employee gender. Combined this suggests that improving the accuracy of manager information could close the gender gap in performance evaluations, even in high stakes settings.

This paper comprehensively investigates how people expect labor market behaviors by men and by women to be viewed. Men conforming to the stereotypical male behavior—to be confident, to be assertive, to be ambitious, to be competitive, to be leaders, to take risks, to negotiate, and to self-promote—are expected to be praised and not criticized. Since the opposite is expected for men not conforming to the male stereotype, we say that men face a “restricted path." We instead say women face a mission impossible because there is no behavior for women that is expected to secure praise and avoid criticism. Any behavior women pursue is also less likely to secure praise than men who conform to the male stereotype. Additional results investigate whether evidence for men’s restricted path and women’s mission impossible differs when we ask about the expected views of groups that vary in gender, age, race, political identity, education, and profession. Additional results also reveal that—for prosocial traits—no mission impossible arises for either gender; being prosocial is expected to be praised and not criticized.

This paper presents an axiomatic characterization of the Grether (1980) updating rule and Bayes’ rule that is centered around the preservation of the Monotone Likelihood Ratio Property (MLRP). I show that Bayesian updating can be characterized by the preservation of MLRP from signals and the martingale property, while the Grether (1980) model is characterized by the preservation of MLRP from signals and priors. This framework allows us to identify a class of non-Bayesian updating rules where we can obtain useful comparative statics across different signal realizations and prior beliefs in canonical belief updating problems. I also conducted an experiment to test the axioms that can be used to characterize Bayes’ rule to identify why people are non- Bayesian. I find that subjects’ broadly comply with the preservation of MLRP, providing validation for the widely used Grether (1980) model.

Forecasts about research findings affect critical scientific decisions, such as what treatments an R\&D lab invests in, or which papers a researcher decides to write. But what do we know about the accuracy of these forecasts? We analyze a unique data set of all 100 projects posted on the Social Science Prediction Platform from 2020 to 2024, which received 53,296 forecasts in total, including 66 projects for which we also have results. We show that forecasters, on average, over-estimate treatment effects; however, the average forecast is quite predictive of the actual treatment effect. We also examine differences in accuracy across forecasters. Academics have a slightly higher accuracy than non-academics, but expertise in a field does not increase accuracy. A panel of motivated repeat forecasters has higher accuracy, but this does not extend more broadly to all repeat forecasters. Confidence in the accuracy of one's forecasts is perversely associated with lower accuracy. We also document substantial cross-study correlation in accuracy among forecasters and identify a group of "superforecasters". Finally, we relate our findings to results in the literature as well as to expert forecasts.

The process of diagnosis in medicine depends on high dimensional, complex belief formation, repeated experimentation and data aggregation from multiple sources (e.g. labs, images, discussions with patients). In this paper we explore the role of point-in-time cognitive capacity on decision making of physicians. Using detailed audit log data from emergency rooms, we access how cognitive load during the shift affects future choices of tests as well as the treatment. We draw upon theories of inattention and information acquisition to infer how much is physicians’ knowledge refined when making a particular choice. For instance, we find that high cognitive load implies that physicians make less informed choices in the future, perhaps choose to think less, and instead substitute this with more orders of diagnostic tests.

We relate the degree of understanding a decision-maker has in binary-choice decision problem to both the degree of randomness in choice from that problem, and the degree of confidence that the decision-maker expresses about their decision. We introduce novel partial orders over Blackwell experiments in order to rank them in terms of understanding (i.e., informativeness), induced choice randomness, and induced confidence in choice. We show that in general, knowing the relative ranking of two Blackwell experiments in two of the three orderings does not tell you anything about how to rank them in the third order. We then provide sufficient conditions to relate the orderings. Our results suggest caution in interpreting the relationship between shocks to understanding of a decision problem (such as changes in complexity) and measurements of choice randomness and choice confidence, such as recent work on cognitive uncertainty.