Michael G. Cuna

We study the hidden curriculum: informal, tacit strategies crucial for academic and career success. These unobserved skill gaps are a critical barrier to information acquisition and limit upward mobility. Using observational data from over 100,000 U.S. college students, we document significant gaps in academic and labor market outcomes, with first-generation students falling behind continuing-generation peers in ``hidden'' actions like proactively engaging with faculty and joining organizations. Through a field experiment at UC Berkeley, we identify two main channels driving this gap: lack of awareness and low subjective beliefs about returns. Information treatments substantially increase willingness to invest in hidden curriculum actions for first-generation students, closing the gap with their more informed peers. Finally, we develop an AI college advisor to further isolate these mechanisms in an online experiment. At baseline, first-generation students devote less search effort to hidden curriculum topics and exhibit lower propensity to switch topics. However, when the AI actively steers conversations toward hidden curriculum actions-increasing awareness-these gaps narrow, providing additional evidence that information constraints drive hidden curriculum underinvestment.

From newborns to the elderly, exposure to violence and conflict has been found to have deleterious effects. In this study, we explore a unique type of violence: exposure to the Taliban. Pairing a field experiment with a field survey among citizens in Khyber Pakhtunkhwa (KP), Pakistan, we examine how exposure to violence affects general trust, subjective well-being, and confidence in institutions. In our field experiment, we observe that exposure to conflict significantly alters the relative valuation of monetary rewards for oneself compared to those for a comparable peer. Specifically, individuals subjected to violence demonstrate a marked tendency to prioritize their own financial gain over that of a similar other.  In the survey, we find that exposure to violence is associated with reduced general trust, trust in informal institutions, and subjective well-being. Interestingly, being exposed to violence increases trust in formal institutions.  Our combined results highlight that the interplay between violence and trust dynamics is complex and highly consequential.  In turn, the policy implications highlight the need for a multifaceted strategy to support individuals and communities affected by violence, ensuring both immediate relief and long-term resilience.

Understanding the role of preferences, beliefs, and constraints on social and wealth inequities is a key unlock for economic growth.  This study focuses on the inter-relationship between risk and ambiguity preferences of mothers, their early childhood investments, and their children's outcomes.  To do so, we jointly elicit ambiguity attitude and risk aversion preference parameters from more than 6000 randomly sampled mothers from nearly 500 villages in the district of Udaipur in Rajasthan, India. Across several measures of mothers' investment in nutrition of their children between the ages of 0-6, we find a robust and stable positive correlation of estimated ambiguity attitude and risk aversion parameters with maternal investments in children:  the more risk and ambiguity averse the mother, the greater her maternal investments.  Such investments are correlated with better children's cognitive and non-cognitive skills, as mothers with greater risk and ambiguity aversion have children with superior skills, even after accounting for socio-economic differences. Importantly, the positive effect of ambiguity and risk aversion on early-life outcomes can attenuate the negative impact of proxies of socio-economic disadvantage, such as illiteracy of the mothers, belonging to historically discriminated social groups, no exposure to radio, television, or zero access to mobile phones for all measures of cognitive and non-cognitive early-life skills. 

Using a field experiment with mothers in Rajasthan, India, we document substantial divergence between stated early childhood vaccination intentions and actual vaccination uptake. The information intervention significantly increased positive survey responses on stated willingness to take up the remaining vaccinations for their children, but had negligible effects on actual vaccination rates. We develop a model to make precise predictions regarding mechanisms - social signaling and construct validity- that might influence discrepancies between survey responses and real outcomes.

We study human AI interaction in job expectations and job search behavior by partnering with an AI-powered job-tech startup. The increasing availability and sophistication of AI tools in the labor market—from resume builders to personalized cover letter generators—has reshaped how job seekers navigate the job search process. A prevailing assumption is that AI tools should increase the speed of job applications and provide users with informative signals about what types of jobs to apply for and how to apply. Users would then use the information generated by AI to update their job application material and beliefs in a Bayesian fashion. However, users may have reasons to override recommendations, either because they correctly or incorrectly believe themselves to be more knowledgeable than the AI tools, or because they have ego concerns or motivated reasoning to reinforce their existing beliefs. We first document human-AI interaction in the job search process using a proprietary dataset on job searches, job application material (CVs, cover letters, etc.) and job outcomes. We find that AI job search tools are associated with an increase in the number of jobs applied to, but don’t necessarily lead to more job offers and job matches. To understand the mechanisms behind this pattern, we are in the process of fielding an experiment that will randomize the type and presentation of AI-generated advice to examine how recommendations affect user expectations, application quality, and longer-term employment outcomes.