Research

Children’s mental health is the defining public health crisis of our time. Using insurance claims for a national sample of 8 million privately insured children, I provide the first systematic quantification of the drivers of variation in children’s mental health prescribing in the United States. I separate variation in pediatric ADHD medication and antidepressant prescribing due to differences in: 1) primary care provider (PCP) prescribing intensities, 2) regional practice environments, and 3) child health and demand. I find that eliminating differences in PCP prescribing intensities would reduce the variance of provider prescribing rates by 50 percent for ADHD medication and 65 percent for antidepressants. Geographic variation analyses understate the extent of treatment variation and the role of providers in driving overall treatment variation. I also find suggestive evidence that higher-quality PCPs tend to have higher ADHD prescribing intensities but lower antidepressant prescribing intensities.

Patients often lack information about the out-of-pocket cost of medical care. Recent federal rules requiring hospitals and insurers to disclose negotiated prices and personalized cost estimates are premised on the idea that better-informed patients will shop for cheaper care. The effects of providing that information are theoretically ambiguous, however, because insurance cost-sharing distorts the prices patients face. Better information may make patients more price-elastic, but it may also reveal that insurance compresses the out-of-pocket gap between high- and low-price options, inducing substitution toward expensive care and raising total spending. This paper studies the interaction between limited patient information and moral hazard in the market for emergency care. Using commercial claims from the Colorado All Payer Claims Database, I estimate a model of demand for emergency departments and urgent care centers. I use a moment inequality estimator that partially identifies preferences while allowing patients to be unobservably uninformed about negotiated prices, care intensity, and cost-sharing rules. I then simulate spending and patient welfare under counterfactual information and cost-sharing policies.

Private equity (PE) firms have expanded rapidly into mental health treatment, with little evidence on patient outcomes. We build a new dataset of residential mental health facilities acquired by PE firms from 2013 to 2024, link it to national insurance claims, and estimate stacked event studies comparing treatment spells at the same facility before and after acquisition.

Evaluating changes to social programs requires predicting how take-up responds, which means separating the roles of application costs and potential applicants’ limited information about their eligibility and benefits. We develop a moment inequality framework for estimating a model of take-up that allows agents to have unobservably limited information and heterogeneous application costs, using observational data feasibly obtainable in most program settings. We make three methodological contributions: we extend moment inequality methods to random coefficients; show how survey data on average misinformation can relax the standard rational expectations assumption; and construct bounds on counterfactuals without imposing a particular information set. We apply the framework to Supplemental Nutrition Assistance Program (SNAP) take-up among low-income elderly households, combining administrative application data from Pennsylvania with an original survey.