Session 16: Economics of Health Inequality

We provide the first systematic quantification of how decision-making frictions—such as failing to claim government benefits, choosing dominated insurance plans, not saving for retirement, and not quitting smoking—aggregate to affect inequality in income, consumption, and wealth. We review the existing literature and combine it with original analysis of survey data to estimate the prevalence and financial impact of 18 frictions across the income distribution. To make these frictions comparable, we develop a framework in which each friction is characterized by three parameters: the share of the population at risk, the share affected by the friction, and the average loss conditional on being affected. Aggregating across the frictions with dollar-loss estimates, the estimated impact on annual income is 7.8% for the bottom quartile of the income distribution relative to 4.2% for the top quartile; the total loss for low-income households is approximately 7.5 times larger than the impact of a major EITC expansion. We then incorporate these frictions into a life cycle model with realistic institutional features, including tax-advantaged retirement accounts, progressive taxation, portfolio choice, and a social insurance system. The model reveals that removing frictions tends to reduce inequality in lifetime consumption, with the largest effects coming from smoking and attending for-profit colleges. Our results suggest that decision-making frictions are a quantitatively important contributor to inequality in income, consumption, and wealth.

We document the individual, organizational, and field-wide consequences of a public disclosure revealing substantial male premiums in competence assessments of recent biomedical PhDs applying for a major individual grant from the Swedish NIH. Following the disclosure, three key changes occurred: (i) male-only review committees were rapidly phased out; (ii) individual reviewers adjusted their decision-making; and (iii) the average male premium was eliminated. We follow applicants’ publications, promotions, and earnings for up to 18 years after application. After the disclosure, the share of female grantees rose and, consistent with improved selection on research potential, the long-run output gap between funded and unfunded applicants widened. However, the disclosure of bias not only improved fairness and allocative efficiency; but also triggered coordinated responses with broader implications for womens’ health. Female enrollment in biomedical PhD programs rose by 10 pp relative to other fields, which in turn increased female health-focused research by 20 percent, without crowding out attention to men’s health.

Worldwide, depression is more prevalent among women than men, but it is unclear whether this reflects differences in underlying mental health or diagnosis. We find that the gender gap is smaller in large-scale representative screening surveys than in recorded-case data, suggesting that part of the gap reflects differential entry into diagnosis. We test three plausible mechanisms in an online experiment: depression recognition, severity assessment, and help-seeking. For equivalent symptoms, men are less likely than women to recognize depression and to seek help from specialists, while women are more likely to overestimate symptom severity. These findings suggest that men with depressive symptoms may be less likely to appear in depression statistics, while women may be more likely to do so even when symptoms are mild.

Generative artificial intelligence (AI) may reshape expert–client decision-making in two ways: by reducing the informational asymmetries that underlie expert authority, and by transmitting the institutional logic encoded in AI training. We study these mechanisms in healthcare, a canonical credence-good market characterized by substantial information asymmetry between physicians and patients. In a preregistered field experiment involving more than 10,000 outpatient visits in China, we randomly assign patients access to a generative AI chatbot before their consultation. Access to the chatbot changes clinical practice: prescription rates decline, particularly for Traditional Chinese Medicine and antibiotics, while diagnostic testing increases and short-term revisit rates fall. Analysis of the patient–AI conversation logs shows that the chatbot consistently raises cautions when discussing medications but issues clean recommendations for diagnostic testing, consistent with liability-driven guardrails encoded in AI training. Our results highlight that these guardrails propagate into real clinical decisions at scale. Beyond shifting healthcare utilization, survey results show that AI usage reduces patient compliance and satisfaction, shifting the balance of authority between experts and clients.

Every child transitions to adulthood. At age 18 this predictable transition severs a longstanding tie: pediatricians typically stop seeing patients, forcing adolescents to rebuild care in the adult system. We use a regression discontinuity design to estimate the consequences, following 5.8 million continuously enrolled Medicaid beneficiaries through their 18th birthday. We find that the week after a beneficiary turns 18, pediatrician visits decrease by 14% and emergency department visits increase by 8%. Prescriptions for nearly every drug class fall with one exception: new opioid prescriptions rise by 15%. Exploiting large, exogenous increases in Medicaid reimbursement, we show that raising payments to adult providers offsets roughly half the decline in office visits, pointing to a potential policy lever for smoothing the transition.

In the U.S., federal law mandates that contraceptives are available at zero cost to patients, yet nearly one in three women of reproductive age still report inadequate access to contraceptive services. With demand-side affordability barriers largely removed, supply-side factors—the actions of healthcare facilities, providers, and insurers—may be critical in explaining this unmet need. This paper examines supply-side determinants of long-acting reversible contraceptive (LARC) use among Medicaid beneficiaries, a population for whom both LARCs and oral contraceptives (OCs) are covered at zero out-of-pocket cost. Using nationwide Medicaid claims data from 2015 to 2019 and a movers design that exploits cross-area migration to isolate supply-side from demand-side variation, we find that local supply-side factors explain 56 percent of geographic variation in LARC use—substantially more than the 35 percent explained for OCs. We then estimate the causal effects of two supply-side policies.Clinic participation in the 340B Drug Pricing Program increases LARC use by 4.6 percent without displacing OC use or increasing LARC removals. By contrast, Medicaid reimbursement changes for immediate postpartum LARC insertion increase uptake by 167 percent but are accompanied by a comparable rise in removals, raising concerns about whether induced insertions reflect informed patient choice. Together, these findings highlight that supply-side factors are critical determinants of LARC use, but the clinical context of insertion matters for whether expanded access translates into persistent, patient preference-concordant use.

Marriage and health are positively correlated due to selection and protective effects, but little is known about how health affects marital dissolution. Using data on unanticipated health shocks from the PSID, I construct matched control groups of couples with similar expected marriage durability and exposure to health shocks and find that adverse health shocks raise divorce hazard: wives’ health shocks raise ten year divorce incidence by 69% while husbands’ raise it by 23%. The effect varies by diagnosis: cancer diagnoses do not significantly affect divorce hazard, while psychiatric and cardiovascular health shocks elevate it. The increase in divorce hazard cannot be explained by income loss or medical expenses. Instead, heterogeneity by household composition suggests that spousal care burdens mediate the effect of health shocks on divorce. Descriptive evidence on remarriage also suggests that both experiencing a health shock and being married to a spouse who experiences a shock affect spouses’ outside options in the event of divorce.

Vaccination forms a cornerstone of public health, combining private protection with reduced transmission to others. Motivated by this, we study how prior adverse health events affect vaccine hesitancy. We use new, nationwide Swedish administrative data that link individual vaccination records, reports of adverse drug reactions, and detailed healthcare utilization across the entire population. We first examine how individuals draw on their past experiences when making new vaccination decisions. To do this, we analyze a severe, well-identified case: narcolepsy, a chronic neurological disease plausibly induced by the 2009–2010 swine-flu vaccine. We find large reductions in COVID-19 vaccination more than a decade later, with spillovers to close family members, suggesting that individuals rely extensively on own experiences when making decision about their health. The effects do not attenuate among those with high health literacy but show some evidence of attenuation with extensive prior healthcare contact. Second, we assess the broader social costs of routine adverse events by studying serious events from all licensed pharmaceuticals. The average effects of experiencing such events on COVID-19 vaccination outcomes are small. While there are some effects for vaccine side effects, there are virtually none for adverse events from other drugs. This highlights that individuals draw on only a narrow set of experiences in future decisions and suggests that the overall social costs of adverse events are likely limited. In the final part of the paper, we show that similar patterns extend to well-established childhood vaccines. This suggests that severe adverse events can erode trust in official safety communication, rather than merely reflecting learning from experience. While the overall impact of routine adverse events on vaccination is limited, rare but severe vaccine-related events can meaningfully lower uptake, providing a cautionary tale for future vaccination campaigns.

How harmful is stigma in the “real world”? Answers are elusive because stigma is difficult to measure in observational data, and isolating its effects requires exogenous variation in stigma without variation in the stigmatized trait. This study addresses these challenges by focusing on a widespread form of stigma — weight stigma — in the high-stakes setting of inpatient healthcare. BMI categories are displayed prominently to providers in electronic medical records, and obesity is heavily stigmatized socially. The “obese” cutoff may thus discretely shift stigma while keeping constant the underlying trait. Using a regression discontinuity design that exploits this institutional feature, I find a discontinuous increase in in-hospital mortality at this cutoff, though patient health does not change. Two patterns suggest stigma-based discrimination as the mechanism. First, just-obese patients receive lower diagnostic effort than almost-obese patients. Second, a physician- validated LLM identifies a rise in stigmatizing language in clinical notes at the cutoff — specifically, statements that impose moral judgment, undermine patient credibility, and stereotype patients — that closely tracks mortality effects. Overall, this paper establishes stigma as a powerful social force that can have life-or-death consequences.

Despite having health insurance, patients in low-income countries like Ghana often face significant out-of-pocket expenditures (OOPE) for covered health services and drugs. This study examined whether a targeted information campaign can empower patients to reduce these unauthorized fees, especially in contexts of low health insurance literacy. Through a randomized controlled trial across 41 health facilities and 2,284 patients, it was found that informing insured patients about their benefits increases their understanding of health insurance and leads to a 25% reduction in OOPE. The specific delivery method of the information (verbal vs. verbal and written, general vs. specific information) does not differentially affect outcomes, with all information messages being similarly effective. Importantly, this intervention does not compromise access to health care or the quality thereof. The results imply that information provision leads to a reduction in individual-level rent-seeking rather than facility-level cost recovery. While the results are encouraging, they also reveal a persistent challenge: 65% of insured patients still incur OOPE, underscoring that patient education alone is insufficient to fully eliminate the problem. The study provides strong evidence for a simple, cost-effective intervention to combat petty corruption and improve financial protection in health care.

This paper studies whether hospital performance for Black patients is malleable and how it shapes racial disparities in healthcare. Using local Black Lives Matter protests as an exogenous shock to providers’ racial awareness and social pressure, we find that exposed hospitals experienced significant declines in 30-day mortality among Black heart attack patients, with no detectable change for non-Black patients. Evidence from emergency department admissions and non-exposed patients supports a supply-side response. We observe no changes in resource utilization but substantial productivity gains, partly driven by an increased likelihood of Black patients being treated by high-productivity physicians.

We investigate the evolution of health from birth to death and its transmission across generations using comprehensive Swedish administrative data that span multiple decades. We link parental characteristics, including health, income, and education, with the long-term health outcomes of their offspring, enabling a detailed examination of how these factors shape health trajectories throughout life. We create a novel health index based on hospitalization records, which captures both the extensive and intensive margins of health care utilization and is consistently measured over the life course. We find that parental health is a stronger predictor of health at birth than SES, with children of unhealthy mothers having lower health and higher mortality in the first year of life. Thereafter health improves during childhood but then worsens throughout adulthood, albeit at very different rates depending on health levels. Notably, children in poor health before age 20 live significantly shorter lives. We find a sharp deterioration in our measure of health that is detected well before mortality occurs. As children grow older, the influence of socioeconomic factors and parents’ health, particularly paternal income, becomes more prominent. Nevertheless, intergenerational correlations in health rise sharply as children enter their late forties.

Do policies that restrict abortion access change contraceptive choices? We use Planned Parenthood data in a difference-in-difference design to show that both realized and expected restrictions increase take-up of more effective contraception. We then estimate a dynamic model of contraceptive choice in which women trade off efficacy against monetary costs and preferences for method attributes. Restrictive policies induce women to upweight efficacy relative to other characteristics. Eliminating abortion access and insurance coverage for contraception reduces realized lifetime utility for 57% of women with an average loss of $2,080, while providing free abortion and contraception increases utility for 85% of women.

We use novel Medicare data that link spouses to assess the value of spousal informal care. We provide new evidence on substitution between formal and informal care by examining how a spousal caregiver’s sudden incapacitation affects one’s own take-up of formal care. A spouse’s health shock causes their partner to be 30% more likely to visit any nursing home and 20% more likely to visit a skilled nursing facility; a pattern that reflects both a change in health and – primarily – a shift from informal care to formal care. We similarly find that after one spouse is incapacitated, the other spouse becomes less sensitive to the price of formal care, revealing higher valuation and less moral hazard when informal care is absent. Our findings suggest that optimal health insurance contracts should be designed at the household level with more generous coverage to beneficiaries whose spouses are incapacitated. We estimate that doing so would yield meaningful welfare improvements compared to Medicare’s current individual-level coverage structure.

Congress began taxing excessive executive compensation at tax-exempt organizations to discourage the diversion of resources from their social missions in 2017. I find that the policy instead increases executive compensation while reducing social benefit spending over time. In a dynamic agency model with moral hazard and social benefit provision, optimal contracts reward high earnings realizations by allocating a larger share of financial assets to CEO compensation, crowding out resources available for social benefit spending. Excise taxes erode financial assets. To deliver promised compensation, CEO claims must grow faster, further constraining social benefit provision. Empirically, using nonprofit hospitals as the setting, I find that affected organizations reduce charity care and community benefit expenditures while increasing CEO com-pensation following the tax introduction. Structural estimation shows that a 21% excise tax is equivalent to increasing the severity of the moral hazard friction by 2.2%, whereas governance-based interventions that directly reduce agency frictions increase organizational social value.