Do Places That Spend More on Health Care Actually Improve Health?

Finkelstein and Gentzkow (2026) asked whether places that increase Medicare spending also increase life expectancy for older adults. They analyzed Medicare claims and mortality data for millions of beneficiaries, using migration (“mover”) designs to separate the effects of places from the characteristics of the people who live there. They found that places with larger causal effects on health care spending did not produce longer life expectancy. Place-based factors explained about 50–60% of geographic variation in Medicare spending but only about 15% of geographic variation in life expectancy. A high-performing area increased life expectancy at age 65 by about 1.1 years.

Why This Article Was Selected for The Policy Scientist

Geographic variation in health care spending and outcomes remains one of the most important topics in health policy because it informs how health systems allocate resources while seeking to improve population health. The question is especially timely as aging populations place increasing pressure on Medicare and other public insurance programs. Finkelstein and Gentzkow have made sustained contributions to this literature, and this study extends the influential Dartmouth research by applying a rigorous causal inference framework. Published in the American Economic Association’s AEA Papers and Proceedings, a highly regarded journal in economics, the article draws on exceptionally large Medicare claims and mortality datasets. Its migration-based mover design, combined with two-way fixed effects, rich pre-move health controls, and adjustments for selection bias, provides substantially stronger causal evidence than conventional multivariate regression. The findings should be broadly informative for other high-income countries with geographically diverse health care systems.

Full Citation and Link to Article

Finkelstein, A., & Gentzkow, M. (2026). Reexamining geographic variation in health and health care. AEA Papers and Proceedings, 116, 138–143. https://doi.org/10.1257/pandp.20261072

Central Research Question

This article revisits one of the most influential questions in health economics: why health care spending varies so dramatically across geographic regions and whether higher spending actually improves health outcomes. Earlier research documented that Medicare spending differs substantially across regions even among patients with similar observable characteristics, yet areas with higher spending often experience no better—and sometimes worse—health outcomes. These findings have shaped decades of debate over the efficiency of the U.S. health care system.

The authors seek to answer a more precise causal question than earlier studies. Rather than asking whether high-spending regions have healthier populations, they ask whether places that causally increase Medicare spending also causally increase life expectancy. In other words, after separating the influence of a place from the characteristics of the people who live there, do regions that induce greater medical spending also improve health? This distinction is critical because observed geographic differences may reflect patient health, preferences, or socioeconomic conditions rather than the performance of local health care systems. By isolating place effects, the study evaluates whether the longstanding relationship between spending and health remains after accounting for these alternative explanations.

Previous Literature

Research on geographic variation in health care spending has been dominated for more than two decades by work originating at Dartmouth. Those studies documented remarkably large differences in Medicare expenditures across hospital referral regions while finding little evidence that patients living in higher-spending areas enjoyed superior health outcomes. This work became highly influential because it suggested that substantial health care spending might produce little additional health benefit.

Although these findings shaped policy discussions, critics argued that simple geographic comparisons could not distinguish whether differences reflected local medical practice or differences in the underlying populations. Patients in higher-spending regions may simply have been sicker, possessed different preferences regarding treatment, or faced different social conditions. If so, reducing spending in high-cost regions might produce unintended consequences.

The present study builds directly upon a series of earlier investigations by the same authors examining the effects of geographic location using migration-based research designs. Their previous work demonstrated that place-based factors explain approximately one-half of geographic variation in Medicare spending and contribute meaningfully to differences in life expectancy. However, those studies did not determine whether places that increase spending also improve health. This article addresses that remaining question by comparing estimated causal place effects on both spending and longevity.

The study therefore extends, rather than replaces, the existing literature. Instead of relying on simple cross-sectional comparisons, it employs a stronger identification strategy designed to isolate causal effects while preserving the original policy question that motivated decades of research.

Data

The analysis combines exceptionally large administrative datasets covering Medicare beneficiaries throughout the United States. Spending estimates are derived from a 20 percent random sample of traditional Medicare beneficiaries aged 65 and older between 1998 and 2008. Beneficiaries enrolled in Medicare Advantage are excluded because comparable claims data are unavailable.

Health care utilization is measured using Medicare spending adjusted for geographic differences in administratively determined prices. This adjustment ensures that observed differences primarily reflect variations in medical services delivered rather than differences in payment schedules.

The analysis is conducted at the level of 306 hospital referral regions (HRRs), which approximate regional hospital markets. These geographic units have become the standard framework for studying regional variation in health care delivery.

Life expectancy estimates are based on a separate Medicare sample extending through 2014. To preserve statistical independence between spending and mortality estimates, individuals used in the spending analysis are excluded from the mortality analysis. Mortality models incorporate detailed demographic information, including race and sex, together with indicators for approximately two dozen chronic medical conditions.

The resulting dataset encompasses millions of Medicare beneficiaries and provides one of the most comprehensive administrative data resources available for evaluating regional differences in health care spending and outcomes.

Methods

The principal contribution of the article lies in its causal inference strategy. Rather than comparing different populations living in different places, the authors exploit migration among Medicare beneficiaries. Individuals who move between geographic regions effectively serve as their own controls, allowing researchers to estimate how health care spending and mortality change after relocation.

This “mover design” isolates place effects from person effects. If spending changes immediately after a move, the difference can reasonably be attributed to characteristics of the destination rather than to fixed characteristics of the individual. Likewise, changes in mortality risk following relocation provide evidence regarding the causal influence of geographic location on health outcomes.

The analysis of spending employs two-way fixed effects models incorporating both individual and geographic fixed effects. This approach controls for stable individual characteristics while estimating the independent contribution of each hospital referral region.

The mortality analysis extends this framework by controlling for origin location, demographic characteristics, numerous chronic medical conditions, and extensive measures of pre-move health status. The authors also implement a novel adjustment designed to account for residual selection bias arising from differences in the types of individuals who choose particular destinations.

Finally, the estimated place effects are adjusted using empirical Bayes methods to reduce sampling error before examining the relationship between spending and life expectancy. Collectively, these methods represent a substantial improvement over conventional observational analyses that rely primarily on multivariate regression. Although the study is not a randomized controlled trial, the migration-based design constitutes a rigorous quasi-experimental approach that substantially strengthens causal interpretation.

Findings/Size Effects

The study confirms that large geographic differences exist in both Medicare spending and life expectancy. Simple cross-sectional comparisons continue to show that regions with higher Medicare spending generally do not experience superior health outcomes. Indeed, when spending is measured on a logarithmic scale, areas with higher expenditures often exhibit lower life expectancy.

The central contribution of the paper emerges after estimating causal place effects. The authors find that places producing higher Medicare spending do not also produce longer life expectancy. Regions that causally increase medical spending fail to demonstrate corresponding improvements in longevity.

The magnitude of earlier place-effect estimates provides important context. Previous work by the authors found that place-based factors account for approximately 50 to 60 percent of observed geographic variation in Medicare spending. By contrast, place effects explain only about 15 percent of geographic variation in life expectancy, indicating that individual characteristics remain the dominant source of mortality differences.

Nevertheless, geography does matter for health. Moving from a relatively low-performing region to a high-performing region is estimated to increase life expectancy at age 65 by approximately 1.1 years, representing roughly one-half of the observed difference between the 10th and 90th percentile places. Thus, places exert meaningful causal effects on longevity even though higher-spending places are not systematically healthier places.

These findings eliminate one important alternative explanation for earlier research. Because the analysis isolates causal place effects, the absence of a positive relationship between spending and health cannot simply be attributed to sicker patients residing in higher-spending regions. At the same time, the authors emphasize that the results do not prove that additional medical spending lacks value. Differences in health care productivity or efficiency across regions could still explain why higher-spending places fail to achieve better outcomes. Consequently, the evidence narrows the set of plausible explanations without definitively resolving the underlying policy debate.

Conclusion

This article represents an important methodological advance in the literature on geographic variation in health care. By combining large administrative datasets with a migration-based causal inference design, the authors move beyond descriptive comparisons toward estimates of the independent effects of place on both spending and health.

The results indicate that geographic regions capable of increasing Medicare spending do not necessarily increase life expectancy. This finding persists even after accounting for patient selection and numerous observable differences among beneficiaries. At the same time, the study demonstrates that geographic location itself exerts measurable causal effects on longevity, underscoring the importance of local environments and health systems.

Rather than resolving the longstanding debate over the productivity of medical spending, the article refines it. The findings eliminate one major source of potential bias while highlighting the need for future research examining differences in health care efficiency and productivity across regions. As a result, the study substantially strengthens the empirical foundation for future investigations into the relationship between health care spending and population health.