Do Healthier Individuals Systematically Select Into Medicare Advantage?

Bhai and Hughes (2026) ask whether individuals who enroll in Medicare Advantage at age 65 are systematically different, especially in health, from those who do not. They use administrative claims data (2007–2017) tracking commercially insured individuals transitioning into Medicare, leveraging the age-65 eligibility cutoff. They find strong advantageous selection: healthier individuals are more likely to enroll in Medicare Advantage. For example, having diabetes without complications reduces enrollment by about 1.5 percentage points (from a 9.6% baseline), and higher out-of-pocket spending and utilization similarly decrease participation, with the highest-cost individuals significantly less likely to enroll.

Why This Article Was Selected for The Policy Scientist

This article addresses a central issue in public policy: how individuals sort into publicly subsidized insurance programs and the resulting fiscal and equity implications. Selection dynamics affect not only Medicare Advantage but any mixed public–private system, making the topic broadly consequential. The study is timely given the continued growth of Medicare Advantage enrollment and ongoing concerns about program costs and risk adjustment. Bhai and Hughes have contributed repeatedly to this literature, and this paper extends prior work by leveraging rich pre-enrollment claims data. The dataset is unusually strong, though limited to one insurer, which may constrain generalizability. The quasi-experimental design improves inference.

Full Citation and Link to Article

Bhai, M., & Hughes, D. R. (2026). Estimating self-selection in Medicare Advantage. Journal of Policy Analysis and Management, 45(2), e70090. https://doi.org/10.1002/pam.70090

Central Research QuestionThis article examines whether individuals who enroll in Medicare Advantage (MA) at the point of Medicare eligibility are systematically different from those who do not, with particular attention to health status, utilization, and financial characteristics. The central question is whether advantageous selection—where relatively healthier individuals disproportionately choose MA—persists at the critical transition from commercial insurance to Medicare at age 65. By focusing on this initial enrollment margin, the study seeks to isolate baseline selection effects rather than outcomes shaped by subsequent plan exposure or switching behavior. The authors also explore whether observable pre-enrollment characteristics can predict which individuals are most likely to benefit from MA, thereby shedding light on forward-looking selection behavior.

Previous LiteratureThe study builds on a substantial body of research documenting selection in insurance markets, particularly in Medicare Advantage relative to Traditional Medicare. Earlier work has relied heavily on comparisons of enrollees after plan entry or on “mover” designs that track individuals switching between MA and Traditional Medicare. These approaches have consistently identified evidence of advantageous selection, with MA enrollees appearing healthier and less costly. However, such designs face significant identification challenges, including confounding from treatment effects (differences in care delivery across plans), nonrandom switching behavior, and upcoding practices that inflate risk scores in MA. The literature has also drawn on foundational economic models of selection, including the Rothschild-Stiglitz framework and subsequent empirical applications such as the positive-correlation test developed by Einav and Finkelstein. This article extends that literature by focusing on the pre-enrollment period, thereby addressing longstanding concerns about disentangling selection from treatment effects. It contributes to a more precise understanding of how selection arises at the point of entry rather than as an artifact of post-enrollment dynamics.

DataThe analysis uses administrative claims data from Optum’s de-identified Clinformatics Data Mart Database, covering the period from 2007 to 2017. This dataset includes over 50 million individuals and provides detailed, longitudinal information on medical claims, utilization, and plan characteristics. The authors construct a cohort of 560,037 individuals who were continuously enrolled in commercial insurance for the 12 months prior to turning 65, allowing them to observe baseline health and spending before Medicare eligibility. Approximately 9.6 percent of this cohort transitions into a Medicare Advantage plan within the dataset. The richness of the data is a central strength: it includes granular measures such as the Charlson Comorbidity Index (CCI), out-of-pocket spending, and detailed utilization across care settings. Importantly, the dataset captures individuals before they enter Medicare, which avoids distortions from upcoding and plan-induced behavioral changes. However, the data are drawn from a single insurer’s network, which may limit representativeness. Individuals who enroll in MA plans outside the dataset or who choose Traditional Medicare cannot be fully observed, introducing potential downward bias in estimated selection effects and constraining external validity.

MethodsThe empirical strategy exploits the sharp eligibility cutoff at age 65 as a quasi-experimental design. The Initial Enrollment Period for Medicare provides plausibly exogenous variation in plan choice, enabling the authors to examine selection at the moment individuals first choose between MA and Traditional Medicare. The primary approach involves estimating linear probability models of MA enrollment as a function of pre-enrollment health, utilization, and demographic characteristics, with state and year fixed effects. The analysis incorporates multiple measures of health, including the CCI, out-of-pocket costs, and service utilization, often modeled in nonlinear forms (e.g., quintiles) to capture distributional differences.

In addition, the authors implement the positive-correlation test, a standard method for detecting selection by examining whether individuals with lower expected costs are more likely to enroll in a given plan. The study also develops a conceptual framework based on a Roy model of self-selection, formalizing how individuals sort into MA versus Traditional Medicare based on health and preferences. While the quasi-experimental design improves internal validity relative to prior studies, the approach does not fully achieve causal identification in the sense of randomized controlled trials or stronger natural experiments such as instrumental variables or regression discontinuity designs with tight bandwidths. The reliance on observational regression methods, even with rich controls, leaves open the possibility of residual confounding from unobserved factors.

Findings/Size EffectsThe results provide consistent evidence of advantageous selection into Medicare Advantage. Individuals who enroll in MA are systematically healthier and lower-cost than those who do not, based on multiple pre-enrollment measures. For example, higher CCI scores—indicating worse health—are associated with a lower probability of MA enrollment. Similarly, individuals with higher out-of-pocket spending and greater utilization of medical services are significantly less likely to enter MA. These effects are nonlinear: individuals in the highest quintiles of spending and utilization are substantially less likely to enroll compared to those in lower quintiles.

At the condition level, the analysis shows that certain chronic conditions meaningfully reduce MA participation. For instance, having diabetes without complications lowers enrollment by approximately 1.5 percentage points relative to a baseline enrollment rate of 9.6 percent, representing a sizable proportional reduction. Other conditions, such as AIDS/HIV, are associated with even larger declines in participation. Conversely, individuals with low utilization and minimal health expenditures are more likely to select into MA.

The study also finds that prior insurance type is a strong predictor of MA enrollment. Individuals enrolled in managed care plans such as HMOs before age 65 are more likely to transition into MA, suggesting that familiarity with network-based care structures influences plan choice. This pattern is consistent with forward-looking selection, where individuals anticipate the relative advantages of MA based on their prior experiences and expected health needs. Overall, the magnitude and consistency of these effects across multiple measures provide robust evidence of positive selection into MA at the point of entry.

ConclusionThe article demonstrates that advantageous selection into Medicare Advantage is both substantial and observable at the initial enrollment stage, prior to any plan-induced treatment effects. By leveraging pre-enrollment claims data and the age-65 eligibility threshold, the study offers a clearer view of baseline selection dynamics than prior research relying on post-enrollment comparisons. The findings reinforce the conclusion that MA attracts relatively healthier individuals, which has implications for program costs, risk adjustment, and the broader functioning of mixed public–private insurance systems.

Methodologically, the study advances the literature by improving measurement of pre-enrollment characteristics and reducing biases associated with upcoding and switching behavior. However, the absence of stronger causal identification limits the extent to which the results can fully isolate selection mechanisms from unobserved heterogeneity. Future research using more rigorous causal designs, such as randomized interventions or stronger quasi-experimental approaches, would further strengthen the evidence base. Despite these limitations, the study provides a detailed and credible account of selection into Medicare Advantage and contributes meaningfully to the empirical understanding of insurance market behavior.in