Oregon’s Medicaid Lottery: Lessons From a Landmark Study

About 35,000 people were randomly chosen to apply for the state’s expanded Medicaid program. Those who were not selected formed a comparison group with similar underlying characteristics.

The resulting evidence unsettled several easy assumptions about Medicaid expansion. Coverage improved access to care. It reduced financial strain. It lowered the probability of a positive depression screening by 9.15 percentage points, a relative reduction of roughly 30 percent. It also increased emergency department use by about 40 percent. Yet after approximately two years, researchers found no statistically significant improvement in several measured physical health indicators, including blood pressure, high cholesterol, and average glycated hemoglobin.

That combination is the point. The Oregon Health Insurance Experiment outcomes were neither a simple endorsement nor a simple indictment of Medicaid. They showed how a policy can produce substantial gains in financial security and healthcare access while failing to generate measurable short-term changes in population health metrics.

The 2008 Lottery: A Rare Natural Experiment in Public Policy

Most policy evaluations begin with a difficult problem: the people who receive a program are not randomly distributed across the population. Medicaid recipients may have worse health, lower incomes, less stable housing, or weaker access to providers than people who remain uninsured. If their outcomes later differ, it becomes difficult to determine how much of that difference was caused by Medicaid and how much was already present.

Oregon’s enrollment system created an unusual opportunity to separate those effects.

The state had more eligible people than it could immediately enroll in its expanded Medicaid program. Rather than selecting applicants through a conventional administrative ranking system, Oregon used a lottery. The lottery did not randomly assign people to receive insurance in a strict experimental sense. It randomly determined who could apply. Some selected individuals did not complete enrollment, while some people in the comparison group could obtain other forms of coverage. But the lottery still generated a strong source of variation in access to Medicaid.

Researchers could therefore compare two groups that were broadly similar at the starting point:

• People selected in the lottery, who gained a substantially higher probability of obtaining Medicaid.
• People not selected, who remained subject to the same broader economic conditions but generally lacked that new route into coverage.

In the first year, lottery winners experienced a 25 percentage point increase in the probability of being insured compared with the control group. That distinction matters. The study measured the effects of being given access to Medicaid, not the effects of forcing every eligible person into the program.

The design also placed a boundary around what the study could establish. It could identify the consequences of expanding access for a low-income, uninsured adult population in Oregon under the conditions of the late 2000s. It could not automatically answer how Medicaid would affect every age group, every state, or every long-term health outcome.

Still, the structure was powerful because it reduced a central source of bias. The research did not simply compare people who chose Medicaid with people who did not. It compared people whose opportunity to apply was shaped by random selection.

The Oregon study did not ask whether insurance and health were correlated. It asked what changed when access to insurance was altered by lottery.

This is why the Oregon Medicaid lottery study became a reference point in public policy analysis. Randomized evidence is rare in healthcare. Health systems contain too many interacting variables: provider capacity, patient behavior, local labor markets, baseline disease, transportation, prices, and administrative rules. A lottery cannot eliminate all of that complexity. It can, however, make the causal starting point far more credible.

Medicaid Changed the Financial Exposure of Being Sick

The clearest gains appeared outside the conventional clinical dashboard.

Medicaid significantly reduced financial strain. It lowered the amount of medical debt sent to collection agencies and virtually eliminated catastrophic out-of-pocket medical expenditures. For low-income households, this is not a secondary outcome. It is a direct change in economic security.

Uninsured patients face a difficult allocation problem when medical needs arrive. They may delay care, borrow money, use high-cost credit, skip rent or utilities, or accept debt that remains on their record long after the original treatment. Even when a person does not enter bankruptcy, medical bills can reduce the household’s ability to absorb other shocks.

Insurance changes that exposure. It does not make healthcare free, and it does not erase every cost associated with illness. But it can place a ceiling on the financial consequences of an acute event. That ceiling has policy value even when laboratory measurements do not move.

The Oregon Medicaid study’s financial-strain findings also challenge a narrow definition of health policy success. A program may be judged through blood pressure, cholesterol, glucose, hospital admissions, or mortality. Those indicators are essential. They are not exhaustive.

Financial instability can affect health through several channels:

• Debt can restrict spending on food, housing, transportation, and medications.
• Fear of medical bills can delay diagnosis and treatment.
• Unpaid bills can damage credit access and increase economic volatility.
• Household members may absorb unpaid caregiving or work losses when illness is unmanaged.
• Reduced financial risk can make it easier to seek care before a condition becomes acute.

The study directly measured some of these consequences rather than treating them as theoretical possibilities. Medicaid coverage reduced medical debt sent to collections and nearly eliminated catastrophic out-of-pocket spending. That is a meaningful structural effect even if the two-year clinical data did not show broad improvements in average physical health.

The result also illustrates why insurance coverage should not be evaluated as though it were a single medical intervention. Medicaid is an institutional arrangement. It changes the price of care, the set of providers a patient can approach, the probability that prescriptions are filled, and the financial penalty attached to a diagnosis. Those mechanisms operate on different timelines.

Financial relief can appear quickly. Disease progression may take years to change. The causal chain is not synchronized.

More Coverage Produced More Care, Including More Emergency Care

The experiment found that Medicaid increased healthcare utilization across multiple settings. Recipients were more likely to use outpatient services, prescription drugs, preventive care, hospitals, and emergency departments.

The emergency department result was the most politically disruptive. Medicaid increased the probability of an emergency department visit by about 7 percentage points, equivalent to roughly a 40 percent increase in total emergency department visits. The increase persisted over the two-year observation period.

That finding contradicted a common policy expectation: if insurance expands access to primary care, emergency department use should fall because patients will shift toward less expensive outpatient treatment. In Oregon, the opposite pattern appeared. Primary and preventive care increased, but emergency use increased as well.

The mistake is to treat emergency departments and primary care as perfectly substitutable channels. They are not. A person who gains insurance may use more of every available healthcare setting because the price barrier has been reduced. New coverage can uncover previously unmet demand rather than immediately reorganizing that demand into the most efficient setting.

Several mechanisms can produce this outcome:

1. Previously deferred care enters the system.

People who postponed treatment while uninsured may seek care once coverage becomes available. Some of that care will be routine. Some will arrive through emergency departments because the underlying condition has already worsened.

2. Insurance lowers the marginal cost of seeking help.

When the financial penalty for a visit falls, the threshold for seeking treatment falls as well. This can increase appropriate and inappropriate use simultaneously.

3. Primary care capacity may be limited.

Coverage expansion does not automatically create more physicians, appointments, evening clinics, transportation options, or culturally competent services. Insurance can expand demand faster than the provider network expands supply.

4. Emergency departments remain operationally accessible.

They provide care outside standard office hours and do not require the same appointment process. For people with unstable schedules or limited access to regular providers, that matters.

5. New diagnosis generates follow-up activity.

Insurance can increase the detection of conditions such as diabetes. Detection then creates medication use, monitoring, and additional encounters. Utilization rises because the system is doing more of its intended work.

The emergency department increase therefore cannot be reduced to a claim that Medicaid caused waste. Nor does it prove that emergency care became more medically necessary in every case. It demonstrates that access expansion interacts with existing system capacity in a non-linear way.

A policy that increases coverage without expanding primary care supply may produce a predictable bottleneck. More patients enter the system. The most accessible institution absorbs the overflow. The result is not a failure of coverage alone. It is a capacity mismatch.

This is one of the central public health insurance experiment lessons: utilization is not a single scale running from good to bad. More care can represent improved access, pent-up demand, inefficient delivery, or some combination of all three.

Mental Health Improved Before Physical Health Indicators Moved

The study found a statistically significant reduction in depression screening results among Medicaid recipients. The probability of a positive depression screening fell by 9.15 percentage points, a relative reduction of approximately 30 percent.

That improvement was larger and more immediate than the changes observed in physical health measures. The contrast is analytically important because it demonstrates that insurance can affect health through pathways that are not captured by short-term biomarkers.

Depression is influenced by financial pressure, uncertainty, untreated illness, and the ability to obtain care. Medicaid reduced several of those pressures at once. It lowered the risk of catastrophic medical spending, increased contact with healthcare providers, and expanded access to prescription drugs and treatment. The study does not establish that every improvement in mental health was caused by a single mechanism. It does show that the coverage intervention had a meaningful effect on measured depression.

Physical health outcomes behaved differently. After approximately two years, researchers found no statistically significant effects on hypertension, high cholesterol, or average glycated hemoglobin. These findings are often presented as a contradiction: if Medicaid increased care and improved mental health, why did blood pressure and glucose not improve?

The answer begins with measurement and time horizon.

A two-year window may be sufficient to detect changes in healthcare use, debt exposure, or screening results. It may be too short to detect changes in chronic disease at the population level, particularly when treatment is uneven, baseline disease varies, and the intervention affects only a fraction of the eligible population. The statistical signal may also be smaller for physical health measures than for financial or psychological outcomes.

The diabetes findings make this especially clear. Medicaid significantly increased diabetes detection and diabetes medication use. Yet average glycated hemoglobin did not show a statistically significant reduction.

That is not a logical failure. It is a distinction between process and outcome.

Detection is an upstream event. Medication initiation is another upstream event. Glycated hemoglobin is a downstream measure that reflects blood glucose over time and depends on diagnosis, treatment choice, adherence, follow-up, diet, comorbidities, and disease severity. A system can improve the first two stages without producing a detectable population-wide change in the final measure over a short observation period.

A useful way to read the findings is through a sequence:

• Coverage expands the ability to seek care.
• More encounters create more opportunities for screening.
• Screening identifies previously undiagnosed conditions.
• Diagnosis makes treatment possible.
• Treatment may require sustained adherence and follow-up.
• Clinical improvement may emerge only after a longer period, and not uniformly across the population.

Each stage introduces structural friction. Coverage removes one barrier. It does not remove every barrier downstream.

Insurance is an access intervention first. Its long-term health effects depend on what the healthcare system can do with the access it creates.

This is why the study should not be used to claim that Medicaid is useless for physical health. The evidence covered a limited set of indicators over roughly two years. It was not designed to measure every disease trajectory, rare outcome, long-term complication, life expectancy, or mortality effect.

Nor should the absence of statistically significant improvement be treated as proof of no effect whatsoever. In policy research, a non-significant result means that the study did not establish a sufficiently precise difference under its design and time frame. It does not transform uncertainty into a negative finding.

Coverage Expansion Did Not Automatically Change Employment

The experiment found no statistically significant impact on labor market outcomes such as employment status or total earnings.

This result also cuts against a common assumption. Public insurance may improve health, reduce financial stress, and make it easier to seek treatment. But those effects do not automatically produce a measurable increase in employment within the study period.

The labor market has its own constraints. Low-income workers may face unstable schedules, limited transportation, inadequate childcare, weak local demand, or jobs that do not offer a path to higher earnings. Health insurance can remove one constraint while leaving the others intact.

There is also no reason to assume that every effect operates in the same direction. Better access to treatment might support work for some people. Medicaid’s income eligibility structure could alter incentives for others, depending on the specific rules and the availability of jobs with employer-sponsored coverage. The Oregon evidence did not show a statistically significant aggregate change in employment or earnings, so stronger claims would exceed the findings.

This is a broader lesson about policy bundling. Health insurance is often expected to solve problems that originate in the labor market, housing system, education system, or transportation network. It can interact with those systems. It cannot substitute for them.

A person may obtain coverage and still be unable to attend appointments because of an inflexible shift. A patient may receive a diagnosis but remain unable to afford stable housing or nutritious food. A prescription may be covered while transportation to the pharmacy remains unreliable. These are not arguments against Medicaid. They are descriptions of the system around it.

The Oregon results show that coverage has real effects, but those effects are domain-specific. Financial security improved. Mental health improved. Healthcare utilization rose. Employment and earnings did not significantly change in the observed period. Physical health indicators did not significantly improve across the measures examined.

That is a more useful conclusion than a single verdict about whether the program worked.

What the Oregon Experiment Can—and Cannot—Tell Policymakers

The study’s value lies in its separation of policy mechanisms. Medicaid did not operate as a universal treatment with one expected outcome. It altered several conditions simultaneously, and the results appeared at different points in the causal chain.

The evidence supports several conclusions.

First, expanding public insurance improves financial protection. This is not an incidental benefit. Reducing medical debt and catastrophic spending changes household resilience and may prevent health shocks from becoming long-term economic shocks.

Second, coverage expands healthcare use. That expansion includes preventive services, outpatient care, prescriptions, hospitalizations, and emergency departments. The increased use of emergency care means that policymakers should not assume that insurance expansion alone will reduce high-cost settings.

Third, insurance can improve mental health even when short-term physical health indicators remain stable. Financial security and access to care are themselves health-relevant outcomes.

Fourth, early increases in diagnosis and medication use do not guarantee immediate improvement in average disease markers. Chronic disease management requires time, continuity, provider capacity, and patient stability.

Fifth, Medicaid expansion is not a stand-alone labor market intervention. It may influence work indirectly, but the Oregon findings do not show a statistically significant improvement in employment or earnings over the study period.

The study also defines the limits of confident extrapolation. The long-term physical health effects beyond the initial two-year evaluation remain uncertain. The experiment does not establish whether Medicaid expansion changes life expectancy or mortality across broader populations. It also does not resolve whether labor market effects might emerge over a longer period.

Those unknowns are not weaknesses to conceal. They are part of the evidence.

Policy debates often force complex programs into binary categories: success or failure, efficient or wasteful, transformative or pointless. The Oregon Health Insurance Experiment resists that compression. Its findings are more precise and therefore less convenient.

Medicaid coverage improved the economic position of low-income adults. It reduced depression screening results. It increased contact with the healthcare system and expanded treatment activity. It did not produce statistically significant short-term improvements in several physical health measures. It did not significantly change employment or total earnings. Emergency department use rose rather than fell.

The correct interpretation is not that coverage failed because biomarkers did not shift. Nor is it that every increase in utilization represented an efficient investment. The evidence points to a system in which insurance removes a major financial barrier but leaves provider capacity, continuity, diagnosis, adherence, and social conditions to determine what happens next.

The Real Policy Lesson Is About Sequencing

The deepest lesson from the Oregon Medicaid lottery is not a headline about Medicaid. It is a lesson about sequencing.

Coverage is an upstream policy. It changes who can enter the healthcare system and what financial risks they face. The downstream outcomes depend on the system’s ability to convert access into sustained care.

If policymakers expand eligibility while leaving primary care capacity unchanged, demand can spill into emergency departments. If coverage increases screening without ensuring follow-up, diagnosis may rise faster than disease control. If medical debt falls but housing and labor markets remain unstable, broader economic gains may remain limited. If financial stress improves, mental health may respond before chronic physical indicators do.

This is a feedback structure, not a single transaction. Insurance changes behavior, demand, provider workloads, diagnostic rates, household finances, and institutional pressure at the same time. Some effects reinforce one another. Others expose bottlenecks.

The Oregon experiment remains valuable because it makes those interactions visible. It replaces several assumptions with observed outcomes:

• More coverage did not mean less healthcare use; it meant more use across the system.
• More primary and preventive care did not displace emergency department care.
• More diagnosis and medication use did not immediately lower average glycated hemoglobin.
• Better financial protection did not automatically increase employment.
• Physical health measures did not capture the full value of insurance.

That is the analytical standard the study sets. Public policy should be judged by the mechanisms it changes, the outcomes it can plausibly affect, and the time required for those effects to appear.

The Oregon lottery did not deliver a neat conclusion. It delivered something more valuable: a map of where coverage expansion works quickly, where it encounters structural friction, and where the evidence still requires a longer view.