Applied economics research on healthcare spending, Medicaid program integrity, and evidence-based policy evaluation. We bring the same data-driven rigor to health policy that we apply across all our research areas.
Health policy decisions affect millions of lives and billions of dollars in public spending. Yet the data underlying these decisions is often more complex than it appears. Our health policy research applies applied economics methods to questions about healthcare spending, program integrity, and the gap between what data shows and what commentators claim.
Medicaid serves over 90 million Americans and accounts for roughly one-sixth of national health spending. When HHS released 227 million rows of billing data in February 2026, amateur analysts claimed to find billions in fraud within hours. Our four-part series examines what that data can and cannot actually support.
The findings challenge quick conclusions: the dataset contains only 7 columns with no diagnosis codes, 40% of provider exclusions in the federal LEIE database are not fraud-related, and billing volume more closely proxies patient need than fraudulent behavior. A simple logistic regression matched the performance of complex machine learning models when validated honestly.
Health policy analysis sits at the intersection of several challenges: large administrative datasets with limited variables, strong political incentives to find dramatic results, and real consequences when analysis goes wrong. Providers flagged as suspicious may lose their ability to serve vulnerable populations. Genuine fraud may go undetected when screening methods optimize for volume rather than behavioral anomalies.
Our approach treats health policy questions the same way we treat food security or education policy: start with the data limitations, document every assumption, and let the analysis speak for itself.
Our health policy research draws on:
Summary of our health policy research findings
The largest Medicaid dataset in HHS history contains billing amounts and provider IDs, but no diagnosis codes or service documentation.
Four in ten LEIE exclusions stem from license revocation or default on student loans, not fraudulent billing.
High billing amounts often indicate providers serving high-need populations, not fraudulent behavior.
Logistic regression matches complex ML on honest validation. Model complexity adds noise, not signal.
Our health policy research organized by focus area
Examining what public Medicaid data can and cannot tell us about program integrity and provider behavior.
Data validation, classification approaches, and methodological considerations for health policy research.
Future health policy research directions including healthcare access, cost-effectiveness analysis, and program evaluation.
HHS published the largest Medicaid dataset in history. Within hours, amateur analysts claimed billions in fraud. We walk through what the data can actually support, from the raw columns to an honest classifier.
The LEIE database is the standard source for identifying excluded providers. But exclusion reasons range from fraud convictions to unpaid student loans.
High billing volume is the most common flag in fraud screening. But providers who bill the most often serve the highest-need populations.
When we validate honestly with proper train-test splits, a simple logistic regression performs as well as gradient-boosted models.
New health policy analysis delivered directly. No spam, unsubscribe anytime.
Subscribe for Free