Embedded Files
Dissertation Papers
Dissertation Papers
Dissertation Paper 1 — Job Market Paper. Under review.
The Trump administration's 2019 expansion of the public charge rule made Medicaid receipt a negative factor in certain immigration determinations and generated widespread concern that eligible families would avoid health coverage. I study whether that fear translated into measurable coverage losses among populations formally exempt from the rule. Using American Community Survey data on 1,622,582 Medicaid-eligible U.S. citizen children below 200% of the federal poverty level from 2016 to 2024, I estimate difference-in-differences and event-study models comparing citizen children in mixed-status families with those in citizen-only families. I complement the national child analysis with restricted-use California Health Interview Survey data on 220,925 adults from 2015 to 2024, which permits a four-level decomposition of citizenship status (U.S.-born, naturalized, green-card holder, no green card). In the national analysis, citizen children in mixed-status families experienced a 2.0 percentage point decline in Medicaid enrollment after the September 2018 public charge announcement (p < 0.001), with no evidence of pre-policy divergence in the preferred event study (joint Wald test p = 0.92). Effects persisted through the policy reversal period and deepened during the 2023-2024 Medicaid redetermination restart ("unwinding"). Two additional ACS specifications further refine the chilling channel: (i) decomposing the citizen-only ACS control group reveals that the chilling effect concentrates in families with a noncitizen parent (–1.81 pp Medicaid relative to U.S.-born-parent baseline, p = 0.003) rather than across immigrant-connected families broadly (naturalized-parent × post: +1.41 pp, p = 0.07); (ii) excluding seven jurisdictions that extended Medicaid (or near-equivalent state-funded coverage) to undocumented adults during the paper window (California, Illinois, Oregon, New York, Colorado, Washington, and the District of Columbia) tightens the chilling estimate to –2.6 pp (p < 0.001), consistent with state policy environments partially counter-programming the federal rule. In California, naturalized citizens experienced a 1.3 percentage point Medi-Cal decline (p = 0.030), and the restricted-CHIS data identify large fear-based avoidance gradients: undocumented adults are 20.5 percentage points more likely than naturalized citizens to report having declined a non-cash benefit because of immigration concerns (p < 10⁻³⁰). California's contemporaneous full-scope Medi-Cal expansions for undocumented adults (January 2020 ages 19–25, May 2022 ages 50+, January 2024 ages 26–49) preclude a clean within-state estimate of coverage-chilling, even in the cleanest within-state cell (age 26–49 in 2019–2021): Medi-Cal coverage among adults without green cards rose +6.8 pp rather than fell. Taken together, the clearest coverage losses appear among eligible citizen children — concentrated in mixed-status families and in non-expansion states — while the California adult evidence provides supporting evidence on spillovers, fear-based avoidance, and administrative vulnerability rather than a coverage-chilling estimate for directly exposed adults.
Dissertation Paper 2. Under review.
When individuals are incarcerated, states may either suspend or terminate their Medicaid eligibility. Under suspension, coverage is paused during incarceration and reactivated upon release without a new application; under termination, eligibility is ended entirely and the individual must reapply after release, navigating documentation requirements, processing delays, and bureaucratic complexity at a moment of acute vulnerability. I estimate the intent-to-treat association between state Medicaid suspension policies and post-release Medicaid enrollment, employment, earnings, and mortality among state prison release cohorts, exploiting the staggered rollout of suspension across U.S. states between 2005 and 2022. Linking a newly constructed state-year policy panel triangulated from six independent sources to the Census Bureau's Criminal Justice Administrative Records System (CJARS) Justice Outcomes Explorer---which provides administratively linked aggregate outcomes for state prison release cohorts across 31 states---I implement two-way fixed effects, heterogeneity-robust local projections difference-in-differences (LP-DiD), and Callaway-Sant'Anna group-time average treatment effect estimators. Because the outcome is measured at the release-cohort level, the estimates should be interpreted as intent-to-treat effects of state suspension policy on the full release cohort, not as individual-level treatment effects among those whose Medicaid was actually suspended. Suspension is associated with a 22.4 percentage-point increase in 1-year post-release Medicaid enrollment (LP-DiD, p = 0.001), with similar estimates from TWFE (22.2 pp) and Callaway-Sant'Anna (25.8 pp), and similarly large estimates at 3-year (18.7 pp) and 5-year (20.7 pp) horizons. I find no significant effects on W-2 employment or earnings. Mortality estimates are suggestive but not robust across estimators and should be treated as exploratory. The enrollment finding survives leave-one-out checks (including dropping each never-treated state and dropping South Carolina), policy-coding sensitivity, weighted estimation, and a pre-ACA-only restriction; design-based randomization inference is more conservative and does not reach conventional thresholds. These results provide multi-state, state-level, longer-horizon evidence on Medicaid suspension complementing prior single-state individual-level evidence from South Carolina (Packham and Slusky, 2024) and Wisconsin pre-release enrollment-assistance evidence (Burns et al., 2022), and they offer an implementation benchmark for the federal prohibition on terminating Medicaid eligibility solely because of inmate status that took effect on January 1, 2026 under Section 205 of the Consolidated Appropriations Act, 2024.
Dissertation Paper 3. Under review.
Thirty-three states participate in interstate Medicaid drug purchasing pools, yet there is almost no causal evidence on whether pooling actually raises the supplemental rebates states extract from manufacturers. I assemble a primary-source panel of pool membership for all 50 states from 2002 through 2024 — reconciled from state Medicaid manuals, archived web materials, agency correspondence, and public-records requests — and link it to CMS-64 supplemental rebate data and State Drug Utilization Data drug-mix measures. Using estimators that are robust to staggered timing and to pool exit (Sun-Abraham for the absorbing Sovereign States Drug Consortium, or SSDC; de Chaisemartin-D’Haultfœuille for the reversing vendor pools), with wild-cluster-bootstrap inference appropriate to the small number of treated clusters, I find that joining a pool modestly raises supplemental rebate capture. Among SSDC later adopters (first treated in 2012 or after), the average post-adoption effect on the supplemental rebate share of drug spending is 3.4 percentage points (wild-cluster p = 0.051; it sharpens to 0.03–0.04 under valid outcome-construction covariates), and the estimate is stable across five alternative estimators and leave-one-state-out. The effect is heterogeneous by pool type: the National Medicaid Pooling Initiative (NMPI) and the state-owned SSDC are positive (NMPI 1.9 pp; any-pool 1.2 pp), while The Optimal PDL Solution (TOP$) estimates are centered near zero and rule out effects as large as those estimated for later SSDC adopters, but modest positive or negative effects remain possible. Smaller states gain about twice as much as larger states, consistent with a countervailing-buyer-power mechanism, though the size and switcher cuts are not statistically distinct under small-cluster inference. A coarse state-year test of the class-substitutability prediction
is non-confirmatory. The binding constraint on precision is structural: by 2024 most states had pooled, leaving only thirteen never-pooled comparison states, so several confidence intervals hug zero and I bound this imprecision explicitly rather than overclaim. The headline is a measured, mechanism-consistent positive effect, not a clean win — and the governance-controlled specifications are shown to over-condition on post-treatment mediators and to be identified on a restricted post-2010 sample, so the uncontrolled total effect is the primary specification. This study contributes a primary-source multi-pool membership panel as a public good and the first quasi-experimental multi-pool estimates of Medicaid supplemental rebate capture. The paper studies rebate capture, not net drug spending; the language of “savings” is reserved for explicit comparisons against net spending. However, a back-of-the-envelope calculation where all non-SSDC states join SSDC and obtain their rebate levels indicates up to ~2 billion dollars in savings per year.
Other Manuscripts in Review
Other Manuscripts in Review
I evaluate whether removing Medicaid prior authorization for buprenorphine changes prescribing for opioid use disorder. I assemble a balanced 2011-2024 state-quarter panel for all 50 states, linking Medicaid State Drug Utilization Data, CDC WONDER mortality files, MBES enrollment, and verified state policy dates. Twelve states removed prior authorization between 2015 and 2022. Using Gardner's two-step difference-in-differences estimator, I find that removal increased OUD-indicated buprenorphine prescribing by 23,180 prescriptions per state-quarter, about 92 percent of treated states' pre-removal mean, with similar estimates from local-projections DiD and conventional two-way fixed effects. Per-enrollee prescribing effects are positive but imprecise, while state-year opioid mortality estimates are null and not directionally robust. I conclude that prior authorization remains a binding barrier to buprenorphine access, but that prescribing gains should not be read as evidence of downstream mortality effects without patient-level evidence on treatment retention and overdose outcomes.
I provide a first multi-state causal evaluation of Electronic Visit Verification, the Medicaid home-care monitoring mandate introduced by the 21st Century Cures Act. Using staggered implementation timing across 51 jurisdictions, I estimate effects on personal care services and home health care spending, utilization, provider participation, and service patterns in T-MSIS claims and CMS Financial Management Report data. EVV compliance has no statistically significant effect on quarterly PCS claims spending, total HCBS claims spending, prices per claim, claims per provider, provider counts, or beneficiary counts. FMR estimates corroborate the null. The design is underpowered to rule out the smaller savings projected by CBO, and home-health estimates are sensitive to measurement conventions and fail pre-trend checks. I interpret the strongest evidence as a bounded null on large aggregate spending reductions and large provider-exit effects. EVV may still affect documentation, compliance, or smaller utilization margins, but the public aggregate data do not support large fiscal effects.
I use California's January 2024 full-scope Medi-Cal expansion for adults ages 26-49 regardless of immigration status as a public-data screen for state-funded immigrant coverage policy. I build a 2018-2024 IPUMS ACS state-year panel for low-income noncitizens and estimate ridge-augmented synthetic control and synthetic difference-in-differences against a donor pool excluding states with comparable adult immigrant-coverage policies. Across two co-primary income bands, 1-200 percent FPL and the Medicaid-relevant 1-138 percent FPL, California's measured uninsurance is 7.0-8.6 percentage points lower in 2024 than its synthetic counterfactual. Estimated Medi-Cal gains are positive but attenuate in the narrower income band. Placebo-in-space and HonestDiD diagnostics are weak, with Fisher-exact rank statistics that do not distinguish California from placebo states and sensitivity bounds crossing zero for the 1-138 percent uninsurance estimate. I present the evidence as a cautious directional screen, not a definitive causal estimate.
Working Papers
Working Papers
I study how local Social Security Administration field-office closures interact with the federalist SSI-Medicaid linkage. In Section 1634 states, SSI approval automatically confers Medicaid; in Section 209(b) states, Medicaid eligibility runs through a separate state process. Using 2005-2023 ACS and BRFSS data in a Callaway-Sant'Anna staggered difference-in-differences design, I compare disabled-adult Medicaid coverage around office closures across these two institutional regimes. The pooled effect is close to zero, but it decomposes sharply by linkage type: coverage falls in 1634 states and rises in 209(b) states, producing an estimated 1634-versus-209(b) contrast of roughly four percentage points. Small-cluster procedures agree on sign and magnitude, though inference is close to conventional thresholds. I interpret the finding as evidence that field-office access matters differently when federal SSI administration is the gateway to Medicaid, and that reductions in SSA infrastructure may have larger coverage consequences in automatic-linkage states.
I evaluate the long-run consequences of capped federal Medicaid financing in Puerto Rico and the other U.S. territories. I assemble a 1950-2024 panel of territorial and state Medicaid financing, coverage, and population-health outcomes, combining CDC WONDER, NCHS mortality files, vital-statistics volumes, CMS financing data, and newly extracted Puerto Rico registry materials. Synthetic-control analyses do not detect a differential 1968 cap effect on Puerto Rico's infant-mortality decline, a null that I interpret as a level-mismatch and power problem rather than proof that capped financing was harmless. The 2011 ACA funding bump produces no detectable coverage effect, while the FY2020 funding cliff is confounded by Hurricane Maria, COVID-19, and bridge appropriations. The clearest contribution is prospective: applying cap regimes to states would mechanically remove very large federal Medicaid transfers, and literature-based mortality elasticities imply policy-relevant infant and adult mortality consequences. I frame territorial caps as a structural fiscal-architecture problem with limited historical causal leverage but large contemporary stakes.
I revisit the contested evidence on Title VI hospital desegregation and Black postneonatal mortality in the American South. Extending the state panel back to 1959 with NCHS microdata and audited vital-statistics records, I re-estimate the desegregation effect using five modern staggered-adoption estimators. On the extended eleven-state Confederate-South panel, every estimator rejects zero, with mortality reductions ranging from roughly 1.9 to 5.1 Black postneonatal deaths per 1,000 live births depending on the estimator. I decompose the earlier disagreement between Almond, Chay, and Greenstone and Anderson, Charles, and Rees into two issues: cohort anchoring, because the dominant 1967 cohort has no pre-period in post-1968 public data, and forbidden comparisons in conventional two-way fixed effects. The result is a methodological adjudication as much as a substantive historical estimate: modern estimators and a longer pre-period reconcile much of the dispute.
I estimate the coverage effects of state-funded Medicaid-equivalent expansions for undocumented immigrants and other income-eligible noncitizens excluded from federal Medicaid. Using ACS data from 2018-2024, I combine a within-California age-based difference-in-differences design with a national triple-difference design exploiting citizenship status, state policy timing, and eligibility phase-in rules. The California design, which compares age groups exposed at different times to Medi-Cal expansion, yields consistent coverage gains of about 1.3-1.7 percentage points. Broader national specifications produce larger gains, roughly 4.5-9.5 percentage points, but placebo tests reveal residual identification concerns: fake expansion dates and U.S.-born citizen placebo groups also move. I therefore treat California's phased rollout as the clearest evidence and the national design as supporting context. The results point to modest but real coverage gains, while recent program terminations and enrollment freezes highlight how fragile those gains may be.
I evaluate whether state Medicaid doula-coverage policies improved population-level birth outcomes or narrowed racial disparities during the recent wave of state adoption. I construct a 2019-2024 state-year panel from NCHS Vital Statistics Rapid Release tables and CDC WONDER race-specific birth tabulations, coding staggered Medicaid doula coverage dates across states. Callaway-Sant'Anna estimates show no clear beneficial pooled effect on the main birth outcomes. Cesarean estimates are unexpectedly positive but not robust to denominator choices, while preterm and low-birthweight effects are statistically indistinguishable from zero. After replacing constructed race denominators with direct WONDER all-births denominators, previously apparent improvements for non-Hispanic Black outcomes no longer reach conventional significance; the strongest race-stratified signals instead appear among non-Hispanic AIAN births and are mixed in direction. I conclude that state-year aggregate data do not yet show detectable population-level benefits, and that individual-level birth records are likely necessary to evaluate service uptake and disparity channels.
I study whether ACA Medicaid expansion improved post-release outcomes for people leaving state prison. Using the Census Bureau's Justice Outcomes Explorer, which links criminal-justice records to federal administrative data on mortality, Medicaid enrollment, and employment, I estimate release-year exposure effects in a staggered difference-in-differences design. The clearest result is a one-year Medicaid enrollment first stage of about 15.9 percentage points under local-projections DiD. Longer-horizon coverage estimates are sensitive to estimator choice and pretrend specification. Mortality estimates are statistically indistinguishable from zero at the one-year horizon across estimators, and pretrend tests fail at all horizons, so I interpret the mortality evidence as a bounded null rather than a clean causal estimate. Employment estimates are also fragile and reported cautiously. My main contribution is to show that expansion increased coverage for release cohorts, while the available public aggregate data cannot yet identify mortality effects with confidence.
I examine whether states' 1974 decisions to retain stricter Medicaid eligibility rules for SSI recipients under the Section 209(b) option still shape coverage, SSI participation, and mortality. I assemble a 1968-2020 county-year border-pair panel, CPS coverage measures, and a newly constructed SSA county-year SSI uptake panel. Across border counties, 209(b)-side counties show lower disabled-adult Medicaid coverage and lower SSI participation in both survey and administrative data. Mortality estimates point in the same direction over the long run, but the static border-pair estimate is fragile to clustering and weighting choices; the event study suggests a shift from short-run negative coefficients to long-run positive divergence. I therefore frame the evidence as descriptive triangulation rather than a sharp causal mortality estimate. I show that a half-century-old SSI-Medicaid linkage choice remains visible in multiple data systems, while stronger identification is needed to quantify causal health effects.
I estimate the association between Federally Qualified Health Center capacity and county under-65 uninsurance during the post-2016 funding plateau. I build a 2017-2022 county panel from HRSA Uniform Data System records, Area Health Resource File capacity measures, HRSA Section 330 grant revenue, and Census SAHIE uninsurance rates. A corrected ZIP-to-county allocation retains substantially more UDS sites than the prior implementation, and inference is verified using both analytic state-clustered IV2SLS and a pairs cluster bootstrap. The composite shift-share instrument, combining HPSA, MUA, and baseline poverty exposure, implies that one additional FQHC site per 10,000 residents is associated with a 3.89 percentage-point lower under-65 uninsurance rate, though the bootstrap p-value is borderline. A narrower shortage-area instrument is smaller and imprecise. I treat the finding as an inference-sensitive association, not a causal effect, and emphasize proper clustering, allocation, and cross-implementation verification.
I study early labor-supply responses to state Section 1115 waivers extending children's Medicaid continuous eligibility from the federal 12-month standard to multi-year protection through early childhood. Using the staggered adoption of waivers in Oregon, Washington, and New Mexico, I combine cross-state difference-in-differences, a within-state age-based triple difference, and a SIPP sibling fixed-effects specification. The main estimates suggest that parents of children ages 0-6 in waiver states work about 0.7 more hours per week and are 1.7 percentage points more likely to work full time after adoption, with little movement on labor-force participation. Effects survive controls for unwinding intensity and heterogeneity-robust event-study specifications. Coverage effects for children are smaller and do not survive all controls, while sibling fixed-effects estimates are positive but imprecise. I interpret the pattern as an intensive-margin Medicaid-lock response: parents already attached to work appear more willing to increase hours when children's coverage is protected from income fluctuations.
I test whether Medicaid's exclusion of federal matching funds for adult inpatient psychiatric care in institutions for mental diseases contributed to the mid-century decline of state mental hospitals. I digitize and validate the 1947-1968 NIMH state-and-county mental-hospital census, producing an annual state-year resident-patient panel that was previously unavailable in machine-readable form. Combining these data with Medicaid adoption dates, mortality files, prison data, and SSI controls, I estimate staggered-adoption event studies on log resident-patients per 100,000. Cohort-supported pre-period leads are flat, while the first post-adoption coefficient implies a roughly 22 percent decline in state mental-hospital residents. The result is robust to OCR uncertainty and estimator choice. Downstream deaths-of-despair analyses are less clean because under-22 placebo outcomes are not flat, so I lead with the institutional first stage and report mortality channels cautiously. The contribution is an IMD-specific test of Medicaid adoption using newly digitized annual data.
I estimate how optional state Medicaid coverage of GLP-1 receptor agonists for obesity affects prescription utilization. I link Medicaid State Drug Utilization Data from 2018-2025 to a primary-source coverage adoption panel and compare obesity-indication prescriptions with diabetes-indication prescriptions, which are mandatorily covered. Staggered difference-in-differences and within-state triple-difference models show that obesity coverage is associated with an increase of about 2.4 log points in obesity-indication prescriptions per 1,000 Medicaid enrollees, roughly a tenfold increase from a near-zero pre-Wegovy baseline. The diabetes-indication placebo is null, and the result survives restriction to states with exact primary-source adoption dates. Because several states rolled back coverage in early 2026, I treat the 2022-2025 expansion as a potentially reversible access shock. The findings show that state coverage choices, rather than demand alone, strongly ration Medicaid access to obesity-indication GLP-1 medications.
I argue that the U.S. evidence on Medicaid work requirements is better understood through two separate state case studies than through a pooled treatment estimate. Arkansas conditioned continued coverage on work reporting for an already-enrolled expansion population, while Georgia made work documentation a condition for entry into a new partial expansion. Using ACS, CPS, and BRFSS state-year aggregates, I compare policy-eligible adults with eligibility-defined control groups in each state and report conventional, wild-cluster bootstrap, Fisher randomization, and Romano-Wolf adjusted inference. Arkansas shows directionally consistent coverage losses in the affected population, but state-year aggregates lose statistical significance under small-cluster inference. Georgia shows essentially flat Medicaid, employment, and uninsurance outcomes, consistent with very low Pathways to Coverage enrollment. I conclude that Arkansas speaks to disenrollment friction in an existing benefit, Georgia to low take-up of conditional new coverage, and neither should be pooled into a single national work-requirement effect.
I study who is actually constrained by state nursing-home staffing floors and whether responses vary with Medicaid dependence and rural labor-market conditions. I assemble a 2017-2025 facility-quarter panel from CMS Payroll-Based Journal data, statutory staffing-floor timing, LTCFocus Medicaid-day shares, and Care Compare rurality. A simple binary average effect is contaminated by treated-state secular trends, but within-state-quarter heterogeneity specifications show the expected pattern. Facilities with the largest pre-policy staffing shortfalls respond more strongly, and the bite gradient steepens among high-Medicaid-share facilities. The bite-by-Medicaid-share-by-post interaction is positive and statistically strong; rural facilities show an even steeper gradient than urban facilities. Resident-mix controls do not explain the pattern. I conclude that staffing floors bind most where financing and labor-market constraints are most severe, implying that future federal or state staffing mandates should pair compliance rules with targeted technical assistance and rate-cell adjustments for Medicaid-heavy and rural facilities.
I evaluate whether CMS-concurred procedural-disenrollment pauses during the Medicaid unwinding were associated with better renewal outcomes. I link CMS unwinding renewal metrics, enrollment performance indicators, the September 2023 CMS compliance assessment, and CMS documentation of state-elected pause timing into a 51-jurisdiction state-month panel from February 2023 through July 2024. Difference-in-differences estimates compare 15 pause states with never-paused states. Pause states completed renewals at higher rates while pauses were active: the two-way fixed-effects estimate is about 8 percentage points and the DID2S estimate is larger and more precise. Procedural termination and total disenrollment effects are directionally favorable but underpowered. Intent-to-treat estimates based on the compliance assessment are not statistically distinguishable from zero, and formal pretrend tests reject for renewal completion. I therefore interpret pauses as evidence that additional processing time may matter during coverage transitions, while emphasizing that state selection and modest power limit causal claims.
I evaluate whether Medicaid state-directed payments, now exceeding $110 billion in annual projected spending, improve hospital operating margins. I construct a longitudinal panel of CMS-public SDP preprints for rating periods beginning in 2023 or later and link first-approval cohorts to hospital-year HCRIS outcomes. A planned event study of the 2025 statutory cap has no usable identifying variation in the public corpus because most in-scope arrangements are likely grandfathered and parseable cap doses are zero. The primary design is therefore a staggered difference-in-differences analysis of first SDP approval, with multiple estimator and sensitivity checks. The pooled Callaway-Sant'Anna effect on hospital operating margin is essentially zero and robust to net-of-tax weighting, HonestDiD bounds, cluster bootstrap, and random-dose permutation. Placebo outcomes indicate broader post-period volume changes, not a Medicaid-specific margin channel. I establish the first public longitudinal SDP dataset and find no detectable pooled operating-margin effect before the new federal caps bind.
I ask whether state bans on pharmacy benefit manager spread pricing reduced Medicaid managed-care generic drug spending between 2018 and 2024. I link CMS State Drug Utilization Data, National Average Drug Acquisition Cost files, Medicaid enrollment data, and a new state-year policy panel coding spread-pricing prohibitions and related cost-containment actions. Staggered difference-in-differences estimates for log MCO generic spending per enrollee are small, negative, and imprecise, and a generic-versus-brand triple difference does not isolate a clear mechanism. A pre-registered negative control fails because fee-for-service generic spending declines more than managed-care spending, suggesting broader bundled cost containment rather than a spread-pricing-specific effect. Treated states also adopt roughly half an additional cost-containment action per state-year. HonestDiD bounds contain zero throughout the post-period. I conclude that public SDUD/NADAC data cannot identify the spread-pricing mechanism in isolation; decisive evidence requires claim-level Medicaid data.
I develop a public-data linkage playbook for Medicaid program-integrity triage. Public records cannot establish fraud, but they can help oversight teams convert broad provider universes into more targeted queues before subpoena, claims review, chart audit, beneficiary interview, or bank-record analysis. The framework distinguishes public setup traces, including enrollment records, addresses, officers, sanctions, litigation, ownership, public-company disclosures, and site records, from restricted records needed to validate claim execution. A New York Medicaid demonstration across six provider categories shows how public data can produce related-entity maps, signal domains, confidence-tiered linkages, and targeted restricted-data requests. In a 928-NPI home-care subset, the workflow reduces a 441-provider public baseline queue representing $49.4 billion in payment exposure to a 66-provider public-linkage priority queue representing $8.5 billion. These figures are not fraud estimates or overpayment findings. I present the workflow as an operating model for pre-investigation prioritization.
I use the first post-settlement Express Scripts insulin formularies to evaluate public observability of a behavioral antitrust remedy in pharmacy benefit management. The February 2026 FTC consent order prohibits Express Scripts from placing a low-list-price authorized-generic insulin on a less favorable tier than the same-manufacturer high-list-price version, while leaving CVS Caremark and OptumRx outside the order. I construct a public list of qualifying within-manufacturer insulin pairs and audit the April 2026 Express Scripts Standard Offering formularies. Express Scripts achieves tier parity for the Lilly Humalog and Viatris Semglee pairs, excludes the Novo Nordisk NovoLog franchise in a way that makes the rule moot, and appears to place the Sanofi Lantus/unbranded insulin glargine pair in a violation-direction configuration. The result is not a claims-based access study. I show that formulary-design remedies are publicly auditable, partially effective, and suitable for ongoing Section V monitoring.
I estimate early effects of conversion to the Rural Emergency Hospital provider type, which trades inpatient capacity for outpatient payment and a monthly facility supplement. I link public CMS data from HCRIS, Provider of Services, and Hospital General Information files to a hand-built Medicare CCN crosswalk for 38 dated conversions in 2023-2024. A Callaway-Sant'Anna staggered-adoption design compares converters with propensity-trimmed non-converting Critical Access Hospitals. Inpatient days fall by about 1,485 per hospital-year, confirming the statutory no-inpatient mechanism and passing pretrend and HonestDiD checks. The year-1 financial-proxy estimate, based on winsorized net income over operating revenue, is positive but statistically indistinguishable from zero. Operating-margin estimates are not treated as causal because they appear contaminated by HCRIS Worksheet G-3 reclassification. I conclude that the inpatient-elimination side of the REH trade is clear, while the financial stabilization side remains inconclusive in public data.
I revisit the 2006 Deficit Reduction Act citizenship-documentation rule, which ended self-attestation for Medicaid enrollment while leaving citizen eligibility rules formally unchanged. Using state-year interrupted time series models for 2002-2009 and ACS-derived citizen-population weights, I estimate post-rule changes in past-year Medicaid coverage among modal-citizen-proxy adults and children. Adult Medicaid coverage falls by about 0.9 percentage points after July 2006, while child coverage falls by about 1.7 points. ACS reweighting strengthens prior estimates slightly, consistent with residual noncitizen contamination in the race-as-citizenship proxy used in earlier work. The 2007 final rule's electronic-match safe harbor produces no detectable incremental shift, supporting the interpretation that the 2006 documentation burden was the binding friction. I conclude that documentary proof requirements reduced Medicaid coverage among likely citizen populations, and that the prior race-proxy approach is highly but not perfectly accurate against the ACS benchmark.
I evaluate whether the 2020-2023 wave of state Medicaid adult dental benefit expansions changed past-year dental visits, cost-related care barriers, or self-rated health at the state adult population level. I link BRFSS state-year aggregates with a state dental-benefit scope panel constructed from CareQuest, KFF, and state Medicaid materials. Because BRFSS fields the dental-visit outcome only in even years, the primary panel has 153 state-year observations. Staggered difference-in-differences estimates are positive but imprecise: the main dental-visit coefficient is about 1 percentage point and does not survive Romano-Wolf multiple-testing correction. A supplementary NHIS trend analysis among Medicaid-enrolled adults does not reveal a hidden Medicaid-specific increase. I conclude that state-population data do not detect dental-visit gains from recent benefit expansions, though the design is underpowered for the diluted effect sizes expected when Medicaid-enrolled adults are observed inside the full adult population.
I estimate coverage changes after Oregon extended full Oregon Health Plan benefits to all income-eligible residents regardless of immigration status. Because earlier policy already covered children, young adults, and adults 55 and older, the newly exposed adult target group is low-income noncitizens ages 26-54. Using a 2018-2024 ACS state-year panel and ridge-augmented synthetic control, I find that Oregon's 2024 uninsurance rate for this group was 21.0 percent, compared with a synthetic counterfactual of 46.4 percent. Medicaid or means-tested public coverage rose by about 19 percentage points. Synthetic difference-in-differences corroborates the uninsurance decline but yields a less precise Medicaid gain. The 2024 post year also includes Oregon's Basic Health Program launch, and the design has only one post-treatment ACS year. I therefore interpret the estimates as strong directional public-data evidence, not a mature causal estimate of Healthier Oregon alone.
I re-examine the late-1980s OBRA Medicaid pregnancy expansions using the simulated-eligibility design associated with Currie and Gruber and public state-aggregate vital-statistics data. I construct a 1980-1995 panel for 50 jurisdictions, excluding Arizona to match the original study, with birth and infant-death outcomes from NCHS Natality and Linked Birth/Infant Death files. The treatment variable is the state-specific simulated change in Medicaid eligibility for women ages 15-44 between 1986 and 1990. The canonical simulated-bite specification yields a right-signed infant-mortality coefficient of -1.30 deaths per 1,000 live births, but the confidence interval is wide and crosses zero. Race-stratified mortality, Black-White mortality gaps, and low-birthweight outcomes are also statistically null. I conclude that public aggregate data cannot precisely replicate the original infant-mortality benefit; power and cluster count, rather than the core design, are the binding constraints.
I study whether the 2023-2024 Medicaid unwinding translated into financial stress for Federally Qualified Health Centers. I originally test whether pre-pandemic state renewal automation could instrument for procedural disenrollment, using a 2018 KFF/Georgetown survey of automated renewal capacity and HRSA Uniform Data System outcomes for 2019-2024. The candidate instrument proves empirically weak: first-stage F-statistics are well below usable thresholds, and bin means show little relationship between 2018 automation capacity and unwinding-era procedural disenrollment. A planned T-MSIS reduced-form check also contradicts the proposed channel, with higher automation tiers predicting lower FQHC claims volume and spending share. I therefore reframe the project as a descriptive analysis of the association between unwinding patterns, FQHC payer mix, and operating outcomes, with the failed first stage and mechanism contradiction reported transparently. The contribution is as much methodological as substantive: plausible administrative instruments can fail when crisis-period implementation overwhelms pre-existing infrastructure differences.
I revisit regression-kink designs built around the 50 percent federal Medicaid matching-rate floor. Modern panels include the ACA Medicaid expansion, ARRA, and FFCRA enhanced-match episodes, all of which create additional federal-generosity margins correlated with state characteristics. Rebuilding the FMAP-kink design on a 2008-2023 public state-year panel from MACPAC, HHS Federal Register, BEA, and CMS/Mathematica LTSS sources, I estimate local-polynomial kink models with exact two-way fixed-effect residualization and state-clustered inference. The FMAP mechanical slope change is recovered, confirming that the public-data pipeline is parsed correctly, but substantive spending and HCBS-share estimates are imprecise and always include zero. Dropping ACA-expansion state-years does not recover a stable effect. An earlier negative-and-significant result is withdrawn as an implementation artifact from sequential one-pass demeaning on an unbalanced sample. I recommend exact fixed-effect residualization and caution in interpreting public-data FMAP-kink estimates for 2008-2023.
I test whether states with more intense procedural disenrollment during the 2023-2024 Medicaid unwinding experienced detectable declines in adult coverage or self-reported health. I combine CMS state-month unwinding data with BRFSS state-year outcomes for 2019-2023 and ACS microdata for 2019-2024. Continuous-intensity difference-in-differences estimates show no worsening in adult self-rated health, any coverage, or cost barriers in BRFSS; if anything, the fair-or-poor-health estimate moves slightly in the opposite direction but does not survive Romano-Wolf correction. ACS estimates likewise show null uninsurance effects among all adults, low-income adults, and low-income adults in expansion states, with HonestDiD intervals containing zero. I interpret the public-data evidence as an honest null, consistent with either selective procedural removal of lower-baseline-health enrollees or survey-window limitations. I argue that restricted NHIS and T-MSIS Analytic Files are needed to test individual-level health consequences.
I build the analytic infrastructure for evaluating Section 1115 Medicaid reentry demonstrations, which allow participating states to provide Medicaid-covered services before release from incarceration. I construct a 2018Q1-2025Q4 state-quarter panel for 50 states and DC, coding waiver approval and implementation timing and linking those measures to overdose mortality and Medicaid enrollment outcomes. The primary mortality question cannot yet be answered because, by the end of the observed period, only California had implemented its waiver and post-treatment mortality data were not available. Enrollment specifications suggest a positive response after implementation, but the estimate rests on a single treated state and a short post-period. Descriptively, approved waiver states have higher baseline overdose burden, larger Medicaid programs, and are disproportionately expansion states. I therefore present this as a launch-ready evaluation framework and early mechanism screen, not as causal evidence that prerelease coverage reduces overdose mortality.
I map California counties' continuing responsibility for indigent care under Welfare and Institutions Code Section 17000 as Medi-Cal coverage becomes less certain. After the ACA, Medi-Cal expansion, and AB 85, county backstops receded from view, but state and federal policy changes scheduled for 2026 and 2027 could increase county exposure to adults losing coverage, including some noncitizens. Using legal, budget, and county program documents, I classify the current county landscape. Thirty-five counties participate in the shared County Medical Services Program. Among the 23 self-administered counties, 19 maintain visible county-branded indigent-care or uninsured-adult access programs, while four rely on county hospitals, behavioral-health systems, FQHCs, and payment-assistance structures rather than a named program. I argue that Section 17000 was submerged, not superseded. As Medi-Cal retrenchment unfolds, county institutional capacity and program design will shape local equity, affordability, and fiscal strain.
I examine whether 12-month postpartum Medicaid extensions changed outpatient contraception and broader chronic-care utilization. I build a balanced 51-jurisdiction state-quarter panel from Medicaid State Drug Utilization Data, public Medicaid provider-spending files, and MBES enrollment data, then estimate staggered difference-in-differences and triple-difference models. Standard estimates show no detectable increase in outpatient LARC use or LARC removals, and negative-control drug categories reveal that the Medicaid unwinding confounds raw post-2022 treated-control comparisons. After relative-scale triple-difference adjustments, LARC estimates remain small and imprecise. In contrast, several non-LARC outcomes move in a coherent positive direction: postpartum blood-pressure monitoring, established primary-care visits, insulin, naloxone, antipsychotics, mood stabilizers, UTI antibiotics, HIV ART, and HCV antivirals. Multiple-testing correction prevents treating any single drug category as definitive, but the consistent positive direction suggests broader chronic-care continuity effects. I conclude that public aggregate Medicaid data detect chronic-care signals better than postpartum-specific contraception effects.
I use IPUMS USA ACS microdata from 2014-2024 to ask whether the Consolidated Appropriations Act, 2021 restoration of Medicaid eligibility for migrants from the Marshall Islands, Federated States of Micronesia, and Palau is visible in public coverage data. The treatment group is low-income COFA noncitizen migrants identified by birthplace; the comparison group is low-income non-COFA Asian/Pacific Islander noncitizens. A national low-income difference-in-differences model shows an 8.8 percentage-point increase in Medicaid or means-tested public coverage after 2021, while uninsurance and any-coverage estimates move in the expected direction but are not precise. A stricter triple-difference specification is smaller and statistically indistinguishable from zero. Arkansas is substantively important but underpowered in ACS: year-specific COFA cells are small and synthetic-control diagnostics have poor pre-treatment fit. I present the results as a short public-data screen: ACS suggests national Medicaid gains after COFA restoration, but Arkansas-specific causal evaluation requires administrative enrollment data or pooled survey designs.
I estimate whether ACA Medicaid expansion changed health insurance coverage differently across disaggregated Asian American, Native Hawaiian and Pacific Islander, West Asian/MENA proxy, and American Indian and Alaska Native adults. Using 2008-2019 IPUMS ACS microdata aggregated to state-year-subgroup cells, I estimate pairwise triple-difference models comparing each target subgroup with non-Hispanic White adults while absorbing state-year, state-subgroup, and year-subgroup fixed effects. AIAN adults show the clearest expansion-associated coverage gains: uninsurance falls 2.6 percentage points more and Medicaid coverage rises 2.9 points more than among non-Hispanic White adults. West Asian/MENA proxy adults show a 3.5-point larger Medicaid gain, and NHPI adults show a 3.7-point larger uninsurance reduction. Detailed Asian subgroup estimates are mixed, including offsetting Medicaid and uninsurance patterns among Vietnamese adults. I argue that pooled AANHPI categories can hide both coverage gains and adverse subgroup-specific patterns.
I build a public-source screened census using SEC Form D of Medicaid-oriented digital health and health-technology companies from 2011 to 2025. The current screen identifies a 95-company discovery universe, a conservative 9-company analytic core, and an 18-company core-plus-sensitivity frame. Venture capital has increased in Medicaid significantly, barring a dip during the COVID-19 pandemic. The core sample clusters around behavioral health and substance use disorder care, home and community-based models, eligibility and renewal infrastructure, medication optimization, and social support. I frame the market as one shaped by Medicaid managed care, procurement, and measurable operating problems, not as a broad substitute for public capacity.
I assemble current public Medicaid managed care formularies and preferred-drug-list sources and link them to the top 100 Medicaid drugs by 2024 State Drug Utilization Data gross spending. The source universe contains 103 plan/source rows across 19 states, including 74 plan rows in states without a binding common preferred drug list. In that no-common-PDL sample, 56.0 percent of plan-drug cells are not listed or unmatched, 21.8 percent are preferred with utilization management, 7.8 percent are nonpreferred, 7.6 percent are preferred without utilization management, and 6.8 percent have mixed within-plan matches. Same-pharmacy-administrator pairs are more aligned than other pairs but still not identical. I frame plan choice in Medicaid managed care as a pharmacy-access and consumer-transparency issue, not only a provider-network choice.
I use Colorado's delayed CHIPRA Section 214 child implementation as a public-data case study of immigrant child coverage. HCPF and CMS materials document July 1, 2015 MAGI Medicaid/CHP+ implementation for lawfully residing children, followed by July 1, 2016 non-MAGI cleanup. Using ACS state-year contrasts and ridge-augmented synthetic control, I compare recent immigrant children with US-born children after treating 2015 and 2016 as transition years. In 2017-2020, Colorado's recent-immigrant child coverage contrast improves relative to the synthetic counterfactual by 18.1 percentage points for any coverage and 18.3 points for Medicaid or means-tested public coverage. The any-coverage result is the most stable; Medicaid is more donor-sensitive. I frame the evidence as a cautious Colorado case study, not a revived pooled CHIPRA estimate. Direct HCPF enrollment counts and a future Colorado APCD extension are suggested future extensions.
I evaluate whether waiving adoption fees for older shelter animals improves adoption and live-release outcomes. I exploit Austin Animal Center's sharp fee schedule discontinuity, where adoption fees for dogs and cats fall from $80-$120 to zero at exactly seven years of age. Using 173,775 intake records linked to outcomes from 2013-2025, I implement a sharp regression discontinuity design with robust bias-corrected local-linear estimation and donut-hole specifications for age heaping. For dogs, the headline margin, fee elimination has no detectable effect on adoption, live release, euthanasia, or length of stay, and the null survives multiple bandwidth, subsample, and covariate-balance checks. For cats, apparent adverse effects are explained by compositional selection: the share entering in sick, injured, or aged condition jumps at the cutoff. I conclude that senior fee waivers do not move key outcomes in this open-admission shelter, and that shelter policy evaluations must account for selection at administrative age cutoffs.
Page updated
Google Sites
Report abuse