Medicaid Care Economy: Concentration of State Job Growth

Purpose

This visualization answers one question that, until now, has been almost impossible to see clearly: how much of each U.S. state's job growth is coming from Medicaid-funded care work?

The national story has been told. We've covered it ourselves, and partnered with reporters to study it at the national level. Across the country, employment in care industries that lean heavily on Medicaid funding has become one of the most reliable job-growth engines in the U.S. economy. Services for the elderly and persons with disabilities alone has added 2.16 million jobs over twenty years, a 266% growth rate against just 17% for total nonfarm payrolls. In recent months, this single industry has been responsible for the majority of all U.S. job growth.

The national picture, though, masks where the exposure actually sits. A 35% national share of growth means very different things in California (where the absolute number is enormous) and in Wyoming (where it isn't). Until this tool, there was no public way to see, state by state, which job markets had quietly become dependent on these eleven NAICS codes. That matters because the funding source these jobs run on, Medicaid, is set to be cut starting in January 2027. Anyone trying to understand what that means for jobs, by definition, has to start at the state level. Medicaid is administered by states.

This page documents how we built the tool. It is meant to be specific enough that you could rebuild the visualization yourself with public data and the formulas here.

What this page is

Every chart we publish should be something you can check, question, and rebuild yourself. This page lays out every methodological decision behind the heat map: the data sources, the industry definitions, the formula, the thresholds, and the calls we made along the way. None of it is proprietary. The classification system is industry-standard NAICS, the data is public, and the math is arithmetic. Our contribution is putting the pieces together in a way that lets you see the state-level pattern at a glance.

We also want to be direct about what we don't know. This is the first state-level view of this pattern, and not every methodological decision has a single right answer. Where we made judgment calls, we say so. Where there are known limitations, we list them in the "Honest notes" section near the bottom.

The data sources

We use two public series from the U.S. Bureau of Labor Statistics (BLS), each doing a specific job:

• State and Area Employment (CES, also referred to as SM in BLS file naming) provides our state total employment denominator. CES reports monthly nonfarm employment for every state. It is survey-based, statistically estimated, and published with a roughly three-week lag. We use it because total state employment isn't subject to the cell-suppression problems that complicate QCEW at the state-by-industry level.
• Quarterly Census of Employment and Wages (QCEW) provides our industry-level numerator. QCEW is a near-complete count of jobs from state unemployment insurance records. It's far more granular than CES, reaching down to the 6-digit NAICS level that defines our industry groupings. It is published quarterly with a roughly five-month lag.

Using two series introduces some friction. CES and QCEW are constructed with different methodologies, and at the national level, state-summed CES totals diverge slightly from BLS's national CES (the most recent gap is roughly 290,000 jobs out of 159 million, or about 0.2%). This is an artifact of how CES is benchmarked annually against QCEW: between benchmarks, state estimates drift slightly from what a full count would show. We accept this trade-off because the alternative, using QCEW for both numerator and denominator, would mean state totals get hit by cell suppression in smaller industries and smaller states. In an earlier draft of this work, that approach pushed Iowa's calculated share of growth to 100% and Oklahoma's to 99%, neither of which is defensible. Switching the denominator to CES brought every state to within 6% of its published BLS total for 2024.

Step 1: Define "Medicaid Care Economy"

There is no official BLS category called "Medicaid Care Economy." We built one.

We selected eleven 6-digit NAICS codes that, based on our reading of available literature, are predominantly funded by Medicaid in some combination of direct payments, waiver programs, and state matching funds. We grouped them into four sub-categories:

Behavioral Health & Disability Support

• 621420: Outpatient Mental Health and Substance Abuse Centers
• 623210: Residential Intellectual and Developmental Disability Facilities
• 623220: Residential Mental Health and Substance Abuse Facilities

Facility-Based Long-Term Care

• 623110: Nursing Care Facilities (Skilled Nursing Facilities)
• 623311: Continuing Care Retirement Communities
• 623312: Assisted Living Facilities for the Elderly

Home & Community Based Services (HCBS)

• 621610: Home Health Care Services
• 624120: Services for the Elderly and Persons with Disabilities
• 624310: Vocational Rehabilitation Services

Family & Community Safety Net

• 624190: Other Individual and Family Services
• 624210: Community Food Services

A note on construction: we placed 621610 (Home Health Care Services) in the HCBS sub-group, even though NAICS classifies it under "Ambulatory Health Care" rather than "Social Assistance." This is a policy classification, not a NAICS one. Home health care is overwhelmingly delivered through Medicaid's HCBS waiver programs, which is the structural connection that matters for the question we're asking. The NAICS hierarchy would put it elsewhere; we don't think that hierarchy is the right organizing principle for this analysis.

As of publication, the full code list is under review by a Medicaid policy expert. If any of the categorizations change as a result, we will update this page and the visualization, and note the change.

What we explicitly did not do is try to estimate the Medicaid funding share for each code. Every code in our list receives significant non-Medicaid funding too, whether from Medicare, private insurance, out-of-pocket payment, or state and local programs. The viz does not claim these are "Medicaid jobs." It claims they are jobs in industries where Medicaid is a major, often dominant, source of payment. That is a softer claim, and an honest one.

Step 2: Do the arithmetic

For each state, for each quarter from 2015 onward, and for each industry sub-group (plus the aggregate "Medicaid Care Economy"), we compute:

Share of state job growth = (industry net job change) ÷ (state total net job change)

over the user-selected window (1, 2, or 3 years). The window ends at whichever quarter the user picks; the start is the same quarter the chosen number of years prior. We compare same-quarter to same-quarter to avoid contaminating the comparison with seasonality.

A few specifics. Both numerator and denominator are net changes over the window, not levels. Both are quarterly averages computed from monthly employment values: CES reports monthly state totals, which we average across the three months of the relevant quarter; QCEW reports three monthly employment levels per quarter, which we average the same way, producing a quarterly average comparable to the CES denominator.

We did not adjust for inflation, population, labor force, or any other normalizer. The metric is intentionally direct: of every X jobs the state added (or lost) over this window, how many were in these eleven industries?

Step 3: Edge case classification

The simple ratio only describes the cleanest case: state grew, industry grew. Real data produces five cases, and we treat each one explicitly.

1. State grew, industry grew. Standard ratio, displayed as a percentage in the relevant tier of the color scale.
2. State grew, industry shrank. The state expanded despite the sector, not because of it. We label this "industry declined" and color the state in a white hatched tier so it's visually distinct from the share-of-growth scale, which doesn't apply.
3. State shrank, industry grew. Industry was the only thing growing while everything else combined was net negative. We label this "industry-only growth" and color the state in the darkest red, with diagonal hatching, to mark it as the most concerning case on the map. Mathematically, the share-of-growth ratio would be misleading here (a negative denominator produces values that read like collapse, when the story is the opposite).
4. State shrank, industry shrank. Both moved the same direction. We group these with case #2 (industry declined), since the editorial point is the same: this industry was not load-bearing in this state over this window.
5. State net change was exactly zero, industry moved. Rare but it happens, particularly in smaller states where CES rounding produces flat YoY totals. (We've verified two of these, Kansas Q4 2024 to Q4 2025 and Montana Q2 2024 to Q2 2025, by comparing the underlying monthly CES values, which differ between the two years but happen to sum to identical quarterly averages. These are real data coincidences, not pipeline bugs.) Where the denominator is zero and industry grew, we treat the state as industry-only growth. Where the denominator is zero and industry shrank or was flat, we treat the state accordingly.

The classification logic is deterministic. There is no smoothing, fitting, or modeling. Every state's color on a given window is the direct output of the formula and the edge-case rules above.

Step 4: Color thresholds

The color tiers are fixed and do not change as the user slides the time window or switches industry groups. This is deliberate. A relative-color scale that adjusts to the current view would make every map look equally alarming and obscure the magnitudes that actually matter.

The thresholds:

• 0–2%
• 2–5%
• 5–10%
• 10–20%
• 20–30%
• 30–100%: Critical reliance
• >100%: Industry growth exceeded total state growth
• Hatched (darkest red): State lost jobs overall while industry grew
• Hatched (white): Industry declined

Two of these thresholds carry the most weight in the analysis:

30% (critical reliance). This is the softest of our anchors. The 30% line comes from regional-economics literature on single-sector dependence, where studies of oil-and-gas, automotive, and similar concentrated industries treat a 25–35% share as a rough threshold beyond which a regional economy starts to behave like a single-sector economy. We chose 30% because it sits in the middle of that range and produces a meaningful, but not overly inclusive, group of states. This is defensible but not from a single canonical paper. Reasonable analysts could pick 25% or 35% and not be wrong. We picked 30% and would rather tell you so than pretend it was inevitable.

100% (industry growth exceeds total state growth). This is the unimpeachable arithmetic anchor. By construction, a state with a share above 100% has an industry that has added more jobs than the state's net total, which mathematically means every other sector combined lost jobs. There is no value judgment in this threshold. It is the line where the math itself flips meaning.

Step 5: Window lengths

The viz offers three windows: 1, 2, and 3 years. Each pulls a slightly different story out of the data.

The 1-year window shows the most recent acute conditions. The 2- and 3-year windows reveal whether what we're seeing is a recent slowdown effect or a structural pattern that has been building. We dropped a 5-year window because it stretches across regimes (mid-pandemic recovery distortions, Biden-era infrastructure spending, the early Trump administration cuts) that introduce so much noise that the underlying signal becomes harder, not easier, to read.

Step 6: Time range

The data begins in 2015, giving a clean ten-year span through the most recent full year. We chose a round starting point on purpose: a 2015 baseline is easy to reason about, where an odder start year invites the reader to wonder why that particular year. Two things also make 2015 a sound baseline on the merits. It sits comfortably after the Affordable Care Act's main Medicaid expansion provisions took effect in 2014, so our starting point reflects the post-expansion structure of the program rather than the transition into it. And it sits clear of an earlier NAICS reclassification of code 624120 (Services for the Elderly and Persons with Disabilities), one of the largest codes in our group, that affected reported employment counts in prior years. Starting in 2015 gives us a consistently classified series for every code.

Updating

We refresh the chart quarterly, following BLS's QCEW release schedule. New QCEW data drops roughly five months after the close of each quarter. The CES denominator updates more often (monthly), but the limiting factor for the viz is the QCEW industry data.

The process is mechanical: download the new QCEW quarterly singlefile from BLS, update the CES file, run our prep script, and rebuild the HTML. The thresholds, classifications, and date ranges all extend automatically.

Honest notes and limitations

We'd rather tell you the edges of this than have you find them.

This is a working definition, not an official one. "Medicaid Care Economy" is a label we constructed. The eleven NAICS codes in our group are not all 100% Medicaid-funded, and reasonable analysts could add or remove codes. A policy expert is reviewing our list as of publication; if anything changes, we will update.

Our numerator is private-sector only; our denominator is all employers. Industry employment (the numerator) counts only privately-owned establishments (QCEW ownership code 5), while state total employment (the denominator) is CES total nonfarm, which includes government workers. We use private-only for the industry count because Medicaid-funded care is overwhelmingly delivered by private providers: nationally, 96.4% of all employment in these eleven NAICS codes is private, and only 3.6% (roughly 343,000 of 9.5 million jobs in late 2025) is government-owned, concentrated in state-run facilities for people with intellectual and developmental disabilities, public social-services agencies, and state vocational rehabilitation programs. Excluding that 3.6% is the conservative choice: counting those workers would raise each state's Medicaid Care Economy share, not lower it. Our published figures, if anything, understate the concentration. A future version may incorporate public-sector employment to give a fully complete count.

Not all eleven codes face the same exposure to 2027 cuts. The proposed Medicaid changes will hit different funding streams differently. HCBS waiver programs are often cited as the most vulnerable; mandatory benefits like nursing home care for low-income elderly are structurally more protected. Our visualization treats all eleven codes as equally exposed, which is a simplification. A more refined view would weight the codes by Medicaid funding share and waiver-program reliance. That work is beyond the scope of this first cut and is the kind of analysis our reporting partners are better positioned to do.

Cell suppression affects six small jurisdictions. QCEW suppresses individual-cell employment counts when a single employer would be identifiable. In the most recent QCEW release (Q4 2025), six small states have at least one of the eleven NAICS codes suppressed: Delaware, Hawaii, Kansas, Mississippi, South Dakota, and Wyoming. In each case, the suppression affects only one or two codes out of eleven, and the codes in question are necessarily small (suppression is triggered by employer concentration). The impact on the state-level aggregate is minimal. Forty-five of fifty-one jurisdictions report all eleven codes in full.

The CES denominator drifts between annual benchmarks. As noted in the data sources section, state-summed CES totals diverge slightly from BLS's national CES figures between annual benchmarks. This is a known feature of CES, not a flaw in our pipeline. The gap is small (well under 1% of total employment) and does not change the directional story for any state.

A small number of states show identical CES totals across adjacent years. This is a real, if striking, feature of CES at the state level. Total nonfarm employment is reported in thousands; three monthly values that sum to the same quarterly total are not impossible. Kansas Q4 2024 and Q4 2025, and Montana Q2 2024 and Q2 2025, are two such cases we've verified by hand. The viz handles this by routing zero-denominator cases into the appropriate edge-case tier, but it's worth knowing that the underlying data is, in these specific instances, genuinely flat year-over-year.

The most recent QCEW quarter is preliminary. When QCEW first releases a quarter, some private-sector cells at the 2-digit NAICS level can be partially suppressed or pending, and may fill in over the next one or two release cycles as state reporting catches up. The eleven 6-digit codes in our index are large enough that this affects them rarely, but readers comparing our most recent quarter to a future revision may see small differences. The directional story does not change.

This is a public-data tool, not a private database. Every input series is published free by BLS. If a researcher wanted to verify any number on the map, the underlying data is downloadable. We did not invent this approach to measuring sectoral concentration, and we don't claim to have. Our contribution is assembling the public pieces into a state-level view that didn't exist before.

Reproduce it yourself

If you want to rebuild this visualization, you need:

1. CES state-level total nonfarm employment for the years you want to cover, from the BLS State and Area Employment series (file naming: sm.data.0.Current, sm.series, related metadata files).
2. QCEW state-level quarterly singlefiles for the same years, from BLS.
3. The eleven NAICS codes listed in Step 1, filtered by own_code = 5 (private), agglvl_code for 6-digit NAICS state-level reporting, and the state FIPS codes you care about.
4. The formula in Step 2 and the edge-case logic in Step 3.

That is everything. There is no proprietary data, no model, no fitted parameters, and no secret sauce. If you replicate our process and get different numbers, we want to know. Tell us, and we'll look.