Six numbers, and how to not misread them
Every figure on this site comes from a named CMS file. That makes them defensible, but it also means they carry the quirks of how CMS collects them. These are the six that get misread most often, what each one actually counts, and the mistake to avoid.
Lesson 01
MA & other health plan share: two right answers, and they differ by seven points
What it is. The share of Medicare beneficiaries in a Medicare Advantage plan rather than Original Medicare. Enrollment divided by eligible population.
The trap. There are two defensible denominators and they do not agree. The CMS monthly enrollment file counts MA and other health plans, which includes cost plans, PACE and demonstrations, and lands at 51.6% nationally on the April 2026 county file. Sources that use an eligible-only denominator — excluding people with Part A alone, who cannot enroll in Medicare Advantage — report a higher percentage for the same period. Neither is wrong. Quoting one against a competitor's other is.
How to read it here. Every page states its basis. County pages use the CMS county definition throughout, so counties are comparable with each other. If you are pasting a figure into a deck next to a number from elsewhere, check the other source's basis first.
Worked example. Fresno County, CA reads 48.5% on the April 2026 file. That is MA-and-other over all Medicare beneficiaries in the county. It is not the same quantity as a plan-year MA-only figure for Fresno, which would be lower.
Lesson 02
Momentum: percentage points and percent are different numbers
What it is. Two separate movements. Penetration change is measured in percentage points. Enrollment change is measured in percent.
The trap. They can point in opposite directions and both be true. A county whose Medicare population is growing fast can add MA members every month while its penetration falls, because the denominator grew faster. Reading "+2.7% enrollment" as "+2.7 points of share" is the single most common error in this category.
How to read it here. Percentage-point movements always carry
pp. Percentages always carry %. The what-changed panel on every county
page shows both side by side for exactly this reason.
Worked example. Fresno County over twelve months: MA enrollment +2.7%, MA & other health plan share +0.4 pp. The plans grew, and they grew slightly faster than the eligible population. A market can post the first number and a negative second number in the same release.
Lesson 03
Duals and D-SNP capture: the denominator is the whole argument
What it is. The D-SNP capture proxy (Aug 2026 SNP enrollment / Apr 2026 dual population) is D-SNP enrollment divided by the dual-eligible population in the same geography. It answers whether the dual opportunity in a market is already taken.
The trap. "Dual eligible" is not one population. Full-benefit duals get full Medicaid; partial-benefit duals get help with premiums and cost sharing only. A capture proxy computed against all duals looks far lower than one computed against full-benefit duals. This site uses full-benefit duals and says so on the page.
The second trap. A capture proxy above 100% means the denominator does not describe the same population, not that a market is oversubscribed. Puerto Rico is the standing example: Medicaid there runs under a block grant and the dual flag in the Medicare file counts something different. The site suppresses that rate rather than printing a nonsense number.
Worked example. National D-SNP capture is 51.8% on the August 2026 SNP report. Roughly half the addressable dual population is already in a D-SNP. The national figure is far less useful than the spread underneath it, which runs from under 20% to over 95% by state.
Lesson 04
SNP types: three products that share a name
What it is. Special Needs Plans are MA plans restricted to a population. D-SNP serves dual eligibles, C-SNP serves people with a qualifying chronic condition, I-SNP serves people who are institutionalized or need an equivalent level of care.
The trap. Treating "SNP enrollment" as one market. The three behave nothing alike. D-SNP is roughly three quarters of the book and where the growth and the regulation sit. C-SNP is dominated by a single cardiovascular-and-diabetes bundle. I-SNP is about 1.5% of enrollment and operationally the most intensive per member.
The integration layer. Within D-SNP, the labels matter commercially: FIDE (fully integrated), HIDE (highly integrated) and CO (coordination-only) describe how far the plan integrates Medicare and Medicaid. Integration status is a live policy pressure, so its direction of travel is worth more than its level.
Worked example. August 2026: 8,437,333 SNP members in total, of which D-SNP 6,500,322, C-SNP 1,806,608 and I-SNP 130,403. Between May and August, FIDE rose 3.8% and coordination-only fell in absolute terms. That shift, not the headline total, is the story.
Lesson 05
Star ratings: the year on the label is not the year measured
What it is. CMS publishes contract-level quality ratings each October, on a one-to-five scale, and they drive quality bonus payments.
The trap. A "2026 star rating" is published in October 2025 and reflects performance measured mostly across 2023 and 2024. There is a two-year lag between what a plan did and what the rating says. Reading a rating as current-year performance overstates how quickly an intervention shows up.
The second trap. Ratings are annual and only the official October values are real. Any quarterly or interpolated star figure is a construction. This site displays official October ratings only, deliberately.
The money. Bonus payments lag again: a rating change shows up in revenue roughly two years later. A plan slipping below 4.0 today is describing a revenue problem for the year after next.
Lesson 06
Provenance tags: how much weight a number can carry
Every figure on this site carries one of four tags. They are not decoration. They tell you whether you can cite the number, and to whom.
| Tag | Means | Safe to cite as |
|---|---|---|
| live | Pulled from a named CMS file by an automated pipeline, refreshed on the CMS cadence | A CMS figure, with the vintage stated |
| static extract | From a named CMS file, extracted once because the file is not machine-readable on a schedule | A CMS figure, with the vintage stated |
| curated | Assembled by hand from published sources, each cited | MedicareInsights analysis, sources named |
| modeled | Estimated with a published formula. Not observed anywhere | An estimate, never as a CMS figure |
The rule. Anything tagged modeled shows its formula and its
inputs, and can be reproduced with a spreadsheet. The Market Opportunity Score is the clearest case: it
currently uses four of eight planned components, because the other four need CMS
files not yet ingested, and it says so in place rather than quietly dropping them. Read it as
market attractiveness, not as competitive difficulty.
Every number has a vintage. If you are comparing a figure here against one from elsewhere, match the vintages before you conclude anything. Most apparent contradictions between Medicare data sources are two different months.
Read the full methodology See the Market Opportunity Score method