Primary Market Area, defined

What the PMA is

A facility's Primary Market Area is the straight-line radius ring around it — a circle centered on the facility's CMS-published coordinates. Distances are great-circle (“as the crow flies”) — we do not currently model drive time.

The radius is set per facility, not per portfolio, because one number rarely fits a mixed portfolio: an urban building competes inside a few miles, while a rural one may be the only skilled nursing for twenty-five. Where you haven't set a radius, we seed a default from CMS's own urban/rural classification — 25 miles for rural, 10 miles for urban, and 15 miles where CMS doesn't classify the facility. You can override any facility individually, or apply one radius across the portfolio, from either the Demographics or the Competitors tab — it is one saved radius per facility, so a change on either tab moves both. Every market-area figure — bed supply, occupancy, all demographics, and the Competitors tab's comparison — is recomputed against that facility's own ring.

The facility dashboard's adaptive PMA

Every PMA benchmark on the public facility dashboards (the Rating Build-Up's PMA columns, the star-matrix and drill-chart PMA series, the Snapshot tab's Competition card, and the Comp Set tab's default set) uses one shared adaptive market area rather than a fixed ring:

• Comparable peers only. The default set keeps facilities in the subject's own profit class — for-profit facilities benchmark against for-profits; non-profit and government facilities (whose cost structures and missions align more with each other than with proprietary operators) form one class and benchmark together. Hospital-based SNF units are excluded from freestanding facilities' sets — they carry hospital cost structures and file a different cost report. Excluded facilities are disclosed by name in the market-set roster and can be re-included through the set editor. Facilities CMS doesn't classify by ownership are never excluded, and hospital-based subjects keep their whole nearby market. Because membership keys on CMS's current ownership classification, a facility whose ownership type changes (for example after a sale) re-benchmarks against its new class — history series apply today's membership throughout.
• Thin-market fallback. When fewer than 2 comparable peers exist within the 25-mile cap, the set keeps every nearby SNF instead (disclosed on the surface) — an honest local benchmark beats none.
• The ring starts at the same urbanicity defaults as the portfolio PMA — 10 miles for urban facilities, 25 for rural, 15 where CMS doesn't classify — per CMS's own urban/rural indicator.
• When fewer than 5 comparable SNFs sit inside the start ring, the ring extends to the 5th-nearest comparable neighbor, never past 25 miles.
• When the ring holds more than 25 comparable SNFs (dense metros), the set keeps the nearest 25 — a fixed ring in Los Angeles would hold 250+ facilities, which is not a primary market.
• Members are active facilities only — delisted or likely-closed SNFs are excluded. The subject facility is part of its own market area, and benchmark averages include it (each surface states this).
• A market with fewer than 3 facilities (subject included) of any kind within the 25-mile cap is suppressed: no PMA benchmark renders, and comparisons fall back to the state. The comparable-peer rules never cause suppression — the fallback fires first.

Market areas can cross state lines (about 15% of them include at least one out-of-state member) — referral patterns do too. They are also asymmetric by construction: a rural facility's 25-mile reach can include an urban neighbor whose own 10-mile market doesn't reach back. Every surface discloses the market's size and reach — the Competition card and Comp Set tab in their labels and roster, the benchmark charts in their PMA toggle tooltips and collapsed notes. The Comp Set tab shows the full roster on a map, with preset rings available as manual overrides.

The portfolio Competitors tab uses a simpler, explicit basis. Its local-market comparison takes every active SNF within the facility's market-area radius above (the nearest 25 in the densest rings) — no automatic comparability filtering — narrowed only by the competitor filters you set yourself (ownership type and an Overall star band), either for all facilities or per facility. Its market columns describe that set excluding the subject: average, median, 25th and 75th percentile for occupancy (average bed-weighted) and Medicare census share (average patient-day-weighted), plus the range of Overall star ratings. Its per-facility competitive summary cards seed with the nearest active facilities from that same filtered set (same radius and filters), and are curated per portfolio: you can add competitors by search, remove any, and restore removals; curated sets are saved to that portfolio and apply nowhere else. This intentionally differs from the facility dashboard's adaptive market area above — the portfolio tab is a complete radius census you can adjust by hand, while the facility surfaces resolve the comparable-peer set.

Monitoring's Comp Set means

Portfolio Monitoring reports quote the Comp Set as a paired mean: for each metric (overall rating, occupancy, staffing hours), only peers with a value at both of the report's endpoints contribute, so the mean moves because facilities moved — never because the roster of reporters changed between the two dates. Peers whose staffing basis changes between the endpoints are excluded the same way the subject's own staffing comparison would be.

A Comp Set mean is quoted only when it represents the set: at least 70% of the set's members must be paired reporters for that metric, and never fewer than 4. Four of four peers reporting in a rural market is that market's mean; ten of twenty-five in a metro is not, however many ten is. When the mean covers fewer members than the set holds, the sentence says so (“9 of 11 other SNFs rated”). Below the gate, monitoring prose reports the subject's own movement without market context rather than quoting a mean that silently describes a different set.

Bed supply & occupancy

PMA bed supply % is the facility's certified beds divided by the total certified beds of every operating SNF whose coordinates fall inside the ring (the facility included). It reads as the facility's share of local licensed bed supply — a proxy for how concentrated or fragmented its market is.

PMA occupancy is bed-weighted across the ring — total average daily census divided by total certified beds, not an average of the facilities' individual occupancy percentages. Only facilities with a reported census contribute to the denominator, so a facility that hasn't reported a census can't drag the market figure down. Supply comes from the same CMS bed and census fields used elsewhere in the product; closed / decertified facilities are excluded, including facilities whose staffing reporting has gone dark for two or more quarters (likely closed) — the same hygiene every comp-set surface applies.

75+ population & median home value

The demographic figures come from the U.S. Census Bureau's American Community Survey (ACS) 5-year estimates at block-group grain — the finest geography ACS publishes for these variables. We take every block group whose centroid falls inside the ring and then:

• PMA 75+ population — sum the population aged 75 and over (ACS table B01001) across those block groups.
• PMA median home value — a household-weighted average of the block groups' median owner-occupied home values (ACS table B25077, weighted by ACS table B11001 households).

Two deliberate approximations are worth stating plainly. First, a block group counts in full if its centroid is inside the ring and not at all otherwise — we do not split a block group's population by how much of its area overlaps the circle. At a 15-mile radius the edge effect is small; at very small radii in dense areas it is coarser. Second, a household-weighted average of medians is not a true pooled median — the true figure needs the full value distribution, which ACS does not publish at this grain — so read it as a representative market value, not an exact median.

Demographic trends over time

The Demographics tab charts each facility's market area across three ACS vintages — 2013, 2018, and 2023. Those three are chosen deliberately: an ACS 5-year vintage is a five-year pooled sample, so the 2023 vintage covers 2019–2023, the 2018 covers 2014–2018, and the 2013 covers 2009–2013. They are strictly non-overlapping and contiguous, which makes the movement between them a real change rather than an artifact.

We deliberately do not publish an annual series. Consecutive ACS 5-year vintages share four of their five sample years, so a year-over-year difference is mostly sampling noise dressed up as demographic change. There is also no annual alternative at this geography: the ACS 1-year product is only published for areas of 65,000+ people, so block groups exist only in the 5-year estimates.

Census redrew block-group boundaries for the 2020 census, so block-group identifiers before and after do not correspond. That does not break these trends, because the market area is defined geographically: for each vintage we aggregate that vintage's own block groups, using that vintage's own centroids, inside the same ring around the same facility. The ring never moves, so the totals stay comparable across the redraw.

The “Combined market” row is a union, not a sum. Facility market areas overlap — two buildings on the same street share nearly all of their ring — so adding the rows together would count the same residents once per facility. Instead every facility's ring is combined into one set of block groups, each counted once, and the metric is computed over that set. The row is therefore normally smaller than the column appears to add up to, and it is labelled “deduped” whenever at least two rings genuinely overlap. On a real 21-facility portfolio this removed about 57,000 double-counted residents aged 75+, roughly 7% of the naive total. A useful consequence: because the union is a real set of block groups, medians and shares have a well-defined combined value and are no longer left blank.

How each metric aggregates. Counts (population, households) are summed across the ring. Shares (65+ share, homeownership 65+, 65+ living alone) are computed as total numerator ÷ total denominator across the whole ring — never an average of per-block-group percentages, which would let a 20-household block group count as much as a 2,000-household one. Medians (income, home value, gross rent) cannot be summed or plainly averaged, and ACS does not publish the underlying distribution needed for a true pooled median, so we report a weighted average of block-group medians — weighted by the population each median actually describes (owner-occupied units for home value, renter-occupied for gross rent, households with a 65+ householder for the 65+ income). Treat those as close approximations, not exact medians.

The trailing column reports compound annual growth over the full 2013–2023 span for counts and dollar figures. For metrics that are already percentages, a compound growth rate of a percentage is not a meaningful number, so we report the simple change in percentage points instead.

Nominal vs real dollars. Dollar metrics default to nominal — each vintage as published, in its own year's dollars. Consumer prices rose about 31% between 2013 and 2023, so a large share of nominal income and home-value “growth” is simply inflation. The Real toggle deflates every vintage to the newest vintage's dollars using the CPI-U annual average (BLS series CUUR0000SA0: 232.957 for 2013, 251.107 for 2018, 304.702 for 2023) and recomputes the growth rate from the deflated endpoints, so the levels and the rate always describe the same basis. The newest column is unchanged by definition; only history moves.

This is not a double adjustment. The ACS already inflation-adjusts dollar values within each 5-year period to that period's final year, so each vintage is internally consistent but expressed in a different year's dollars; converting between vintages is the missing step, not a repeated one. Two limits worth knowing: the Census Bureau performs its own within-period adjustment with the CPI-U-RS research series rather than headline CPI-U (we use CPI-U because a reader can look it up and reproduce our arithmetic), and home value was not inflation-adjusted within the period in the older releases our 2013 and 2018 vintages come from — so real-dollar home value is a good approximation rather than an exact restatement. Income and rent do not carry that second caveat.

Non-SNF portfolio members. A portfolio can include assisted living, memory care or independent living buildings that aren't in CMS data. Market demographics apply to them unchanged, because the calculation needs only a geocoded location and a radius — nothing from CMS. They default to a 10-mile ring rather than the SNF rural/urban defaults: a private-pay senior-housing move is local and discretionary, while skilled-nursing admissions arrive from a wider hospital catchment, so an SNF-width ring would overstate the market. That default is adjustable per building like any other. Clinical and financial metrics remain unavailable for them — only the market-area figures apply.

Chain pages. The same matrix appears on chain rollups, with two differences. The market-area radius is read-only there — a chain is derived from the operator name CMS publishes rather than a saved list, so there is no per-facility setting to store against it; every facility uses its rural/urban default. Save a chain as a portfolio to tune the rings. And because market areas are computed per facility, chains above 100 facilities show the largest 100 by certified beds, stated on the card — nine of roughly 635 chains are affected.

How we flag a low-confidence growth figure. ACS suppresses medians in block groups with too few responses, and that suppression rate is not stable across vintages — median income for householders aged 65+ is published for roughly 91% of block groups in the 2013 vintage but only about 68% in 2023. When a median's reporting coverage inside a ring moves by more than 10 percentage points between the endpoint vintages, part of the change between the two levels reflects which block groups reported rather than the market itself. We still show the growth figure — a caveated number can be weighed, a blank cannot — but render it muted with a ° marker, and hovering gives the actual coverage on each endpoint (for example 65% → 46%). Counts and shares come from tables that are not suppressed this way and are never flagged.

The stronger flag: a different basis. One case is different in kind rather than degree. When the endpoint vintages cover different numbers of facilities — because a vintage is missing data for part of the portfolio — the combined-market figures are not a noisier version of the same quantity, they describe different portfolios. The rate is still shown, but marked ‡ rather than °, and a banner names the affected vintage and how many facilities it covers. Per-facility rows are unaffected — each is accurate for its own market.

We show both kinds of flagged figure rather than blanking them, on the view that a labelled number can be weighed and discounted while a blank cell tells the reader nothing about why it is empty. The marks exist so that a figure resting on a moving sample, or on a changing set of facilities, is never mistaken for one that isn't.

CON state

CON state reports whether the facility's state requires a Certificate of Need to add or substantially expand SNF beds. A CON regime constrains new supply, which is a material backdrop for any bed-supply read. This is a state-level designation (it does not vary with the radius), sourced from the same curated registry that drives the state posture pills on the industry pages — primarily the NCSL Certificate-of-Need tracker, cross-referenced with state carve-outs.

Where the data comes from

• Bed supply, census, coordinates — CMS Nursing Home Provider Information (monthly), the same source as the rest of the facility data.
• 75+ population, households, median home value — U.S. Census Bureau American Community Survey 5-year estimates, block-group grain (tables B01001, B11001, B25077). Public domain.
• Block-group centroids — Census TIGER/Line internal points, joined to the ACS estimates by block-group GEOID.
• CON status — curated from the NCSL Certificate-of-Need State Laws tracker and state advocacy sources.

The demographic figures reflect Census-modeled survey estimates, not a licensed commercial demographics product. We do not publish forward population projections or household net-worth estimates — those are not available from a free public source at this grain.

Known caveats

• Straight-line, not drive time. A 15-mile ring can span a river, a mountain, or a metro edge that a resident would never cross for care. Drive-time market areas are a possible future refinement.
• ACS is a survey. Block-group estimates carry sampling error — larger for small populations — and lag the current year by the 5-year window's midpoint. Median home value is suppressed by ACS in block groups with too few owner-occupied units, and those block groups simply don't contribute to the weighted average.
• Centroid apportionment. As noted above, block groups are counted whole-or-not-at-all by centroid. This is standard for a first-pass market read and is most approximate at small radii in dense geographies.
• Supply is SNF-only. Assisted-living, memory-care, and hospital swing beds are not counted in the bed-supply denominator — the column measures skilled-nursing supply specifically.