Where the need is, and where the money lands

Need. Six measures from the Census Bureau's American Community Survey 5-year estimates (2019-2023) for every county and county equivalent in the fifty states and the District of Columbia: the poverty rate, the unemployment rate, the uninsured rate, the share of adults 25 and over without a high-school diploma, median household income and the share of people with a disability. Each is ranked nationally across all 3,143 counties and turned into a percentile; the six percentiles are averaged into one score, and the score is cut into ten equal groups. Decile 1 is the tenth with the most need. No measure is weighted above another, and nothing is modeled. Connecticut reports its need on its eight legacy counties from the 2017-2021 release, the last one published on those boundaries, because the grant crosswalk and the map still use them.

Money in. Grant lines from IRS Form 990 and 990-PF e-file filings for fiscal years 2020 to 2023, the four complete years since electronic filing became mandatory. Each grant is placed in the county of its recipient's ZIP code through the Census ZCTA-to-county crosswalk, then summed, divided by four and divided by the county's residents. A ZIP code that is not a ZCTA, which is usually a post-office box, is placed by the nearest ZCTA number in the same three-digit prefix, the same sectional center and almost always the same county; that is $99.3B of the dollars. A grant is placed where its recipient is based, which is the recipient's address on the filing, so a national organization's grants count toward the county of its headquarters; that is the whole of why the District of Columbia reads $8,863 per resident.

Money out. The same grant lines, summed by the county of the funder's own filing address. Donor-advised fund sponsors are excluded as givers: a sponsor's grants are its donors' decisions, made everywhere, and counting them would make Boston the most generous place in America. The rest of the gate is the same on both sides, so in and out net against each other county by county.

What is excluded, and why. Before anything is summed, four kinds of payment come out: deposits to donor-advised fund sponsors ($81.6B), which are transfers into an account rather than grants to a community; transfers between related organizations ($148B), which are a hospital system or a university moving its own money between its own entities; grants to intermediaries that regrant them ($156B), which would otherwise be counted twice; and grants to individuals ($57.3B), which carry a person's home address. Payments classed as non-philanthropic ($21.0B) come out too. Together that is $463B of $943B. The community-organizations view further removes any recipient whose median revenue is $100M or more, and any education or health recipient whose name says university, college, hospital, medical center or health system.

The economy. County GDP is the Bureau of Economic Analysis's current-dollar estimate for 2023 (CAGDP1), a public domain US Government work. BEA publishes Virginia's independent cities and Kalawao, Hawaii folded into combination areas ("Albemarle + Charlottesville"); those are split across their constituent counties by population, and the 53 counties concerned are marked in the data.

Since 2020. A same-store comparison: the funders that gave in both FY2020 to 2021 and FY2022 to 2023 are fixed, and each county's dollars from that set are compared across the two halves. New filers and departed filers are out of both halves, so the change is giving moving. Counties with under $500,000 from those funders in the first half are not scored.

The sector. Every organization whose latest e-filed return (FY2021 onward) carries a ZIP code in the county: a count, total revenue and total assets from that return, and net assets of the private foundations among them. The operating figures, charity income and spending, take the public charities alone, so a county's foundations are counted as endowment and not as spending. Donor-advised fund sponsors are excluded throughout. Hospitals and universities dominate assets and spending almost everywhere.

The similarity space. Each recipient organization carries a 768-dimension text embedding of its name, cause and mission from the site's semantic layer. A county's vector is the dollar-weighted average of its recipients' embeddings, and UMAP projects the 3,055 counties that have one onto a plane. Recipient text carries place names, so the raw projection is largely a map of states; the default view first centers each county's vector on the mean of its state's counties, which takes out what the counties of one state share and leaves what they are funded for. In the raw projection two in five of a county's nearest neighbors are in its own state; after centering, one in twelve. Counties funded the same way then sit together, tribal nations, college towns and retirement coasts each finding each other. Distance on the plane is a visual aid, not a measurement, and the dominant cause coloring is the largest of eight NTEE groupings among the county's classified dollars. The columns and the scatter rank or plot any of the measures above; a measure's columns are its deciles, largest first.

Reading it. Two counties in Alaska created in 2019 have no polygon in the map data and appear in the poster and the similarity space only. Small counties are noisy: one grant to one organization can move a county of 5,000 people from the bottom band to the top, which is why the named lists carry a population floor and the skyline caps height at the 99th percentile. The need score ranks counties against each other; it says nothing about need inside a county, and a large county with pockets of deep poverty can rank in the middle. Association only: nothing here says a grant caused, or failed to cause, a change in any of the six measures.