This page documents how the net worth figures on fatfirewoman.com are produced, publishes the full dataset for download, and offers the charts for reuse. Everything here is computed from the Federal Reserve’s own public microdata and checked against the Fed’s own published tables on every run. If you cite a number from this site, this is the page that tells you where it came from.
The dataset in one paragraph
Source: Federal Reserve Survey of Consumer Finances (SCF) 2025, public summary extract, 4,360 households, survey-weighted, in 2025 dollars. Contents: household net worth at 21 percentile points (1st to 99th) for all households, for seven age bands, for every education, race and ethnicity (with Asian households split out), marital status, children, and family-structure group, and for every age-by-group combination the survey covers with at least 100 households. Each row states how many surveyed households it rests on and the highest percentile it is allowed to publish. License: CC BY 4.0, attribution “fatfirewoman.com, from Federal Reserve SCF public data.”
Download the data
- CSV (long format, one row per percentile point): networth-percentiles.csv. Columns: group, segment, age_band, percentile, net_worth_usd, households_surveyed, highest_percentile_published.
- JSON (everything the calculators read, including validation results): networth-data.json.
Both files are regenerated by the data pipeline whenever the Fed releases a new survey, so the URLs are stable and safe to link. The next survey (SCF 2028) is detected automatically on release.
How the numbers are made
- Download. The pipeline fetches the Fed’s public SCF summary extract directly from federalreserve.gov. There is no API for net worth; the extract is a Stata file inside a zip, five implicates per household, with survey weights.
- Compute. Weighted percentiles use the Fed’s convention: the first value whose cumulative weight reaches the target. This reproduces the Fed’s published medians exactly. Age bands use continuous age (so 18 to 24 and 25 to 34 are real groups, not the Fed’s single “under 35” class). Race uses the five-way classification the public file has carried since 2025, which separates Asian households from “other.”
- Validate against the Fed. Every computed median and mean for the 41 groups in the Fed’s public Table 4 must match the Fed’s own spreadsheet: 82 of 82 statistics reproduced on the current run. Press-release anchors (median $215,900, mean $1.24 million) and the Bulletin’s five-way race medians must match too. If any check fails, nothing is published.
- Validate the calculator. A separate gate takes every published curve and, for 18 test net-worth values from negative $50,000 to $50 million, compares the percentile the calculator would display against the exact weighted share of that group below that value in the microdata. The worst error on the current run is 3 percentile points, at the pile of households sitting at exactly $0, across 98 curves. A curve that would mislead a reader blocks the publish.
- Publish with verification. The JSON is written to the site, read back, and byte-compared before the cache is cleared. Page copy that quotes a number pulls it from the same file at render time, so the text and the calculator can never disagree.
Sample-size tiers: why some combinations say “not enough data”
A percentile is only as good as the households behind it. A “top 1 percent” line computed from three surveyed households is noise dressed as precision, so every cell publishes only what its sample supports:
| Households surveyed in the cell | Highest percentile published |
|---|---|
| Fewer than 100 | Nothing. The calculator says “too few surveyed households.” |
| 100 to 199 | 90th |
| 200 to 399 | 95th |
| 400 or more | 99th |
On the 2025 survey this means, for example, that Asian households (328 surveyed) publish up to the 95th percentile, and no Asian-by-age combination qualifies. Of 126 possible age-by-group combinations, 72 do. We would rather show a smaller true number than a larger invented one.
What this fixes
The common shortcut for “net worth by age and race” is to take a group’s median, divide by the overall median, and multiply every percentile by that ratio. It is wrong at the top end: the ratio of medians is not the ratio of 99th percentiles. Applied to Asian households (median $677,400, about three times the overall median), that method implies an Asian 99th percentile above $50 million. Computed directly from the microdata, the Asian 95th percentile is $8.39 million. Our calculator used the shortcut until November 1, 2026. It no longer does, and this page exists so that nobody has to take that on faith.
Charts you can reuse
All charts are CC BY 4.0. Right-click to save, or copy the embed code. Attribution: “fatfirewoman.com, from Federal Reserve SCF public data.”
Median, 75th and 90th percentile net worth by age
<a href="https://fatfirewoman.com/net-worth-data/"><img src="https://fatfirewoman.com/wp-content/uploads/2026/11/chart-net-worth-by-age-scf2025.png" alt="Net worth by age, SCF 2025" width="1600" height="896"></a>
Full percentile curves by age band
<a href="https://fatfirewoman.com/net-worth-data/"><img src="https://fatfirewoman.com/wp-content/uploads/2026/11/chart-net-worth-percentile-curves-scf2025.png" alt="Net worth percentile curves by age, SCF 2025" width="1600" height="896"></a>
Median and 90th percentile by race, education and family structure
<a href="https://fatfirewoman.com/net-worth-data/"><img src="https://fatfirewoman.com/wp-content/uploads/2026/11/chart-net-worth-by-group-scf2025.png" alt="Net worth by group, SCF 2025" width="1600" height="1024"></a>
How to cite
fatfirewoman.com (2026). Net worth percentiles by age and demographic group, United States, computed from the Federal Reserve Survey of Consumer Finances 2025 public extract. https://fatfirewoman.com/net-worth-data/ (CC BY 4.0). Please also cite the primary source: Board of Governors of the Federal Reserve System, Survey of Consumer Finances.
Primary sources
- Federal Reserve, Survey of Consumer Finances (survey home, documentation, and the public extract download)
- Federal Reserve press release, October 9, 2026 (SCF 2025 headline figures)
- SCF 2025 public-data historical tables (real dollars), the spreadsheet every computed median and mean is checked against
- SCF 2025 summary extract public data, the microdata itself
Known limits
- The SCF is a survey. Even weighted, the top tail is thin; the tiers above are the honest response, not a fix.
- Percentiles between published points are linearly interpolated by the calculator. Measured error is under four percentile points everywhere and concentrated at exactly $0, where many households sit.
- Groups are shown side by side, never multiplied together. If you need “Asian, college-educated, married, 45 to 54,” the survey cannot support it, and neither can anyone else’s page, whatever it shows.
- Net worth in the SCF includes vehicles and the value of businesses and real estate; it excludes defined-benefit pension wealth and future Social Security. Compare like with like.
Questions or corrections
If you find an error, write to me. Corrections are logged here with the date.
Changelog: November 1, 2026: group and age-by-group curves, sample tiers, calculator accuracy gate, dataset and charts published. October 2026: re-based on SCF 2025.