A survey built to see the whole ladder
Most household statistics are good at the middle and weak at the top. A survey that samples addresses at random will find plenty of renters, wage earners and retirees, and very few of the families that own large businesses or large stock portfolios. Yet those few families hold a large share of the country’s wealth, so leaving them out distorts the total. The Federal Reserve’s Survey of Consumer Finances (SCF) was built around that problem. It asks a sample of U.S. families about income, assets, debts and borrowing experiences, and it deliberately reaches up the wealth ladder as well as across it. [1][2][4]
The newest results, for the 2025 survey, were released on October 9, 2026, with a Board report titled Changes in U.S. Family Finances from 2022 to 2025. The Board said the survey gives “detailed insights into the economic condition of American families.” [2]
The headline numbers are easy to repeat and easy to misread. This article walks through how the survey is run, how it defines its terms, why the median and the mean can point in different directions, and what a single survey can and cannot say. Where it reports numbers, they are the Fed’s, stated in 2025 dollars unless noted. Where it offers interpretation, it says so.
Who runs it, and how it began
The survey is sponsored by the Federal Reserve Board. The Board’s own description says it is sponsored “in cooperation with the Department of the Treasury.” [6] The fieldwork is done by NORC, a social science research organization at the University of Chicago, which the Board’s filing with the federal Office of Management and Budget names as the contractor for the 2025 survey. The Board’s older description says NORC has collected the data since 1992. [2][4][6]
The program began in 1983. The Board’s filing describes the 2025 survey as “the 15th triennial survey, which began in 1983.” The Board’s description adds that respondents to the 1983 survey were re-interviewed in 1986 and 1989. The Board’s press release says the survey has been run every three years since 1989, which is why the modern series is usually dated from then. [4][6][2]
Participation is voluntary. The Board’s participant page says NORC interviewers attempt to contact everyone selected, and that the 2025 fieldwork was expected to run from March through December 2025. The Board’s filing estimates an average of 110 minutes per response. [12][4]
The unit being counted: a family, not a household
The report says its definition of “family” differs from that typically used in other government studies. It uses what it calls the primary economic unit: an economically dominant single individual or couple, together with anyone financially dependent on them, with everyone else living at the address left out of that family’s figures. A one-person family counts. [3]
Because wealth and income are reported for the family, rather than for each adult, comparisons with studies of individuals or of Census households are not like for like. A median for families will differ from a median for people for reasons that have nothing to do with the economy. [3]
Two samples: an area sample and a tax-record list sample
The SCF uses what its documentation calls a dual-frame design. One part is a national area-probability sample, a geographically structured random sample that provides good national coverage and is gathered by the contractor. [4][5]
The second part is a list sample drawn from statistical records derived from individual tax returns, which the Board obtains under an agreement with the Internal Revenue Service’s Statistics of Income division. The list sample is designed to oversample wealthy households: in the Board’s earlier methodological paper, observations are sampled at progressively higher rates from strata that correspond to higher levels of an index of likely wealth. [4][5]
The Board’s filing describes privacy firewalls around this arrangement. The Board does not receive respondent names, the tax division does not receive data that would link survey answers to tax records, and the contractor does not receive tax-derived income data. [4]
The reason for the second sample is statistical. The Board’s methodological paper says estimates of tail-sensitive statistics, such as the mean of net worth, would be biased if the area sample were used alone. A small number of very wealthy families can move an average a great deal, and a random sample of addresses will usually contain too few of them to measure it. [5]
One limit remains. A Board paper on the Distributional Financial Accounts notes that privacy restrictions keep the SCF from sampling families on the Forbes 400 list, so the survey’s top is covered by a statistical correction rather than by those families themselves. [8]
Weights, nonresponse and imputation
The two samples are combined using sample weights so that the result represents U.S. families in the survey year. The Board’s filing says the 2022 SCF represented 131.3 million households. In the Board’s earlier methodological paper, the analysis weights are built in stages: each sample is adjusted to reflect its original selection probabilities, cases are classified by gross assets, and the weights are rescaled within those groups. [4][5]
Not everyone selected agrees to take part, and the shortfall is not random. In that earlier paper, the area sample’s response rate was below 70 percent of eligible cases in every year since 1989, the list sample’s overall rate was about 30 percent, and cooperation was substantially lower in the higher wealth strata. The paper calls nonresponse a very important problem with clearly nonrandom components. These are historical figures from a 2000 paper, and this article has not found a published 2025 response rate in the materials it reviewed. [5]
Respondents also skip questions or answer “don’t know,” particularly about amounts. The Board fills these gaps by multiple imputation: its program page says missing data in the survey have been imputed five times, and the methodological paper describes a system called FRITZ that imputes each missing value five times. That is why the Board warns that standard errors have to be computed with the provided instructions rather than treating the file as a simple sample. [1][5]
Sample sizes are modest compared with the population they describe. The 2025 survey interviewed 4,367 families and the 2022 survey interviewed 4,602. The full 2025 data set contains 21,835 records, five imputed versions of each family, and the public version has 21,800 records for 4,360 families after seven observations were removed to avoid disclosing identities. [1][3]
How the survey defines income and net worth
Income in the SCF is before-tax income for the calendar year before the survey, so the 2025 survey describes 2024 income. It includes wages, business income, capital gains, Social Security and retirement withdrawals, among other sources. Including capital gains matters: swings in stock and business values can move reported income for high earners from one survey to the next. [3]
Net worth is assets minus liabilities, measured as of the interview. Assets in the report include a family’s home, financial assets, retirement accounts, vehicles and privately held business equity; liabilities include mortgages, credit cards, vehicle and student loans and other debts. Because the home is valued by the owner, and business equity is estimated by the owner, net worth rests partly on self-assessed values. [3]
Dollar amounts are adjusted for inflation to 2025 dollars using the Bureau of Labor Statistics’ current-methods research series (CPI-U-RS), so that the 2022 and 2025 figures can be compared at constant purchasing power. The report describes the period as one of real growth of around 2.5 percent a year, unemployment moving from 3.7 percent to 4.2 percent, and consumer price inflation averaging 3.6 percent a year. [3]
Median and mean: two answers to different questions
The median family sits exactly in the middle: half of families have less, half have more. The mean is the total divided by the number of families, so a handful of very large values pull it up. The Board’s report explains that medians are less sensitive to extreme values, but are not additive, meaning the medians of subgroups cannot be combined into a median for the whole; means are additive and better for comparing groups. Neither is the correct number. Each answers a different question. [3]
In 2025 the two measures of net worth were far apart, and for income they moved in opposite directions. The table shows the Fed’s published figures, in 2025 dollars. [3]
Dividing the published figures gives a rough sense of the gap. Mean net worth was about 5.5 times the median in the 2022 survey and about 5.8 times in 2025; mean income was about 2.0 times the median in 2022 and about 1.8 times in 2025. That ratio arithmetic is this article’s, not the Fed’s. The Fed’s own reading is that income inequality decreased slightly between surveys and that wealth inequality appears little changed since 2022. [3]
The mean income decline illustrates why the choice matters. The Board attributed the 6 percent drop in mean income largely to families at the top, and excluding the top 1 percent of earners, mean income declined only 1 percent. For families in the top decile of usual income, median net worth grew 31 percent to $3,669,900. The report also noted that mean income for families headed by someone aged 35 to 44 fell from $186,000 to $139,800, which it linked to lower capital gains. [3]
Scroll horizontally to see all columns.
What changed from 2022 to 2025, as published
Reported changes were not uniform. The table collects some of the Board’s published group medians for net worth, in 2025 dollars. [3]
On other measures, the Board reported a homeownership rate of 66.1 percent in 2022 and 65.6 percent in 2025; median net housing value of $218,900 and then $230,000; and stock ownership, direct or indirect, of 58 percent and then 56 percent. The Board also reported that median net worth for families in the bottom quartile was $3,800 in 2022 and $1,700 in 2025. Among families holding stock, the report puts the median holding up 36 percent, from $56,900 to $77,400. Retirement plan participation among families whose reference person was 64 or younger was 64.4 percent and then 65.1 percent. [3]
Behind the group medians are sample sizes that shrink as the groups get narrower. The Board’s tables show standard errors for the first row only, and the full set is on its website, so a reader looking at a subgroup change needs those errors before reading much into its size. [3]
Scroll horizontally to see all columns.
| Group (median net worth, 2025 dollars) | 2022 survey | 2025 survey | Change as published |
|---|---|---|---|
| All families | $211,100 | $215,900 | +2% [3] |
| Families headed by someone under 35 | $42,700 | $33,000 | −23% [3] |
| Families headed by someone 75 or older | $367,300 | $504,900 | +37% [3] |
| White non-Hispanic families | $311,900 | $332,100 | +6% [3] |
| Black non-Hispanic families | $49,100 | $36,800 | −25% [3] |
| Hispanic families | $67,400 | $79,600 | +18% [3] |
| Families without a high school diploma | $41,600 | $18,600 | −55% [3] |
| Top decile by net worth | $4,152,700 | $4,935,200 | +19% [3] |
Debt burden: what the shares do and do not show
The Board reported that the share of families holding any debt was 77.4 percent in 2022 and 77.0 percent in 2025, with median debt among debtors of $87,800 and then $88,100. Beneath that stability, the Board found different measures of burden pointing different ways. Debt as a share of assets, a leverage measure, fell for the median debtor from 29.2 percent to 26.9 percent, while the payment-to-income ratio for the median debtor rose from 13.4 percent to 15.4 percent. The share of families whose debt payments exceeded 40 percent of income rose from 6.5 percent to 8.6 percent, which the Board called a level last seen in the 2013 survey. [2][3]
Those two readings are consistent. Asset values rose, which lowers debt relative to assets. Payments relative to income rose, which the Board’s report ties to higher borrowing costs: it cites average rates between survey years of 4.2 percent to 6.7 percent on 30-year fixed mortgages, 4.9 percent to 7.7 percent on new-vehicle loans and 14.6 percent to 21.4 percent on credit cards. The link between those rates and payments is the Board’s interpretation of the pattern rather than a tested finding. [3]
A share of families is not a count of hardship. A payment above 40 percent of income can reflect a family with a large mortgage and a high income as well as one that is struggling, and late payments are self-reported by the family. The ratio uses the payments and income the family reports, which depends on the family’s recollection and on the one-year income window. [3]
Scroll horizontally to see all columns.
| Debt measure | 2022 survey | 2025 survey |
|---|---|---|
| Families with any debt | 77.4% | 77.0% [3] |
| Median payment-to-income ratio, debtors | 13.4% | 15.4% [3] |
| Families with payments above 40% of income | 6.5% | 8.6% [3] |
| Families late on a loan payment in the past year | 12.2% | 19.6% [3] |
| Families 60 days or more late | 4.9% | 8.2% [3] |
| Families with a balance | 7% | 12% [3] |
| Families turned down for credit | 10.1% | 10.1% [3] |
What other data shows, and how it fits
The Federal Reserve publishes Distributional Financial Accounts that start from the SCF but fill in the years between surveys. They combine the quarterly Financial Accounts of the United States, which give aggregate balance sheets for sectors, with the SCF’s household microdata. The SCF is used to allocate the aggregate totals across groups in three steps: reconcile the two datasets’ wealth concepts, interpolate and forecast the three-yearly SCF balance sheets into quarterly ones, and apply each group’s share to the aggregates. The estimates run quarterly from the third quarter of 1989, and the Board’s page says the two sources use somewhat different wealth concepts and are measured at different frequencies. [7][8]
The tradeoff is directness. Quarterly values between surveys are interpolated, and values after the latest survey year are forecasts. A Board paper on the data also notes that items such as accrued defined-benefit pension rights are not directly measured in the SCF, and that its consumer-durables total runs about 61 percent of the Financial Accounts figure. A survey measurement and an accounts-based estimate can therefore differ without either being wrong. [7][8]
The Census Bureau’s Current Population Survey, which is cosponsored by the Bureau of Labor Statistics, is best known for monthly labor force statistics and gathers other topics through supplements. It is a different instrument built for a different purpose, so comparisons of income with the SCF involve differences in the unit counted, the income definition and the reach of the sample at the top. This article does not compare specific CPS figures. [10][3]
The Board also released a companion note on spending in October 2026. Using a new total-spending question in the 2025 survey, its authors reported that the top 1 percent of spenders accounted for 9 percent of spending, against 19 percent of income and 32 percent of wealth, and that the SCF figure for aggregate spending was just below that of the Consumer Expenditure Survey. The authors caution that a single cross-section cannot establish trends, and that measurement error is likely larger for consumption than for income. [9]
The limits of one survey
The SCF is a cross-section: separate samples in 2022 and 2025, not the same families followed over time. When median net worth for young families falls from $42,700 to $33,000, it does not mean any particular family lost that amount. The people in the group may differ, and families age into and out of the category. [3]
Sampling error matters especially for small changes. The Board reports a standard error of $8,200 on 2022 median net worth and $6,400 on 2025 median net worth. The published change, from $211,100 to $215,900, is $4,800. As a rough illustration of this article’s own, combining the two errors for independent samples gives about $10,400, so a $4,800 change is small relative to the sampling uncertainty the Board reports. For mean net worth the Board reports standard errors of $25,400 and $31,000, against a published increase of about $77,400, again by this article’s arithmetic. [3]
Measurement matters too. Home and business values are owner-reported, the highest-wealth families are the least likely to take part, and imputation fills in missing amounts. A change in how the wealthiest families answer, or are drawn, can move a mean without moving a median. Timing matters: income describes the calendar year before the interview, while net worth describes the moment of the interview, and a 2025 stock market or housing level affects the second but not the first. [3][5]
Finally, the survey period is a backdrop, not a cause. The report places the 2022 to 2025 changes against inflation averaging 3.6 percent a year and higher interest rates. A single survey can show what changed alongside those conditions, but it cannot show that the conditions caused the change. [3]
What would help clarify the picture
Several dated sources can add evidence. The Distributional Financial Accounts update quarterly, and later quarters can draw on the 2025 survey. The full 2025 microdata and the chartbook, which carries 1989 to 2025 estimates in 2025 dollars, let outside researchers compute standard errors and subgroup results the report does not show. [1][7][11]
The next triennial survey would normally follow in about three years. Until then, comparisons of the 2022 and 2025 figures are comparisons of two samples, each with its own error, rather than of a tracked group of families. [1][3]
Sources
- Federal Reserve Board — Survey of Consumer Finances (SCF) program pageOfficial sourceBack to text: ↑1↑2↑3↑4↑5
- Federal Reserve Board — Press release on 2025 Survey of Consumer Finances results, October 9, 2026Official releaseBack to text: ↑1↑2↑3↑4↑5
- Federal Reserve Board — Changes in U.S. Family Finances from 2022 to 2025 (SCF report), October 2026Official source · PDFBack to text: ↑1↑2↑3↑4↑5↑6↑7↑8↑9↑10↑11↑12↑13↑14↑15↑16↑17↑18↑19↑20↑21↑22↑23↑24↑25↑26↑27↑28↑29↑30↑31↑32↑33↑34↑35↑36↑37↑38↑39↑40↑41
- Federal Reserve Board — Supporting Statement for the Survey of Consumer Finances (OMB filing, reginfo.gov)Official sourceBack to text: ↑1↑2↑3↑4↑5↑6↑7↑8
- Arthur B. Kennickell, Federal Reserve Board — Wealth Measurement in the Survey of Consumer Finances: Methodology and Directions for Future Research (prepared for the May 2000 AAPOR meetings)Official source · PDFBack to text: ↑1↑2↑3↑4↑5↑6↑7
- Federal Reserve Board — About the Survey of Consumer FinancesOfficial sourceBack to text: ↑1↑2↑3
- Federal Reserve Board — Distributional Financial Accounts overviewOfficial releaseBack to text: ↑1↑2↑3
- Batty and others, Federal Reserve Board — Introducing the Distributional Financial Accounts of the United States (FEDS 2019-017), March 2019Official source · PDFBack to text: ↑1↑2↑3
- Aladangady, Bricker, Chang, Goodman and Li, Federal Reserve Board — Measuring Spending in the Survey of Consumer Finances (FEDS Notes), October 9, 2026Official sourceBack to text: ↑
- U.S. Census Bureau — Current Population Survey (CPS): AboutOfficial sourceBack to text: ↑
- Federal Reserve Board — SCF chartbook (1989–2025 data visualization)Official sourceBack to text: ↑
- Federal Reserve Board — 2025 Survey of Consumer Finances participant information pageOfficial sourceBack to text: ↑