How the Breathing Room Index measures household financial position

Breathing Room Index compares U.S. households at similar ages across different periods. Each published estimate identifies the metric version, survey, cohort, source years, inflation basis, sample size, uncertainty status, and reliability status.

Core measures

Compute the household result first

Each cash-flow result is calculated for an individual surveyed household. The weighted median is taken from those household-level values.

for each household:
    baseline = income_after_all_personal_taxes
        - retirement_contributions
        - household_paid_health_insurance_premiums
        - life_and_other_non_health_personal_insurance
        - direct_housing_cost
        - household_operations_and_services
        - utilities_and_telephone
        - home_internet
        - paid_television_and_video
        - groceries
        - medical care excluding health-insurance premiums
        - fuel_and_motor_oil
        - transportation_insurance_maintenance_leasing_licensing_and_local_transit
        - ordinary_apparel
        - personal_care

    historical_norm = baseline
        - household_observed_or_imputed_paid_childcare
        - observed_household_vehicle_payment
        - observed_or_reconstructed_household_student_loan_payment

    counterfactual_scenario = baseline
        - reader_selected_paid_childcare
        - reader_selected_vehicle_loan_payment
        - reader_selected_student_loan_payment

weighted_median(breathing_room)

The cash-flow table uses weighted cohort means, with rounding reconciled so displayed rows add to the displayed subtotals. These additive means are distinct from the household medians in the chart and percentile panel.

CEX recurring and episodic expenditures are averaged across a consumer unit’s available interviews only while its household type, number of children, and housing tenure remain unchanged. A move, separation, or child-count change starts a new smoothing state so spending from one Explorer cohort is not assigned to another.

The housing, vehicle, student-loan, paid-video bridge, and paid-childcare donor models use that same consumer-unit state as their assignment key. All interviews in an unchanged state receive the same modeled draw; the model does not treat each quarterly interview as a new household.

Default same-age cohorts

The first wealth analysis uses Survey of Consumer Finances households whose reference person is ages 33–37.

Enter the age of the survey reference person, not automatically either partner’s age. In the SCF, the reference person is the man in a mixed-sex couple and the older partner in a same-sex couple, even if someone else answered the interview. CEX uses its own reference-person convention. A five-year reference-age group can therefore differ from a group defined by your own age.

Parent selections use children under 18 rather than the SCF’s unfiltered KIDS count, which includes adult children. The public household grid supplies child ages for all three SCF waves. Coupled-parent selections require minors and no adult children, matching the CEX oldest-child-under-18 boundary; single-parent selections require at least one minor and may also include adult children. Adult-child-only families remain in “All households,” not in the childless selections.

These are the closest available source categories, not identical families matched across surveys. CEX “couple” means a married couple only; its two-parent category means a married couple with own children only. AHS likewise identifies husband-and-wife households. SCF married categories also include cohabiting couples, and the child-age rule does not exclude every other relative or household member. SCF amounts describe its financially interdependent primary economic unit; CEX measures a consumer unit and housing surveys measure a physical household. Public SCF relationship recodes also limit distinctions between child types. These boundaries remain limitations after child-age harmonization.

The first Consumer Expenditure Survey analysis pools 1988–1990, 2006–2008, and 2021–2023 to improve cell stability.

These are repeated cross-sections, not the same families tracked over time and not a causal test of a generation’s advantage. The 2007 SCF snapshot precedes the main financial-crisis losses, while its CEX pool extends into 2008. The 2022 SCF snapshot and 2021–2023 CEX pool span pandemic-era policy and asset-price conditions. SCF income refers to the calendar year before the interview, while balances describe the survey period. Changing household composition, tenure, and economic conditions can affect comparisons even at the same age.

CEX supplies the income and spending cohort. SCF supplies observed historical debt burdens for Historical Norms and conditional debt-payment distributions for Scenario Lab.

Included and excluded uses

The core layer covers ordinary costs of maintaining a household, working life, connectivity, and access to paid television and video. Mortgage principal is part of direct housing cost. Historical Norms then subtracts household-level observed or imputed childcare spending, observed household vehicle-payment burden, and observed or reconstructed household student-loan payments. Vehicle purchases are excluded because an irregular acquisition outlay is a poor monthly cash-flow measure. Restaurants and delivery, entertainment other than paid television and video, travel, alcohol and tobacco, gifts, furniture and upgrades, extra debt repayment, and non-retirement saving and investing remain available uses of breathing room.

The original cash-flow release separately reconstructs mortgage, vehicle-loan, and student-loan payments. The cash-flow extension adds remaining nonmortgage SCF payment categories without counting those three again, and exposes the full source payment breakdown. All mortgage-classified payments are excluded from the added layer to avoid overlap with direct housing; other-property mortgages are not deducted again from potentially net property income. Credit-card payments use the Fed’s 2.5%-of-balance assumption. Source payment totals include reported and imputed amounts, not uniformly observed contractual minimums. Neither release should be described as exact money left after every possible obligation. The separate SCF debt-payment-share measure uses before-tax income; the extension labels its modeled after-tax denominator separately.

Home internet is outside the BLS utilities category, which covers utilities and telephone service. The 2007 and 2022 comparisons capture home-internet expenditure separately from CEX UCC 690114 and remove it from household operations. The separate field is unavailable for 1989, so the site shows that line as unavailable; the public-use data cannot distinguish no connection from a bundled payment, payment by someone else, missing data, or an amount recorded elsewhere. In years with the separate field, a missing UCC expenditure record contributes zero to the population-average expenditure calculation; it does not establish that the household had no connection, subscription, or payment. Paid television and video combines cable or satellite service with recorded streaming-video subscriptions or rentals.

Reader-selected payment scenarios

Scenario Lab begins with no paid-childcare, car-loan or modeled education-debt payment. Its controls deliberately impose the same selected payment scenarios and education-debt profiles across years. This answers a counterfactual cost question, not whether those profiles or two car payments were equally common in 1989 and 2022.

Historical Norms uses all-household SCF student-loan and vehicle-payment distributions, including zero-payment families. The 1989 student-loan payment is reconstructed from education-purpose loans in the full public loan grid because the Fed’s summary macro sets the payment field to zero; the reconstructed balances reconcile to the summary balance field. Paid childcare uses household-level observed or imputed CEX spending in the outcome distribution; the additive cash-flow table shows the weighted cohort mean of those household values. These year-specific burdens are the default on /explorer/.

One- and two-car-payment scenarios use personal cars, trucks, vans and SUVs, and exclude other vehicles such as boats or aircraft. They count positive individual car payments, not every outstanding loan. The source lists three individual cars in 1989 and four in 2007/2022. Its residual-car payment aggregate counts as one additional payment only when the number of owned cars (NOWN, equivalent to raw X2202) establishes exactly one remaining car; otherwise a positive residual aggregate has an unknown payment count and is excluded from exact-count pools. A two-payment result is the source household’s car-payment total, not two copies of a household-total payment. Historical Norms retains the full observed/imputed all-vehicle payment distribution, including boats and aircraft, and is labeled separately. One-child paid-care scenarios draw from positive CEX household-total childcare spending among matched one-child households. The two-child scenario sums two independently assigned draws from that same one-child paid-care distribution, with each draw representing one child receiving paid care; their donor-sampling uncertainty is combined in quadrature.

The one- and two-car settings draw from different observed payment profiles. They do not keep the first car, its price, remaining balance or financing terms fixed. Two inexpensive payments can therefore total less than one expensive payment, especially in a thin source pool. This is not evidence that adding a loan makes the same household better off. Scenario Lab shows the modeled payment amounts and both source counts when the two-payment profile is cheaper; the raw amounts are not floored to force an ordering. Each positive-payment profile must still reduce the corresponding no-car-payment remainder by its own payment cost.

The modeled education-debt controls use the 2022 full SCF loan grid, which asks whose education was financed, not who legally owes the loan. The one-adult profile draws the total balance financing one respondent’s or spouse/partner’s education. The two-adult profile adds one positive individual-beneficiary balance drawn from households financing both relationship-coded beneficiaries’ education. Those codes do not guarantee current household presence or age. This is an education-beneficiary proxy for a modeled adult profile. Debts financing others’ education are outside this proxy, but remain in Historical Norms’ full household education-debt payments. Parent PLUS and cosigned obligations cannot be assigned to legal borrowers from these fields; use the custom combined payment to represent them. Each balance is converted to a standard ten-year payment at 4.99%; no donor draw is floored or rewritten. The same 2022-dollar modeled profile applies to every comparison column.

Direct housing source

Direct housing comes from official housing microdata rather than the CEX-SCF bridge. The model uses the 1989 and 2007 American Housing Survey national files and the 2022 ACS one-year housing PUMS. Renter costs are contract rent. Mortgagor costs are principal and interest plus property tax and homeowners insurance when those amounts are paid separately. Outright-owner costs are property tax and insurance. Utilities, condo fees, maintenance, and repairs remain outside this line.

Each direct-housing estimate uses the selected five-year age band, household structure, and tenure. Nonempty eligible samples remain visible, with a low-precision warning below 50 independent housing units; an empty matched source remains unavailable. The Explorer displays the actual housing-source sample beside the calculation. Other donor hierarchies retain their established preference for larger pools when one is available. If none meets that preference, the closest nonempty declared pool is used with its scope and count disclosed. A preferred pool size is not a publication cutoff.

Transportation and personal-care source layers

CEX Interview records supply fuel and motor oil separately from insurance, maintenance and repairs, leasing, licensing, and local transportation. Out-of-town travel spending is excluded. The SCF vehicle-balance and full debt-payment breakdowns retain the source’s broader all-vehicle scope; they are not car-only balances.

Personal-care products and services come from matched CEX Diary households. The displayed line is a weighted monthly expected cost. The Diary records two one-week periods, so a raw two-week median would turn many ordinary but infrequent purchases into zeros. The 1989 comparison uses the adjacent 1990 Diary release because earlier Diary public-use microdata are unavailable.

The 1988 and 1989 public packages also omit service-level paid-television expenditures. The model uses the observed 1990 CEX distribution for matching households in the same comparison pool, retaining zero-spending households. This bridge captures the cost of cable-era participation without treating television equipment as a monthly service.

Inflation-adjusted grocery spending describes real outlay; it cannot establish equal food quantity, quality, nutrition, convenience, or product mix. Historical apparel and personal-care results display the source-period cash equivalent alongside the 2022-dollar comparison value.

Personal household percentiles

Compare Yourself uses the same age, household, and tenure match as Historical Norms. The release stores weighted household quantiles at five-percentile intervals for each supported income, cost, subtotal, and remainder. Reader-entered values are interpolated between those points and labeled as approximate percentiles. Categories modeled only as cohort means do not receive a personal percentile.

The reader’s financial entries remain in the browser tab. They are not put in the URL or sent to the site. Only the demographic match (age, household type, and tenure) can appear in a shareable URL.

The reader must select the calendar year of the entered budget. Annual BLS R-CPI-U-RS values convert those entries to 2022 dollars before calculating totals or percentiles. The stored series supports budget years 2022–2025. Blank entries count as zero only for the reader’s arithmetic and receive no category percentile. Remainder percentiles are conditional on that zero-cost assumption.

Accounting classification

The accounting register distinguishes consumption, interest, principal reduction, asset acquisition, tax, insurance, transfers, and saving where the source supports the split. The published cash-flow view counts required principal payments as money leaving the household. Cash after recorded flexible spending is a separate cash-flow remainder, not a pure consumption measure or an estimate of saving; survey income and expenditure timing can differ.

A negative modeled remainder does not establish that most actual families spent more than their income. Calibrating spending categories to official population means and combining costs from separate surveys can change the sign of that remainder. The original CEX total-expenditure balance is shown separately for comparison. Model checks also retain the source spending distributions as an alternative to the mean-only categories.

Personal income taxes use separately reported wages, self-employment, Social Security, pensions, transfers, and other income, with taxpayer and spouse ages. Total income is not treated as wages. The public files still cannot fully reconstruct tax units, qualified dividends, taxable pension fractions, or every mixed-income category; alternate earner and tax-unit assignments are sensitivity checks. TAXSIM 35 extrapolates state tax law after 2021. The separate CEX-estimated Social Security and Medicare contribution line includes self-employment contributions, not just amounts withheld from wages; it is deducted once rather than adding TAXSIM’s payroll-tax output on top. Retirement contributions use direct contribution fields, smoothed and calibrated separately from that contribution line.

The source income totals include food-stamp benefits. These are kept as nontaxable resources, not wages, but cannot be spent freely on every bill. The CEX expenditure concept also includes food purchased with those benefits. The modeled remainder is therefore not a literal checking-account balance. The CPS income cross-check uses a different money-income definition as well as a different household universe. BLS income and expenditure concepts.

Inflation basis

The SCF summary extracts use the Federal Reserve’s 2022-dollar values. CEX income and expenses and the historical housing files use the BLS Retroactive CPI-U Research Series to convert survey-year dollars to 2022 dollars.

Published-table sanity checks

The official BLS LABSTAT age tables provide a broad external check on component trends. They report annual population means for all consumer units in seven reference-person age bands, so each benchmark still describes a wider group than any household and tenure selection in the Explorer.

The model estimates component-specific factors separately for under 25, 25–34, 35–44, 45–54, 55–64, 65–74, and 75-and-older consumer units. Each CEX record receives only the factor for its own age band. Age-specific internet and paid-video series are applied in 2022; equivalent series are unavailable for 1989 and 2007. This corrects age-specific extraction drift while preserving the distinction between an official broad-age mean and a narrower Explorer cohort. If a component’s calibration factor falls outside the registered 0.5-to-1.5 range, the site supports only its calibrated cohort mean. The aggregate household distribution therefore applies that mean as a constant to every household instead of allowing the unsupported calibrated source shape to affect the percentiles.

Sources: 1989 BLS age table and 2022 BLS age table.

The transportation total was essentially flat after general inflation. Gasoline and public-transportation spending rose materially. Vehicle-purchase net outlay fell, but that line measures household purchasing incidence and outlay; it does not measure vehicle prices. Food-at-home outlay was also nearly flat in real terms. The apparel row confirms that the high historical inflation-adjusted amount is a real feature of the source.

Weighting and uncertainty

Each source uses its own survey weights and variance procedures. SCF wealth estimates use all five multiple-imputation implicates, the public analysis weight, and 999 replicate weights. The exporter reproduces the Federal Reserve’s published all-family median net worth for 1989, 2007, and 2022. Its 95% normal confidence intervals combine imputation and sampling variance using the median procedure in the Federal Reserve codebook.

SCF wealth and debt comparisons also report Kish effective sample size and the largest independent family’s share of survey weight. These diagnostics sum eligible WGT records within each family across the five implicates; five imputed records are not five independent families. Conditional groups (such as families with positive debt, negative net worth or primary-home ownership) have their own diagnostics, not the all-family sample’s precision label. A raw sample of 50 or more families can still have low effective size or concentrated weights, so those warnings remain necessary. They describe weight concentration, not a replacement confidence-interval calculation: the median intervals still use the replicate-weight and imputation procedure above.

Normal-approximation intervals are not constrained to the possible balance range. A lower endpoint below zero for total debt or retirement accounts reflects that approximation, not a negative debt or retirement balance. Net worth, home equity and liquid net worth can genuinely be negative. Participation-share intervals are bounded to 0–100%.

Housing spending share divides each household’s recorded housing outlay by positive before-tax income before aggregation. Zero or negative income is excluded from this ratio, never replaced by an invented denominator; eligible counts and the excluded weighted share are shown. A household can spend more on housing than its income, for example by using savings, borrowing or temporary support, so a ratio above 100% is possible. It is not a share of after-tax income or the narrower direct-housing line in Breathing Room.

Large changes between neighboring ages are not necessarily life-course changes: these are different selected families, and an age window can gain or lose an unusually large payment or gift. Component notes identify when up to three independent households supply at least half of a weighted spending amount. Such means are not typical bills.

The added component-donor uncertainty is approximate, not complete design-based uncertainty for the cross-survey remainder. In zero-heavy payment distributions, a quartile-based donor-error approximation can be zero even when some families pay substantial amounts. The intervals do not fully capture uncertainty in payer prevalence, donor assignment or every cross-survey modeling choice.

A singleton source cannot establish sampling variance. Its point estimate and household-distribution quartiles remain visible, but the affected aggregate sampling standard error and confidence interval are shown as unavailable instead of as a zero-width interval. Two-adult education-debt profiles disclose the independent family count and donor scope for both the first and additional beneficiary pools.

For a one-household match, the personal comparison retains its modeled point but claims no percentile advantage over other households. An extreme negative remainder can also reflect negative net income, such as a business loss offsetting wages or benefits, rather than exceptionally large household bills. Adjacent age windows overlap and can include that same household; repeated appearances are not independent confirmation of a typical generational experience.

Breathing Room also carries a model-sensitivity range. The release gate reruns publication-critical cells with alternative income-response treatments, age and survey-year pooling rules, household-composition weights including number of earners, donor-dependence structures, missing-component imputations, calibration strengths, student-payment specifications, and a split-tax-unit specification. Cross-year claims require a stable direction across every registered variant. The interface shows direct sampling intervals and model ranges separately because they answer different questions.

Published cohort estimates carry:

• Unweighted sample size
• Weighted population estimate
• Standard error or confidence interval
• Suppression or warning flag
• Source and survey years
• Cohort definition

Nonempty CEX cohorts are not withheld simply because their sample is small. Each comparison column shows n, the number of distinct source households, not quarterly interview records or weighted population. Fewer than 50 households triggers a visible small-sample warning. Kish effective sample sizes below 150 and individual survey-weight shares above 2.5% also trigger precision warnings. Small samples can be unstable and should not support confident generational claims. Empty samples or unsupported required component sources remain unavailable, with the actual CEX sample count and reason disclosed. SCF, housing and donor layers keep the separate safeguards described above. The site does not silently widen the selected CEX age, household, or tenure slice to manufacture a result. Broader component donor pools are labeled separately from that selected sample.

Version registry

Definitions and source mappings are versioned. A historical result will never change silently because a variable name, survey category, or accounting rule changed.

metric_id
metric_version
source
source_year
variable_mapping_version
cohort_definition
deflator
inclusion_rules
exclusion_rules
known_limitations

Evidence status

Breathing Room and income after all personal taxes are labeled as modeled estimates. Housing spending share, real net worth, liquid-account balances, and debt-payment share are validated survey estimates. “Validated” describes the release checks, not freedom from sampling error or source limitations. Nonempty SCF point estimates remain visible with their actual independent-family counts. Below 50 families they are low precision. An empty replicate leaves the confidence interval unavailable, not the point estimate.

Coverage boundaries

The current release covers national comparisons at ages 18–80, five household groups, and four housing-tenure groups. A selectable combination is not a guarantee of an available estimate: empty selections and missing required sources remain unavailable.

Housing spending share describes existing households, not the cost for a first-time buyer entering the market. Home-entry scenarios separately finance each year’s annual median-priced new home at 10% and 20% down, using official mortgage rates and clearly labeled tax, insurance and closing-cost assumptions. They are not observed buyer outcomes.

The cash-flow extension calculates breathing room as a share of each household’s own after-tax income and the weighted share with zero or negative breathing room. Nonpositive income is excluded from ratios, with that excluded share shown; those households remain in shortfall rates. Medians and quartiles are not substituted for population shares. Its broader-debt remainder deducts additional payment categories and retains a separate donor-dependence sensitivity. Sampling intervals are conditional on the donor assignments and do not represent total model uncertainty.

The consumption-basis remainder uses original CEX total expenditures less that same source’s pension/Social-Security category and cash contributions. It excludes principal repayments and financial-asset purchases by the CEX accounting definition, but counts goods when acquired, including durable purchases and goods bought as gifts. It is a CEX-based proxy rather than a full national-accounts consumption measure. It differs from breathing room in flexible-spending coverage, source substitutions and transfers as well as principal; the difference between their medians is not principal repayment. Consistent observed principal/interest splits are unavailable for every historical loan. The home-entry model makes that split explicitly for its assumed mortgages.

The observed principal-outlay tables additionally expose CEX owned-home, vacation-home and other-property principal reductions for the 2007 and 2022 pools, across the same age, household and tenure filters. The EMRTPNO, EMRTPNV and EMISCMT previous/current-quarter pairs identify those outlays. They can include lump sums and home-equity debt. These source amounts are not a matched principal split of the AHS/ACS housing model and are not subtracted again. Comparable respondent-level fields are unavailable for the full 1989 pool, so that column is unavailable, not zero.

The SCF supplement adds selected-asset liquid net worth, primary-home equity, retirement-account assets, total financial assets, total debt, education loans, vehicle loans, credit-card balances, and debt relative to annual before-tax income. The data guide defines each universe and exclusion. Liquid net worth is a defined asset-minus-debt proxy, not a promise that assets can be converted to cash immediately or without taxes or penalties. Retirement-account coverage changes for accounts already paying benefits; the definition records that historical limitation.

Debt balances and asset values include zero holdings unless the selected household or tenure group excludes them. Debt-to-income uses only positive-income households and does not replace a missing denominator with an arbitrary ratio. Debt balances, debt payments, before-tax income, and after-tax income are different quantities; the debt-payment share uses the SCF income denominator, not modeled CEX after-tax income.

Each SCF comparison also reports the fraction meeting the stated condition and the median within that group, with sampling and imputation uncertainty. A zero all-household debt median can coexist with substantial borrowing among fewer than half of families. Home-equity context uses all primary-home owners, including owners with zero or negative equity. The owner/renter mix uses the same selected population as the metric. Debt-to-income context therefore also excludes nonpositive-income families. Nonempty conditional medians remain visible with eligible-family counts and a low-precision warning below 50. Their confidence intervals remain unavailable when an implicate or replicate cannot support the calculation.

The 2022 observations are historical, not a description of conditions at launch. They include the federal student-payment pause and the balance-sheet conditions measured by that SCF wave. The Federal Reserve independently reports a large 2019–2022 increase in real median wealth; that does not establish that a new buyer could afford a home more easily or that every family benefited. SCF period context; student-payment pause context.

SCF population counts sum the summary-extract weights across all five implicates. Those weights already contain the Federal Reserve’s one-fifth adjustment. Independent sample sizes count surveyed families, not imputed records. See the population-count correction for the release history.

Independent household-well-being check

The Federal Reserve’s separate 2022 Survey of Household Economics and Decisionmaking (SHED) challenges how these figures can be interpreted. We compare broad populations once, rather than averaging overlapping five-year cohorts. These are plausibility comparisons, not targets used to force the model to agree.

Source universes: 4595 independent 2022 SCF families, including zero holdings; 812 families have positive education debt in at least one implicate. CEX uses 26180 independent consumer units with an adult reference person from the 2021–2023 pool. SHED uses adult-weighted October 2022 self-reports; its education balances refer to July 2022. The CEX before-tax consumption-proxy shortfall is 33.7%; using after-tax income and only central-year interviews gives 41.0%. These bridges do not quantify the remaining cross-survey differences.

SHED education-debt source; SHED savings source; SHED spending source; download the numerical check and source hashes.

SHED’s education-debt range offers a magnitude check, not proof that family debt and an individual borrower’s debt are identical. The cash-buffer and spending comparisons materially differ. Checking accounts can contain money committed to bills, and adult self-reports do not have the same population, timing or accounting rules as SCF family holdings or CEX consumer-unit estimates. Those are possible explanations; we have not measured how much of either gap each explanation accounts for.

Consequently, liquid-account balances do not by themselves measure emergency funds available to spend, and a negative modeled remainder is not an observed overspending or missed-payment rate. SHED separately reported that 82% of adults expected to pay all their October 2022 bills. The historical tools measure a different accounting quantity and cannot establish who had an easier life. SHED expenses.