Did households around age 35 have more money left after ordinary costs?

The Historical Norms point estimates are $1,020 a month in 1989, $1,440 a month in 2007, and $1,520 a month in 2022, all in 2022 dollars. These are descriptive estimates for similarly aged households in three periods, not the same families followed over time. The model-sensitivity comparisons below determine whether an ordering survives the tested assumptions.

The calculation starts with each CEX consumer unit’s income before personal taxes. Income and other personal taxes plus Social Security and Medicare payroll tax are removed before the income-after-all-personal-taxes subtotal. It then subtracts retirement contributions, household-paid health-insurance premiums, life and other non-health personal insurance, direct housing, household operations, utilities and telephone service, separately captured home internet where the source field exists, paid television and video services, groceries, medical care excluding insurance premiums, fuel and other operating transportation costs, apparel, and expected personal-care costs from matched CEX Diary households. The 1989 files do not support a separate home-internet line and cannot establish where or whether a payment was recorded. Historical Norms also uses each year’s observed student-loan payment distribution, vehicle-payment distribution, and household-level observed or imputed paid-childcare spending. The chart reports the weighted household-level median and distribution. The cash-flow table uses additive cohort means so every displayed subtotal reconciles exactly.

The table then shows mean recorded flexible spending and transfers (restaurants and other food away from home, other entertainment, furnishings and equipment, travel transportation, education, alcohol, tobacco, gifts, miscellaneous spending, and reading) and the mean unallocated cash remaining after those categories. At this point, retirement contributions were already deducted with the paycheck-like deductions. The official-budget accordion provides a separate cross-check against the CEX all-category expenditure total.

The original CEX total-expenditure balance produces average remaining cash of about $70 in 1989, $1,050 in 2007, and $1,510 in 2022 per month in 2022 dollars. This is not measured saving: expenditure definitions, balance-sheet transactions and survey residuals all affect it.

A negative modeled remainder is not evidence that most actual families overspent. Calibration to official spending means and combining costs from separate surveys can change its sign. The CEX total-expenditure balance in the official-budget accordion is not a complete saving measure or a replacement for the reconstructed cash budget.

Scenario Lab separates counterfactuals from history

Scenario Lab can select education debt financing one or two coded adult beneficiaries, one or two observed car-payment profiles, or paid care for one or two children. The education and care controls impose a common 2022-dollar obligation definition across all three years. Car-payment profiles instead draw from each year’s observed households with the selected number of positive car payments, and they exclude payments identified only as other vehicles, such as aircraft. The one- and two-payment pools describe different observed households, not identical vehicles or loan contracts: two cheaper payments can total less than one expensive payment, and different distributions can move means and medians differently. Each profile’s cost is deducted from its own no-car baseline; a higher remainder in the two-payment profile is not evidence that adding debt improves the same family’s finances. Historical Norms answers a different question: what mix of obligations did households in each period actually face?

The education-debt scenarios use the 2022 full SCF loan grid, whose relationship codes identify whose education was financed, not who legally owes the debt. The one-adult profile draws the household’s total balance financing exactly one coded respondent or spouse/partner beneficiary. The two-adult profile adds one positive individual-beneficiary balance drawn from households financing both coded adult beneficiaries. A spouse/partner code does not establish that the person is currently present or reveal their current age. Debt financing other people’s education is excluded from these modeled profiles but retained in Historical Norms. Each modeled balance is converted to a standard ten-year payment at 4.99%, a historical modeling assumption rather than a current loan quote; no donor draw is floored or rewritten.

What the historical comparison includes

Mortgage principal counts because it leaves the household’s checking account. Direct housing costs come from same-age, same-household, same-tenure official housing microdata: the 1989 and 2007 American Housing Survey national files and the 2022 ACS one-year housing file. Historical vehicle payments use the full SCF household distribution, including households with no reported vehicle payment and the source’s other-vehicle payments.

Home internet and paid television or video access count as ordinary participation costs. The 1988 and 1989 public files do not expose service-level television spending, so observed 1990 CEX households supply the donor distribution for those records. Restaurants and delivery, other entertainment, travel, alcohol and tobacco, gifts, furniture and upgrades, extra debt repayment, and non-retirement saving or investing remain outside the measure.

How much confidence to place in the result

The 95% normal confidence intervals include CEX sampling and income-imputation variance plus approximate donor-sampling uncertainty for housing, SCF debt components, paid childcare, and the 1989 paid-video bridge. This is not complete design-based uncertainty for the cross-survey remainder: zero-heavy payment distributions can have zero quartile-based donor error despite substantial payments among some families. Separate model ranges include declared specification variants and three alternate within-region imputations for suppressed public-use states; they do not guarantee that all uncertainty has been measured.

The release runs the production estimate against alternative income-response treatments, fixed-age and central-year samples, leave-one-year-out pools, 2022-standardized household composition, independent and inverse-rank donor assignments, distribution-preserving missing-value imputation, partial and zero component calibration, retained source spending shapes, observed and alternative student-payment specifications, and alternate tax-unit, income-taxability, wage-assignment and suppressed-state specifications.

A range spanning zero does not support a robust ordering. These bounds summarize registered model alternatives; they do not cover every possible modeling choice.