Using the current observation weights , determine the rescaling factor for each consumption, income or wealth category k. Multiply the reported amount for observation i by , with , so that its weighted survey total matches the national-account target (e.g. NIPA). The rescaled incomes are then used for reweighting.
Pareto tail coefficients over time
PSID and SCF over time
Before adjustment, aggregate coverage differs across surveys. Averaging raw aggregates relative to scope-matched macro benchmarks over waves after 2005, expenditure coverage is 72.5% in PSID; the comparable SCF figure is 68.6%, available only for 2025. Total-income coverage is 83.3% in PSID and 89.3% in SCF. Wealth coverage is 82.2% in PSID and 123.2% in SCF; a value above 100% means that the raw aggregate exceeds its benchmark. CEX spending uses controls matched to its Meyer baskets. Expenditure and income benchmarks use matched NIPA controls; wealth uses Federal Reserve Financial Accounts controls. See Survey Adjustments and the paper’s Supplemental Appendix for definitions and procedures.
US results table
No estimates match these selections.
Consumption and wealth
Normalized bin statistics in the top 20%; adjacent intervals are combined to retain at least 10 observations per bin.
HFCS & SHIW
No estimates match these selections.
Show numerical results
† Luxembourg is shown, but excluded from the benchmark because of its small sample.
Rescaling and reweighting: iterative procedure
This procedure follows Section 2.3 of the paper. The iterative algorithm is explained below.
Under-representation adjustments change observation weights to match a benchmark income or wealth distribution. For income, use the current rescaling factors λ to calculate income relative to its weighted median and form bins indexed by j. The survey’s population share in bin j is , and the benchmark share in that same bin is . Update each observation’s weight from to using the benchmark share divided by the survey share, .
Full algorithm: joint rescaling and reweighting
Let S denote the study survey and R the dataset providing the benchmark distribution. For observation i and category k, is a reported amount, its national-account target, its current weight and its rescaling factor. Superscripts S and R identify the datasets, and 0 denotes original design weights. When benchmark microdata are available, their observations are indexed by h. The formulas for R below apply to these microdata. When R provides published distributional data, such as WID or IRS income statistics, use its relative-income brackets and population shares directly in Step 4.
In the steps below, we repeatedly rescale the income components used to assign households to income bins, then update their weights to match the benchmark distribution. Once the weights and rescaling factors stabilize, we use the final weights to rescale the remaining consumption, income and wealth categories to their national-account totals.
Initialize weights and factors
Start from the original study-survey weights and set every rescaling factor to one. If the benchmark is constructed from microdata, keep its original observation weights fixed throughout.
Rescale amounts to national accounts
Using the current weights, calculate the factor that matches each ranking-income category to its account target. Do this separately for the survey data and, where applicable, the benchmark microdata:
Updates are damped for numerical stability:Express income relative to its weighted median
Add the rescaled income components to obtain income , then divide by the survey’s own weighted median to obtain relative income .
Compare population shares in the same income bins
Use the benchmark’s relative-income brackets to define common bins indexed by j. Let equal one when study observation i is in bin j, and zero otherwise; is the corresponding benchmark-data indicator.
PSID example. The benchmark bins are P0–P50, P50–P70, P70–P80, P80–P85, P85–P90, P90–P95, P95–P97.5, P97.5–P99, P99–P99.5 and P99.5–P100. These are benchmark percentiles. Their income cutoffs, divided by the benchmark median, define the same relative-income bins in PSID.
depends on the current rescaling factors and weights. is the benchmark share. Each bin with a positive benchmark share must contain study observations.
Reweight observations using the bin-share ratio
For each study observation i in bin j, multiply its current weight by the benchmark share divided by the survey share, so weights increase in under-represented bins and decrease in over-represented bins:
For stability, move a fraction toward this target, then normalize the new weights to preserve the total population:
Return to rescaling and check convergence
Use the new study weights and return to Step 2. Recalculate the rescaling factors, weighted median and bin shares. Continue until both the relative factor changes and the gaps between survey and benchmark bin shares meet the convergence tolerance ( in the PSID baseline).
With the converged weights, rescale the remaining consumption, income and wealth categories.
At convergence, the final weights and rescaling factors make the survey match the NIPA aggregates and the benchmark population shares across the common bins.
Contributions to the adjustment
No decomposition is available for this selection.
Income distribution relative to WID
Income ranking and the spending correction
| CEX year | SCF reference | Threshold | Reported mass | Rescaled mass | Final mass |
|---|
Adjustments over years
Consumption
Income
Wealth
Sequential decomposition: change weights first, then amounts.
Superscripts 0 and * denote original and adjusted values. Shapley gives essentially the same allocation.
SCF has reporting adjustment only. In levels, the charts show the reported, representation-only and fully adjusted aggregates.
Component shares before and after adjustment
Raw data
Adjusted for reporting
Adjusted for reporting and representation
Each point is a component’s average share within a weighted percentile bin across the displayed survey waves. Lines show LOESS-smoothed average shares. SCF has two panels because its adjustment is reporting-only. PSID uses the NIPA consumption basket; final consumption flows are constructed after NIPA adjustment (see the paper).
Replication package
Data are currently available upon request.