Methodology

For each month t and archetype a, the Fixed Basket Index is the weighted sum of price relatives between month t and the base period (January 2000), expressed as an index with base = 100. Item-level price relatives are aggregated across strata using the arithmetic weighted mean of relatives. Within multi-SKU strata, the arithmetic mean (Carli) is used; the geometric mean (Jevons) is not used because Jevons mathematically corresponds to consumer behavior in which spending shares stay constant as relative prices change — an embedded substitution assumption.

The annual rate is the change in the Fixed Basket Index over a window, annualized to percent per year — trailing 12 months for the rolling rate, or the geometric mean across the full window for the headline long-run rate. The purchasing power of the dollar is the reciprocal of the index level; it is a direct reading, not a forecast.

Basket weights are anchored to the BLS Consumer Expenditure Survey (CEX) 2000 standard tables. Weights are frozen at the January 2000 base period and do not update when the source agency revises CEX tabulations. Five demographic archetypes (National Average, Homeowner, Renter, Working Parent, Retiree) are derived from CEX 2000 Tables 3, 5, and 7; five income-quintile archetypes are derived from CEX 2000 Table 1. Per-archetype weight schedules sum to 1.0 and are version-locked.

One augmentation modifies pure CEX: the healthcare premium stratum is augmented with the employer-paid private group health insurance contribution from NIPA Table 7.8 line 17 ($329.199 billion across 109.367 million consumer units = $3,010 per consumer unit per year for 2000). The augmentation is added to both the stratum numerator and the basket denominator and is set to zero for the Retiree archetype. Mortgage interest and personal insurance and pension lines are excluded from the basket denominator because they are debt service and deferred income, not consumption.

The shelter stratum is priced from observed market rents, with no Consumer Price Index input. From January 2015 onward the driver is the Zillow Observed Rent Index (ZORI), a repeat-rent, mean-of-the-middle measure of asking rents on the stock of advertised units. For periods before 2015 — where ZORI does not extend — the series is spliced to the Census Housing Vacancy Survey median asking rent for vacant rental units, with the two series chain-linked at the splice boundary so the level is continuous. Owners' Equivalent Rent, the imputed quantity CPI-U uses for owner-occupied housing, is not used. Owned-shelter operating costs (property tax, maintenance, services) are constructed as a composite with the property-tax slice sourced from Lincoln Institute state effective property tax rates; mortgage interest is excluded from the consumption basket and is reserved for the planned Asset Price Index in v1.1+.

Neither BLS CPI Rent of Primary Residence nor HUD Fair Market Rents are used as a driver, fallback, or triangulation source. The CPI rent series is excluded under the zero-CPI doctrine; FMR is excluded because it is a different conceptual quantity — the 40th percentile of recent-mover gross rents (the Section 8 voucher ceiling), aggregated from 5-year-smoothed ACS data.

Health insurance premiums are priced at total cost — the worker-paid premium plus the employer-paid premium. The price series is the Kaiser Family Foundation Employer Health Benefits Survey family-coverage total premium, published annually since 1999. The basket weight is augmented to include the per-consumer-unit employer contribution from NIPA Table 7.8 line 17 so that both the price level and the weight share reflect the employer-inclusive burden. The Retiree archetype is excluded from the augmentation because retirees are predominantly enrolled in Medicare.

At the state level, the price source is the Medical Expenditure Panel Survey Insurance Component (MEPS-IC), which KFF aggregates for its State Health Facts state breakouts. MEPS-IC and KFF EHBS measure the same construct through different operational definitions and persistently disagree at the US level by approximately 3–5% in levels and 0.3 percentage points per year in growth. v1.0.3 applies a cross-source calibration adjustment so state values aggregate to the national KFF EHBS series; the calibration ratio is held identical across states within a year (see Cross-source calibration, below).

The Fixed Basket does not apply hedonic adjustments to source prices. When a current-period model carries features that the base-period model did not, the price is recorded as the price; it is not deflated by an imputed quality differential. The replacement protocol for discontinued SKUs is matched-model: the closest still-available specification by stated criteria (size, type, grade) is selected, and the substitution bumps the basket version with a public changelog entry. Continuity ratios are published so older basket versions remain reproducible.

For technology-heavy strata (e.g., communications, household electronics) this convention produces a higher measured rate than a hedonic-adjusted measure. The convention is documented per stratum.

Index values are computed from observed source data. Temporal interpolation across a single series to fill missing months is not used. Where a primary source's coverage starts after January 2000, a next-tier fallback source is used for the pre-coverage period and a chain-link splice is applied at the first month of primary coverage so the level series remains continuous; splice ratios are computed once at basket-version time and frozen. Cross-source level calibration (e.g., the MEPS-IC to KFF EHBS state calibration in v1.0.3) is documented as a separate, named operation distinct from within-series interpolation.

Concrete example — the 2025 shutdown gap. The 2025 federal appropriations lapse left BLS unable to collect October 2025 prices, so the food-at-home source has no value for that month. The index omits October 2025 rather than estimating it — no carry-forward of the September value, no interpolation between September and November. It appears as a one-month gap in the food-at-home series; the long-run headline (anchored on full-coverage endpoints) is unaffected. If BLS later republishes the month, the monthly refresh ingests it automatically.

The geographic fallback ladder is walked per period, not per reference window: for each month, the lowest-priority geography with data for that month wins, and a chain-link rescale is applied when the chosen geography changes between consecutive months. This protocol matches BLS chain-link rebase at sampling-frame changes.

The Fixed Basket publishes a measured quantity. The methodology does not assign the measured erosion to a particular causal mechanism — monetary expansion, fiscal policy, supply shocks, demand dynamics, expectations, and velocity are all consistent with the published series. Source selection between candidate data series uses methodology purity as the criterion, not the direction the source moves the headline. The annual rate and the purchasing-power reading are outputs of the methodology, not targets.