What changed in how the CPI is collected — and what it means here

A benefit for tens of millions of people is indexed to a survey. During 2025 that survey got smaller. This page sets out what is documented, what BLS measured, and — the part that actually decides how this site is built — what BLS explicitly did not measure.

What BLS measured, and found

under 0.01pp

The average difference between reduced-collection estimates and the published national all-items index. That is the one thing BLS affirmatively assessed.

Areas fully suspended: 3during 2025
Sample cut elsewhere: about 15%on average across remaining areas
National all-items: measureddifference under 0.01pp
Subnational and component: not measuredBLS states this explicitly

What is documented

Price collection was suspended entirely in three metropolitan areas during 2025 — two from April and one from June.
The sample in the remaining areas was reduced by roughly fifteen percent on average.
Imputation in the commodities and services survey rose to several times its pre-2025 norm and remained elevated month after month for more than a year.
One reference month was lost entirely and permanently. October 2025.

The asymmetry that decides how this site is built

BLS did two different things with these changes, and the difference matters more than either one on its own.

What BLS assessed, and what it did not
Index level	BLS assessment	Published on this site?
National, all items	Measured. Reduced-collection estimates differed from the published figures by less than one hundredth of a percentage point on average.	Yes — this is the only CPI series here
Subnational — metro and region	Explicitly not measured	No
Item and component level	Explicitly not measured	No

That is not a stylistic scoping decision. Publishing a metro-level or component figure here would mean publishing a number whose own producer has declined to characterise its accuracy under current conditions. Staying on national all-items CPI-W keeps every index value on this site inside the one zone BLS has vouched for, and it is written into the methodology as a limit rather than a preference.

What this does and does not call into question

It does not call the 2026 adjustment into question. The 2.8% was computed from the published national all-items CPI-W in the ordinary way, and this site reproduces that computation exactly from the raw index.

What it does raise is a genuine measurement question: an index built from a smaller sample is a noisier estimate of the same underlying quantity, and a noisier index feeding a benefit formula produces a noisier benefit. BLS’s own measurement suggests the noise is small at the national all-items level. That is the honest position, and it is neither reassurance nor alarm.

The base period, which is a separate question

BLS rebased around sixty CPI sub-component series in May 2026. The all-items series used here was not among them, and releases published afterwards still quote CPI-W at 1982-84=100. Any code that hardcodes a base period for a component series will break; the base for this one has not moved since 1988 and is read from the catalog rather than assumed.

What is not changing

No change to the index used for the adjustment is enacted or scheduled. Proposals to use a different index exist and are sitting in committee. Where those proposals actually stand.

Common questions

Is the CPI still reliable?

For the national all-items index, BLS says yes and published a measurement to back it: the reduced-collection estimates differed from the published national figures by less than one hundredth of a percentage point on average. For subnational and item-level indexes it did not publish a measurement at all, which is a different situation.

What is imputation?

Where a price cannot be observed in a given month, BLS estimates it from the behaviour of similar items. Some imputation is normal and always has been. What changed in 2025 is the share of the sample being handled that way, which rose several-fold and stayed elevated for over a year.

Did this affect the COLA?

The 2026 adjustment of 2.8% was computed from the published national all-items CPI-W in the ordinary way. Whether the underlying index is a slightly noisier measurement than it was in 2019 is a real question; it is not a question about whether the arithmetic was applied correctly.

Why does this site not publish regional or component CPI?

Because BLS vouched for the national all-items series specifically and stated that it did not assess the others. Publishing a metro-level figure here would mean publishing a number whose producer has explicitly declined to characterise its accuracy under current collection conditions.

Sources and freshness

U.S. Bureau of Labor Statistics — CPI-W (CWUR0000SA0), all items, U.S. city average, not seasonally adjusted — verified 10 August 2026
SSA Office of the Chief Actuary — CPI-W table — verified 10 August 2026

Page figures last verified against the sources above on 10 August 2026. BLS data retrieved 10 August 2026. BLS.gov cannot vouch for data or analyses derived from these data after retrieval. Corrections: the correction log · support@inventum.com.au