How is CPI actually calculated and what are its limitations?

The short answer

CPI is not a single price — it is an average of averages. Each month, BLS field economists collect prices from urban retailers, service providers, and rental units, then aggregate them into a weighted basket designed to represent typical urban consumer spending.

The weights themselves come from the Consumer Expenditure Survey, which asks households what they actually buy. Until 2023, weights were updated every two years; the BLS now updates them annually to reduce the lag between observed spending shifts and their reflection in the index.

The index is published in two main forms: headline CPI (everything) and core CPI (excluding food and energy, which are volatile). Markets typically watch year-over-year percentage changes rather than the level itself.

→ New to inflation measurement? Financial education hub

What the data shows

Key figures (BLS, FRED, 2021-2024):

• Approximately 94,000 prices collected monthly across 75 urban areas
• 8 major categories: food, housing, apparel, transportation, medical care, recreation, education, other
• Shelter accounts for roughly 36% of headline CPI and 44% of core CPI
• CPI inflation peaked at 9.1% YoY in June 2022 — the highest reading since November 1981
• Core CPI peaked later, at 6.6% YoY in September 2022, reflecting shelter’s lagged dynamics

The composition matters because two CPI prints with the same headline number can mask very different underlying dynamics. In 2022, the headline was driven by energy and food; in 2023, the persistence came from shelter and services.

→ Dataset: US Core CPI inflation

Why it happens — the macro mechanism

Three structural choices in CPI construction shape what the headline number captures and what it misses.

The basket and its weights. The CPI basket reflects past spending, not present substitution. When beef prices rise sharply, households shift to chicken — but the basket only adjusts later. This creates a documented upward bias in CPI relative to true cost-of-living changes, estimated by the 1996 Boskin Commission at roughly 1.1 percentage points per year. See why PCE differs from CPI.

Hedonic adjustments. When a $1,000 laptop in 2024 is twice as fast as a $1,000 laptop in 2020, the BLS treats part of that price as a quality improvement rather than a price increase. Hedonic adjustments are most aggressive for technology categories. Critics argue this understates true price pressure for households who don’t necessarily benefit from quality gains.

Owners’ equivalent rent (OER). Roughly 24% of headline CPI comes from a survey asking homeowners what they would pay to rent their own home. OER tracks market rents with a lag of approximately 12-18 months. In 2021-2022, OER underestimated true housing inflation; in 2024, it has been overestimating it relative to current market rents. See owners’ equivalent rent explained.

The CPI is a constructed average, not a price tag — it tells a story about consumption, not a single truth about cost.

→ Framework: Inflation regimes pillar

What it means for different economic actors

Savers who use CPI to estimate the erosion of their cash holdings can be misled in either direction. When OER lags, headline CPI may understate real shelter costs for renters facing immediate market rent increases, and overstate them for owners with locked-in mortgages.

Investors in TIPS and other inflation-linked instruments receive coupons indexed to headline CPI, not core CPI. This means their real returns depend on the volatile components — energy and food — that policymakers tend to look through. See breakeven inflation rates explained.

Policymakers at the Fed officially target headline PCE inflation, not CPI, partly because PCE captures a broader scope of spending and uses chain-weighted aggregation that reduces substitution bias. CPI nonetheless remains the reference for Social Security adjustments, tax bracket indexation, and TIPS.

A common misreading is treating the monthly CPI print as a measurement rather than an estimate. Each release carries revisions to seasonal adjustment factors, sample updates, and weight adjustments that can change the trajectory months after the fact.

Practical observation

Go deeper

Related questions

• Why does PCE inflation differ from CPI inflation?
• What is owners’ equivalent rent and why does it matter?
• Why does the Fed watch median and trimmed-mean CPI?
• What is the difference between core and headline inflation?

Frequently asked questions

Is CPI the most accurate measure of inflation?

CPI is widely used but is not necessarily the most representative measure for every purpose. The PCE price index covers a broader scope of household spending and uses a chain-weighted formula that reduces substitution bias. The Fed officially targets PCE for its 2% mandate. CPI remains the legal reference for Social Security adjustments, tax bracket indexation, and TIPS coupons, which gives it institutional weight regardless of methodological debates.

Why does CPI sometimes diverge from people’s lived experience?

Personal inflation rates differ from CPI because individual consumption baskets differ from the BLS basket. Households spending heavily on healthcare, education, or rent may experience inflation rates substantially above headline CPI, while retirees with low transportation expenses may experience inflation rates closer to it. The BLS publishes alternative indices, including R-CPI-E for elderly households, which historically runs slightly higher than headline CPI.

How often is the CPI methodology revised?

Major methodological changes occur roughly every decade. The most recent significant change came in 2023, when BLS moved from biennial to annual weight updates. Before that, the 1999 introduction of geometric means within item categories was a major shift, partly responding to the Boskin Commission critique. Each change creates measurement breaks that complicate long historical comparisons.

Last updated — 1 May 2026