CPI Explained: What the Consumer Price Index Measures and How It Is Calculated

What the consumer price index measures

BLS defines the CPI as a measure of the average change over time in the prices paid by consumers for a representative basket of goods and services. The level only has meaning against its base: an index of 336.940 means the basket costs 3.37 times what it cost in the 1982 to 1984 base period, which is set to 100. The area indexes work the same way, so the CPI cannot say whether Denver is more expensive than Dallas, only how fast each has risen.

The basket covers what people buy for day-to-day living: food, clothing, shelter, fuel, transport, medical care and services. Sales and excise taxes tied to a purchase are included. Income and Social Security taxes are not, and neither are investments such as stocks, bonds, real estate and life insurance, which BLS treats as saving rather than consumption.

The index is old. BLS began publishing price indexes for 32 cities in 1919, started a national index in 1921 and estimated it back to 1913. On the official series, the September 2026 level is about 34 times the 1913 average, which is another way of saying a 1913 dollar bought what about $34 buys now. The inflation calculator runs that conversion for any year.

How BLS calculates the CPI

Every month BLS data collectors record the prices of about 80,000 items. They cover 75 urban areas, about 6,000 housing units and roughly 22,000 retail and service establishments, from supermarkets to hospitals, by personal visit, phone, web and app. Fuel and a few other items are priced monthly everywhere; most other items are priced monthly in the three largest areas and every other month elsewhere.

The calculation runs in two stages:

1. Basic indexes. Prices are grouped into 243 item categories in 32 geographic areas, giving 7,776 basic indexes, such as apples in Chicago. Since January 1999, BLS has averaged prices within most of these cells with a geometric mean, which allows for some switching between similar products when one gets dearer.
2. Aggregation. The basic indexes are combined into the all items index using spending weights. For the CPI-U and CPI-W the quantities are held fixed for 12 months, which assumes households do not shift spending between categories within the year.

The weights come from the Consumer Expenditure Survey. Since January 2023 they have been updated every year from a single calendar year of spending, with a two-year lag: the 2023 index used 2021 spending, and the 2026 index uses 2024 spending.

Prices also have to be adjusted for quality. When a product is replaced by a new version, BLS has to separate the part of any price change that reflects a better or worse product. In categories with heavy quality change, such as apparel, appliances and electronics, it uses hedonic regression: a statistical model that estimates a value for each characteristic of an item and adjusts the price when those characteristics change.

The CPI is a sample, so it carries sampling error. BLS puts the standard error of the monthly all items change at 0.04 percent.

CPI-U vs CPI-W vs chained CPI

BLS publishes several all items indexes from the same prices. They differ in whose spending sets the weights and in the formula.

• CPI-U (all urban consumers) covers over 90 percent of the US population. It is the headline index, introduced in 1978. In September 2026 it rose 3.7 percent over 12 months.
• CPI-W (urban wage earners and clerical workers) covers about 30 percent of the population: households earning at least half their income from clerical or wage jobs, with someone employed at least 37 weeks of the year. It excludes retirees, the self-employed and salaried professionals. Social Security cost-of-living adjustments use it. It rose 3.9 percent in the year to September 2026.
• Chained CPI-U (C-CPI-U), published since July 2002 data, uses a superlative formula and spending data from the same month, so it captures shifts between categories, from beef to chicken for example. It is published as preliminary and revised three times. Federal tax brackets have been indexed to it since 2017. It rose 3.5 percent in the year to September 2026.

BLS also calculates a research index for Americans aged 62 and older, the R-CPI-E. BLS cautions that its weights rest on about a fifth of the spending survey's urban sample and that it prices the same stores and areas as the CPI-U, so conclusions drawn from it should be treated as tentative.

The populations matter because their baskets differ. In the December 2025 weights, the CPI-W gave energy 8.1 percent against 6.4 percent in the CPI-U, and gasoline 4.0 percent against 2.9 percent. That heavier energy weight helps explain why the CPI-W ran ahead of the CPI-U in the year to September 2026, when energy prices rose 20.4 percent. The 2027 Social Security increase of 3.6 percent came from the CPI-W's third-quarter average.

The CPI weights: where the basket's money goes

Each category's weight, which BLS calls its relative importance, is its share of total consumer spending. Housing is the largest by a distance.

Two features stand out. Shelter alone is over a third of the CPI-U, and owners' equivalent rent, a figure no household actually pays, is more than a quarter. Services other than energy make up 60.7 percent of the CPI-U, against 19.2 percent for goods other than food and energy, so rents, insurance, medical care and other services drive most of the index.

Weights also drift between updates. When one category's prices rise faster than the rest, its share grows: energy went from 6.4 percent of the CPI-U in December 2025 to 7.4 percent by August 2026, without any change to the basket.

Criticisms of the CPI: housing, substitution and quality

Most serious criticism of the CPI comes down to three methodological choices, plus the limits of any average. Each choice has a defensible logic, and each can make the index differ from what a particular household feels.

Owners' equivalent rent

The CPI does not price owned homes. BLS, like most other countries' statistical programs, treats a house as a capital asset and measures only the shelter service it provides: the implicit rent an owner would pay to rent the same home, without furnishings or utilities. Mortgage interest, property taxes, real estate fees, house prices and improvements are left out. BLS adopted this rental equivalence approach in January 1983 for the CPI-U and January 1985 for the CPI-W.

The case for it is consistency: a home is partly an investment, and the CPI excludes investments. The case against is that buyers pay house prices and mortgage rates, which the index leaves out, so periods when home prices and borrowing costs jump faster than rents are understated for anyone trying to buy.

Substitution

A fixed basket assumes people keep buying the same things when relative prices change. In practice they switch to cheaper options, so a fixed basket tends to overstate the rise in the cost of keeping the same standard of living. In 1996 the Boskin Commission, appointed by the Senate Finance Committee, put the upward bias of the CPI, from substitution between items and between stores, quality change and new products, at 1.1 percentage points a year, within a range of 0.8 to 1.6. The Social Security Administration's archive notes the stakes: adopting the commission's recommendation would have reduced future cost-of-living increases. The geometric mean in 1999 and the chained CPI in 2002 followed.

The counter-argument is that switching from steak to chicken is itself a loss of living standard, which a substitution-adjusted index scores as no loss at all. Each tenth of a point removed from the measured rate compounds in every payment indexed to it.

Quality adjustment

When a new model costs more but does more, BLS counts part of the increase as quality rather than inflation. That is correct in principle. Critics point out that the value of a feature is estimated by the statistician, while the higher price is paid in full at the till.

One average, many households

BLS itself says a national average reflects millions of individual price experiences and seldom mirrors any one consumer's. A retired homeowner who drives a lot and spends heavily on medical care has a different basket from a young renter in a city.

How to use the CPI to measure inflation

Any change in prices between two dates is the ratio of the two index levels, minus one. The CPI-U was 324.800 in September 2025 and 336.940 in September 2026: 336.940 divided by 324.800 is 1.0374, a rise of 3.7 percent. The same arithmetic converts dollars across time. $1,000 in September 2025 needed about $1,037 in September 2026 to buy the same basket.

For comparisons with the past, use the unadjusted CPI-U, which is final when published. Seasonally adjusted figures are revised for up to five years. One gap affects most series: BLS published no October 2025 index for most CPI series because of the 2025 government shutdown, so comparisons that need that month have to use a neighbouring one.

New figures arrive each month in the CPI report, and the full annual record since 1914 is in US inflation rate by year. The Federal Reserve's preferred gauge is covered in PCE vs CPI.