How the Consumer Price Index Is Actually Measured: One Price, From Shelf to Headline
"Inflation was 3.4%" sounds like one number. It's really about 94,000 prices a month, 243 item categories, 75 cities, a survey of what households spend, and a rent sample that is only ever one-sixth fresh. Here's the whole pipeline, station by station.
Evergreen explainer, checked against the primary sources listed below. Full disclaimer
On this page6 sections
Around the middle of every month, at 8:30 in the morning Eastern, the Bureau of Labor Statistics publishes a number that moves bond markets, anchors wage negotiations, and resets tax brackets and benefit checks. When the August 2026 report came out on September 11, 2026, the headline was that the Consumer Price Index for All Urban Consumers had risen 3.4% over twelve months.
That one figure is the last step of a long assembly line. Most coverage treats it as a thermometer reading. It's closer to a manufactured product — and like any manufactured product, it's easier to trust, and to argue about sensibly, once you've walked the factory floor. So let's follow a single price through every station.
Station one: deciding which shelf to stand in front of
The CPI starts from a sampling problem. BLS can't price everything everyone buys, so it builds a probability sample at three levels: places, stores, and specific items.
Places first. The index is built from 75 urban areas, which BLS then consolidates into 32 "index areas" for calculation. The population the headline CPI-U describes is all urban consumers, which BLS puts at about 93% of the U.S. population. The older CPI-W, which covers households of wage earners and clerical workers, now represents about 29% — a detail that matters later, because it's the one Social Security uses.
Then stores. Outlets aren't picked because they're convenient for a data collector. They're drawn from survey data on where households actually shop, so a big-box store that gets a lot of real spending is more likely to land in the sample than a boutique that doesn't.
Then items. A data collector arriving at a sampled store with a category to price — say, a given kind of bread — lists every product matching the category's definition, groups them, and selects one with probabilities proportional to their sales. That exact product, down to size and brand, becomes the item priced there month after month. BLS says the survey collects about 94,000 prices and 8,000 rental housing unit quotes each month.
How often depends on where you live. Fuels and a few other items are priced every month in all 75 areas. Most other goods and services are priced monthly in the three largest areas — Chicago, Los Angeles, and New York — and every other month everywhere else.
The CPI assembly line
How a single observed price becomes part of the headline index
Diagram: The Explainer Desk, drawn from BLS methodology. Source: BLS Handbook of Methods, CPI Design and Calculation chapters
Station two: the small indexes
Every price that comes in is compared with the price of the same item at the same store last time. Those comparisons are rolled up into what BLS calls basic indexes — one for each combination of 243 item categories and 32 index areas, for 7,776 in all. Think "apples in the Midwest's smaller cities" or "men's suits in Los Angeles."
Inside most of those cells, BLS combines price changes using a weighted geometric mean rather than a simple average. The practical effect is that the formula assumes shoppers shift a bit toward whichever version of an item got relatively cheaper — if one brand of cereal jumps and another doesn't, some people switch. A limited set of categories, including selected shelter services, utilities, government fees, and medical care services, instead use a modified Laspeyres formula, which assumes people keep buying the same things. Those are areas where switching within the category isn't realistic: you don't substitute your way out of your water bill.
Two complications live here. The first is quality change. When the TV the data collector has been pricing is discontinued and replaced by a newer model with a bigger screen, the price difference isn't all inflation. For some categories — apparel, consumer appliances and electronics among them — BLS uses statistical models that value individual features so the price comparison is apples to apples.
The second is missing prices. When an item can't be priced, BLS imputes it in a set order: first from the average change for similar items in the same area (a "home cell"), then from the same category in a wider geography (a "different cell"), and only as a last resort by carrying the previous price forward. That ladder became newly visible in 2025, when BLS suspended collection entirely in Lincoln, Nebraska, and Provo, Utah, in April, and in Buffalo, New York, in June. BLS said the effect on the national indexes was minimal but that local and item-level indexes could get more volatile, and that the number of imputed items rose.
Station three: weights, or why your inflation isn't the inflation
The small indexes are then combined using weights that reflect how households spend their money, taken from the Consumer Expenditure Surveys. BLS refreshes those weights every year with the January data; the indexes published in 2026 use 2024 spending. Updating them is not cosmetic: BLS calculated that January 2026's all-items monthly change was 0.37% with the new weights, versus 0.40% if it had kept the old ones.
BLS publishes the result as each component's "relative importance" — its share of the index. Here's what the December 2025 CPI-U looked like.
What the CPI is made of
Relative importance of selected components in the CPI-U, December 2025 (% of the all-items index)
Source: BLS, Relative Importance of Components in the Consumer Price Indexes, December 2025 - vehicles combines new (3.8) and used (2.8); categories overlap and are not meant to sum to 100.
A worked example: how much can one item move the headline?
Relative importance lets you do the arithmetic yourself. Gasoline (all types) carried a weight of 2.895% in December 2025. Suppose gasoline prices rose 10% in a month and nothing else changed. The contribution to the all-items index is roughly the weight times the change:
- 2.895% x 10% = about 0.29 percentage point added to the all-items change.
Now try shelter, at 35.625%. A 3% rise in shelter, with everything else flat, contributes about:
- 35.625% x 3% = about 1.07 percentage points.
A modest move in the heaviest component outweighs a dramatic move in a light one. (This is a first-order approximation — weights drift month to month as relative prices change — but it's how analysts estimate "contributions" on release day.) It's also why a month of cheaper gas can make headlines while barely denting the index, and why grocery prices can feel far worse than an index in which food at home is about 8% of the weight.
Station four: the rent you don't pay
The biggest single piece of the CPI is a price nobody actually pays. Owners' equivalent rent, at 26.2% of the CPI-U in December 2025, is how the index handles the cost of living in a home you own.
BLS deliberately doesn't use house prices. Its reasoning is that a house is a capital asset — an investment — and what the CPI tracks is consumption, which for a homeowner means the shelter the house provides. So the weight comes from a question in the expenditure survey: "If someone were to rent your home today, how much do you think it would rent for monthly, unfurnished and without utilities?" The price change then comes from actual rents in the CPI Housing Survey, reweighted to represent owner-occupied homes. That's also why a surge in home prices, the thing the housing shortage is usually measured by, shows up in the CPI only indirectly and with a lag.
It also produced one of the strangest episodes in the index's recent history. During the 2025 lapse in appropriations, BLS did not collect CPI data from October 1 through November 12, 2025. The October and November 2025 indexes were affected — and, because the October rent panel is the same one that gets priced again in April, BLS said the missing data also affected the April 2026 rent and owners' equivalent rent indexes. A gap in the fall reappeared, on schedule, in the spring.
Station five: which CPI, for what?
"The CPI" is several indexes built from the same raw prices, and different laws are wired to different ones. That wiring is policy, and it has real money attached.
| Index | Who it describes | What it's used for |
|---|---|---|
| CPI-U | All urban consumers, about 93% of the population | The headline inflation number; many contracts and escalators |
| CPI-W | Urban wage-earner and clerical households, about 29% of the population | Social Security's annual cost-of-living adjustment |
| C-CPI-U (chained) | Same population as CPI-U | Federal income tax bracket indexing since the 2017 tax law took effect in 2018 |
The chained CPI, which BLS began publishing with July 2002 data, uses a formula that also lets shoppers substitute across categories — beef for pork, say — not just within them. It tends to rise a bit more slowly: BLS puts the average gap in December-to-December changes from 2001 to 2023 at about 0.2 percentage point. Because it depends on spending data that arrives late, it's first published in preliminary form and not final until 10 to 12 months later. Small annual differences compound, which is why the choice of index for any program is a recurring, legitimately contested debate — each option builds in a different assumption about how people respond to prices.
What to take into the next release
And keep the trade-off in view. The CPI is a carefully engineered compromise between precision and timeliness, between a fixed basket and real behavior, between what homeowners spend and what homes cost. The Federal Reserve watches it alongside other measures precisely because no single design answers every question. The index doesn't claim to know what your month cost. It claims to know, consistently and transparently, what happened to the average urban household's — and it publishes its own recipe so you can check.
Primary Sources
- U.S. Bureau of Labor Statistics, Handbook of Methods: Consumer Price Index - Design
- U.S. Bureau of Labor Statistics, Handbook of Methods: Consumer Price Index - Calculation
- U.S. Bureau of Labor Statistics, Relative Importance of Components in the Consumer Price Indexes, December 2025
- U.S. Bureau of Labor Statistics, How the CPI Measures Price Change of Owners' Equivalent Rent and Rent
- U.S. Bureau of Labor Statistics, 2025 Lapse in Appropriations: Impact on the CPI
- U.S. Bureau of Labor Statistics, Chained CPI: Questions and Answers
Keep explaining: related from the desk
Why Grocery Prices Don't Come Back Down
Inflation cooling doesn't mean prices fall — it means they climb slower. Here's the machinery that makes grocery prices a one-way ratchet, from sticky wages to the 11.8-cent farm share.
How the Fed Actually Moves Your Interest Rates
The Fed doesn't set your mortgage rate, your credit-card APR, or even "the" interest rate. Here's the actual chain — from an FOMC vote to two administered rates to the prices banks charge you — walked end to end.
The US Housing Shortage
America builds fewer homes than it forms households, and has for years. A data story about the lost decade of construction, why estimates of the gap range from 1.5 to 5.5 million homes, and the mechanics that keep supply stuck.