Every month a single number lands, moves bond markets within seconds, and is then argued about by people who are certain it is wrong. Understanding how inflation is measured explains most of that argument, and most of the distance between the figure and the experience.
Inflation is measured by pricing a basket of goods and services each month and comparing the total against an earlier month. The US Bureau of Labor Statistics collects tens of thousands of prices for the Consumer Price Index, weights each item by how much households actually spend on it, and publishes the change.
The disputes are rarely about the prices. They are about the weights, and about one large component that is not a price anyone pays.
What the index actually is
The consumer price index is not a measure of the cost of living, though it is constantly described as one. It is a measure of what a fixed pattern of spending would cost this month against what it cost before.
Government staff collect the prices, in person, by phone and online, across a sample of retail outlets and rental units in metropolitan areas. Roughly 80,000 price quotations go into a month.
Each quote is for a tightly specified thing. Not “coffee” but a particular brand, size and grind at a particular shop, so that this month is compared against the same item last month rather than against a different purchase.
That specificity is the whole design, and it is also the first place the index parts company with real life. Households substitute. The index, within a given period, largely does not.
The weights do most of the work
Prices alone would tell you nothing. A basket needs to know how much each item matters, and those proportions come from the Consumer Expenditure Survey, in which households record what they actually spend.
The effect is straightforward. Americans spend far more on chicken than on tofu, so a chicken price rise moves the index and an identical tofu price rise barely registers.
This is why arguments about “the basket” are usually arguments about weights. Nobody disputes that petrol prices rose; they dispute how much of a household’s budget petrol represents, and whose household is being described.
Airline fares illustrate the point neatly. They are one of the most volatile lines in the index and one of the smallest by weight, which is why a dramatic swing in what it costs to fly can move the headline number almost not at all — even though it dominates the experience of anyone trying to find a cheaper flight.
The largest component is a price nobody pays
Housing is the biggest thing in the index by a wide margin. Shelter makes up nearly a third of the CPI, and the single largest line inside it is not rent paid by tenants.
It is owners’ equivalent rent — an estimate of what homeowners would pay to rent the homes they already own. On BLS figures it has run at close to a quarter of the entire index, around 23.5% as of late 2021.
No money changes hands. Roughly a quarter of the most-watched economic statistic in the world is an imputation.
There is a defensible reason for it. A house is partly an asset and partly a service, and an index of consumer prices is supposed to capture the cost of the shelter consumed, not the investment. Mortgage payments mix the two together, which is why the BLS stopped using them in 1983 and moved to rental equivalence instead.
Before that change, a rise in mortgage rates pushed measured inflation up directly — which meant the main tool for fighting inflation mechanically increased the thing it was fighting.
The weight and the price come from different places
Here is the part that even careful coverage gets wrong, and it matters for how much trust the number deserves.
The weight for owners’ equivalent rent does come from homeowners. The Consumer Expenditure Survey asks them, in effect, what they think their home would rent for monthly, unfurnished and without utilities, and the answers set how large the component is.
The price movement does not come from those guesses. Month to month, the component moves with observed rents collected from the same sample of actual rental units used for the tenant rent index, matched to comparable owner-occupied housing.
So the common complaint — that inflation is built on homeowners speculating about their own house prices — is wrong about the mechanism. Homeowners’ estimates size the box. Real leases, signed by real tenants, determine how the contents move.
Whether observed rents are a good proxy for the cost of owning is a fair question, and economists disagree about it. That is a different objection, and a much stronger one.
Shelter arrives late
Rents in the index move slowly, and the reason is contractual rather than statistical.
Most tenants are on leases. In any given month only a small share of them are signing a new one, so the market rate being quoted to new tenants today reaches the index gradually, as existing leases come up for renewal over the following year or more.
The consequence is a systematic lag in the largest component of the index. When market rents turn, the published shelter figure keeps travelling in the old direction for months.
This produces a familiar and genuinely confusing pattern: official inflation still rising while the rental market has already cooled, or still elevated after prices have stabilised. Neither is an error. The index is reporting what tenants collectively pay, not what a new tenant would be quoted.
Why your inflation is not the published number
The headline figure describes an average household that does not exist. Everyone’s basket is their own, and the divergence is not small.
A household that rents in a tight market, drives a long commute and has children in childcare is exposed to three of the fastest-moving categories at once. A homeowner with a fixed mortgage taken out years ago, working from home, is insulated from all three — and the index counts their shelter cost as rising anyway, through an imputed rent they never pay.
Both are accurately measured by an index that describes neither. The practical consequence lands on savings, because the erosion rate that matters for the cash in an emergency fund is the one from your own basket, not the published average.
There is a further wrinkle worth knowing. The published number is seasonally adjusted, and quality changes are accounted for, so that a laptop with twice the memory at the same price registers as a price decline. That is defensible statistics and it is also, reliably, the part people find hardest to accept — because the household still paid the same money.
There is more than one CPI
The figure in the headlines is one of several the agency publishes, and the differences between them are not academic.
CPI-U covers all urban consumers and is the series quoted in the news. CPI-W is narrower, covering urban wage earners and clerical workers, and it sets the annual cost-of-living adjustment to Social Security payments. Retirement income is therefore indexed to a basket built from the spending of working households, weighted differently — less medical care, more commuting.
A third version, the chained CPI, allows for substitution between categories. If beef becomes expensive and shoppers shift to pork, the chained index registers the switch rather than holding the original basket fixed, and it consistently rises more slowly than the standard measure as a result.
Small gaps compound over decades. Which index is used to uprate a payment or a tax threshold is a distributional decision worth a great deal of money, which is why the choice gets fought over in legislation rather than settled quietly by statisticians.
None of that changes what prices did. It changes which measurement is treated as the official one.
The Fed does not target this number
A detail that reframes a lot of coverage: the central bank setting interest rates is not steering by the CPI at all.
The Federal Reserve targets inflation of 2 percent as measured by the price index for personal consumption expenditures, a different series produced by a different agency.
| CPI | PCE | |
|---|---|---|
| Weights | From household spending surveys, updated on a fixed schedule | From business sales data, reweighted as spending shifts |
| Shelter | Around a third of the index | Roughly half that share |
| Scope | What households pay directly | Also includes spending made on their behalf, such as employer-funded healthcare |
Because PCE lets its weights shift as households substitute, and because shelter counts for less in it, PCE inflation usually runs a little below CPI inflation. The gap is not a contradiction and neither series is the corrected version of the other.
It does mean that a headline built on the CPI is describing something the rate-setters are not directly responding to.
What to look at in the monthly release
The headline rate is the least informative number published, because it blends components that behave nothing alike. Three things carry more signal:
- The month-on-month change, not the annual one — the annual figure is dominated by what happened up to eleven months ago
- Shelter separately from everything else, given both its weight and its lag
- Core inflation, which strips out food and energy, not because those do not matter but because they are volatile enough to obscure the trend
And when the published figure conflicts with what a month of spending felt like, the reconciliation is usually one of two things. Either the basket is not the reader’s basket, or the largest component is an imputed rent moving on a year-old signal.
For a related mechanism, see our explainer on how clinical trials work.
