Field report

The Average Has No Kitchen Table

A national number can be accurate, useful, and nowhere near the mix of bills sitting in one home.

The average household does not have an address. It does not stand in a grocery aisle, renew a lease, fill a prescription, or notice that the bus route changed. It is a statistical construction made from many households because a country still needs a coherent way to describe what prices are doing.

That construction is useful. It is also easy to misuse.

When a national inflation number and a household’s experience do not seem to match, public conversation often chooses one of two lazy answers. Either the official number is called fake, or the household is told that its feeling is wrong. A better answer starts with the basket. Different households buy different mixes, and a sound average can move differently from the mix on a particular kitchen table.

A small annual gap can be a real pattern

Bureau of Labor Statistics researchers built a research Consumer Price Index by equivalized-income group for 2006 through 2023. Over that period, average annual inflation for the lowest income quintile was 0.28 percentage point higher than for the highest quintile.

A percentage point is the direct difference between two rates. If one rate were 3.00 percent and another 3.28 percent, the gap would be 0.28 percentage point. The BLS finding is an average across the full 2006–2023 period, not a claim that the gap was exactly 0.28 in every year.

The number looks small because annual rates are usually written small. Its importance is not theatrical size. The research found a persistent difference in the price change of the mixes purchased by two large household groups.

The grouping method matters. BLS ranked households using equivalized income: before-tax household income divided by the square root of household size. That adjustment recognizes that $50,000 supports one person differently from four, without pretending that four people require exactly four times the income.

Then the researchers reweighted detailed CPI components using spending patterns for five income groups. Lower- and higher-income households placed different shares of their spending in categories whose prices did not move identically. The result is not a moral ranking. It is a measurement of different baskets.

Research is not the headline series

The income-group index is called R-CPI-I. The “R” matters: it is a research series, not an official subgroup CPI that replaces CPI-U or the other published headline indexes. BLS warns that it may contain more measurement error than official indexes and that it can be revised.

It also reuses the CPI-U geography, stores, rental units, items, and prices. The researchers did not send a separate national price-collection army to every income group. They changed expenditure weights around a shared price infrastructure. That design makes the comparison possible, but it limits claims about differences in the exact stores, brands, neighborhoods, prices, or product quality experienced by each group.

This is not a reason to discard the research. It is a reason to quote it correctly. “BLS research found an average group difference” is accurate. “The government has calculated your household’s true inflation rate” is not.

CPI never promised to total every pressure

BLS itself says CPI does not quantify every factor that affects the cost of living. It is sometimes described as a conditional cost-of-living index because conditions involving government services, environment, health, education, safety, water quality, crime, and other parts of life are not all priced into one complete measure.

The agency’s language is less dramatic than many arguments built around it. CPI is a powerful, carefully governed measure of consumer price change inside a defined scope. It is not a national anxiety gauge, an affordability guarantee, or a complete household balance sheet.

That boundary is ordinary statistical hygiene. Trouble begins when a scoped measure is used as a full answer to a broader question. “Prices rose by this much for the target population and basket” is different from “your standard of living changed by this much.”

Housing shows the distinction in concrete form. For owner-occupied shelter, CPI measures the service a home provides through rental equivalence rather than pricing the purchase of the house as a consumer good. That method is appropriate to CPI’s consumption purpose. A household trying to enter homeownership is asking about down payments, interest rates, prices, insurance, taxes, and repairs—a different question with a different set of numbers.

A measurement can be right and still be the wrong tool for the job in front of you. A thermometer is not defective because it does not tell you the rent.

The income total has a distribution too

The Bureau of Economic Analysis now publishes distributional personal-income statistics that divide the national aggregate across household groups. Its analysis tool can produce tables by income concept, year, and quantile. In June 2026, BEA also added an income-stratified Personal Consumption Expenditures price index showing inflation rates by income decile.

Those additions are an institutional admission of something simple: a national total and its distribution answer related but different questions. “How much personal income exists?” is not the same as “which households received it?” “What was national consumer inflation?” is not the same as “how did price change differ across ten income groups?”

Distributional statistics do not eliminate the boundary. An income decile still contains millions of different lives. BEA identified its 2024 distribution figures as provisional and its 2025 estimates as experimental nowcasts in the June 2026 release. The statistics combine sources, rely on methods, and can be revised.

Moving from one national average to five quintiles or ten deciles improves the view. It does not produce a household diary. A family can change income group, move, add a child, lose insurance, refinance debt, or alter what it buys. Group data provide context, not a personal invoice.

There is also no contradiction when a group measure and one household move in opposite directions. A national food category can rise while a family changes stores and spends less. A headline shelter measure can ease while a new tenant signs a sharply higher lease. Someone can experience a lower personal price increase and still feel worse because income fell. Price, spending, income, and margin are connected, but they are not interchangeable. Keeping those columns separate makes the story more demanding and far more honest.

Three readings of the same number

The same official number can be used in three very different ways:

  1. Description: “This index rose by this amount for its stated population, basket, and period.” That is the number doing its job.
  2. Comparison: “This research index moved differently across income groups because their expenditure weights differed.” That is a bounded extension with its own method and caveats.
  3. Verdict: “Therefore your household should feel this much pressure.” That leaps from a statistical target to a private life the dataset does not observe.

The first two can be excellent analysis. The third is where the average starts wearing someone else’s shoes.

For a clearer household conversation, name the index, period, population, basket, and excluded question. Then add the relevant distributional measure if one exists. If the question is personal affordability, use the household’s actual recurring costs, income, debt, and margin rather than asking CPI to impersonate them.

The defensible conclusion is not that averages lie. It is that averages require an address label. Headline CPI can accurately describe price change for its target population. BLS research can show that income groups experienced different weighted inflation over a long period. BEA can show how national income and price measures distribute across groups. None of them can sit at one kitchen table and know which bill was postponed.

The sharpest public argument is often the least useful one: official truth against lived truth. The evidence supports a better arrangement. Let the official measure state exactly what it knows. Let the household state what it paid. The interesting work begins where the two stop pretending to be the same record.

Evidence boundary

This report uses BLS methodology and research records and BEA’s June 2026 distributional statistics release. The 0.28-percentage-point finding, equivalized-income method, research-series status, shared CPI price infrastructure, CPI scope, and BEA distributional tools are sourced. The three-reading framework and kitchen-table language are bent.WTF’s bounded interpretation.

The report does not calculate cumulative group loss from the annual average gap, infer a causal mechanism beyond documented expenditure weights, provide a household-specific inflation rate, or claim that every lower-income household experienced higher inflation than every higher-income household. It does not invalidate CPI. It asks the measure to keep its name tag on.

Continue through the record

Keep the source record beside the claim.

The Library keeps definitions, exhibits, caveats, and related work close enough to check.

Start with an active file. Open the exhibits when a claim does real work.

Open the Library