By almost every measure economists watch, this is a good economy. Unemployment is 4.2%. Retail sales are running 5.6% ahead of last year. High-yield credit spreads sit near the low end of their range, the yield curve is positively sloped, and the St. Louis Fed’s financial stress index is below normal. Six of the eight recession tripwires we track read benign.
And consumer sentiment just printed 44.8 — a record low for the University of Michigan index. It has since recovered to 49.5. The pre-2021 average was 92.9.
That gap gets explained away a lot: people are misinformed, or partisan, or just miserable. This week we went looking for a duller explanation, and found one. People are describing a different basket of goods than the one in the headline — and theirs went up more.
Inflation coming down does not mean prices came down. This is the part that gets lost, and it isn’t a small point. The inflation rate is the speed prices are rising. When it falls from 9% to 3%, prices are still rising — just more slowly. Nothing gives anything back. The rate has retraced most of the ground it lost since 2022. The price level has retraced none of it, because that isn’t a thing price levels do.
Since January 2020, the overall price level is up 28.3%. That is the number nobody quotes and everybody pays.
And it is not spread evenly. Here is what actually happened, category by category, since January 2020 — with the headline marked so you can see which things beat it.
Everything a household transacts in weekly sits to the right of that line. Groceries are up 32%. Eating out is up 36.8%. Gasoline is up 41.1%. Electricity is up 43.2%. The published figure — all items, 28.3% — is below every one of them. Core inflation, the measure that strips out food and energy specifically because they are volatile, is lower still at 26%.
Food is the one that does the work here. Not because it rose most — electricity and gasoline beat it — but because you cannot defer it. A car purchase waits. A doctor’s visit waits. Dinner does not. Groceries are about 8% of the price index and roughly 150 shopping trips a year, and every one of those trips is a small reminder priced in real money.
Then there is the largest single component of the index, and almost nobody pays it.
Owners’ equivalent rent is 26.2% of the Consumer Price Index — more than a quarter of the whole thing. It is the Bureau of Labor Statistics’ estimate of what a homeowner would pay to rent their own house. No such payment exists. No check is written. For a household with a fixed mortgage taken out in 2019, that quarter of the inflation number corresponds to a cost that has not changed at all.
This is not a scandal and it is not hidden. BLS documents the method openly, and the logic is defensible — homeowners genuinely do consume housing services, and leaving that out would understate the cost of living in a different direction. But the effect is real: the most heavily weighted piece of the published number is an estimate of a transaction that never occurs, while the categories people meet several times a week make up a much smaller share.
Put a household’s actual weekly basket together and the gap is measurable. Groceries, eating out, gasoline and electricity together are 19.1% of the index. Weighted by their real shares, that basket is up 36.2% since January 2020, against a headline of 28.3%.
Almost eight percentage points, between what gets published and what people touch most often.
That is not a perception problem. It is two accurate measurements of two different things.
One complication, because it cuts against the easy version of this story. Average hourly earnings are up 32.4% since January 2020, against prices up 28.3%. On that measure, wages did not fall behind — they ran about four points ahead. The simple claim that people are poorer than in 2020 is not supported by the aggregate data, and we are not going to make it.
What we can say is narrower and less satisfying: an average is not a household. Average hourly earnings is a mean across the whole workforce, and it tells you nothing about who kept up and who didn’t. And the cushion is thin right now — wage growth and inflation are currently running within a rounding error of each other, so whatever ground was gained is not currently being extended.
One thing worth understanding. Every measurement has a blind spot, and this week’s is the difference between what a number measures and what it is used to explain.
The price index is built to measure the cost of a representative basket, and it does that well. It was never built to predict how people would feel, and the mismatch we have described is a known effect with a name — economists call it frequency bias, and it has been demonstrated experimentally and in work on how grocery prices shape inflation expectations. People weight what they buy often far more heavily than the basket weights justify. That is not a mistake on their part. It is what having a memory of prices does.
We should also name what this analysis cannot see. It cannot see distribution — whether the household in question kept up. It cannot separate the effect of prices from the effect of an election year, and consumer sentiment surveys have a large, well-documented partisan split that this panel has no way to measure. And nothing here is a forecast. We have described what prices did and what a survey said. We have not established that one causes the other, and the honest position is that both are true at once and the arrow between them is not something this data can draw.
Filter Lab is descriptive analysis, not investment advice. Price data from the Bureau of Labor Statistics via FRED, seasonally adjusted, January 2020 base, latest observation June 2026. CPI relative importance weights are BLS, December 2025 reference. Consumer sentiment is the University of Michigan index. Prices, not budget shares — the category figures describe individual prices and do not sum to a cost of living. No price targets, no ratings, no buy or sell.