Whose inflation is it anyway? How the number is made, and why it never feels like yours

Every month the Office for National Statistics publishes a single number that stands in for the cost of living, and every month a large share of the country does not believe it. In August, consumer price inflation was 3.1%. Asked in the same month what they thought inflation was, the public's median answer was 4.9%. Asked whether their own cost of living had gone up in the past month, 56% said yes, and 94% of those blamed the price of food, whose official inflation rate was 1.3%, the lowest since 2021.¹¹⁸¹⁹

Someone must be wrong. The usual explanations are that the public cannot do arithmetic or that the statisticians are cooking the books. Neither survives contact with the evidence. What the evidence shows is that the official measure answers a narrower question than most people think it does, that the question it answers is not the one that matters to a household, and that for several years the difference between the two was large and fell hardest on the people with the least room to absorb it. This piece explains how the number is built, what it leaves out, why it feels wrong, and what a better version would look like.

How the number is made

The Consumer Prices Index tracks a basket of 760 goods and services, revised each March. This year 27 items went in, among them houmous, alcohol-free beer, pet grooming, motorhomes and dashboard cameras, and 19 came out, including sheets of wrapping paper, which the ONS found hard to price consistently. The choice is not scientific in the strict sense; the ONS says it rests on judgement about what people spend money on, whether an item is available all year, and whether its price behaves like others in its category.⁵

Prices for those items are collected in about 180,000 quotations a month from around 20,000 shops in 150 locations, plus central collection for items priced nationally. Since March, the largest single change in the index's history has been under way: for roughly half the grocery market, the ONS now uses supermarket scanner data, about 300 million price points a month drawn from more than a billion units sold, in place of 25,000 prices noted by hand. Scanner data records what people actually paid, including promotions, and tracks price per unit of weight, which is how the statistics catch shrinkflation. Early testing suggested the switch changes the headline rate by a few hundredths of a point on average.⁷⁸⁹

The item price changes are then weighted by how much the country spends on each category, using national accounts data from two years earlier. This is the step that matters most for the question in this article, and it is worth looking at what those weights are.⁶

Two things stand out. Food is 11% of CPI, which will surprise anyone whose weekly shop is the biggest bill they notice. And recreation, restaurants and hotels together are 29%, larger than food and housing combined. That is not a mistake. It reflects what the country as a whole spends, and the country as a whole includes a lot of people eating out and going on holiday. The weights are what statisticians call plutocratic: a household that spends £100,000 a year counts ten times as much as one that spends £10,000. The index is a pound-weighted average, not a person-weighted one.⁶¹¹

The rate is falling. The level is not

The single largest source of confusion is not statistical at all. Inflation is a rate of change. When it falls from 11% to 3%, prices are still rising, just more slowly; nothing has got cheaper. People, reasonably, remember what things cost before, and the honest comparison is with the level. On the ONS's own index, the CPI basket cost 28% more in August 2026 than in August 2021. Food cost 38% more. Real average pay, after inflation, was up 1.7% over the same five years. When a politician says inflation is back under control and a shopper says everything costs a fortune, they are both reading the same chart, one looking at its slope and one at its height.²³⁴

Does it describe the average person? No, and it is not meant to

Because the weights are pound-weighted national averages, CPI describes a household that does not exist. The ONS knows this, and since 2017 has published a second family of measures, the Household Costs Indices, built the other way round: each household's spending pattern counts equally, and costs are counted when they are paid rather than when a good is acquired, which brings in mortgage interest, student loan repayments and insurance premiums. The HCIs are broken down by income decile, tenure, retirement and children. They are still classed as official statistics in development, and they are published quarterly and three months in arrears, which is why nobody quotes them on the news.¹⁰¹¹

What they show is that the gap between rich and poor is real, but that it opens and closes with energy and food. In the year to March 2026, the latest available, all-household costs rose 3.6% against CPI's 3.3%, with low-income households (the second decile) at 3.7% and high-income (the ninth) at 3.5%: a gap, but a small one. At the peak of the crisis in October 2022 the same two groups faced 13.5% and 11.5%, the widest gap the series has recorded, because gas and electricity are a far larger share of a poor household's budget. The Resolution Foundation put the poorest tenth at 11.9% and the richest at 9.2% in early 2023, and in April this year projected the gap opening again to 3.8% against 2.9% by the end of 2026 as energy bills climb, noting that the second-poorest decile spends 11% of its income on energy and the second-richest 6%.^10121314

There is a subtler effect that the official measures only partly capture, and it goes to the heart of why the poor felt the crisis more. Research by Xavier Jaravel and Martin O'Connell at the Institute for Fiscal Studies, using scanner data on tens of thousands of grocery products, found that between 2021 and 2023 the cheapest varieties within each category, the value pasta and the own-label butter, rose in price far faster than the premium versions. They called it cheapflation. Households in the bottom quarter of grocery spending faced food inflation 5.6 percentage points higher than the top quarter, roughly £100 a year on a shopping basket, and almost all of it came from which products people bought within a category rather than which categories they bought. The ONS's own experimental tracker of the cheapest supermarket lines found them up 17% in the year to September 2022, with vegetable oil up 46% in five months. That tracker stopped in 2022.¹⁵¹⁶¹⁷

Housing: the biggest blind spot

If you own your home with a mortgage, the official inflation measures have almost nothing to say about your largest cost. CPI, the number the Bank of England targets, excludes owner-occupiers' housing costs entirely. CPIH, the ONS's preferred headline measure, includes them, but through a method called rental equivalence: it estimates what you would pay to rent a home like yours and treats that as the price of the housing you consume. Mortgage interest is excluded from both, on the argument that a house is partly an asset and that interest is a financing cost rather than a price. Statistically that is defensible. To the household whose monthly payment rose by several hundred pounds when rates went up in 2023, it means the index recorded nothing.⁷²²

Renters fare a little better, because actual rents are in the index. Private rents rose 3.8% in the year to August across the UK and 4.0% in England, where the average is now £1,459 a month, and renters were the group with the highest inflation in the latest Household Costs Indices. But rents are a small share of the CPI basket, because the basket is weighted by all households, most of whom do not rent. For a young private tenant spending 40% of income on rent, the official number is a poor guide. The HCIs, which do include mortgage interest, were built to answer exactly this. The trouble is that they arrive late and carry a health warning.¹⁰²³

Why it feels higher than it is

Even allowing for all of that, people overestimate current inflation. The Bank of England's survey has asked the same question for decades and the median answer has run above the official rate in almost every period. In August the public put inflation at 4.9%, at a time when CPI was running at about 3%. The Bank switched survey provider this year, which it warns makes comparisons with earlier rounds unreliable, but the perception figure was unchanged across both providers in a parallel run.¹⁸

The research on why is fairly settled. People judge inflation from the prices they see most often, and the items they buy most often, food and fuel, are a small share of the basket but almost all of the experience. In an experiment in which subjects were shown price changes and asked to estimate overall inflation, estimates tracked the frequently purchased items and largely ignored the rest. People also notice rises more than falls, and remember them longer. Motor fuel is up 23% on a year ago, which every driver has seen on a forecourt sign; the items whose prices fell over the year are noticed by nobody. The ONS's own survey finds that when people say their cost of living has risen, 94% cite food, more than any other item, in a month when official food inflation was 1.3%.¹¹⁹²⁰

Shrinkflation feeds the same suspicion. Eight in ten adults told YouGov last year they were concerned about it, up from three-quarters in 2023, and three-quarters of those said they had noticed it in crisps and confectionery. The statistics do try to catch it: when a product's size changes, the ONS adjusts the price per unit, and the new scanner data does this automatically by pricing per gram. But the adjustment only happens when the change is detected, and the public's impression that they are paying the same for less is, item by item, often accurate. The same is true of quality adjustment in the other direction: when a better phone replaces an older one at the same price, the index records a price fall, which is right in theory and invisible in a wallet.⁷⁸²¹

Is the number trustworthy?

The fair answer is that it is honestly produced, occasionally wrong, and never revised. In June last year the ONS admitted that faulty vehicle tax data from the Department for Transport had overstated April's CPI by 0.1 percentage points; under its long-standing policy the published figure was left uncorrected, and the ONS apologised and said it would review how it checks data supplied by other departments. This month's release carries a note about a classification error in the new scanner data affecting some sub-categories by up to 0.2 points, though not the headline. The Office for Statistics Regulation's 2026 review of the statistical system describes an ONS still working through a recovery plan for the quality of its economic statistics. None of this suggests the number is manipulated. It does suggest the public is right to treat one decimal place with some scepticism.¹²⁵²⁶

The one measure that genuinely misleads is the oldest. The Retail Prices Index, still used to uprate some rail fares, student loan interest and index-linked gilts, was declared "not a good measure, at times significantly overestimating inflation and at other times underestimating it" by the then chair of the UK Statistics Authority, Sir David Norgrove, in 2019, and it lost its accredited status years before that. It runs above CPI, 3.4% against 3.1% in August, because of a flaw in how it averages prices. Governments have kept it alive because index-linked gilts pay out on it and changing the formula would cost bondholders; the fix is finally due in February 2030, when RPI will adopt CPIH's methods.¹²⁴

What a better measure would look like

Nobody serious proposes replacing CPI. Central banks need a consistent, timely, economy-wide price index, and CPI is one of the better ones in the world. The argument is about what sits beside it. Three changes would close most of the gap between the statistic and the experience. First, publish the Household Costs Indices monthly, alongside CPI, and promote them out of development; a headline that reads "inflation for the poorest fifth was 3.9% and for the richest 3.2%" would end the argument about whether the figure reflects ordinary people, because it would say. Second, revive the tracker of the cheapest grocery lines, which the scanner data now makes trivial to produce, so that cheapflation is visible as it happens rather than two years later in an academic paper. Third, report the price level as routinely as the rate, so that "prices are 28% above 2021" appears in the same sentence as "inflation is 3.1%".²¹¹¹⁷

So, does inflation reflect the average person? It reflects the average pound, which is a different thing, and it does so well. Do people feel price rises higher than the figure says? Yes, and for three reasons that are all partly justified: they are remembering levels while the figure reports rates; they are weighting their own basket, which for anyone poor, renting or paying a mortgage is heavier in the things that rose most; and they are noticing the frequent purchases that did rise while forgetting the occasional ones that fell. The number is not lying. It is answering a question most people never asked.¹⁰¹⁸