Step 0Money = stuff × price
Start with one good. Spending on bread is the number of loaves times the price of a loaf. So, near enough, growth in spending is growth in loaves plus growth in the price.
That split is the heart of Giles Wilkes's argument about UK growth. The Bank of England steers total spending. The supply side (here, how many loaves the bakeries can make) decides how much of it becomes more stuff and how much becomes higher prices.
With one good and an obvious unit, splitting spending into stuff and prices is easy. Everything below is about what happens when it isn't.
Step 1You can't add loaves to bulbs
Now there are two goods. Pounds add up: £80 of bread plus £20 of bulbs is £100. Loaves and bulbs don't. To add them, statisticians value both at the same set of prices, and the prices they pick change the answer.
Here bulbs get cheaper every year and people buy far more of them. Move the slider to change whose prices are used, and watch real growth change, even though nothing in the world has.
- The base year matters. At early prices bulbs are dear, so the big rise in bulbs bought carries a heavy weight and real growth looks fast. At late prices bulbs are cheap and it looks slow. Same world, same pounds.
- Chain-linking fixes that. ONS values each year's change at the previous year's prices, then multiplies the yearly links together. Growth no longer depends on the year you pick: the solid line just slides up or down to meet spending in that year.
- The price of chain-linking is parts that don't add up. Chain-linked bread plus chain-linked bulbs equals the chain-linked total only in the year you pick and the year after. That is why ONS warns that its chained volume components are not additive.
- Prices come last. The implied deflator is spending ÷ real. Nobody measures it directly: it is whatever is left once real has been built.
So far each good comes in an obvious unit: a loaf, a bulb. The next steps ask what happens when the unit itself is the question.
Step 2What is one unit?
A chocolate bar shrinks from 200 g to 180 g and still costs £1. People buy the same 100 bars. Did anything change?
It depends on what you count. Per bar, no. Per gram, people got less chocolate and paid more for it. The money is the same either way: the unit decides how it splits.
Almost nobody argues that per gram is wrong, and UK consumer price statistics do adjust for pack size. Even per gram is a claim, though: that the last 20 g of a bar are worth as much as the first 20 g. Choosing a unit is ordinary. The next step shows it can be startling.
Step 3The bulb that lasts longer
Between 2008 and 2020, British homes swapped incandescent bulbs for LEDs, pushed along by regulation. An LED gives the same light from about a seventh of the electricity, and lasts about fifteen times as long. So what happened to the amount of lighting people bought?
Three ways to count it:
- Count the stuff: bulbs and kilowatt-hours, each in its own unit.
- Link the LED in: treat the LED as a new product. It joins the price index at its launch price, and its price gap with the old bulb at that moment is counted as quality. Later price cuts count; the jump in value at launch doesn't.
- Count the light: lumen-hours, the thing the bulbs and electricity are for.
- Most of the LED's gain is electricity saved. Counted in kWh, electricity can't show it, so the first two ways miss most of the gain whatever they do with the bulbs. If people take the saving as more light (the default), electricity use hardly changes at all.
- Counting the light dissolves the new-product problem. The LED isn't a new product at all, just a cheaper way to buy an old one: light.
- But that is built into this toy. Its world is generated from a demand for light, so counting light is "right" here by construction. For light that is easy to accept. For bits, later on, it is much harder.
- No industry sells light. Households make it from bulbs and electricity, so the national accounts have no light product to count. Step 4 shows what that means.
- Simplifications: two kinds of bulb (no halogen or CFL); bulbs bought at a steady replacement rate, though real households buy an LED's fifteen years of light in one go; one electricity price for the whole period; illustrative bulb specifications and prices.
Step 4What it does to GDP and inflation
Now put lighting in a whole economy of bread and light. Lighting is a small slice of spending: the toy uses 0.72%, a long-run estimate by Tsao and colleagues. Total spending is held flat, so every change you see comes from how light is counted.
A statistical office can only re-price products it can see: bulbs and electricity. The nearest it can get to counting light is to price each bulb by the light it gives over its life. Whether any of that reaches GDP depends on where the bulbs are made.
- Big for light, small for the economy. Light's weight is its share of spending, so even a sixfold rise in the light people get moves real consumption by about a point and a half over twelve years.
- Equal and opposite. Whatever re-measuring adds to real growth, it takes off prices. Spending, the part the Bank steers, never moves.
- Re-pricing the bulb catches only a sliver. Bulbs are well under a tenth of what light costs; the rest is electricity, and most of the LED's gain is electricity saved. That saving becomes more light inside people's homes. No industry sells light, so the gain sits outside GDP's production boundary. Only counting light itself shows it.
- Made here or abroad. GDP counts what is made in the UK. If bulbs are imported, re-pricing them raises real imports as much as real consumption, and GDP doesn't move. UK telecoms services, by contrast, are sold by UK firms, which is one reason their re-measurement did reach GDP.
- The Wilkes link. Re-measuring an imported good moves consumer prices, the inflation the Bank targets, but barely moves the GDP deflator, which is Wilkes's split. Re-measuring a UK-made product moves the GDP deflator. And because national accounts deflators for consumer goods are largely built from CPI components, a change in how CPI counts something flows into both.
Step 5Where does utility get in? The laptop that got better
So far, choosing a unit looked like a physical question: grams, lumens. It is really a question about what counts as the same thing, and the answer usually rests on what people value.
The "economic approach" to price indices defines the ideal consumer price index in terms of well-being: how much more money would someone need to be exactly as well off as before? Hedonic adjustment was invented to get closer to that. Its founders wanted to link prices "to a yardstick more nearly relevant to its intrinsic utility" (Adelman and Griliches, 1961, quoted by Abdirahman and colleagues, 2020).
Here, five laptop models are on sale each year. The typical one costs about £1,000 every year, while speed doubles every two years. Within any year, faster models cost more; the dial sets how much more.
- One formula, three choices. All three lenses fit price = (that year's level) × speedβ. Per box sets β to 0, so speed is worth nothing. Per unit of speed sets β to 1, so double the speed is double the stuff, like the lumen and the byte. Hedonic reads β from the slope of prices across the models on sale.
- Every lens is an assumption about what counts as the same thing. Per box says speed is worthless. Per unit of speed says value rises in proportion to speed. Per gram (Step 2) says the last 20 g are worth as much as the first. Per byte (Step 7) will say a byte of phone call is worth a byte of video.
- Value to the buyer, or cost to the maker? For consumer prices the textbook test of "the same thing" is value to the buyer. For producer prices, including the telecoms deflator in Step 7, it is the cost of making it (Triplett, 1983).
- Market prices aren't pure utility. The slope reflects what buyers will pay at the margin, what sellers need, and mark-ups where firms have market power, which economists have long argued about (Rosen, 1974; Pakes, 2003). Here one dial stands in for all of that. And the statistician still chooses which characteristics count: real laptops differ in memory, screen, battery and weight, and leaving one out biases the index.
- Expectations move. Abdirahman and colleagues note that someone might be "equally satisfied in 2012 and 2017, given what is available on the market and her (socially-influenced) expectations". If software gets heavier as hardware gets faster, part of the gain is relative.
- Matched models. Statistical offices mostly track the same model's price over time. In this toy every model sits on its year's price curve, so that gives the hedonic answer. The two part company when quality arrives mainly through new models launched at old price points.
- Laptops are imported. In the UK, re-measuring them moves consumer prices and real consumption, not GDP (Step 4).
Step 6When does volume track welfare?
Back to light, and one question: how much better off are people because light got cheaper? The toy knows, because its world has a stated demand curve for light. The gain is the area under that curve between the old and the new price of light, what economists call the change in consumer surplus.
Here light is counted as a product (the household view from Step 3), and lighting is 0.72% of all spending. How close do official-style measures get?
- With the right unit, chain-linked growth tracks the gain. This runs against a common claim that GDP misses all the benefit of things getting cheaper. Chain-linking values each year's extra light at prices from that year or the one before, which is close to the right value for that year's small change. A chained Fisher index lands almost exactly on the true gain, and the ONS's chained Laspeyres a little above it.
- The two fixed-base answers bracket the truth. Valuing today's light at the old price overstates the gain; valuing the old light at today's price understates it. That is the textbook result, and it rests on the assumptions below.
- Where volume and welfare part company:
- The wrong unit. Counting bulbs and kWh records nothing while people are better off. Over two centuries this is Nordhaus's (1996) argument: conventional price indices badly overstated the price of light since the early 1800s, so real wages and output growth were understated.
- Brand-new goods. When something genuinely new appears, the gain between what people would have paid for it and its launch price is never recorded once it is linked in.
- Free goods. A zero price means zero weight (see the coda).
- Total value. GDP measures changes at the margin, not what light is worth altogether: Adam Smith's water, cheap but priceless. Under this toy's default demand curve that total is infinite, which is why nothing here tries to measure it.
- The assumptions behind the bracket: preferences that scale with income, people buying what suits them at the prices they face, and all output consumed. If light is a necessity, or prices change a lot, the link loosens.
- A history check on the toy. The second chart is the real UK price of light: down over 99.9% since 1800. Look at its last decade, though. Fouquet and Pearson's series covers all UK lighting, including offices and streets already lit by efficient tubes, and electricity got dearer, so the average price of light fell only about a quarter over 2008–2020. The toy's sixfold fall is the household bulb swap alone, at a fixed electricity price: an upper-end illustration.
Step 7Bits: UK telecoms, 2010–2017
Between 2010 and 2017 the data moved over UK networks grew about 24-fold. Yet the official statistics of the time had the telecoms industry's real value added falling, with some of the slowest productivity growth of any industry (Abdirahman, Coyle, Heys and Stewart, 2020).
Revenue is not in doubt: Ofcom's figures have it slipping from £40.5bn to £38.1bn. The question is how to split it into stuff and prices, and that depends on what one unit of telecoms is. Diane Coyle and colleagues produced six answers, each a published price index.
- From economist to engineer. The old official index priced a basket of calls, texts and line rental. The improved and "A" options add broadband and mobile data as separate products, weighted by what people spend on each. Price per byte treats everything as one product: bytes moved.
- ONS did not adopt price per byte. In Blue Book 2021 it adopted option A.1's method: unit values (revenue ÷ quantity) for voice minutes, texts and terabytes of data, weighted by revenue. "Per byte" applies within data, not across services. ONS's own series, built with later data, may differ from the paper's.
- Price per byte makes real output equal bytes moved, by definition: real = revenue ÷ (revenue ÷ bytes). It treats a switch from expensive voice bytes to cheap video bytes as a price fall, which assumes a byte of phone call is worth a byte of video: Step 5's question again.
- An arguable reading. Triplett's rule says output prices should judge quality by the cost of making it. Carrying a byte costs a network roughly the same whatever it carries, which arguably favours the engineer for an output deflator. The authors expected prices per byte to converge, which "would make a data usage based unit value index a more meaningful proxy deflator".
- This is real output, not value added. Telecoms firms buy inputs, and many of their customers are other firms. Step 8 follows that through to GDP.
Step 8Bits are mostly an input
Much of what telecoms firms sell goes to other businesses: a bank's network, a shop's card machines. So re-pricing telecoms changes two sets of figures: telecoms, which sells it, and every industry that buys it.
Since Blue Book 2021, ONS has used double deflation: each industry's real value added is its real output minus its real inputs. Before that, an industry's real value added simply followed its real output (single deflation).
- Huge in the industry, modest in GDP. Two reasons. Telecoms is a small share of the economy (Step 4's lesson). And the part bought by other industries cancels out: cheaper telecoms raises the buyers' real inputs and cuts their real value added by the same amount. Only what households buy (in a fuller model, also exports and investment) reaches GDP. Set the household share to zero and re-pricing telecoms doesn't move GDP at all.
- Double deflation amplifies. Real value added is a difference: real output minus real inputs. When output prices fall faster than input prices, value added grows far faster than output. That is a known weakness of the method: where the margin is thin, double-deflated value added swings wildly.
- Single deflation overstates the GDP effect. With nothing to offset telecoms' gain, GDP takes all of it, and adding industries up this way need not match real spending.
- What ONS reported. Its Blue Book 2021 article found much faster real growth in telecoms after the change, lower growth in industries that buy telecoms (finance and insurance, and gambling and betting, were the most affected) and a modest upward revision to GDP growth.
CodaComputing, cloud and free goods
- The cost of computation has kept falling. Coyle and Hampton (2024) find the decline slowed after 2010 but not before, while the productivity slowdown they ask about began in the mid-2000s. They put any link to it in complementary inputs such as skills and organisation, not in computing itself.
- Cloud computing has its own price indices (Byrne, Corrado and Sichel; Coyle and Nguyen for the UK), and they fall fast too.
- Where the price to the user is zero (search, maps, social media), there is no spending to weight at all. GDP records the advertising that pays for these services, as a cost to advertisers, not their value to users. That is the end of this ladder: the unit question can no longer be answered with a price.
Further reading: Diane Coyle, GDP: A Brief but Affectionate History and The Measure of Progress; Brynjolfsson and colleagues on measuring free goods.