From 'prices rose' to a single number
In the last guide you met inflation as a sustained rise in the [[price-level|price level]] — not this or that price, but prices in general. That definition is honest but slippery. *Whose* prices? Bread fell while rent soared; your phone got cheaper while your haircut got dearer. To say anything useful, we need to squeeze the whole noisy cloud of millions of price tags into one comparable figure — and then watch how *that* number moves over time. The most famous attempt is the [[consumer-price-index|consumer price index]], or CPI, the number you see in the news when a country reports 'inflation came in at 3.2 percent.'
The trick is to stop chasing every price and instead follow one carefully chosen shopping basket — a fixed list of the goods and services a typical household actually buys, in roughly the amounts they buy them. Each month, government statisticians send price-collectors (and now web-scrapers) to record the cost of that exact basket: the rent, the loaf, the litre of petrol, the streaming subscription, the bus fare. The CPI is simply the cost of buying the same basket today, expressed relative to its cost in a chosen base period set equal to 100. If the basket that cost 100 in the base year now costs 112, the index is 112, and the price level has risen 12 percent since then.
Building the basket, weighting the basket
A basket is not a flat list — its items carry weights. Housing eats a huge share of a household's budget, so a 5 percent rise in rent moves the CPI far more than a 5 percent rise in the price of salt. Statisticians learn these weights from household-spending surveys: roughly a third of the typical basket might be housing, a sixth food, a tenth transport, and so on. The headline inflation number is therefore a *weighted average* of thousands of individual price changes — each price change pulling on the total in proportion to how much families actually spend on it.
A toy 3-item basket (base-year quantities, fixed) Item Qty Base price Base cost This-year price New cost Bread 20 $2 $40 $2.20 $44 Rent 1 $600 $600 $630 $630 Bus 40 $1.50 $60 $1.80 $72 ---------------------------------------------------------------------- Total $700 $746 CPI now = (746 / 700) x 100 = 106.6 Inflation since base year = 6.6% Note: rent is ~86% of the basket's cost, so its 5% rise dominates -- bread's 10% jump barely registers in the total.
Here is the subtle, important choice baked into the table above: we held the *quantities* fixed at base-year levels and only let prices change. That is what makes it a clean price comparison — like the fixed base-year prices that turned nominal into real GDP, only flipped: there we froze prices to see quantities, here we freeze quantities to see prices. Holding the basket still is exactly what lets the index isolate inflation. But, as the next section shows, a frozen basket is also where the CPI's honest flaws are born.
The biases: why the CPI overstates inflation
A fixed basket is a fiction, and economists know it. The most studied flaw is substitution bias. When beef gets dear, real households buy more chicken; the index, clinging to the base-year basket, keeps pricing the old, now-expensive beef as if shoppers never flinched. By ignoring the very switching that people do to dodge price rises, a strictly fixed basket overstates the true rise in the cost of living. The same logic applies across shops — if a discounter opens and families flock to it, the index can miss the saving entirely.
The thornier flaw is quality-change and new-goods bias. A laptop today costs about what one cost a decade ago, but it is vastly faster — so part of that 'same price' is really a price *cut* per unit of computing you get. If the index records the dollar price as unchanged, it misses a quality improvement that made you better off, again overstating inflation. New goods are harder still: a product that did not exist in the base year (the smartphone, the mRNA vaccine) simply has no slot in the old basket, so the enormous value of its arrival is invisible until statisticians fold it in years later. Untangling 'the price went up' from 'the thing got better' is one of the genuinely hard, contested problems in the whole field.
These are not idle worries. The famous 1996 Boskin Commission in the United States estimated the CPI overstated inflation by roughly 1.1 percentage points a year — a gap that, compounded over decades, is enormous for anything tied to the index. Statistical agencies have fought back with chained indices that update the basket far more often (letting substitution show up), and with painstaking 'hedonic' methods that strip out quality gains. The biases are smaller now, but no one claims they are zero. Be honest about this: the CPI is a careful estimate, not a thermometer reading — and it most likely still leans toward overstatement.
Two rulers: CPI versus the GDP deflator
The CPI is not the only price ruler. From the macro guides you already met the [[gdp-deflator|GDP deflator]] — nominal GDP divided by real GDP, times 100 — a price index built straight out of the national accounts. The two answer subtly different questions. The CPI asks, 'what is happening to the cost of the things a household *buys*?' The deflator asks, 'what is happening to the prices of the things this country *produces*?' Those baskets are not the same.
Three differences fall out of that. First, imports: a pricier imported car shows up in the CPI (households buy it) but *not* in the deflator (the country did not produce it). Second, capital and government goods: factory machinery, tanks, and bridges are in the deflator but never in a consumer's shopping cart. Third, the basket itself: the CPI clings to a fixed basket and so suffers substitution bias, while the deflator's basket re-weights automatically each year toward whatever the economy actually made — dodging that particular bias but introducing its own quirks. In an oil shock, where energy is heavily imported, the CPI can spike while the deflator stays calmer; the gap between them is itself informative.
Why the number bites: indexing wages and pensions
The CPI would be a curiosity if it only sat in newspapers. Its real power is that it is *wired into contracts*. [[indexation|Indexation]] means automatically adjusting a payment to track the price index — so its real value, its purchasing power, holds steady as prices rise. Many pensions, some union wages, social-security benefits, rent caps, and inflation-linked government bonds all rise each year by the measured CPI. When the index ticks up 3 percent, a pension cheque indexed to it rises 3 percent — protecting the retiree from quietly growing poorer.
This is precisely why the biases above are not academic nitpicking. Suppose, as the Boskin Commission argued, the CPI overstates true inflation by 1 percentage point a year. Then every indexed pension quietly rises 1 percent *faster* than the cost of living it is meant to track — a small over-payment that compounds into vast sums across an entire nation's retirees over decades, paid out of public budgets. Trim the measured number and you cut those payments; inflate it and you raise them. The dry question of how to handle a new smartphone in the basket becomes, three steps later, a multi-billion-dollar fight over the public purse. Measurement is never politically neutral.
Indexing carries a deeper risk too. If wages, pensions, and prices are *all* mechanically tied to last year's inflation, a one-off price shock can lock in and feed on itself: prices rise, indexed wages rise to match, higher wages push prices up again — a self-fuelling spiral that helped make the inflation of the 1970s so stubborn. So indexing is a double-edged tool. It is the humane way to keep promises in real terms rather than nominal ones, yet widespread automatic indexing can make inflation harder to stop once it starts. Like much in this rung, the honest verdict is 'it depends' — and the next guides turn from measuring inflation to reckoning with what it actually does to people.