Substitution bias
Substitution bias occurs in a Laspeyres-type price index when it fails to account for consumers switching from expensive goods to cheaper alternatives. By holding the spending basket fixed at an older period, the index overstates inflation — consumers would not actually pay as much as the index implies, because they would substitute toward cheaper substitutes.
The mechanics of substitution bias
A Laspeyres price index calculates today's price of yesterday's basket. The problem: when the price of one good rises relative to another, rational consumers buy less of the expensive good and more of the cheap alternative. A fixed-basket index does not adjust for this — it continues to price the expensive good at its original (higher) weight, treating the consumer as if they made no substitution. The result is a systematic upward bias: the index overstates the cost of maintaining a given standard of living, because the consumer can maintain the same utility at lower cost by substituting.
A classic example: if beef prices rise 20% while chicken prices rise only 5%, consumers buy less beef and more chicken. A fixed-basket CPI that was 50% beef and 50% chicken would record a ~12.5% price increase. But a consumer who shifted to 70% chicken would only face a ~8% increase. The CPI overstates by 4.5 pp in this example. This is substitution bias.
Upper-level vs lower-level substitution bias
Substitution bias operates at two levels:
Upper-level (between categories): Consumers substitute between major expenditure categories — from housing to transport, from food to electronics — when relative prices shift. The traditional US CPI used basket weights updated only every 10 years (now 2 years), meaning it could not capture how spending patterns shifted as housing became more expensive relative to goods. The chain-linking adopted by the BLS and Eurostat's annual HICP weight updates are the primary solution.
Lower-level (within categories): Consumers substitute between items within the same category — from brand A to brand B cereal, from cheddar to mozzarella. The BLS uses geometric averaging within most CPI categories (the geometric mean formula) rather than arithmetic averaging, which provides a first-order correction for within-category substitution. The EU's HICP uses geometric averaging throughout. This is one reason HICP is generally considered a slightly better measure of actual cost-of-living changes than older CPI formulations that used arithmetic averaging throughout.
Why PCE has lower substitution bias than CPI
The Federal Reserve prefers the PCE price index over CPI partly because PCE has lower substitution bias. PCE uses a chain-weighted Fisher ideal index formula — which averages the current and previous period weights — rather than a fixed-weight Laspeyres formula. This means PCE automatically adjusts for substitution between categories on a quarterly basis. Studies suggest PCE runs roughly 0.3–0.5 pp per year below CPI for this reason. Over a decade, this produces meaningfully different assessments of cumulative inflation. The BLS comparison of CPI and PCE quantifies the formula effect in detail. For policy, this means the Fed's 2% PCE target is roughly equivalent to 2.3–2.5% CPI — a distinction that matters when interpreting headline CPI data.