Consider a deceptively simple question: was an English labourer in 1750 better off than one in 1550? The historian's instinct is to gather wage series, deflate by a price index, and report real wages. Yet decades of scholarship have produced estimates that diverge by factors of two or three for the same period, using overlapping data. The disagreement is not empirical carelessness—it is structural.

The problem lies in index number theory itself. A price index is not a neutral measurement instrument; it is a construct that embeds assumptions about consumer behaviour, product identity, and the stability of preferences. When we compress the price movements of hundreds of heterogeneous goods into a single scalar, we necessarily discard information. The question is which information we choose to discard, and whether our choices systematically bias comparisons across long time horizons.

This matters because much of what we claim to know about the Industrial Revolution, the Great Divergence, and pre-modern living standards rests on real wage series. Robert Allen's high wage economy hypothesis, Gregory Clark's Malthusian model, and Bob Allen versus Jan Luiten van Zanden debates over European welfare ratios all hinge on methodological choices about index construction. Understanding these choices is not a technical footnote—it is prerequisite to reading the literature critically.

The foundational tension in price index construction was formalised by Irving Fisher in 1922, though the practical implications for historians remain underappreciated. A Laspeyres index, which fixes the consumption basket at base-period quantities, systematically overstates the cost of living over time because it ignores consumer substitution away from goods whose relative prices rise. A Paasche index, fixing quantities at current-period levels, understates it symmetrically.

For historical work spanning centuries, the divergence between these bounds can be substantial. Consider English data from 1500 to 1850: computing a real wage series using a Laspeyres index anchored in 1500 versus a Paasche index anchored in 1850 can yield welfare trajectories that differ by 30 to 50 percent at the endpoints. Neither is wrong; they answer different questions about hypothetical consumers.

The Fisher ideal index, computed as the geometric mean of Laspeyres and Paasche, satisfies the time reversal test and approximates the theoretically preferred Divisia index under standard assumptions. Yet applying it to historical data requires quantity information at multiple benchmark dates—information that survives patchily for pre-industrial economies. Most historical price indices are therefore Laspeyres-type by necessity, with the well-known upward bias baked in.

The problem compounds when relative prices shift dramatically, as they did during the transition from organic to mineral economies. Coal, cotton textiles, and refined sugar experienced order-of-magnitude price declines relative to grains between 1700 and 1850. A basket fixed in 1700 quantities weights these goods trivially; a basket fixed in 1850 weights them heavily. The choice determines whether we measure the Industrial Revolution as a modest or transformative event.

Chain-linking indices at short intervals mitigates but does not eliminate the problem, and introduces new pathologies including chain drift under non-monotonic price movements. There is no purely technical resolution; the choice reflects a substantive judgement about which counterfactual consumer we wish to track.

Takeaway

A price index is not a measurement but an argument. When you read a real wage series, ask which index formula was used and whether the answer would survive a different but equally defensible choice.

Price indices assume that we are pricing the same good over time. This assumption fails almost everywhere it matters. A pound of sugar in 1600 was a luxury refined by artisanal methods and varying substantially in purity; by 1900 it was a standardised industrial commodity. The nominal price fell dramatically, but part of that decline reflects the good itself becoming a different object.

Modern statistical agencies address this through hedonic regression, decomposing observed prices into implicit valuations of underlying characteristics. Applied to housing, computers, or automobiles, hedonic methods can separate the pure price effect from quality improvements. Applied historically, the technique demands characteristic data that rarely survives—we know sugar prices, but seldom the precise refinement standards or contamination levels of individual transactions.

The bias runs predominantly in one direction. Unmeasured quality improvement means conventional indices understate the real gains in living standards. Nordhaus estimated that a lumen-hour of light cost roughly one thousand times more in 1800 than in 1992, once technological improvements from tallow candles to fluorescent tubes are properly counted. Standard price indices capture perhaps a fraction of this decline because they track the price of candles, then lamp oil, then electricity as separate categories rather than as substitutes for the same underlying service.

New goods present the sharpest version of the problem. When a product does not exist in the base period, its introduction cannot be captured within a fixed-basket framework at all. The welfare gain from the introduction of the potato in early modern Europe, or of tea and coffee, is invisible to indices that only track continuously available goods. Hausman's reservation price approach offers a partial solution but requires demand elasticity estimates that historical data cannot always support.

The implication is that long-run real wage series almost certainly understate the growth of material welfare, and that the understatement is largest precisely for the periods and populations experiencing the most rapid consumption transformation.

Takeaway

What looks like price stability may conceal profound quality change, and what looks like a stagnant living standard may hide the entry of goods that would have been unimaginable luxuries a generation earlier.

Every price index requires a consumption basket, and every basket embeds assumptions about who the representative consumer is. Allen's respectability basket and bare-bones subsistence basket, now standard in comparative welfare studies, illustrate the stakes. The same wage data yields different welfare ratios depending on whether we assume workers consumed bread and beer or oats and cheap fats.

Historical baskets face an inescapable circularity. We construct them from surviving budget studies, institutional accounts, or backward projection from later evidence. Yet the composition of consumption is itself endogenous to relative prices and incomes. A basket that describes actual behaviour in 1600 may be a poor guide to what workers would have consumed in 1800 had prices remained constant, and vice versa.

The problem is particularly acute for goods with high income elasticity. Meat consumption, sugar, imported textiles, and manufactured household items all show budget shares that vary substantially with income. Fixing a basket based on institutional records from workhouses or naval provisions builds in a specific income assumption that may fit some populations and periods badly. Van Zanden's critique of Allen's welfare ratios turns partly on this point.

Regional variation compounds the difficulty. A basket calibrated to English consumption imperfectly represents Mediterranean or Baltic patterns, where olive oil, wine, rye, or fish figured differently. Comparative international series therefore require either separate baskets, sacrificing comparability, or common baskets, sacrificing representativeness. Neither choice is neutral, and the estimated Great Divergence looks somewhat different under each.

Sensitivity analysis offers the honest response. Rather than presenting a single real wage estimate, researchers should report a range across plausible basket specifications. When conclusions are robust across specifications, we can hold them with confidence; when they are not, the disagreement is the finding.

Takeaway

The representative consumer is a fiction, and useful fictions require explicit acknowledgement of what they represent and what they exclude. No single basket captures a heterogeneous population.

The methodological problems surveyed here are not reasons to abandon quantitative history. They are reasons to practice it more carefully. Real wage series remain among the most valuable empirical tools we possess for understanding long-run economic change, but their interpretation requires understanding what the numbers can and cannot say.

The productive path forward involves triangulation: comparing indices constructed under different assumptions, cross-checking wage-based welfare measures against anthropometric evidence, mortality patterns, and household-level budget data. When multiple independent methodologies converge, we have knowledge; when they diverge, we have identified where more research is needed.

The cost of living index problem is ultimately a problem of aggregation across heterogeneous experiences. Recognising this pushes historians toward disaggregation—separate series for different social groups, regions, and consumption profiles—rather than the pursuit of a single number that never existed in the world.