A price index looks simple because its final form is simple: 100, 102.4, 115.7. Yet those numbers are not prices. They are compressed statements about how a selected collection of prices has changed relative to an earlier reference, using a particular set of weights, definitions and linking rules.

Index numbers are measurement systems for relative change. They convert many unlike movements into a common scale so that a reader can ask whether a basket of household consumption, producer output, trade prices or quantities has risen, fallen or shifted over time. Their usefulness comes from compression. Their danger also comes from compression: the single headline can hide whose basket was measured, which products changed, how quality was treated, which weights were used and whether the underlying pattern differs across households or sectors.

This article owns the methodological question of how index numbers are built and interpreted. It does not replace the wider institutional owner How Official Statistics Work, nor the Singapore-specific publications of the Singapore Department of Statistics. Instead it opens the machinery beneath familiar numbers such as consumer price indices.

Reading route: begin with why index numbers measure relatives rather than absolute prices, then examine weights, index formulas, chaining and rebasing, quality change, new data sources and finally how to interpret inflation without asking the index to answer a question it was not designed to answer.

1. An index number measures change, not the price level itself

Suppose a loaf of bread costs $2.00 in one year and $2.20 in the next. Its price relative is 2.20 ÷ 2.00 = 1.10, or 110 when expressed on a base of 100. The index is telling us that the price is ten per cent above the base-period price.

Now suppose another product moves from $200 to $210. Its price relative is 105. The second product is still far more expensive in dollars, but its relative increase is smaller. Index numbers deliberately separate the level of a price from its movement.

This is why Singapore’s official CPI documentation states that the CPI measures price movements rather than absolute price levels. The distinction matters whenever someone interprets a higher index as meaning one place is absolutely more expensive than another. A time index and a cross-sectional price-level comparison are different statistical objects.

2. A base of 100 is a reference convention

If an index is set to 100 in the base period and later reads 120, the relevant movement is twenty per cent above the base, not a price of 120 dollars. Changing the reference base from one year to another can rescale the displayed index without changing the underlying measured price movements between adjacent periods.

Rebasing can therefore make a table easier to read without changing the economic history it represents, provided the series is linked correctly. A careless reader can mistake the new numerical scale for a new inflation history. The publication should make the reference period visible.

3. A basket is a model of expenditure

A consumer price index does not follow every possible transaction made by every household. It represents a defined universe of consumption through a selected basket of goods and services, outlets, price observations and expenditure weights.

Singapore’s 2024-based CPI uses expenditure patterns derived from the 2023 Household Expenditure Survey and updated to 2024 values. SingStat also reports that the 2024-based basket includes about 6,800 brands or varieties drawn from about 4,500 outlets. These operational details matter because an index is only as interpretable as the population and expenditure pattern it represents.

The basket is not an attempt to describe one perfectly typical household. It is a statistical representation of aggregate consumption within the defined scope.

4. Weights tell the index what matters more

Imagine two categories. Food prices rise by ten per cent while a rarely purchased item rises by fifty per cent. If households spend much more on food, the food movement should usually exert more influence on a broad consumer price index.

Weights formalise that idea. A category receiving twenty per cent of expenditure receives far more influence than a category receiving one per cent, even if the smaller category has a dramatic percentage price change.

Weights therefore connect the price system to behaviour. They are not decorative percentages printed beside the real calculation; they are part of the index’s definition.

5. A weighted average of price relatives is already an index

Take a deliberately simple basket with two categories. Category A receives weight 0.75 and its price relative is 1.04. Category B receives weight 0.25 and its price relative is 1.20. A weighted average gives 0.75 × 1.04 + 0.25 × 1.20 = 1.08.

On a base of 100, the combined index is 108. The twenty-per-cent rise in B does not dominate because B represents only one quarter of the expenditure weight.

This toy calculation is intentionally simpler than an official CPI. Its value is explanatory: the headline reflects both price movement and economic importance.

6. The same price changes can produce different indices for different populations

Suppose a lower-income household group spends a larger share on food and utilities, while a higher-income group spends a larger share on recreation and travel. If food and utility prices rise rapidly while travel prices fall, the same market prices can generate different experienced inflation profiles because the weights differ.

Singapore publishes CPI series by household income group for precisely this reason: distribution matters. The all-households index remains useful, but it should not be treated as evidence that every household experienced the headline rate identically.

This connects index-number reasoning to How Surveys and Sampling Work: weighting always refers to a population and a design.

7. Laspeyres logic holds base-period quantities fixed

A classic Laspeyres price index compares what the base-period basket would cost at current prices with what that same basket cost at base-period prices.

Laspeyres price index = current cost of base-period quantities ÷ base-period cost of base-period quantities.

The appeal is practical: the base basket and weights are known. The limitation is conceptual: people can change what they buy as relative prices change, so an old basket may gradually become less representative.

8. Paasche logic uses current-period quantities

A Paasche price index asks what the current basket costs at current prices relative to what that same current basket would have cost at base-period prices.

Paasche price index = current cost of current quantities ÷ base-price cost of current quantities.

Its current-period weighting responds to changed behaviour but requires current quantity or expenditure information. In many applications that information arrives later or is harder to collect than prices.

9. A Fisher index combines Laspeyres and Paasche information

The Fisher price index is the geometric mean of Laspeyres and Paasche indices:

Fisher = square root of (Laspeyres × Paasche).

It is often valued in index-number theory because it balances the two weighting perspectives and satisfies useful axiomatic properties. That does not make it automatically easiest to produce operationally. Formula choice belongs to the statistical purpose, available data and publication framework.

10. Substitution is one reason formulas can diverge

Suppose tea becomes much more expensive relative to coffee and households shift some purchases toward coffee. A base-weighted index still gives the old tea quantity substantial influence. A current-weighted index reflects the newer purchasing pattern.

Neither calculation is simply dishonest. They answer slightly different counterfactual questions about the basket. The divergence reveals that consumption behaviour changed while prices changed.

This is why index-number methodology cannot be reduced to “average the prices”. The object being averaged is a set of price relatives inside an expenditure model.

11. A unit-value average is not always a price index

Suppose a shop sells two models of the same product. One month customers buy mostly the expensive model; the next month they buy mostly the cheaper one. Average revenue per unit can fall even if neither model’s price changes.

The change came from product mix, not necessarily from individual prices. A price index tries to separate price movement from changing composition. That is why detailed product identification matters.

12. Elementary indices sit beneath headline indices

Official price systems often begin with many observed prices for narrowly defined products before aggregating them into broader categories. At this lower level, expenditure weights for every exact product may not be available.

Different elementary formulas can therefore be used depending on the available information and statistical guidance. Small choices at a low level can propagate upward because higher-level indices are built from these components.

13. A geometric mean treats proportional changes differently from an arithmetic mean

If two price relatives are 1.20 and 0.80, their arithmetic mean is 1.00 while their geometric mean is the square root of 0.96, approximately 0.98. The formulas encode different aggregation assumptions.

Formula choice therefore needs a stated rationale. A calculation should not be selected merely because one version produces a more convenient inflation rate.

14. Chain linking lets weights evolve

A fixed-base index can become stale when consumption patterns, products or economies change. Chain indices address this by linking short-period comparisons, often with frequently updated weights.

Eurostat explains that the Harmonised Index of Consumer Prices is compiled as a chain-linked Laspeyres-type index. Short links are calculated with current weighting structures and then connected to preserve a continuous long-run series.

Chaining improves representativeness but introduces its own discipline: the link period, classification changes, revised weights and continuity rules must all be documented.

15. Chaining is multiplication of relatives, not addition of rates

If prices rise five per cent in one period and five per cent in the next, the two-period increase is 1.05 × 1.05 − 1 = 10.25 per cent, not exactly ten per cent.

This compounding matters when building long index series. Percentage changes combine multiplicatively because each new movement applies to the level produced by the previous movement.

16. Rebasing and reweighting are related but not identical

Rebasing can mean expressing a series on a new reference scale, such as setting a newer year to 100. Reweighting changes the economic importance assigned to components. An agency can rescale a series without changing weights, or it can introduce new weights as part of a larger rebase exercise.

Readers should therefore look beyond the word “rebased” and ask what actually changed: the displayed reference, the basket, the weights, the classification, the collection system or all of them.

17. Updating weights is an evidence refresh

When households spend differently because of new technology, ageing, regulation or changing tastes, old weights gradually represent an older economy. Updating them is analogous to updating a model with a new estimate of the system’s structure.

The 2024-based Singapore CPI is a current example of this principle. Its weights derive from the 2023 Household Expenditure Survey and were updated to 2024 price values. The statistical object therefore carries both a price history and a periodically refreshed expenditure model.

18. New products create a continuity problem

A base-period basket may contain products that disappear, while entirely new products emerge. The index must incorporate innovation without pretending that every new product existed in the earlier period.

The practical response can include sample maintenance, replacement rules, overlap periods and classification updates. The conceptual challenge is to preserve the meaning of “price change” when the objects themselves are changing.

19. Quality change can masquerade as inflation

Suppose a laptop costs $1,500 in both years but the new model has more memory, a faster processor and a better display. Treating it as exactly the same product would record no price change, but the amount of quality received has risen.

Now suppose the new model costs $1,650. The observed fifteen-hundred-to-sixteen-fifty increase combines a quality difference and a price difference. A price index aims to measure the price component for comparable utility or product characteristics, not every change in the sticker price of a changing object.

20. Direct comparison works when products remain comparable

If the old and new items are sufficiently comparable, the observed price relative can be used directly. This is the simplest case and requires the least modelling.

The difficult cases arise precisely when the market changes fastest: technology products, telecommunications, packages, subscription plans and products whose size or composition changes.

21. Quantity changes matter too

If a package falls from 500 grams to 450 grams while the shelf price stays constant, the price per standard quantity rises. Singapore’s CPI materials explicitly note that the system accounts for shrinkflation.

This is an important educational example because the visible price can remain unchanged while the effective unit price rises. Measurement requires comparing like with like.

22. Hedonic methods model price from characteristics

For products with observable characteristics, a statistical model can estimate how price varies with those characteristics. The model may help separate the value associated with a quality change from the underlying price movement.

Hedonic adjustment is powerful precisely because it is model-based. Its specification, data and diagnostics matter. A fitted relationship should not be treated as though the missing comparable price had been directly observed.

23. Missing prices are not automatically zero change

An item can be temporarily unavailable, permanently discontinued, out of season or replaced by a new model. Each situation has different implications for price measurement.

The international CPI manual devotes substantial attention to temporarily and permanently missing prices and to quality change because these are not peripheral edge cases. They are part of normal index production.

The general statistical logic connects to How Missing Data Analysis Works: missingness should be understood before it is repaired.

24. Outlet substitution can change the observed market

Consumers can move from department stores to online sellers, from full-service retailers to discounters or from one platform to another. An index whose outlet sample remains frozen can gradually measure the wrong purchasing environment.

Sample maintenance therefore applies to outlets as well as products. Coverage must move with the market while preserving continuity and representativeness.

25. Discounts need transaction-relevant treatment

A posted list price can differ from the price actually paid. Sales, loyalty discounts, rebates and temporary promotions may or may not belong in the index depending on their availability and treatment rules.

The objective is not to chase every coupon. It is to measure price change for transactions inside the statistical definition using consistent collection rules.

26. Taxes and administered prices need explicit scope

Consumer prices can be affected by taxes, subsidies, regulated tariffs and administered prices. Whether and how these appear depends on the index concept. A user should not assume a price movement reveals only market supply and demand.

This is another reason an inflation index is descriptive evidence before it is causal explanation. The index says that a defined price measure changed; separate analysis is required to identify why.

27. Owner-occupied housing exposes a conceptual choice

Housing is both a consumption service and an asset. Different statistical systems can represent owner-occupied housing using approaches such as rental equivalence, user cost or acquisition-based concepts depending on the index and institutional purpose.

Singapore’s CPI uses an imputed rental concept for the service flow from owner-occupied housing and explicitly notes that housing purchase prices themselves are not treated as ordinary consumer expenditure in that CPI framework. The statistical choice affects interpretation and should be visible.

28. Scanner data changes the scale of price measurement

Retail scanner data can contain enormous numbers of transactions, quantities and product identifiers. Compared with traditional manual price collection, this can provide much greater coverage and detail.

More rows do not eliminate statistical design. Product churn, classification, returns, discounts, outlet coverage, missing weeks and rapidly changing expenditure shares still require rules. Big data reduces some observation costs while creating new integration problems.

29. Web-scraped prices are observations with a platform context

Online collection can capture prices frequently, but websites may personalise offers, vary by geography, change page structure or display products that are unavailable to many consumers. A scraper sees what the website exposes; the statistical office must decide whether that observation represents the intended transaction universe.

SingStat reports using sources including electronic returns, web scraping, APIs and administrative data for CPI compilation. The interesting point is not that digital collection replaces statistical judgement. It expands the acquisition layer while the conceptual controls remain.

30. Product identifiers are measurement infrastructure

When the same product appears across stores and periods, stable identifiers make comparison easier. When identifiers change for packaging, promotions or catalogue reasons, matching can become uncertain.

This connects price statistics to How Record Linkage and Entity Resolution Work. Before comparing two prices, the system must know whether the observations refer to the same or sufficiently comparable item.

31. Classification changes can break history if not crosswalked

Consumer classifications evolve as economies change. New categories appear; old ones are merged or split. Eurostat notes methodological changes to HICP from February 2026 associated with the new ECOICOP 2 classification aligned with the UN COICOP 2018 structure.

A long series therefore needs mappings between classification versions. Without them, the same label can conceal a changed boundary or a new category can appear to have no history even when its underlying products were previously counted elsewhere.

32. Index revisions are not necessarily errors

Some index systems revise weights, seasonal treatment, classifications or source data according to published rules. A revised number can represent improved information rather than a previous mistake.

The publication should preserve a revision policy and explain whether old values changed because of corrected input data, updated methodology, reweighting or a routine revision cycle.

33. A price index can be precise and still not describe your household

Imagine the all-household index rises three per cent. A household spending heavily on a category that rose ten per cent may experience a larger increase in the cost of its own basket. Another household concentrated in stable-price categories may experience less.

The index is not wrong because individual experience differs. It is a population-level measurement with specified weights. The error occurs when a user silently replaces the defined population with a different one.

34. Inflation over a month and over a year answer different questions

A month-on-month rate compares adjacent months. A year-on-year rate compares a month with the same month one year earlier. An annual average compares average index levels over two years.

All three can be correct and different at the same time. A falling month-on-month rate does not automatically mean the year-on-year rate is already low, because the comparison windows differ.

35. Base effects are arithmetic, not magic

If a price index jumped sharply one year ago, a current year-on-year comparison may fall even when the current index remains high. The denominator has changed.

Base effects do not imply the current price level returned to its old value. They describe how the comparison point affects the rate of change.

36. Headline inflation and core measures are different lenses

Some analytical measures exclude or differently treat volatile or policy-sensitive components to reveal underlying movement. These measures can be useful for particular decisions, but they are not substitutes for the headline index in every context.

A reader should ask which components are included, why exclusions were made and what decision the measure supports. “Core” is not a universal statistical object independent of institutional definition.

37. An index can measure quantities as well as prices

The same index-number logic can be applied to quantities, production volumes and real expenditure. Price and quantity indices can work together to decompose changes in nominal values into price and volume components.

This is important in national accounts, where changes in current-price GDP need to be distinguished from changes in real output. The general framework belongs to official statistics; the index-number machinery provides the mathematical bridge.

38. Deflating a nominal series imports the index’s assumptions

When a nominal expenditure series is divided by a price index to estimate real change, the analyst inherits the index’s scope, weights and quality adjustments. A poorly matched deflator can create misleading real values.

The question is not merely whether a price index is reputable. It is whether that index corresponds to the goods, services, population and time structure of the nominal series being deflated.

39. Index comparisons across countries need harmonised concepts

Two national CPIs can each be well constructed yet differ in scope, treatment of housing, classification or weighting practice. Cross-country comparison becomes stronger when harmonised standards define the object more consistently.

Eurostat’s HICP exists for this comparability purpose within the European system. The general lesson aligns with How Comparative Systems Research Works: compare like with like before ranking unlike systems.

40. Statistical uncertainty exists even when no confidence interval is printed

A price index depends on sampled outlets, sampled products, imputation, quality adjustment, weights and data-processing rules. Not every official release summarises these uncertainties in one conventional sampling-error interval.

The absence of a printed confidence interval should not be interpreted as proof of exactness. Different uncertainty sources are managed through methodological standards, diagnostics, revision analysis and quality reporting.

41. Revision analysis can reveal weak components

If one component repeatedly changes substantially after better data arrive, that pattern may reveal source instability or a weak preliminary estimation method. Index production can therefore be improved by studying its own revision history.

This is the same larger principle described in Data Quality: error control is a loop, not a one-time inspection.

42. Metadata is part of the index

A usable index record should preserve the concept, population, reference period, classification, base or reference scale, weighting source, formula family, treatment of missing prices, quality-adjustment policy and revision state.

Without this information, 105.3 is just a number. With it, the value becomes a structured statistical claim.

This connects directly to Metadata Standards and Interoperability.

43. A CPI is not a cost-of-living oracle

A consumer price index is closely related to cost-of-living questions, but a full cost-of-living concept can involve substitution, household heterogeneity, new goods, public services and changes in welfare that no practical CPI captures perfectly.

The international CPI manual explicitly develops the relationship between index-number theory and the practical statistical task. A careful reader should therefore distinguish the operational index that agencies can compile from a theoretical ideal that may require information unavailable in real time.

44. An index does not explain inflation’s cause

Rising consumer prices can reflect energy costs, wages, exchange rates, taxes, supply constraints, demand, rents, administered prices, profit margins or interactions among these forces. The index measures the movement of a defined price basket. Causal explanation requires additional evidence.

Separating measurement from explanation prevents a common error: treating every visible movement as proof of one preferred narrative.

45. A worked interpretation checklist

• What index is this? Consumer prices, producer prices, import prices, volume or another object?
• Whose basket or output? Which population or sector?
• What period? Month-on-month, year-on-year, annual average or base-period comparison?
• What weights? When were they measured and how often are they updated?
• What formula family? Fixed basket, chain-linked or another approach?
• What changed in the sample? Products, outlets, classifications or data sources?
• How was quality treated? Comparable replacement, package size, hedonic or another adjustment?
• What is the revision state? Preliminary, final, rebased or methodologically revised?
• What does the number not claim? Personal household inflation, absolute price level or causal explanation unless separately established.

46. The educational return: one number can contain a whole measurement architecture

Index numbers are a powerful lesson in statistical humility. They show why a compact output can require an enormous hidden structure of definitions, samples, weights, replacements, quality adjustments, classifications and linking rules.

For a learner, the habit is simple: when you see an index, ask what was held constant, what was allowed to change, and whose experience the weights represent.

47. The Library return: connect price statistics without stealing domain ownership

This article does not own inflation policy, household budgeting, finance, national accounts or Singapore economic policy. Those domains keep their own canonical owners. This page owns the general statistical mechanism by which heterogeneous price movements become index numbers.

That separation matters because one index-number method can support many domains without forcing those domains into one article. The method becomes reusable infrastructure.

Sources and further reading

Source pages and current institutional documentation were checked for this edition on 5 September 2026. This article is an educational synthesis and not a substitute for an official statistical agency’s current methodology notes.

• ILO, IMF, OECD, Eurostat, UN and World Bank — Consumer Price Index Manual: Concepts and Methods, 2020.
• Eurostat — Harmonised Index of Consumer Prices Methodological Manual, 2024 edition.
• Eurostat — HICP methodology and aggregation procedures.
• IMF and Intersecretariat Working Group on Price Statistics — Producer Price Index Manual: Theory and Practice.
• Singapore Department of Statistics — Prices and Price Indices.
• Singapore Department of Statistics — Economy and Prices FAQ.

Continue through eduKateSingapore: return to How Official Statistics Work, then connect to Surveys and Sampling, Data Quality, Data Visualisation and Comparative Systems Research.