How India measures it

Chapter 1 said an inflation rate is a measurement with choices inside it. This is where the choices are made.

Somebody walks to a market

Before any index, there is a person with a list.

The National Statistics Office states the collection in its monthly release: real-time price data are collected from 1,407 urban markets, including online markets, and 1,465 villages, covering all states and union territories, through personal visits by field staff on a weekly roster.

In September 2026 the response rate was 100% in both rural and urban markets.

That is the foundation of the number the Monetary Policy Committee acts on, that your loan is repriced against, and that newspapers lead with. People walking to shops, every week, writing down what things cost.

It is worth sitting with, because the usual intuition about official statistics is that they are modelled or estimated or massaged. The prices are none of those. They are observed, the number of observation points is published, and so is the share that came back.

From prices to an index

A price is not an index. Three steps convert one into the other.

1. Decide the basket. What a representative household buys, established from a household consumption expenditure survey, broken into twelve divisions — food and beverages; paan, tobacco and intoxicants; clothing and footwear; housing, water, electricity, gas and other fuels; furnishings and household maintenance; health; transport; information and communication; recreation, sport and culture; education services; restaurants and accommodation services; and personal care, social protection and miscellaneous.

2. Weight each item by how much of household spending it accounts for. This is the step that does the most work and gets the least attention. Onion carries a weight of 0.7006% of the all-India basket. Potato carries 0.7549%. The whole of onion, potato and tomato together is 1.95% of what the index measures.

3. Express it against a base. The index is set to 100 in the base year and every later month is a ratio to it. India's CPI currently runs on base 2024 = 100. In September 2026 the all-India combined index stood at 109.49, against 108.74 in August.

So the headline inflation rate is not measured directly. It is computed: this month's index against the same month a year earlier.

What the base year is for, and why it changes

A base year is just a reference point — the year the index equals 100. Changing it does not change what prices did.

What matters is that the basket is rebuilt at the same time. A basket fixed in 2012 would still be weighting landline telephony and would have no weight for mobile data, and would be measuring the cost of a life nobody leads any more. Rebasing is the mechanism by which the measure keeps describing the present.

The cost is that indices on different bases are not directly comparable, and a series that looks continuous in a chart may contain a join. When you see a long inflation chart, the base changed somewhere in it.

Two things the method deliberately does not do

Both are criticisms worth knowing and neither is a flaw exactly.

It does not let you substitute. If onion doubles and you switch to potato, your actual cost of living rises less than the index says. The fixed basket keeps buying the same onion. This is a known upward bias in fixed-basket indices, it is why the basket is periodically rebuilt, and it is the price of having a measure that compares like with like over time.

It does not adjust fully for quality. A phone costing the same as one five years ago is a different phone. Some of what looks like a stable price is a quietly improving product, which means measured inflation can overstate the real cost of living in categories that improve fast.

Both biases run the same way: measured inflation is, on these two counts, a little high. Which is an argument for not treating a tenth of a percentage point as meaningful.

Honesty in the document itself

The release contains a note most readers skip and this site likes:

indices at all levels and inflation rates are compiled using the actual index values without rounding off, however the indices published are rounded off to two decimal places — so higher-level indices calculated from the published figures may observe minute differences.

In other words: if you recompute the headline from the published components you will get a slightly different answer, and that is expected rather than an error. The document tells you so in advance.

There is a second, sharper caution about the state-level data on the eSankhyiki portal: item indices for some states "should be interpreted and used with caution, as the indices for some items may have been compiled using thin samples of price quotations." A statistical office telling you which of its own numbers to trust less is the opposite of massaging, and it is the kind of sentence that should raise your confidence in the rest.

The calendar

CPI is published monthly, around the twelfth, for the preceding month. The September 2026 release was dated 12th October 2026 and stated that October's would follow on 12th November 2026.

Two consequences for anyone using it.

The data are old when they arrive, by roughly six weeks from the middle of the month being measured. Chapter 5 shows the Monetary Policy Committee raising rates on 7th October 2026 — five days before the September figure was published. They decided without it.

The first print is provisional. September 2026's 5.54% was published as provisional, while August's was republished as final at 4.83%. Figures get revised, usually slightly, and a chart built from first prints is not quite the same chart as one built from final ones.

The point

India's CPI is built from prices observed by field staff in 1,407 urban markets and 1,465 villages on a weekly roster, with the response rate published each month. Those prices are weighted by how much households actually spend on each item, and expressed against a base year — currently 2024 = 100. The fixed basket deliberately ignores substitution and under-adjusts for quality, both of which push the measure slightly high. And the release tells you its own limits, including which of its numbers to use with caution.