Eleven Years Without a Number

On 14 November 2019, Business Standard published the findings of a report that the government had decided not to publish. Somesh Jha had obtained the draft of Key Indicators: Household Consumer Expenditure in India, the National Statistical Office's report on the seventy-fifth round of the consumer expenditure survey, fielded between July 2017 and June 2018. Adjusted for inflation and expressed at 2009-10 prices, average monthly per capita consumption expenditure had fallen from ₹1,501 in 2011-12 to ₹1,446 in 2017-18, a decline of 3.7 per cent. Rural spending had fallen 8.8 per cent. Urban spending had risen 2 per cent.¹

Indian consumption had last been recorded falling in 1972-73.

The following day the Ministry of Statistics and Programme Implementation issued a statement. It had decided not to release the results, citing data quality issues, and it described the document as a draft on the ground that submissions reaching the ministry are drafts until the ministry says otherwise.² A further Business Standard report on 18 November noted that a committee appointed by the National Statistical Commission had approved the report for immediate release on 19 July 2019.³ The Commission itself was told of the decision to scrap it after the decision had been taken.⁴

The year had a pattern to it. In January 2019 the two non-official members of the National Statistical Commission had resigned over a labour force survey that was approved and not published; that report appeared on 31 May, unchanged, once the general election was over. In November the consumption survey was approved and not published, and this time it was not released later. On 23 November the ministry put out five other survey reports that had been sitting withheld, and kept back the one it had junked.³

What a consumption survey is for

It is worth being precise about what was lost, because the loss was not confined to poverty statistics.

The consumer expenditure survey is the instrument that establishes what Indian households buy. From it come three things. The first is the weighting diagram of the consumer price index, which fixes how much a change in the price of pulses or petrol or school fees moves the headline inflation number. The second is a benchmark for the private final consumption expenditure component of the national accounts, which is roughly three-fifths of gross domestic product. The third is the distribution against which a poverty line is applied to produce a headcount.

Scrapping one round therefore does not leave a hole in one series. It freezes the basket used to measure inflation, removes the independent check on the largest component of national income, and stops the poverty count.

The inflation consequence is the least discussed and possibly the most expensive. A price index measures what happens to the cost of a fixed basket, and the basket has to be refreshed as spending patterns change. Through the whole of the eleven-year gap, the consumer price index carried weights derived from what households were buying in 2011-12, a period before the spread of smartphones and mobile data, before the goods and services tax, and before the shift in rural spending from food towards conveyance and services that the 2022-23 survey went on to record. Every real wage, every deflated output series, and every inflation-targeting decision of the Monetary Policy Committee over that period ran on those weights.

The consequence is visible in the interval chart above. From 1972-73 to 2011-12 the large-sample consumption rounds came at intervals of between four and a half and six years, with an extra round in 2011-12 because 2009-10 had been a drought year and was thought unrepresentative. The next one arrived eleven years later.

The case for scrapping it, taken seriously

There were real objections to the 2017-18 round, and an argument that ignores them is not worth making.

The gap between consumption measured by the survey and consumption implied by the national accounts had been widening for decades, and by 2011-12 the survey captured only about half of what the accounts implied. A round showing an absolute fall would widen that gap further. Survey non-response had been rising, particularly among richer households, which biases measured consumption downward, and the 2017-18 findings showed the sharpest cuts among the top decile in rural areas, which is precisely the pattern that differential non-response produces.³ The fieldwork also ran through an unusual period: the year began eight months after the demonetisation of November 2016 and included the introduction of the goods and services tax in July 2017.

Any of these could support a decision to publish with heavy caveats. None of them supports the decision that was actually taken, which was to withhold the report and the unit-level data together.

That is the specific failure, and it is worth separating from the politics. If the quality problem was real, the remedy available to the ministry was to release the microdata with a note setting out the concerns and let the profession establish whether the estimates were sound. That is what the profession is for. Withholding the data made the quality claim unfalsifiable: nobody outside the ministry could check whether the survey was bad, and nobody could check whether it was fine. Seven years later the unit-level data have still not been released.

What filled the gap

A number that stops being measured does not stop being needed. It gets modelled instead.

Working from the aggregates that had leaked, S. Subramanian estimated that the rural poverty ratio had risen sharply over the six years to 2017-18 and that the number of poor people in India had increased by tens of millions. Others reworked the same fragments to similar effect. In the opposite direction, a 2022 working paper by Surjit Bhalla, Karan Bhasin and Arvind Virmani, using national accounts consumption growth and the labour force survey rather than a consumption survey, concluded that extreme poverty in India had fallen below one per cent. Estimates from the World Bank and from bank research departments landed in between.

The dispersion is the point. These were competent people applying defensible methods and arriving at answers that could not both be roughly right. What separated them was not skill but the assumption each had to make to replace the missing observation. When the measurement exists, the argument is about interpretation. When it does not, the argument is about whose model to believe, and that argument has no resolution procedure.

It also matters who is missing from a modelled estimate. National accounts consumption grows with the formal economy, which is measured through corporate filings and tax returns; the households that a consumption survey reaches and the accounts do not are disproportionately rural, informal and poor. An estimate that replaces the survey with the accounts will therefore tend to find that the people the survey was designed to count are doing better than the survey would have said, and it will find this by construction rather than by evidence.

The survey that came back

Fieldwork for the Household Consumption Expenditure Survey ran from August 2022 to July 2023. The factsheet appeared in February 2024 and the detailed report and unit data in June. A second round followed immediately, fielded from August 2023 to July 2024, with its factsheet released in December 2024 and covering 261,953 households.⁵

At 2011-12 prices, rural monthly per capita expenditure was ₹1,430 in 2011-12, ₹2,008 in 2022-23 and ₹2,079 in 2023-24. The urban figures were ₹2,630, ₹3,510 and ₹3,632.⁵ Rural real consumption therefore grew by about 45 per cent over twelve years, which is a compound rate near 3.1 per cent a year, and urban by about 38 per cent, or 2.7 per cent a year. Measured consumption inequality fell, with the rural Gini coefficient going from 0.283 in 2011-12 to 0.237 in 2023-24 and the urban from 0.363 to 0.284.

Three cautions belong with those numbers, and the ministry states the first two itself.

The survey was redesigned. The 2022-23 round used three separate questionnaires administered over three visits to each household, against a single schedule previously, and a revised item list. Splitting the recall task across visits is known to raise reported expenditure, which means some part of the increase is a property of the instrument.

Items received free through welfare programmes were valued and added, producing a second set of estimates. Rural expenditure with imputation was ₹4,247 in 2023-24 at current prices against ₹4,122 without it.

And the middle observation is missing. The chart above draws the eleven-year stretch as a dotted line for a reason: there is no measurement in it. If the 2017-18 estimate was broadly right, rural consumption fell and then rose steeply, and the recovery from 2017-18 to 2023-24 was faster than 3.1 per cent a year. If it was wrong, growth was steady throughout. The 2022-23 survey cannot distinguish between these, because distinguishing between them requires an observation in 2017-18, and that observation was collected from 100,000-odd households, tabulated, approved, and then set aside.

No line to apply

There is a further absence that follows from the first one. India's last official poverty estimate, using the Tendulkar line applied to the 2011-12 round, put the headcount at 21.9 per cent. The Rangarajan committee reported a different line in 2014, giving 29.5 per cent for the same year, and the government never adopted it. No official poverty line has been applied to the 2022-23 or 2023-24 consumption data.

So the position in 2026 is that India has two recent, well-conducted consumption surveys and no official poverty ratio derived from either, because the line that would convert a distribution into a headcount was last settled for a distribution measured fourteen years ago. The state has resumed measuring consumption. It has not resumed measuring poverty, and those are different acts.

Against that, the multidimensional poverty index, which counts deprivations in health, education and living standards rather than expenditure, has been published and quoted extensively. It is a legitimate measure and it answers a different question. It cannot tell you whether a household can afford to eat, which is the question the consumption survey was built for.

The cost of the gap

Set out plainly, the eleven years produced this. Consumer price inflation was measured throughout on a basket of goods established from spending patterns in 2011-12. The national accounts ran without an independent consumption benchmark. Poverty was contested by parties using models rather than measurements, and the contest ended in no agreed number. The unit records that would settle whether the 2017-18 survey was sound remain unreleased.

The ministry's decision of 15 November 2019 was defended as a matter of data quality, and it may even have been one. What it cost was the ability of anyone else to find out.

1. Somesh Jha, 'Consumer spend sees first fall in 4 decades on weak rural demand: NSO data', Business Standard, 14 November 2019. ↩
2. Ministry of Statistics and Programme Implementation, statement of 15 November 2019 on the Consumer Expenditure Survey 2017-18. ↩
3. Somesh Jha, 'NSO's junked consumer expenditure survey shows inequality gap declining', Business Standard, 18 November 2019. ↩
4. On the sequence in which the National Statistical Commission was informed, see the Carnegie Endowment paper on India's statistical system, 2023. ↩
5. Ministry of Statistics and Programme Implementation, Household Consumption Expenditure Survey 2023-24, press note and factsheet, December 2024, Tables 1 and 2; and HCES 2022-23 factsheet, February 2024. ↩