The Household Consumption Expenditure Survey is the instrument that sets India's poverty line, weights the consumer price index, and supplies the private final consumption benchmark the national accounts are checked against. Between the 2011-12 round and the 2022-23 round, nothing was published. A round was conducted in 2017-18 and withheld. So the published interval is eleven years, and it is eleven years across which nothing in this country's distributional statistics could be measured against a fresh sample.
Two rounds have now appeared, 2022-23 and 2023-24. It is worth being precise about what that restores, because the answer is less than the relief suggests.
Start with what is genuinely new. Average monthly per capita consumption expenditure in 2023-24 was ₹4,122 in rural India and ₹6,996 in urban India. The Gini coefficient of that distribution was 0.237 rural and 0.284 urban. Those are the first state-representative figures on household spending in more than a decade, and every state figure underneath them is usable on its own terms.
Now the complication. Every one of those numbers is published twice.
The survey asks separately about items the household received free through a social welfare programme: the grain from the public distribution system, the mid-day meal, the uniform, the bicycle, the laptop. MoSPI publishes one series that leaves those out of the expenditure total and one that prices them in. For 2023-24 the rural figure is ₹4,122 without imputation and ₹4,247 with it. Urban is ₹6,996 and ₹7,078.
The gap looks small, roughly three per cent rural and one per cent urban. Two things about it are not small.
The first is that the gap is not the same size in the two years. Rural, it was ₹3,773 against ₹3,860 in 2022-23, a difference of 2.3 per cent, and ₹4,122 against ₹4,247 in 2023-24, a difference of 3.0 per cent. So nominal rural consumption growth between the two rounds is 9.3 per cent on the without-imputation series and 10.0 per cent on the with-imputation series. Seven tenths of a percentage point on the headline growth rate, decided entirely by which of two equally official numbers a reader picked up.
The second is that nobody has to declare which one they used. Both come from the same table, both are labelled MPCE, and a figure quoted without the qualifier is indistinguishable from a figure quoted with it. When two reports disagree by three per cent on rural consumption, the first question is not whether one of them is wrong. It is whether they are the same series.
The methodological point underneath this is not about imputation specifically. It is that a statistical system under pressure to be complete tends to publish alternatives rather than choose between them, and the choice then migrates to whoever quotes the number last. That is a worse place for it. The decision about whether a free bag of rice is consumption expenditure is a real analytical judgement with real consequences for the poverty count, and it belongs with the people who designed the survey, stated in one line, once.
What this means in practice, for anyone using these rounds.
State which series you are on, in the table note, every time. Not in the methodology annexe. In the table note, where a reader who lifts the number will see it.
Do not compare across the two. A 2022-23 with-imputation figure against a 2023-24 without-imputation figure understates growth by about half a percentage point, and nothing in either number says so.
Be careful with the eleven-year comparison in particular. It is the comparison everyone wants, and it is the one the 2017-18 gap makes hardest to defend: the recall periods and the questionnaire structure changed between 2011-12 and 2022-23, so the difference between those two points is a mixture of what households did and what the instrument asked. A series with a hole in it is not a series. It is two series that happen to share a name, and the honest treatment says so before it draws the line between them.
None of this is an argument against using the data. It is very good that it exists. It is an argument for handling a restored series more carefully than a continuous one, because the break is invisible in the numbers and visible only in the documentation, and the documentation is not what gets quoted.