The Country That Stopped Counting Itself

Pritiraj BrahmaFounding Partner, Pritiraj & Partners · 12 February 2026

India’s last completed census was in 2011; the next counts its population against a reference date of 1 March 2027. Sixteen years, and almost every consequential decision inside them taken on the old denominator. The food security ceiling is still 81.35 crore, and estimates of who that leaves out range across five crore human beings.

ProvenancePublished 12 February 2026; revised 28 August 2026 to record that Census 2027 has since been notified and begun. Census 2027 dates and design are from the Gazette of India notification of 16 June 2025, the Ministry of Home Affairs notification of 7 January 2026 issued under sections 3 and 17A of the Census Act, 1948, and the Registrar General’s statements of 2026; the self-enumeration uptake figure is as reported on 10 April 2026 and was current then only. Food security coverage is as set by section 9 of the National Food Security Act, 2013 on Census 2011. The three exclusion estimates are from the Union’s August 2023 affidavit to the Supreme Court and from published independent research; they answer different questions and are not reconciled here. Readers should note that results of the 2027 count were not expected before late 2027.

India’s last completed census was in 2011. Phase I of the next one began on 1 April 2026 and the population will be counted against a reference date of 1 March 2027 — sixteen years after the last count, and after almost every consequential decision of that period had already been taken on the old denominator. The food security ceiling is still 81.35 crore. Estimates of how many people that leaves out range across five crore human beings, and no official figure exists.

Almost every important number in Indian public life is a fraction. Per-capita income, doctors per lakh, police per lakh, seats per state, the share of a population entitled to subsidised food — each has a population figure in its denominator. When that denominator is stale, every fraction built on it is wrong in the same direction, and wrong by a growing margin.

The count has now started, and it is worth saying plainly that this is good news. What follows is not an argument that India refuses to count itself. It is an argument about the sixteen years in which it did not, what was decided in them, and what the arrival of a new number does and does not fix.

Six findings anchor this analysis:

1. The count has begun, on a published statutory timetable. The intent to conduct Census 2027 was notified in the Gazette of India on 16 June 2025. The Ministry of Home Affairs notified Phase I — house listing and the housing census — on 7 January 2026 under sections 3 and 17A of the Census Act, 1948, superseding the 2020 notification that the postponed round had rested on. House listing runs between 1 April and 30 September 2026, each state taking a 30-day window of its choosing, with a 15-day self-enumeration option immediately before. This is the sixteenth census in the series, the eighth since independence, and the first conducted digitally. By 10 April 2026 the Registrar General reported that over 5.72 lakh households had used self-enumeration across eleven states and union territories.
2. The gap will be sixteen years, and the decisions that needed the number have already been taken. Population enumeration is scheduled for February 2027 against a reference date of 1 March 2027. India will therefore have gone sixteen years between completed counts, against a ten-year interval maintained without a break from 1881 through 2011. The delay is not the whole cost; the timing is. Entitlement ceilings, seat allocation and per-capita comparisons for that entire period were computed on 2011.
3. Entitlement lists remain capped on 2011 numbers, and the exclusion is contested across five crore people. The National Food Security Act, 2013 fixes coverage at 75 per cent of the rural and 50 per cent of the urban population, determined from the last published census, producing a national ceiling of 81.35 crore computed on Census 2011. How many the growth since has left outside depends on who counts: the economists Jean Drèze and Reetika Khera put it at roughly ten crore; other independent researchers at nearer thirteen; the Union’s own affidavit to the Supreme Court in August 2023, derived from matching e-Shram registrations against distribution records, identified approximately eight crore people holding no ration card. No official figure exists, because establishing one requires the enumeration whose absence created the problem.
4. Census 2027 is the trigger census for delimitation, which raises the stakes on its accuracy considerably. Lok Sabha seat allocation still rests on the 1971 census. The Eighty-fourth Amendment extended the freeze until after the first census taken after 2026 — and a count with a reference date of 1 March 2027 is that census. The same exercise that corrects sixteen years of stale denominators will also become the constitutional basis for redistributing political representation between states. It is the highest-stakes enumeration India has conducted.
5. The count is not a single moment, and in part cannot be. The reference date is 1 March 2027 for most of the country but 1 October 2026 for snow-bound regions — Ladakh, parts of Jammu and Kashmir, Himachal Pradesh and Uttarakhand. This is a long-standing and unavoidable operational necessity. It also means those populations are counted five months before the rest, and any comparison across that boundary carries a gap that should be stated rather than assumed away.
6. The interim was filled by projection, and the frame beneath every other survey aged with it. The Technical Group on Population Projections produces the estimates most official documents now rely on. Projections are sound in aggregate and unreliable at the district level — which is the level at which schools are sited, primary health centres staffed and constituencies drawn. A census is also the sampling frame from which the Periodic Labour Force Survey, the National Family Health Survey and consumption surveys draw. A frame sixteen years old systematically under-represents exactly what changed most: new urban settlements, peri-urban growth, migrant-dense districts.

The interval, and what happened inside it

The bias a stale frame introduces is subtle and compounding. Surveys do not simply become less precise; they become biased toward the India that existed when the frame was drawn, and no increase in sample size corrects it. That is why a delayed census is not a delayed report. It is a delayed instrument, and everything measured with the instrument inherits the delay.

The digital design of the 2027 round is a genuine improvement and deserves credit: self-enumeration, mobile collection, and a published statutory schedule that can be held to. The uptake figure of 5.72 lakh households in the first ten days is a real number, published promptly, which is more than most Indian programmes offer in their first fortnight. None of that retrospectively fixes the sixteen years, and none of it is an argument for treating the gap as closed the moment the enumerators start walking.

The exclusion, counted three ways

The legal history matters as much as the arithmetic. The Supreme Court has directed the Union to devise a formula under which entitlements are not restricted to 2011 figures. The Union’s position has been that broadening coverage would require amending the Act. Both are defensible, and between them sits the structural fact this essay is about: the statute delegated a policy question to a dataset, and the dataset stopped being produced. Nobody decided to exclude ten crore people. Section 9 is operating exactly as drafted, against a census that did not arrive.

This is also the strongest available test of the administrative-register argument. The e-Shram cross-match worked; it found eight crore uncovered people, and it found them because each had registered with the state for something. It could not find, and cannot in principle find, the household that has registered for nothing. A register enumerates the population that has already made contact. A census exists to find the rest, and that is the function nothing else performs.

The counter-case, honestly stated

Three arguments deserve a fair statement, and the first has largely been answered by events.

First, the government’s scheduling case. A census of 1.4 billion people is among the largest peacetime operations any state attempts; the pandemic made 2021 impossible, and subsequent scheduling had to accommodate elections, enumerator availability and a shift to digital collection. That case is now partly vindicated: the round was notified, the timetable published, and Phase I started on the day it said it would. An essay written a year earlier could reasonably have doubted that. It should say so.

Second, administrative data genuinely does provide coverage that did not exist in 2011 — Aadhaar-linked registers, electoral rolls, distribution databases. This is the strongest objection and the most double-edged. Registers count people who have transacted with the state. They are excellent for those inside the system and structurally blind to those outside it, which is precisely the population a census exists to find.

Third, and against our own emphasis: uncapping entitlement coverage from the last census sounds costless and is not. Coverage as a share of a projected population means a fiscal commitment that rises with an estimate nobody has verified, and a government is entitled to want the denominator measured before it is bound by it. The honest version of our recommendation is therefore narrower than “uncap it” — it is that the operative population should be a single published official estimate, defended in public, rather than a figure whose staleness quietly does the rationing.

What we would do

1. Publish one interim denominator, and defend it. Every ministry currently uses some population estimate and they do not all use the same one. One published series, one method, open to challenge, used by all departments until the 2027 results land. This costs nothing and ends a class of quiet inconsistency in Indian public numbers.
2. Fix the entitlement base by statute rather than by dataset. Where a law fixes coverage as a share of population, the operative population should be the best current official estimate with a stated revision rule, not whichever census was last completed. The present drafting delegated a rationing decision to a survey schedule, which is how ten crore people came to be excluded without anyone deciding to exclude them.
3. Publish the census timetable with milestones, and report against them monthly. Phase I dates are notified; processing and release dates for each table are not. A census whose enumeration schedule is public but whose publication schedule is not can still deliver its numbers late without anyone being accountable for the lateness.
4. State the delimitation base before the count, not after it. If Census 2027 is the constitutional trigger for redistributing seats, the rules for how its figures will be used should be settled and published while the count is still neutral. Deciding the method after the numbers are known invites every state to litigate the method rather than the arithmetic.
5. Release micro-data on a fixed calendar, without ministerial discretion, and fund the state statistical machinery. Most states run a Directorate of Economics and Statistics at a fraction of sanctioned strength, and national statistics are assembled from state work. The credibility damage from a withheld survey exceeds the political damage from an inconvenient number; statistical independence is cheap to grant and expensive to rebuild.

A state’s first competence is to know who it governs. India has now begun to recover that knowledge, on a published schedule, digitally, and the exercise deserves to be judged on whether it lands on time. What cannot be recovered is the sixteen years of decisions taken on a photograph from 2011 — and the statutes that still ration entitlements against it will keep doing so until somebody changes the drafting rather than waiting for the data.