Caste and representation
Ninety-Six Years Between Counts
India stopped counting caste in 1931 and starts again in 2027. In the ninety-six years between, reservation policy for a quarter of the population rested on a number extrapolated from a colonial census, and a caste enumeration that was actually carried out in 2011 was never published.
House listing for India's sixteenth census began in April 2026. Population enumeration is due in February 2027, with the reference moment fixed at midnight on the first of March. Question ten on the schedule asks the respondent's caste, and it is open-ended: people state it in their own words rather than choosing from a list.1
The last time an Indian census published a full caste tabulation, J.H. Hutton was Census Commissioner and the year was 1931.
What the republic chose to count
The first census of independent India, in 1951, enumerated Scheduled Castes and Scheduled Tribes and no other caste. The reasoning offered at the time, and repeated by every government until 2025, was that counting caste would entrench it, and that a republic committed to dissolving caste should not build the category into its principal instrument of self-description.
The exception is where the argument becomes interesting. Scheduled Castes and Scheduled Tribes continued to be counted because the Constitution required it. Articles 341 and 342 provide for Presidential lists of those communities, Articles 330 and 332 reserve seats for them in the Lok Sabha and the state assemblies in proportion to their population, and a proportion cannot be calculated without a count. The state therefore kept enumerating precisely those castes toward which it had a justiciable obligation, and stopped enumerating the rest.
That is a general property of statistical systems rather than a peculiarity of this one. What gets measured is what someone is legally obliged to act upon. The corollary is that a category with a claim but no enforceable entitlement tends to have no numbers, and then finds that the absence of numbers is used as a reason to deny the claim.
Governing on a 1931 number
The Backward Classes Commission appointed in January 1953 under Kaka Kalelkar reported in March 1955 with a list of 2,399 backward castes. Its chairman then attached a covering letter dissenting from his own commission's use of caste as the criterion of backwardness, and the government shelved the report.
The second Backward Classes Commission, appointed under B.P. Mandal in 1979, reported at the end of 1980. It needed a population share for the Other Backward Classes and had no data on which to compute one, because none had been collected since 1931. So it took the 1931 figures, adjusted them, and arrived at 52 per cent. On that basis it recommended reserving 27 per cent of central government posts and public sector jobs.
The recommendation sat until V.P. Singh announced its implementation in August 1990. The Supreme Court upheld the 27 per cent quota in Indra Sawhney in 1992, subject to a ceiling of 50 per cent on total reservation and the exclusion of a creamy layer.
Set the sequence out and the oddity is plain. Between 1990 and today, the largest affirmative action programme in the world has been calibrated on a share derived from an enumeration conducted under British administration, in a country that had not yet been partitioned, using caste categories recorded by colonial officials for their own purposes. Every party to the dispute has known this. It has been the standard argument for a caste census and the standard argument against acting on Mandal, depending on who was making it.
The count that happened and was not published
India did conduct a caste enumeration, and recently. It is the part of this story that gets left out.
The Socio-Economic and Caste Census of 2011 was run outside the Census Act, by the Ministry of Rural Development and the Ministry of Housing and Urban Poverty Alleviation, with the caste component under the Ministry of Social Justice and Empowerment. It cost ₹4,893.60 crore. It reached the great majority of Indian households. Its socio-economic and deprivation data were released in 2015 and have been used for beneficiary identification ever since, including under the food security and rural housing programmes.
The caste tables were never published. The stated reason was that the returns were unusable: the enumeration produced something on the order of 46 lakh distinct caste and sub-caste entries, a figure that includes spelling variants, sub-caste fragments, surnames, gotra names, occupational titles and outright duplicates. An expert group under Arvind Panagariya was constituted in 2015 to classify and validate them. No usable caste count emerged.2
The failure was one of design rather than of will, and it is worth being precise about the mechanism, because it bears directly on 2027. An open-ended question about caste does not have a closed answer set. A respondent may give a jati name, a sub-caste, a regional variant, a surname associated with a caste, or the name of a community that is a caste in one state and a sub-group in another. Enumerators recording free text without a code frame produce a list that cannot be aggregated, and no amount of subsequent cleaning fully recovers what was meant. The 2027 census asks the same question in the same open-ended form.
The legal setting matters too. A census under the Census Act, 1948 comes with compulsion and with protection: answering is a statutory duty, and the individual records are confidential, not open to inspection and not admissible in evidence. That combination is what makes people willing to answer questions about themselves. The Socio-Economic and Caste Census had neither, because it was not conducted under the Act. It was an administrative exercise whose caste component sat outside the statutory frame that gives census answers their standing, which is one reason its results could be shelved without anyone having a right to them.
Bihar, and what a code frame does
Bihar did the thing that worked, and it is instructive precisely because it was a state survey with a fraction of the resources.
The state notified its caste-based survey by gazette in June 2022. Data collection began on 7 January 2023, the second phase ran from mid-April to the end of May 2023, and more than three lakh enumerators put seventeen questions to households. The state fixed codes for its castes in advance, a little over two hundred of them, so that what enumerators recorded could be tabulated. The findings were released on 2 October 2023.3
Out of a surveyed population of 13,07,25,310, the Extremely Backward Classes came to 36.01 per cent and the Other Backward Classes to 27.13, together a little over 63 per cent. Scheduled Castes were 19.65 per cent, Scheduled Tribes 1.68, and the unreserved categories 15.52. The single largest caste was the Yadavs at 14.26 per cent.
The difference between Bihar's usable output and the SECC's unusable one is not diligence. It is that Bihar decided in advance what the answer categories were, accepted that any such list is a construction with contestable edges, and lived with it. The Union government in 2011 tried to record caste as respondents gave it, which is more respectful of how people describe themselves and produces nothing that can be counted.
Bihar also showed what a usable number does next. Having found that the backward classes came to 63 per cent, the state raised reservation for the backward classes, Scheduled Castes and Scheduled Tribes to 65 per cent, which with the existing quota for economically weaker sections took total reservation past 75. The Patna High Court struck the increase down in 2024, the ceiling in Indra Sawhney being what it is. So the first serious caste enumeration in ninety years produced, within a year, a direct collision between proportionality and a judicially imposed limit of 50 per cent. A national count will produce that collision in every state at once, and the question of whether the ceiling survives contact with the data is the one nobody in the enumeration debate has had to answer yet.
What rides on 2027
Three separate things now wait on this census, which is a heavier load than any enumeration should carry.
The first is reservation policy. A national caste distribution would replace the Mandal extrapolation, and Bihar's figures suggest the replacement will not be a small adjustment in states where backward class shares have been estimated conservatively. It also bears on the sub-categorisation of the Other Backward Classes, examined by the commission under Justice G. Rohini appointed in 2017, which turns on whether the benefits of the 27 per cent have been captured by a few dominant castes within the list.
The second is delimitation. The allocation of Lok Sabha seats between states has been frozen on the 1971 census since the Forty-second Amendment of 1976, a freeze extended by the Eighty-fourth Amendment in 2001 until the first census taken after 2026.4 The census of 2027 is that census.
The chart above uses the 2011 populations against the seat counts still in force. Rajasthan has about 2.74 million people per Lok Sabha seat and Kerala about 1.67 million, which means a vote in Kerala carries something close to one and two-thirds the weight of a vote in Rajasthan. That gap is the price of the freeze, and it was a deliberate price: states that reduced their fertility fastest were not to be penalised with fewer seats. Unfreezing on 2027 population would move seats from the south and the smaller northern states toward Uttar Pradesh, Bihar, Rajasthan and Madhya Pradesh.
The third is women's reservation. The Constitution (One Hundred and Sixth Amendment) Act, 2023 reserves a third of the seats in the Lok Sabha and the state assemblies for women, and makes commencement conditional on the publication of the relevant census figures and the delimitation exercise that follows. The reservation exists in law and cannot begin until the count is done.
The same instrument, three claims
So a single enumeration will be asked to settle the share of the backward classes, redistribute parliamentary power between states, and trigger a reservation for women, each of which has parties who expect to lose by it.
That is a great deal to hang on a question with an open answer field, asked by an enumerator with a tablet, of a respondent who may reasonably give a different answer to a different official on a different day. The 2011 exercise showed what happens when the design is wrong: the data are collected, the money is spent, and the tables never appear. Whether 2027 avoids that outcome depends on decisions about code frames and classification that are being taken now, quietly, by people whose names are not in the newspapers.
- Ministry of Home Affairs notification S.O. 2681(E), 16 June 2025, fixing 1 March 2027 as the reference date, with 1 October 2026 for snow-bound areas. The decision to include caste enumeration was taken by the Cabinet Committee on Political Affairs on 30 April 2025. ↩
- Socio-Economic and Caste Census 2011, conducted by the Ministry of Rural Development and the Ministry of Housing and Urban Poverty Alleviation, with the caste component under the Ministry of Social Justice and Empowerment; cost of ₹4,893.60 crore; expert group under Arvind Panagariya constituted in 2015. The deprivation data were released in 2015; the caste tables were not. ↩
- Bihar caste-based survey, notified 6 June 2022, fieldwork from 7 January 2023, second phase from 15 April to 15 May 2023, results released 2 October 2023. Total surveyed population 13,07,25,310. ↩
- Constitution (Forty-second Amendment) Act, 1976 and Constitution (Eighty-fourth Amendment) Act, 2001, on the freeze in seat allocation; Constitution (One Hundred and Sixth Amendment) Act, 2023, on women's reservation and its conditionality. ↩