Essay · Section A26 September 2026

What the State counts, it comes to govern.

When H.H. Risley ran the Census of India in 1901, he tried to arrange the castes of each province in order of “social precedence as recognised by native public opinion”. The attempt set off a flood of petitions. Caste associations wrote to census officers asking to be recorded under more honourable names, to be placed higher in the table, to be counted as something other than what their neighbours called them. Historians still argue over how much the colonial census created and how much it only recorded; Nicholas Dirks, in Castes of Mind (2001), argued that colonial rule made caste the master category of Indian society, while others reply that caste was powerful long before any census. On one point they agree. Once the State wrote caste down, caste began to organise itself around what the State had written.

More than a century later, India is counting caste again. The questionnaire for the second phase of Census 2027 asks a caste question for the first time since Independence, using an “open column” in which people who are not Scheduled Castes or Scheduled Tribes write their caste in their own words. Officials say it has not yet been decided how the answers will be sorted and tabulated. This essay argues that counting is never a neutral act of looking: what a State measures, it begins to fund, to regulate and to reshape, and people begin to shape themselves to fit. The real choice, therefore, is not between counting and not counting, but between a count that is governed openly and one that governs in the dark.

Start with the good that counting does. In 1990 Amartya Sen published an essay in The New York Review of Books titled “More Than 100 Million Women Are Missing”, estimating from population ratios how many women and girls should have been alive in Asia and North Africa and were not. A number turned a private practice into a public scandal. In India the Census kept recording the damage: the child sex ratio fell from 927 girls per thousand boys in 2001 to 918 in 2011. The Pre-Natal Diagnostic Techniques Act of 1994 and the Beti Bachao Beti Padhao scheme of 2015 were both answers to what the count had shown. Even the Socio-Economic Caste Census of 2011, whose caste tables were never released, produced deprivation data still used to identify beneficiaries of housing and health insurance schemes.

Not counting also governs, only less honestly. The Mandal Commission estimated the share of Other Backward Classes in the population from the Census of 1931, the last to count every caste. For more than three decades, therefore, reservation and welfare have been allocated on caste lines without current caste numbers. In the absence of data, benefits tend to flow to the groups best able to claim them, not to those most in need. Silence about numbers did not abolish caste; it simply handed the question to whoever could organise the loudest.

Yet counting has a darker side, and it must be faced squarely. In 1933 the Belgian colonial administration in Rwanda issued identity cards recording each person as Hutu, Tutsi or Twa. A category that had once allowed some movement was fixed on paper, and in 1994 those same cards were checked at roadblocks during the genocide. The lesson is general: a category written by the State can become a destiny. Economists know a gentler version as Goodhart’s law, after the economist Charles Goodhart: once a measure becomes a target, people change their behaviour to meet it and the measure stops telling the truth.

Today’s papers carry a small example of how a measure can replace the thing it measures. Karnataka says the Centre has withheld ₹2,186.20 crore in Finance Commission grants for panchayats because about ₹225 crore that the State says reached gram panchayats, and was spent on works, did not appear on the e-Gram Swaraj portal. “Citing such a minor issue as a reason for withholding ₹2,186 crore is an injustice to Karnataka,” the State’s minister said. The Union government’s reply was not reported, and this essay does not judge the dispute. But the principle is visible either way. For the system that releases money, what is not on the portal does not exist.

The case against counting caste deserves its strongest form. France has long refused to collect statistics on ethnic origin, on the principle that the Republic sees only equal citizens, and in 2007 its Constitutional Council struck down a provision of an immigration law that would have opened the door to such data. India’s early leaders chose, after 1951, to count only Scheduled Castes and Scheduled Tribes, hoping that other caste identities would fade if the State ignored them. The fear that a caste count will harden identities and set off competitive demands is not imaginary. Bihar published its caste survey in 2023, raised reservation to 65 per cent, and saw the Patna High Court strike the increase down in 2024 for breaching the fifty per cent ceiling set in Indra Sawhney (1992).

But the experience of the last seventy years is that refusing to look did not make caste disappear. It continued to decide who owned land, who married whom and who was hired, while the State pretended not to know. The honest conclusion is that the danger lies less in counting than in counting carelessly, secretly or for a result already chosen.

What would a well-governed count look like? First, the classification should be fixed before the data comes in, not after. Good statistical practice publishes the coding rules, synonyms and tabulation plan in advance, so that no one can later say the numbers were arranged to suit a result. The chain the officials describe, from the census divisions to the Registrar General, the Home Secretary and finally the Ministry or Cabinet, has no formal place for independent statisticians or social scientists; one should be created. Second, caste should be cross-tabulated with education, occupation and assets, so that policy can see deprivation within castes and not merely produce a new ranking of their sizes. And third, the confidentiality that Section 15 of the Census Act, 1948 promises to every household must be defended absolutely, because a count people fear will not be a true count.

The Karnataka dispute adds a fourth rule. Whoever is counted must be able to answer back. A data system that withholds money should first reconcile its records with those it serves, give reasons in writing, and allow a correction before the penalty falls. Otherwise the portal governs, and the panchayat that built the road is treated as if it had built nothing.

On 25 November 1949, B.R. Ambedkar warned the Constituent Assembly that the Republic would recognise “the principle of one man one vote and one vote one value” in politics, while its social and economic structure would continue to deny “the principle of one man one value”. A census is one of the few instruments a democracy has for checking how far that contradiction has narrowed. It will shape what it counts, as Risley’s tables did. The task is to make sure it shapes it in the open, by rules fixed in advance, for the sake of the person behind the number rather than the category above it.