Ask a foreign visitor what caste is and the answer is usually a ladder with four rungs: priests at the top, warriors, merchants, labourers, and below them the people who are not on the ladder at all. That model comes from texts composed more than two thousand years ago, and it is not how caste works.

The operative unit is something else: the jāti. A jāti is a named group, usually a few thousand to a few million people, usually found in a particular region, usually endogamous, and usually associated with a traditional occupation. There are thousands of them. They are ranked against each other locally, in ways that differ from district to district, and the rankings do not form a single national order. Two jātis can be dominant in adjacent districts and subordinate in each other's.

This distinction is not academic pedantry. Almost every confusion about caste — including the confusion in the nineteenth-century census that hardened it into a national system — comes from treating the four-part textual model as a description of the society rather than as a claim made by people who wanted it to be true.

Two models, one of which is real

The four-part textual model compared with the actual local lattice of jātis, and the three locks that hold it in place THE TEXTUAL MODEL WHAT IS ACTUALLY THERE brāhmaṇa kṣatriya vaiśya śūdra outside the four a claim, made in texts, about how society should be each cluster: a named local group, endogamous, ranked only against its neighbours thousands of them; no national ranking; local order differs district to district Three locks, and only one of them is still mostly shut marriage still shut occupation mostly open shared food partly bars are qualitative, not measured — the point is the difference between them A category that lost its legal basis in 1950 kept its shape because one lock was never forced.
The four-part model is a textual claim. The jāti lattice is the society. Three mechanisms hold it — and marriage, the load-bearing one, has barely moved.

The three locks

Whatever its origins, caste is maintained by three mechanisms, and it is worth naming them separately because they have behaved differently.

The first is endogamy: marriage within the group. This is the mechanism that does the work, and it is worth understanding why. An occupation can change in a generation. A ranking can be contested. But a group that marries only within itself reproduces itself as a group, with a shared network of kin, obligation and inherited reputation, regardless of what its members do for a living. B. R. Ambedkar made this argument in 1916, in a paper that was ignored at the time and is now regarded as the foundation of the sociological study of caste: endogamy is not a feature of caste, it is the mechanism that produces it.

The second is hereditary occupation: the association of a group with a particular trade. This was never as total as the model implies — most villages had more occupational mobility than the four-part scheme allows, and famine, migration and demand moved people between trades constantly. But the association was strong enough that a group's name often is its trade.

The third is commensality: who may eat with whom, and under what conditions. This is the mechanism that produces the most visible daily exclusions, and it is also the most variable — rules about cooked food, about water, about seating, about which hand and which vessel differ so much between regions that no national description survives contact with a single village.

Enforcement, before and after the law

Historically, enforcement was by the jāti council: a body of elders with the power to fine, to boycott, and in the last resort to expel. Expulsion was the severe sanction because it removed a person from the network that provided credit, labour exchange, marriage partners and support in old age. The council did not need the state; it needed the fact that no one could afford to be outside.

The legal basis was removed in stages after independence. Article 15(2) prohibits discrimination in access to shops, restaurants, wells and roads. Article 17 abolishes untouchability outright, and the Protection of Civil Rights Act of 1955 made its practice a criminal offence, followed by the Scheduled Castes and Scheduled Tribes (Prevention of Atrocities) Act of 1989. Reservations in education and public employment created a constitutional route into institutions from which these groups had been excluded.

All of that changed the rules of the game and none of it changed the marriage market. National surveys across three decades — the India Human Development Survey is the most consistent — find that more than nine in ten marriages take place within the jāti, a figure that has moved very little since the 1970s. The law can prohibit discrimination in a restaurant. It cannot make a family accept a daughter-in-law from the wrong group, and it is families that arrange most marriages.

What actually loosened, and how

The three locks did not loosen together, and the order in which they loosened explains a great deal about modern India.

Occupation loosened first and most. Urbanisation, mass education and the shift out of agriculture broke the link between a group and a trade for a large minority of the population, and did so within two generations. A jāti whose traditional work was weaving now contains teachers, engineers, shopkeepers and software developers. The name survives; the occupation does not.

Commensality loosened in cities and much less in villages, and it loosened unevenly by setting: workplaces and public transport do not enforce it, weddings and temples often still do.

Endogamy barely moved. This is the central fact, and it is why caste survives as a social reality in a country where it has no legal standing and where discrimination is a criminal offence. A group whose members marry each other, inherit from each other, and show up at each other's funerals remains a group, whether or not its members share a trade or a table.

Why the census mattered more than the texts

One more piece of the mechanism belongs here, because it is the part that outsiders usually miss and that Indian scholarship has established in detail.

The four-part model became a national fact in the nineteenth century, through administrative practice rather than through scripture. The decennial census from 1871 onward required every person to be classified, and the classification had to be consistent across a subcontinent of thousands of local systems. The result was a project of ranking: local jātis were ordered into a single hierarchy, published in tables, and used for recruitment, for land records and for the allocation of government jobs.

Nicholas Dirks's argument is that this did not merely record a system that already existed; it changed the system's form, from a set of local and contestable arrangements into a documented national order that people could mobilise around. The evidence for that reading is strong, and it has an uncomfortable implication for anyone who wants to blame either tradition or modernity alone: the thing being described and the act of describing it were not separable.

This is also why the numbers are contested. The 2011 census published figures for Scheduled Castes and Scheduled Tribes — 16.6% and 8.6% of the population — but not for jātis as such. The Socio-Economic and Caste Census collected caste data in the same period, and only aggregate tables were released. Anyone quoting a precise count of jātis, or a precise total for "backward" groups, is quoting an estimate that depends on which list was used.

What this leaves

A mechanism, not a verdict. A system whose legal basis was abolished in 1950, whose most visible exclusions are now criminal offences, whose occupational content has largely dissolved in the cities, and which nonetheless reproduces itself reliably because the marriage market was never touched.

That is the state of the question, and it explains why the argument about caste in India is so heated and so unresolved. The participants are not disagreeing about facts. They are disagreeing about how much a society can change while one of its load-bearing structures is left standing.

Sources and checks

  1. Census of India 2011 Primary Census Abstract for Scheduled Castes and Scheduled Tribes; population shares 16.6% and 8.6%
  2. Desai, Sonalde & Vanneman, Reeve India Human Development Survey (IHDS-I 2005, IHDS-II 2011–12), University of Maryland and NCAER — marriage patterns by jāti
  3. Ambedkar, B. R. Annihilation of Caste, 1936; and "Castes in India: Their Mechanism, Genesis and Development", 1916
  4. Dirks, Nicholas B. Castes of Mind: Colonialism and the Making of Modern India. Princeton University Press, 2001
  5. Dumont, Louis Homo Hierarchicus: The Caste System and Its Implications, 1970 — and the substantial critical literature it produced
  6. Government of India The Constitution (Scheduled Castes) Order, 1950; Protection of Civil Rights Act, 1955; SC/ST (Prevention of Atrocities) Act, 1989

Last checked 30 October 2026 Yellowsea Huang Corrections