Fifty Meters Apart
Asher, Jha, Novosad, Adukia, and Tan report that Muslim and Scheduled Caste segregation in India is "only slightly lower than Black-White segregation in the U.S." A segregation index, however, is a measure of unevenness, not an explanation of what produced it or what sort of social reality it represents.
When is clustering concerning?
Communities cluster for many reasons. Some are mechanical: historical settlement patterns that predate the modern housing market. Some are functional: kinship networks, shared language, community institutions, dietary compatibility, marriage markets. Some are concerning: exclusion from housing markets, communal violence, etc. And some are beneficial: under first-past-the-post, geographic concentration lets a minority elect its own representatives; religiously concentrated neighborhoods that are economically diverse can produce the inter-class contact that segregated cities otherwise lack (Wilson, The Truly Disadvantaged, 1987).
In the American case, the number carries weight because it stands for a specific history: restrictive covenants, redlining, white flight, and concentrated poverty. Countries without that history, or the same countries dealing with different groups, have read clustering differently. Canada, for instance, adopted an official multiculturalism policy in 1971 with the explicit purpose of recognizing, accommodating, and supporting cultural diversity (Banting and Kymlicka, 2006). The UK and the Netherlands pursued similar frameworks. The philosophical rationale, developed most fully by Kymlicka (Multicultural Citizenship, 1995), was that cultural membership provides the context within which individuals make meaningful choices, and that supporting community maintenance is a liberal value.
The empirical evidence is also ambiguous. Edin, Fredriksson, and Åslund find that ethnic enclaves in Sweden improve labor market outcomes for less-skilled refugee immigrants (QJE, 2003).
Benign causes in the Indian context
Shahjahanabad in Delhi was built in 1639. The old city of Hyderabad was founded in 1591. The walled city of Ahmedabad dates to 1411. These neighborhoods are Muslim today because they were Muslim when they were built.
Even where neighborhoods are newer, community life generates clustering. Some Muslims may prefer to live within earshot of a mosque whose muezzin calls five times a day, within walking distance of halal butchers and madrassas, and inside the social networks that arrange marriages and organize festivals. The muezzin's call is also relevant in the other direction: non-Muslims who find it disruptive have reason to sort away.
These mechanisms are not specific to Muslims. Marwaris, Tamils, Bengalis, and Jains cluster in Indian cities; jatis cluster in villages. The paper measures clustering for Muslims and Scheduled Castes. It does not measure it for any other group. Without that comparison, there is no way to know whether Muslim clustering is unusual or the Indian baseline.
The same logic extends to the terms on which people share buildings and neighborhoods. Food is an obvious example. For many households, vegetarianism is not an abstract moral identity but part of how they want to live: what is cooked at home and what smells travel through walls and corridors (Outlook). Landlords and housing societies can act on those preferences by restricting tenants to vegetarians. To the extent that diet is correlated with religion, rules of that kind can produce religious clustering. (This isn't to say that there aren't cases where landlords use dietary preferences to exclude people they don't like; see, for, e.g., The Revealer, 2025).
Density and exposure
Even if we set aside the causes of clustering, there is a second problem with reading too much into the index, at least for cities. The same level of measured unevenness need not imply the same degree of lived separation. How closely the two track depends on the size of the spatial units, the density of the city, and the ways people move through everyday space.
An old-Delhi enumeration block of 500 people may be fifty meters on a side, and the adjacent block, demographically distinct, is a short walk away. A US census tract that is 80% Black typically spans a square mile, with the nearest majority-white tract several miles away across a highway. The same index value, applied at both scales, describes walking across the street versus driving across town.
Dense cities pack more people into any given walk. Mumbai averages over 20,000 people per square kilometer. Houston averages around 1,500. Even holding physical distance constant, any walk through a Mumbai neighborhood produces an order of magnitude more encounters than the same walk in Houston. High block-level homogeneity at Indian densities is compatible with substantial daily exposure to other groups in the surrounding blocks.
Indians also spend more of daily life in public. Few own cars. Homes are small and life spills into streets, markets, chai stalls, autorickshaws, and trains. An American car commuter crosses neighborhoods in a sealed private cabin. An Indian commuter walks, cycles, or takes public transit through them, sharing physical space with everyone on the route. The same index corresponds to very different levels of inter-group contact.