The assumption that economic growth, urbanisation, and modernisation would dissolve caste has deep roots in development theory (Deshpande and Darity 2016). But as Thorat and Newman (2007) have argued, caste-based discrimination is structurally embedded and “is not amenable to self-correction”, even as markets expand. Using broad administrative categories of caste groups from the Bihar caste census, Guilmoto and Himanshu (2024) showed significant heterogeneities across and within communities. The Telangana data now provides the most granular empirical test of that claim (IEWG 2023a, 2023b).

What the index shows is not a mild slope of disadvantage but a cliff. The gap is not incremental; as one summary of the Telangana report puts it, “it is exponential”.

These figures are drawn from the Telangana Socio-Economic, Educational, Employment, Political and Caste (SEEEPC) Survey, a census-scale enumeration covering more than 35.4 million individuals—roughly 97% of the state’s population which was carried out in late 2024. (Government of Telangana 2026a, 2026b, 2026c, 2026d). Unlike sample surveys that infer from parts, this dataset allows inequality to be mapped at the population level across 242 castes. It is hard to overstate what this enables: not just broad comparisons of Scheduled Castes (SCs) versus Other Castes (OCs), but also a granular picture of how hierarchy is distributed across communities that live next door, attend the same schools, and compete in the same labour markets.

[Editor's note: A discussion of some of the main features of the Telangana Caste Census, authored by Bhangya Bhukya, has been published in  The Telangana Caste Survey: An Overview on The India Forum.]

At the centre of this effort is the Composite Backwardness Index (CBI), a multidimensional measure built from 42 indicators spanning education, occupation, living conditions, assets, and social integration (IEWG 2026a, 2026b). The motivation is straightforward: single indicators—especially income—miss how disadvantage accumulates and persists across domains and generations. The survey itself is also remarkably detailed, built from 57 main questions and several sub-questions administered by enumerators visiting households across the state.

What the index shows is not a mild slope of disadvantage but a cliff. The gap is not incremental; as one summary of the Telangana report puts it, “it is exponential”. More than the phrase, the point is empirical: even in a fast-growing state, development and caste can run on separate tracks.

Backwardness Gradient

One way to understand what the SEEEPC Survey reveals is to picture the CBI distribution across all 242 castes, ranked from least to most backward. The distribution is not bell-shaped; it has a steep right tail, with a sizeable segment of castes sitting far above the state-average CBI of 81 (Figure 1). In plain terms, extreme multidimensional deprivation is not confined to a small residual group—it characterises a large set of communities.

Figure 1: Telangana: Distribution of Composite Backwardness Index (CBI) across 242 Castes

When we aggregate the index by broad social groups, the hierarchy becomes mathematically precise. SCs record an average CBI of 96.2; Scheduled Tribes (STs) 94.9; Backward Classes (BCs) 85.8; and Other/General Castes (OC/GC) 30.8 (Table 1). This is the survey’s central paradox made visible: modern growth has not produced a post-caste society. It has coexisted with a caste gradient that remains extraordinarily steep.

Table 1: Mean Composite Backwardness Index by Social Group

The comparison is not a matter of small differences. One summary of the government report states the implication directly: by the state’s own multidimensional metric, the average SC household is about three times as backward as the average upper-caste household. This is why income-based poverty lines struggle as a policy compass—the CBI captures gaps that are not reducible to cash shortfalls alone.

Table 2: Population Share of Backwardness

The scale of backwardness is also a political fact. Of the 242 castes assessed, 135—representing 67% of the state’s population—recorded a CBI score higher than the state average (Table 2). According to the SEEEPC report, the majority is backward; what looks like “targeting the vulnerable” is, in practice, about designing institutions that work for a large share of society.

Between-Group Inequality

A second way to read the Telangana findings is to ask a deceptively simple question: is caste inequality mainly about differences within groups—the popular “creamy layer” intuition—or mainly about differences between groups, that is, a structural boundary? The CBI score decomposition by IEWG (2023a, 2023b) answers this with unusual clarity. Of the total weighted variance in caste backwardness, 72% is explained by differences between the four broad social groups—SC, ST, BC, and OC—while only 28% lies within these groups (Table 3).

That proportion matters because it shows where the weight of inequality lies. If most inequality were within-group, broad caste categories would be blunt instruments, and fine-grained targeting based on class-like distinctions within each group could be justified. But when nearly three quarters of inequality is between groups, the main story is not a handful of people moving up—it is the persistence of a categorical boundary between upper-caste and oppressed-caste communities.

These results also assume significance in the context of the underestimation of between-caste inequalities in previous studies. Joshi et al. (2022), for instance, studied jati (sub-caste)-level inequality in Bihar and found that within-jati inequality was actually the dominant component—but they emphasise that this “fractal” pattern emerges only when one moves to finer sub-caste units, not broad categories.

Table 3: Decomposition of Overall Composite Backwardness Index Inequality

What Regression Adds

Descriptive gradients and variance decompositions are powerful because they show the structure of inequality. Regression results add a different kind of clarity: they quantify how strongly caste-group membership predicts multidimensional backwardness, even before other characteristics are taken into account. The regression analysis in Table A1 reports that caste group alone explains 57% of the variation in CBI scores (R-square = 0.572). Adding weights of caste-wise population size raises explanatory power significantly, to an R-square of 0.77.

This is where the Telangana results speak directly to a common policy reflex: that poverty reduction will automatically erode caste differences, on the assumption that caste differences are only income differences in disguise.

Table A1 also reveals a strong and consistent caste gradient in multidimensional backwardness. Compared to the Open Category (OC), the BC, SC, and ST groups exhibit significantly higher log(CBI), with SC and ST showing the largest disadvantages. The magnitude implies more than double the level of deprivation relative to OC. These effects remain stable across ordinary least squares (OLS) and population-weighted models, and after controlling for population size. The positive but modest role of population does not alter caste effects, indicating deeply structural inequalities.

This is where the Telangana results speak directly to a common policy reflex: that poverty reduction will automatically erode caste differences, on the assumption that caste differences are only income differences in disguise. The empirical claim here is the opposite—caste operates independently of class, and disparities persist even when controlling for income.

Extreme Tails

Averages can hide how harshly a hierarchy is structured at its edges. The Telangana caste-wise tables (Table A2) expose those edges. The five most backward castes include Dakkal (SC, CBI 116), Beda (SC, 113), Nakkala (ST, 112), Sindhollu (SC, 112), and Turaka Muslim (BC-E, 111). The five least backward include Kapu (OC, 12), Jains (OC, 13), Raju (OC, 17), Kamma (OC, 19), and Velama (OC, 19).

Two points follow directly from this. First, the hierarchy is not only steep but stretched, with some communities clustered near the survey’s reported theoretical maximum of deprivation while others occupy a low-backwardness floor. Second, Muslim backwardness is not uniform: Muslim communities classified under BC-E feature prominently among the most backward castes, with CBI scores exceeding 105, while Muslims classified under “Other OC” show significantly lower backwardness. This undermines the idea that any religious category can be treated as a monolith for policy purposes—the relevant unit in this dataset is often the sub-community.

Table 4: Five Most Backward Castes in Telangana (Based on Composite Backwardness Index)

Where the Gaps Are

The next question is where that gap is produced and reproduced. Across the observed indicators in the SEEEPC Survey, three mechanisms keep returning: unequal education, a segregated labour market, and social closure through endogamy and discrimination. Table 5 shows the striking results on these aspects.

Education: Telangana’s structural transformation has shifted the axis of stratification from land to human capital, making education the crucial gateway to mobility. Yet access to that gateway remains fiercely unequal, producing what the IEWG (2023a, 2023b) paper calls an “education trap”. The distribution of educational attainment displays a “sticky floor” pattern: enrolment may improve, but drop-off at higher levels is catastrophic for SC and ST populations, while upper castes dominate tertiary education in ways that translate into labour market advantages.

When occupational distributions of this kind persist, income transfers can soften hardship but struggle to change a labour market that keeps secure jobs and professional networks unequally distributed by birth.

The group-level figures capture this with stark simplicity. Only 20% of Other Backward Classes (OBCs), 19% of SCs, and 16% of STs have a diploma or higher education, compared to 32% of OCs. Schooling itself is stratified: private school enrolment for SCs and STs stands at just 10% and 8% respectively, compared to 30% for OCs.

Occupation: Education disparities feed directly into occupational segregation, producing what the IEWG (2023a, 2023b) paper calls a segregated labour market with little evidence of convergence. SC and ST workers are overwhelmingly concentrated in agricultural labour and precarious informal work — jobs with low wages, no social security, and little mobility — while upper castes dominate salaried, professional, and formal sector employment (Pradeep and Goli 2025).

A recent paper by Shah et al. (2025) shows that caste-based occupational segregation extends to climate vulnerability, with marginalised castes facing 25–150% steeper heat-stress exposure during heatwaves because of the occupations they are concentrated in.

The divergence in labour market outcomes was theorised as a secondary effect of educational disparity, and the numbers point strongly to the existence of a separated labour market. Outside OCs, a high concentration can be seen in the daily wage labour market: 46% of SCs, 41% of STs, and 32% of BCs are present there, compared to just 11% of OCs.

Given the persistence of such an unequal distribution, a mere income transfer will not address the causes of such a market structure. When occupational distributions of this kind persist, income transfers can soften hardship but struggle to change a labour market that keeps secure jobs and professional networks unequally distributed by birth.

Land ownership: Land ownership presents a similarly stark disparity across caste groups. While ownership among OCs stands at 31%, it drops to 8% among SCs. Figures for OBCs (15%) and STs (20%) also remain significantly lower than those of the OC group.

Social integration: The most troubling finding is that social practices continue to lock the hierarchy in place across generations. Inter-caste marriages remain extraordinarily rare, at around 5–6%, and endogamy remains nearly universal. Endogamy acts as the “social cement” of the hierarchy, reproducing advantage and disadvantage through marriage markets as much as through labour markets. Goli et al. (2013) similarly find that affirmative action plays only a limited role in promoting intergenerational social mobility, with endogamy acting as a persistent barrier.

Table 5: Education, Occupation, Deprivation, and Social Integration Indicators by Group

Urbanisation

The persistence of caste inequality in cities is not merely about income gaps—it extends to the very public goods that urbanisation is supposed to equalise (Demirguc-Kunt et al. 2018), suggesting that state institutions themselves can reproduce rather than dissolve horizontal inequality.

Telangana’s evidence does not deny that cities improve absolute living standards—it states this directly: a Dalit family in Hyderabad lives better than a Dalit family in a remote village. What it denies is the comforting inference that urbanisation dissolves hierarchy (IEWG 2023a, 2023b). As IEWG states, urbanisation moderates poverty but does not dismantle hierarchy, because upper castes capture a disproportionate share of urban opportunities—from better housing to professional networks—even as overall conditions rise.

The data further reveals that upper castes disproportionately capture the benefits of city growth, while SC and ST households remain concentrated in urban informal settlements with amenities barely superior to rural slums. This is consistent with the broader paper’s conclusion that growth and urbanisation improve averages but do not eliminate caste stratification—the relative gap remains virtually unchanged (Government of Telangana 2023a, 2023b, 2023c, 2023d).

Affirmative Action Not Enough

Schotte et al. (2023), in a systematic review, find that India’s quota system has been effective at increasing representation, but its impact on closing actual achievement gaps remains uneven. Poverty alleviation and income transfers—the mainstays of current policy—are fundamentally insufficient when inequality is structural rather than transitional. The reason is concrete: a family lifted above the poverty line can still face discrimination in hiring, exclusion from higher education, and social closure in marriage markets.

It also links school quality directly to intergenerational reproduction: without high-quality government schools in SC and ST-majority areas, the cycle of occupational segmentation will continue across generations.

Possibly for this reason, the SEEEPC Survey frames its message as a “math test”: income-based targeting has failed because it cannot close a gap that the CBI shows is large, multidimensional, and strongly anchored in categorical group differences.

The immediate lever emphasised across the IEWG (2023a, 2023b) paper is education quality, especially functional government schools. The IEWG states that strengthening government school systems and ensuring quality education is crucial, and explicitly criticises the policy obsession with enrolment over learning outcomes—which remain abysmal for disadvantaged castes. It also links school quality directly to intergenerational reproduction: without high-quality government schools in SC and ST-majority areas, the cycle of occupational segmentation will continue across generations.

Education alone, however, is not the full prescription. We call for a return to rigorous, caste-sensitive, multidimensional affirmative action in education, formal-sector employment, housing, and credit markets—calibrated not to broad categories but to the specific CBI scores of the 242 enumerated castes.

Two further implications follow from the data. First, policy cannot be one-size-fits-all, because not all backward castes are equally backward and heterogeneity within BC categories is substantial. Second, tribal communities face distinct barriers—geographic isolation, language, and cultural distance from mainstream institutions—suggesting targeted rather than generic interventions for ST educational backwardness, which the survey identifies as high even relative to SCs.

Conclusion

The findings based on Telangana’s SEEEPC Survey are not an isolated case. Using primary data from Uttar Pradesh, Tiwari et al. (2022) found nearly identical group rankings across the state: Dalit Muslims are the most deprived, followed consistently by Hindu Dalits at the bottom. What the Telangana SEEEPC Survey adds is the granularity of 242 castes within a single state, moving beyond the broad four-category view through a near-census approach.

Moreover, the Telangana SEEEPC exercise matters because it shifts the caste debate from assertion to measurement. It shows a steep right-tailed distribution of backwardness across 242 castes, with many castes far above the state-average CBI of 81. It shows a backwardness gradient in which SCs and STs remain at the top of the deprivation scale.

It shows that 72% of inequality is between broad groups, making categorical boundaries—not just within-group differences—the core problem. And it shows why growth alone does not free anyone: caste adapts, shifting from land to the classroom, from the village to the office, persisting even as incomes rise.

(I thank Professor Christopher Guilmoto for his detailed comments on an earlier version of this article. The author alone is responsible for any remaining errors.) 

Srinivas Goli is an associate professor at the International Institute for Population Sciences, Mumbai.