Fractal inequality in rural India: class, caste and jati in Bihar
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Oxford Open Economics, 2022, 1, 1–13
https://doi.org/10.1093/ooec/odab004
Advance access publication date 3 February 2022
Research Article
Fractal inequality in rural India: class, caste and jati in
Bihar
Shareen Joshi1 , Nishtha Kochhar2 and Vijayendra Rao3 , *
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
1 Schoolof Foreign Service, Georgetown University, 3700 ‘O’ St. NW, Washington, DC, 20057, USA
2 PovertyGlobal Practice, World Bank, 1818 H Street NW, Washington, DC, 20433, USA
3 Development Research Group, World Bank, 1818 H Street NW, Washington, DC, 20433, USA
*Correspondence address. Development Research Group, World Bank, Washington, DC, USA. Tel: +1 202 458 8034; E-mail: vrao@worldbank.org
Abstract
That inequality varies within and between groups is well understood. We explore how inequality can also be ‘fractal,’ salient not only
between sub-groups of groups but also between sub-groups of sub-groups. We demonstrate this, as a proof of concept using a limited
sample, in the case of Bihar one of India’s poorest states where caste has been a persistent driver of inequality. Caste is generally
analysed with government-defined ‘broad’ categories, such as Scheduled Caste (SC). In everyday life, however, caste is experienced as
‘jati’, a local system of stratification. We explore expenditure inequality at the jati level. Inequality decompositions show much more
variation between jatis than between broad-caste categories. We find that even within generally disadvantaged broad-caste categories
some jatis are significantly worse off than others and that inequality is largely driven by inequality ‘within’ jatis. We show that this
has implications for the implementation of large-scale poverty alleviation programs.
Keywords: caste, inequality, India, jati, policy targeting, Bihar
INTRODUCTION 1119. After 1976, additional orders were passed that
Fractal inequality is generally understood as a phe- enlarged the number of castes by adding more castes as
nomenon where income inequality persists regardless equivalent names and synonyms and sub-castes/tribes
of whether it is measured within a group, between and of existing SCs and STs. In 1990, an amendment was
within sub-groups of a group, between and within sub- passed by the Indian Parliament to include prior SC
groups of sub-groups and so on. The term is sometimes groups that had converted to Buddhism. The original list
attributed to Krugman (1994), but it arguably goes back to as well as updated lists of SC and ST groups for all states
the ‘father’ of fractal geometry, Benoit Mandlebrot (1982) is available through the Ministry of Social Justice and
himself. Though there is a fair amount of discussion in Empowerment, Government of India, at the following
the popular press, scholarly work on fractal inequality website (accessed on June 1, 2017): https://web.archive.
remains quite limited. In this paper, we study the org/web/20120913050030/http://socialjustice.nic.in:80/
archetypical case of caste in India. Inequality within sclist.php.). We take the analysis of inequality to a level
caste has generally been studied on the basis of broad below caste—to locally defined categories known as ‘jati’,
caste categories defined by the government of India, where caste is lived and experienced (Srinivas 1976;
such as Scheduled Caste (SC) and Scheduled Tribe (ST) Béteille 1996). A broad government caste category can
(British colonial administrators defined a broad category contain several jatis. We find that jati level inequality is
of ‘depressed classes’ in the Census of 1921, and then salient and of some policy relevance.
released an official list of socio-economically disad- Jatis are hereditarily formed endogamous groups
vantaged caste-groups in each province of India in the whose identities are manifested through occupational
Scheduled Caste Order of 1936 (Bandhyopadhyay 1992). status, property ownership, diet, gender norms, social
The Government of India Act of 1935 listed 417 groups in practices and religious practices, emphasizing purity and
the list of Scheduled Castes. The Constitution of India of pollution. Each region of India has several hundred jatis.
1950 increased this to 821. In 1956, this was raised to There is no pan-Indian system of ranking them, and the
Received: September 22, 2021. Revised: December 16, 2021. Accepted: December 18, 2021
© The Author(s) 2022. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution
License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original
work is properly cited.
2 | Oxford Open Economics, 2022, Vol. 1, No. 1
local rankings of jatis routinely change (Srinivas 1976; Lanjouw and Rao (2011) use a decomposition method
Bayly 2001; Rao and Ban 2007). Placement of jatis in adapted from Elbers et al. (2008) (known as the ELMO
broad government ‘caste’ categories is complicated and measure) to show that between-jati inequality in income
affected by politics and the level of mobilization achieved in one village, Palanpur in North India, decreased from
by the group (Rao and Ban 2007; Jaffrelot 2010; Cassan 39% in 1974/5 to 29% in 1983/84. More recent estimates
2015). from Palanpur show that this declined even further to
National surveys sponsored by the Indian government 17% by 2008/9 (Himanshu et al. 2013) (Another strand of
generally do not gather information about jati identity this literature uses the Oaxaca–Blinder decomposition,
(There was a Socio-Economic and Caste Census con- and its variants, to identify the structural drivers of
ducted in 2011, but the data have not been publicly inequality (Borooah 2005; Kijima 2006; Gang et al. 2008;
released and do not contain information on consumption Zacharias and Vakulabharanam 2011; Deininger et al.
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
or income and thus cannot be employed to study income 2013). A consistent theme of this literature is the
or consumption inequality.). However, the analysis of persistence of systematic disparities across different
broad caste categories in surveys has shown that disad- caste groups over long periods of time that are not fully
vantage in India correlates with caste status (Deshpande explained by differences in human capital investment,
2001, 2004; Dreze and Sen 2002; Thorat 2009; Govern- labor market returns or location of residence (state or
ment of India 2014, 2017; Jodhka 2017). Yet data on jati, rural/urban residence).).
together with socio-economic status, may hold the key To date, we are aware of no analysis of inequality at
to a deeper understanding of the caste system. This the jati-level using large sample surveys. This paper
lacuna has hampered poverty-alleviation and redistribu- attempts to fill this gap. We explore the intersections
tion programs. The Indian state is increasingly aware that jati, of caste and income class in rural India by analysing
jati is a source of ‘exclusion error’ in program rollout inequality between and within jatis in Bihar, one of
(Government of India 2017: 177). India’s poorest states. The data used in this paper were
Some non-government surveys have gathered some collected in 2011 as the baseline survey for an impact
jati-level identifiers. These data confirm the importance evaluation of a state-run rural livelihoods project called
of jati identity in modern India. Most marriages are con- JEEViKA (This program is also known as the Bihar Rural
tracted within jatis (Desai and Dubey 2012; Banerjee Livelihoods Project.). Consequently, our data have some
et al. 2013). Political mobilization and access to public limitations—it oversamples poorer, lower caste (SC)
services show strong variation across jatis (Banerjee and households and only covers seven districts. It is, thus,
Somanathan 2007; Huber and Suryanaryanan 2016). Jati- not representative of the population of the entire state.
based networks provide an extensive insurance network Thus, our results should be treated as suggestive, a
for credit, transfers and insurance during periods of vul- proof concept, rather than an authoritative statement
nerability (Mazzocco 2012; Munshi 2019). Jati-based dis- on the degree of jati-level inequality. However, the rich
parities are observed in educational attainment (Kumar information on caste, jati, consumption expenditure and
and Somanathan 2017), opportunities for employment vulnerability in the data do allow us to demonstrate
and out-migration (Munshi 2019) and women’s opportu- that jati matters in thinking about caste, class, socio-
nities to participate in markets and community life (Joshi economic status and inequality of access to poverty
et al. 2018). alleviation programs.
A key question that emerges from this literature is Our data confirm the persistence of caste-based
how much jati and income-classes overlap in India inequality, with SC and ST populations generally poorer
today. Are some groups ‘truly disadvantaged’ as a on average than households from higher-ranked castes.
result of both caste and class vulnerability (Wilson Using the ELMO measure of inequality decomposi-
1987)? Or have income differences expanded ‘within’ tion, which permits comparisons when the number
castes? Economists have sought to answer this question of groups varies, we find that decompositions using
by decomposing overall inequality in available sur- government caste categories show that the contribution
veys into two components: the sum attributable to of between broad-caste inequality to total inequality
differences in mean outcomes across caste groups stands at
Joshi et al. | 3
Table 1. Sample descriptive statistics, N = 8973
(a): Districts Percent of Sample (b): Caste Percent of Sample (c): Jati (sub-caste) Percent of Sample
Gaya 3.38 SC 69.93 SC: Chamar 20.44
Madhepura 31.06 ST 1.13 SC: Dobha/Dobh 2.51
Madhubani 5.07 OBC 16.9 SC: Dom 0.67
Muzzafarpur 18.51 EBC 4.65 SC: Dushad 16.72
Nalanda 5.58 Muslim 3.82 SC: Musahar 25.93
Saharsa 19.06 FC 3.58 SC: Pasi 0.89
Supaul 17.34 SC: Sardar 2.08
Other SCs 0.69
ST: Adivasi 0.86
Other STs 0.27
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
OBC: Dhanuk 1.19
OBC: Koeri 0.91
OBC: Kurmi 1.27
OBC: Shershabadia 0.78
OBC: Yadav 6.56
Other OBCs 6.17
EBC: Keuta 0.53
EBC: Mallah 0.76
EBC: Nat 0.99
Other EBCs 2.36
Muslim: Ansari 0.68
Other Muslims 3.14
FC: Brahmin 1.34
FC: Rajput 1.46
Other FCs 0.78
Source: Authors’ calculations based on data collected by Social Observatory, World Bank and Government of Bihar, Odisha and Tamil Nadu, respectively.
We also examine the role of caste in accessing large definitive, but more as a proof of concept that fractal
poverty alleviation programs; we examine the state-run inequality matters.
rural livelihoods program that was the reason behind Our analysis relies entirely on self-reported jati
the efforts to collect these data and the Mahatma identity, as reported by the head, or chief decision-maker,
Gandhi National Rural Employment Guarantee Scheme of the household (We rely on verbatim responses to
(MNREGS). We find that some jatis, even within targetted the jati and caste-status reported by the respondent
SC and ST groups, participate more than others. to the household module, who is the household head
This paper is organized as follows. Section 2 provides in majority cases. Where self-reported caste status
an overview of the survey. Section 3 presents results deviates from the rest of the jati group, we preserve the
on inequality within jatis. Section 4 illustrates how jati- jati response and assign the caste status of the jati’s
based inequality affects participation in poverty allevia- modal household. We have tried to not match these
tion programs. Section 5 concludes. verbatim responses to ‘official’ jati lists in order to let
our data, as much as possible, reflect the labels that
our respondents assign to themselves.). We identify
a jati as a distinct group in our sample if it has at
DATA least 0.5% households in the sample (Jatis with smaller
Our data are drawn from a 2011 survey of 9000 house- sub-samples are grouped together into a category we
holds located in seven districts in Bihar (Table 1). The call ‘Others’ defined separately for each caste group.).
survey was intended to provide baseline estimates of We classify self-reported jatis into broad caste groups
poverty for an impact evaluation of JEEViKA, a large according to government categories: SC, Extremely
anti-poverty program that aimed to empower poor rural Backward Classes (EBC), Other Backward Classes (OBC),
women. Households were randomly selected from major- Forward Castes (FC) and Muslims (The Government of
ity of the SC/ST hamlets within villages. As a result of this India has categorized disadvantaged castes into three
design, the survey oversamples SC and ST households. main categories—Scheduled Castes (SC), Scheduled
One possible advantage of this design is that it provides Tribes (ST) and Other Backward Castes (OBC). The
an opportunity to understand the heterogeneity among Government of Bihar also recognizes an additional
these broad groups along the lines of caste and jati. There group of Extremely Backward Classes (EBC), which were
are important disadvantages, however—the sample is originally included in the broader OBC group.). Table 1,
underpowered in understanding the full extent of jati- panel (b), provides basic descriptive statistics on jatis,
level inequality among higher castes, and it is almost as well as the broad caste groups. Note that SCs, which
certainly not representative of jati inequality at the state include seven main jati groups, account for almost 70%
level. Thus, our results should not be understood as of the sample. The next largest group is OBCs, which
4 | Oxford Open Economics, 2022, Vol. 1, No. 1 consists of five jatis. The EBC group is smaller, including almost entirely landless (0.07 acres of mean land hoding mainly three jatis. We also include two separate Muslim versus the SC average of 0.17 acres of land) and have low groups and two jatis under FCs. STs are
Joshi et al. | 5
Table 2. Sample characteristics by broad caste category
Mean monthly Median MPCE, Mean land Median land Proportion of Gini coefficient
per capita in rupees holding, in holding, in household
expenditure acres acres heads who
(MPCE), in completed some
rupees schooling
Panel (a): Caste 610.04 553.50 0.505 0 0.434 0.203
SC 600.29 546.53 0.174 0 0.368 0.194
ST 580.00 531.80 1.306 0.418 0.474 0.199
OBC 629.36 567.24 1.469 0.501 0.594 0.227
EBC 612.84 550.40 0.652 0 0.496 0.207
Muslim 657.88 605.00 0.424 0 0.452 0.200
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
FC 664.01 600.16 2.050 1.253 0.849 0.224
Panel (b): Jatis 610.04 553.50 0.505 0 0.434 0.203
SC: Chamar 634.55 575.67 0.146 0 0.463 0.199
SC: Dobha/Dobh 625.22 566.45 0.463 0 0.505 0.198
SC: Dom 662.51 581.60 0.015 0 0.237 0.175
SC: Dushad 601.80 548.07 0.270 0 0.449 0.198
SC: Musahar 560.91 520.08 0.075 0 0.226 0.180
SC: Pasi 639.55 579.41 0.534 0 0.563 0.209
SC: Sardar 668.98 623.75 0.501 0 0.333 0.214
Other SCs 619.74 583.43 0.109 0 0.387 0.174
ST: Adivasi 588.96 596.60 1.283 0.418 0.459 0.177
Other STs 551.26 471.02 1.378 0.334 0.522 0.257
OBC: Dhanuk 592.43 554.65 1.232 0 0.452 0.190
OBC: Koeri 575.14 526.78 1.437 0.543 0.728 0.209
OBC: Kurmi 639.90 582.36 1.180 0.752 0.609 0.222
OBC: 648.99 613.44 0.425 0 0.388 0.243
Shershabadi
OBC: Yadav 603.00 533.79 2.266 1.253 0.639 0.238
Other OBCs 667.91 604.84 0.863 0 0.577 0.218
EBC: Keuta 576.51 531.30 1.020 0 0.404 0.224
EBC: Mallah 615.59 548.67 0.251 0 0.258 0.204
EBC: Nat 653.77 583.77 0.617 0.167 0.616 0.224
Other EBCs 603.00 545.90 0.712 0 0.544 0.193
Muslim: Ansari 641.32 596.92 0.526 0 0.424 0.200
Other Muslims 661.46 607.08 0.402 0 0.459 0.200
FC: Brahmin 701.39 637.68 1.533 0.835 0.847 0.213
FC: Rajput 687.27 614.42 2.439 1.670 0.892 0.231
Other FCs 556.43 522.74 2.206 1.670 0.768 0.200
Source: Authors’ calculations based on data collected by Social Observatory, World Bank and Government of Bihar.
Figure 1.1. Kernel density of MPCE, Bihar
as the Chamars, Musahars and Yadavs, are quite well whether the differences between the groups are signif-
represented, at 20%, 26% and 7%, respectively (Table 1). icantly different from zero, we perform an exhaustive
Many other groups are quite small. To better understand pair-wise comparison of the MPCE of each jati with that6 | Oxford Open Economics, 2022, Vol. 1, No. 1
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
Figure 1.2. Distribution of MPCE by jatis, Bihar
of every other jati in the sample. Specifically, we use a Table 3. Tests of the null hypothesis of equality of medians of
monthly per capital expenditure (MPCE)
series of quantile regressions, with bootstrapped stan-
dard errors, to test the null hypothesis of equality of Broad caste and Number of Broad caste and Number of
Jati group rejections of the Jati group rejections of the
the coefficient of MPCE for each pair of jatis (For each
null hypothesis null hypothesis
jati pair (i, j) within the broad caste, we estimate the of equality of equality
quantile regression, MPCEhp = β0 +β1p Jatihp +hp , where h =
SC: Chamar 8 OBC: Dhanuk 0
i + j, and p identifies the jati pair (i, j).Jatih is an indicator
SC: Dobha/Dobh 5 OBC: Koeri 4
variable for household’s jati. The quantile regression is SC: Dom 1 OBC: Kurmi 0
estimated for h number of households that belong to SC: Dushad 6 OBC: 0
the two jatis. We bootstrap standard errors over 1000 Shershabadia
replications. The coefficient on the indicator variable SC: Musahar 11 OBC: Yadav 5
SC: Pasi 1 Other OBCs 3
β1p identifies the difference in median MPCE for jati
SC: Sardar 8 EBC: Keuta 1
pair, p.). The results are presented in Table 3. Note that Other SCs 0 EBC: Mallah 1
for the many groups in our sample—Chamars, Dobhas, ST: Adivasi 2 EBC: Nat 0
Dushads, STs, Musahars, Sardars and Yadavs—we reject Other STs 9 Other EBCs 3
the null hypothesis of equality of coefficients at least Muslim: Ansari 21 FC: Brahmin 23
Other Muslims 22 FC: Rajput 24
at the 10% level of significance. We interpret this as
Other FCs 25
evidence that these groups are distinct from others in our
sample. This is consistent with what we see in Figs 1.1 Notes: (1) Source: Authors’ calculations based on data collected by Social Obser-
vatory, World Bank and Government of Bihar. (2) We test the equality of median
and 1.2. per capita expenditure between all pairwise combinations of jatis, i.e. jati i and
Next, we decompose inequality into within- and jati j (i = 1, . . . ,24 and j = 1, . . . ,24), using quantile regression with bootstrapped
standard errors, over 1000 replications. For each jati pair, we implement this
between-group components. We use two measures. The with the bsqreg command in STATA with MPCE as the dependent variable and
binary for jati as the independent variable. The null hypothesis of equality is
first decomposes Theil’s L or GE(0), which belongs to rejected if the coefficient on jati is significant at least the 10% level. The 24 jati
the additively decomposable General Entropy class of groups in our sample have at least 0.5% representation in our sample.
inequality measures (Bourguignon 1979; Cowell 1980;
Shorrocks 1980) (For a distribution, (y1 , y2 , . . . , yN ), the IB ()
general formula of the General Entropyclass of inequal- I
.
For further details, refer to Cowell and Jenkins
yi α
(1995) and Elbers et al. (2008).). Our second measure,
ity measures is given by GE(α) = 1 N1 N i=1 y −1
α α−1 the ‘ELMO’ statistic (Elbers et al. 2008), normalizes
where y is mean income. An inequality measure, I, from a between-group inequality with the maximum possible
partition can be decomposed as I = IB ()+Iw (), where between-group inequality given the current income dis-
IB () is the inequality from between group differences tribution, relative sub-group size and their rank order. A
and IW ()is the inequality attributed to within group key advantage of this measure over conventional decom-
differences. Given a particular partition of the sample position techniques is that it allows for comparisons
and an inequality measure I, the conventional measure between populations with different numbers of groups
of between group inequality is given by RB () = and different population sizes. For these reasons, theJoshi et al. | 7
discussion that follows focuses on results from the ELMO earlier, affiliation to a broad caste group determines eligi-
measure (The traditional ELMO measure is given by bility for a wide range of government programs and ser-
R
IB () I
B () = = RB () ; where the vices. Our results suggest that some SC groups, for exam-
Max IB (j(n)),J Max IB (j(n)),J
ple, are truly vulnerable on a range of indicators, and
denominator is the maximum between group inequality
they are a small sub-group of the broader SC category.
that can be obtained by reassigning individuals across J
Conversely, many non-SC/ST groups are also equally vul-
subgroups in partition of size j(n). Note that R B () <
nerable.
RB ().Moreover, unlike the traditional measure, R B ()
does not necessarily increase with a finer partitioning
from the original sample (since denominator is the
maximum between-group inequality). A key property of RESULTS: HOW EFFECTIVE IS TARGETING
maximum between-group inequality is that sub-group
BASED ON CASTE?
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
incomes should occupy non-overlapping intervals. In the The results discussed above raise the question of
case of J sub-group partitions, to compute maximum whether aggregate caste groupings are an effective
between-group inequality: take a particular permutation strategy for targeting poverty-alleviation programs, and
of sub-groups {g(1) , . . . , g(J) }, assign lowest income to whether benefits disproportionately accrue to specific
group g(1) , second lowest to g(2) , and so on and compute jatis within these caste groupings. We now address this
the corresponding RB (). Repeat this for all permutations question directly by examining targeting in two anti-
of sub-groups and the highest resulting inequality poverty programs: (i) the Bihar Rural Livelihoods Program
among all permutations is the maximum sought. An (BRLP or JEEViKA) and (ii) the MNREGS.
illustration of this measure can be found in Lanjouw The two programs use two very different approaches to
and Rao (2011). In order to minimize the computational reach the poor. The BRLP requires project implementers
burden in estimating the maximum subgroup inequality, to identify possible beneficiaries and offer the program
we use an alternative measure proposed in Elbers et al. services to them. This approach, which we call ‘program-
(2008), which we refer to as the ‘alternative’ ELMO matic targeting’, assumes that it is up to the policy-
statistic. In addition to fixing the number and sizes maker to induce participation among the population of
of subgroups, it requires that subgroups be arrayed beneficiaries. MGNREGS, on the other hand, is a ‘self-
targeting’ program. It assumes that vulnerable individu-
according to their observed mean incomes—preserving
als who lack better options will choose to participate on
their rank order. This reduces the computational burden,
their own.
since it requires a single calculation, than J! calcula-
tions.).
Results on between-group inequality in MPCE are pre- Program 1: livelihoods programs
sented in Table 4. Note that in panel (a), which features We first analyse the State Rural Livelihoods Program.
broad caste groups, the contributions of ‘between’ caste These programs are community-driven development
inequality to total inequality is just 0.6% using the con- projects that organize women into self-help groups
ventional measure (column b) and 0.8% using the ELMO (SHGs) that enable them to save and access credit.
measure (column c). When we use jati-level groupings, They are now one of the most important anti-poverty
however, these go up >300% to 2.9 and 3.1%, respectively. programs run by the Government of India under the
Thus, considerably more inequality is explained by vari- umbrella of the National Rural Livelihoods Mission.
ations between jatis than variations across government We obtain information on SHG membership using the
caste categories (These estimates are lower than previ- endline survey for the evaluation, conducted in 2013
ously reported estimates of the contribution of between- (Endline surveys were collected at a gap of 1.5 to 2 years
caste inequality to inequality in India (Mutatkar 2005; from the baseline surveys. We merge the SHG member-
Subramanian and Jayaraj 2006; Borooah et al. 2014).). ship from these surveys with the baseline data.). The SHG
Finally, we examine how much inequality ‘between’ membership variable is an indicator for participation in
jatis drives inequality within broad caste groups. Panel the JEEViKA program. Since the programs treated all the
(b) of Table 4 presents decompositions of inequality by villages in treatment areas, with varying levels of take-up
jati for each broad caste group separately. Here, we see at the individual level, we restrict our analysis to house-
that the contribution of between-jati inequality to within holds in treatment villages. A total of 43% of women in
broad caste inequality is relatively low, varying from the treatment areas of our sample reported participating
0.03% for Muslims to 6.5% for FCs. Taken together, these in an SHG (In our sample, although membership under
estimates suggest that in the poorest communities of PVP SHGs is only 21%, SHG membership as a whole is
rural Bihar, inequality appears to emerge from variations almost 50%, which reflects the long history of the SHG
‘within’ caste and jati groups rather than ‘between’ them, movement in Tamil Nadu.).
which is consistent with other literature on the subject We use a simple linear probability model to regress a
(Lanjouw and Rao 2011; Himanshu et al. 2013). dummy variable for participation in JEEViKA on a set of
This observation is critical for an informed debate controls. These include age, age squared, marital status
about targeting poverty alleviation in India. As discussed of the woman, age at marriage, a dummy variable for8 | Oxford Open Economics, 2022, Vol. 1, No. 1
Table 4. Inequality in monthly per capita consumption expenditure
(a) Overall Inequality (GE(0)) (b) Conventional measure of (c) ELMO measure of
between group inequality between group inequality
Panel (a): Caste
6 broad caste groupings 0.0666 0.0064 0.0083
16 jati groupings 0.0666 0.0292 0.0305
Panel (b): Jati
SC (8 groups including others) 0.0610 0.0258 0.0297
ST(2 groups including others) 0.0660 0.0059 0.0118
OBC (6 groups including others) 0.0842 0.0164 0.0207
EBC (4 groups including others) 0.0683 0.0107 0.0114
Muslims (2 groups including others) 0.0641 0.0011 0.0029
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
FC (3 groups including others) 0.0835 0.0483 0.0652
Notes: (1) Source: Authors’ calculations based on data collected by Social Observatory, World Bank and Government of Bihar. (2) In Panel (b), we restrict the
sample to each broad caste category and estimate the inequality between jatis within that broad caste group. The number of jatis per broad caste category is
indicated in brackets.
some schooling of the female respondent, a dummy vari- due to occupational patterns or variations in the status
able for female household-headship, per capita expen- of women, the bias of program implementation teams,
diture, and its square, landholdings, education of the etc. can all make a difference to program participation
household head and household size. We also include rates. We do not examine those possible drivers here—
panchayat level fixed-effects. We do this analysis first we simply highlight that a program that was intended to
by adding government-defined broad caste categories, be for all vulnerable women within a group was, in fact,
with SCs as the omitted category, as additional right- more likely to be utilized by women from specific jatis.
hand side variables. Then we present a separate anal-
ysis for groups of jatis—first for SC/ST jatis, with non- Program 2: participation in MGNREGS
SC/ST as the omitted group, and then for non-SC/ST jatis Next, we examine the caste and jati-level variations in
with SC/STs as the omitted group. Regression results are participation in the MGNREGS, the government of India’s
reported in Table 5.1 (We report full regression results flagship welfare program that offers one hundred days
only for the broad caste regressions. For the jati regres- of work a year to anyone who asks for it at a reasonable
sions we only report the jati coefficients.). Key regression wage. Our surveys asked households about their posses-
coefficients are depicted graphically in the three panels sion of a ‘job card’—a basic prerequisite to access guaran-
of Fig. 2.1. teed employment under this program (Households must
Panel (a) of Fig. 2.1 shows that all broad castes have a have a job card to participate in the program. Each rural
lower probability of participating in SHGs than SCs. This household is entitled to a free job card with photographs
result matches the overall program design—the program of all adult members living in the household. These
was actively promoted among the lowest caste groups. adult members may then apply for employment and the
Figure 2.1b and c, however, shows that uptake within government is obliged to provide work within 5 km of the
these groups was far from uniform. We see that while applicant’s residence within a period of 15 days. Failure
most SC and ST groups in Bihar are more likely to partic- to obtain employment entitles the applicant to unem-
ipate in an SHG than non-SC/ST groups, Doms, Adivasis, ployment insurance (Dutta et al. 2014: 71).). As before,
‘Other SCs’ and ‘Other STs’ are less likely to participate, we use a linear probability model to regress a dummy
indicating a degree of jati-level variation in participation variable for possession of a job card using the same
in these programs. In panel (c), which presents estimates set of baseline controls as in the previous specifications.
for non-SC/ST jatis relative to excluded SC groups, partic- We use both broad caste categories and then jati-level
ipation is much lower for the Shershabedi. This group has regressions that first exclude non-SC/ST jatis and then
a 70 percentage points lower probability of participating exclude SC/ST jatis.
than SC respondents. Brahmins are 19 percentage points Results are presented in Table 5.2 and Fig. 2.2. When
less likely to participate than SCs, and Rajputs are 40 we use broad caste categories, we see the expected pat-
percentage points less likely. There is even heterogeneity tern of lower participation in non-SC/ST groups. However,
among Muslims: Ansaris are 38 percentage points less when we treat these groups as the omitted category and
likely to participate in these programs, but other Muslims examine participation within the SC/ST group, we see
are only 16 percentage points less likely than their SC considerable heterogeneity. We see in panel (b), for exam-
counterparts. ple, that Musahars show the strongest participation, fol-
The key finding here is that even though the program lowed by Chamars and Dushads, when compared to non-
aimed to offer the program to the entire groups, the SC/ST jatis. They are four times more likely to possess
actual take-up shows considerable variation by jati. a job card than Dobha households and almost twice as
There are many possible drivers of this outcome. likely to possess a job card as Sardar households. In
Geographic clustering, unequal demand for the program Figure 2.2(c), we observe heterogeneity in take-up amongJoshi et al. | 9
Table 5.1. Program targeting: JEEVIKA
(a) Government caste categories (b) SC/ST jatis (c) Non-SC/ST jatis
ST −0.160 SC: Chamar 0.140∗∗∗ OBC: Dhanuk −0.122
(0.089) (0.024) (0.075)
OBC −0.078∗∗∗ SC: Dobha/Dobh 0.048 OBC: Koeri −0.104
(0.023) (0.053) (0.073)
EBC −0.076∗ SC: Dom −0.235∗ OBC: Kurmi −0.185∗
(0.037) (0.102) (0.074)
Muslim −0.170∗∗∗ SC: Dushad 0.179∗∗∗ OBC: Shershabadia −0.700∗∗∗
(0.042) (0.024) (0.092)
FC −0.300∗∗∗ SC: Musahar 0.061∗ OBC: Yadav −0.044
(0.039) (0.024) (0.033)
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
SC: Pasi 0.077 Other OBCs −0.023
(0.077) (0.035)
SC: Sardar 0.044 EBC: Keuta −0.041
(0.072) (0.115)
Other SCs −0.352∗∗∗ EBC: Mallah −0.194∗
(0.091) (0.089)
ST: Adivasi −0.028 EBC: Nat −0.023
(0.105) (0.091)
Other STs −0.175 Other EBCs −0.057
(0.161) (0.046)
FC: Brahmin −0.195∗∗
(0.062)
FC: Rajput −0.402∗∗∗
(0.045)
Other FCs −0.266∗∗
(0.087)
Muslims: Ansari −0.379∗∗
(0.123)
Some schooling −0.120∗∗ Some schooling −0.150∗∗∗ Some schooling −0.124∗∗
(0.043) (0.043) (−0.043)
Female headed household 0.045∗ Female headed household 0.038 Female headed household 0.043
(0.022) (0.023) (0.022)
Per capita expenditure 0.118 Per capita expenditure 0.114 Per capita expenditure 0.108
(0.103) (0.103) (0.102)
Per capita expenditure squared −0.037 Per capita expenditure squared −0.040 Per capita expenditure squared −0.030
(0.053) (0.054) (0.053)
Land −0.017∗∗∗ Land −0.020∗∗∗ Land −0.019∗∗∗
(0.005) (0.005) (0.005)
Observations 4187 Observations 4187 Observations 4187
Adjusted R-squared 0.130 Adjusted R-squared 0.136 Adjusted R-squared 0.136
Notes: (1) SC is the omitted caste group in column (a), Non-SC/ST jatis are the omitted group in column (b) and SC/ST jatis are the omitted group in column (c).
(2) Each column represents a separate regression wherein household’s participation in NREGA is regressed on caste identity variables and a set of controls: age,
age squared, marital status, age at marriage, a dummy variable for any schooling; a dummy variable for household female-headship, per capital expenditure,
per capita expenditure squared, landholdings, education of the household head, number of members in the household and panchayat level fixed-effects. (3)
Robust standard errors are in brackets. (4) We report the level of significance: ∗ P value < 0.05, ∗∗ P value < 0.01 and ∗∗∗ P value < 0.001.
OBCs who are less likely to participate in MGNREGS, pre- program participation requirements, geographic rollout
sumably because of their higher socio-economic status, strategies, etc. The participation of specific jatis in each
Shershabedias are significantly less likely to participate program is, however, quite striking and an important
even in comparison to other OBC jatis. topic for further investigation.
In summary, the results from household participation These findings on the significance of jati in the par-
in both these anti-poverty programs suggest that ticipation of two very different poverty alleviation pro-
beneficiaries seem to be concentrated in specific jatis. grams may have important policy implications for the
The contrast between programmatic targeting and design of policies in contemporary India, and of targeting
self-targeting is also interesting. The two leading sets welfare programs more generally (Galasso and Ravallion
of beneficiaries are quite different, suggesting that self- 2005).
targeting has different effects by jati than programmatic
targeting. Moreover, inequality within jatis seems to
matter less for self-targeting than programmatic target- CONCLUSION
ing. We emphasize that the differences in the jati-level This paper has examined the phenomenon of fractal
variations in take-up of these two types of programs inequality, showing that caste inequality in some rural
could be driven by a variety of factors—incentives, areas of Bihar in India has important implications not10 | Oxford Open Economics, 2022, Vol. 1, No. 1
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
Figure 2.1. Regression coefficients on caste groups—JEEViKA
Figure 2.2. Regression coefficients on caste groups—NREGAJoshi et al. | 11
Table 5.2. Program targeting: MGNREGA
(a) Government caste categories (b) SC/ST jatis (c) Non SC/ST jatis
ST −0.099 SC: Chamar 0.239∗∗∗ OBC: Dhanuk −0.194∗∗∗
(0.058) (0.016) (0.047)
OBC −0.223∗∗∗ SC: Dobha/Dobh 0.063 OBC: Koeri −0.235∗∗∗
(0.015) (0.033) (0.045)
EBC −0.187∗∗∗ SC: Dom −0.065 OBC: Kurmi −0.180∗∗∗
(0.024) (0.055) (0.043)
Muslim −0.315∗∗∗ SC: Dushad 0.244∗∗∗ OBC: Shershabadia −0.660∗∗∗
(0.027) (0.017) (0.098)
FC −0.327∗∗∗ SC: Musahar 0.294∗∗∗ OBC: Yadav −0.228∗∗∗
(0.024) (0.015) (0.022)
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
SC: Pasi 0.022 Other OBCs −0.212∗∗∗
(0.049) (0.022)
SC: Sardar 0.166∗∗ EBC: Keuta −0.292∗∗∗
(0.052) (0.068)
Other SCs 0.069 EBC: Mallah −0.135∗
(0.065) (0.060)
ST: Adivasi 0.125 EBC: Nat −0.060
(0.064) (0.050)
Other STs 0.219 Other EBCs −0.229∗∗∗
(0.124) (0.032)
FC: Brahmin −0.405∗∗∗
(0.032)
FC: Rajput −0.330∗∗∗
(0.034)
Other FCs −0.166∗∗
(0.053)
Muslims: Ansari −0.304∗∗∗
(0.076)
Other Muslims −0.320∗∗∗
(0.028)
Some schooling −0.187∗∗∗ Some schooling −0.183∗∗∗ Some schooling −0.187∗∗∗
(0.029) (0.029) (0.029)
Female headed −0.051∗∗∗ Female headed household −0.050∗∗ Female headed household −0.053∗∗∗
household
(0.015) (0.015) (0.015)
Per capita expenditure −0.134∗ Per capita expenditure −0.116 Per capita expenditure −0.134∗
(0.067) (0.067) (0.067)
Per capita expenditure 0.032 Per capita expenditure squared 0.027 Per capita expenditure squared 0.032
squared
(0.034) (0.035) (0.035)
Land −0.021∗∗∗ Land −0.021∗∗∗ Land −0.022∗∗∗
(0.003) (0.003) (0.003)
Observations 8637 Observations 8637 Observations 8637
Adjusted R-squared 0.180 Adjusted R-squared 0.188 Adjusted R-squared 0.183
Notes: (1) SC is the omitted caste group in column (a), Non-SC/ST jatis are the omitted group in column (b) and SC/ST jatis are the omitted group in column (c).
(2) Each column represents a separate regression wherein household’s participation in NREGA is regressed on caste identity variables and a set of controls: age,
age squared, marital status, age at marriage, a dummy variable for any schooling; a dummy variable for household female-headship, per capita expenditure,
per capita expenditure squared, landholdings, education of the household head, number of members in the household and panchayat level fixed-effects. (3)
Robust standard errors are in brackets. (4) We report the level of significance: ∗ P value < 0.05, ∗∗ P value < 0.01 and ∗∗∗ P value < 0.001.
just when it is defined by government-defined broad Keeping this in mind, we find broad caste-based
caste categories, but at the jati level—defined a level differences in average monthly per capita expenditures;
below, where caste is lived and experienced. We use consistent with received wisdom, broadly defined
data from a large sample survey collected in a set of SC and ST populations are, on average, poorer than
rural and relatively poor districts of the state in 2011 to households from higher-ranked castes, but there is
compare patterns of inequality that emerge from the two considerably more variation by jati, which has a large
definitions of caste. The sampling strategy employed to influence on inequality. When we use broad-caste groups,
collect the data was designed to evaluate a women’s SHG the contribution of between-caste inequality to total
program. It is, therefore, not representative of the state inequality is relatively low at just 0.8%. When we use
of Bihar and our results should be treated as a proof of jati-level groupings, however, this number goes up to
concept rather than a definitive statement on the degree 3.2%. Even within some broad-caste categories, we find
of fractal inequality. that a great deal of inequality is driven by jati-level12 | Oxford Open Economics, 2022, Vol. 1, No. 1
variation. A total of 6.5% of FC inequality, for example, is (Besley et al. 2005) and community-based targeting has
explained by between-jati inequality. Some jatis, such as been shown to be very effective in Indonesia precisely
the Musahars, appear to be ‘truly disadvantaged’ within because it draws on local, contextual knowledge (Alatas
the disadvantaged groups. et al. 2012).
All this has implications for targeting anti-poverty pro- In sum, our analysis calls for a more nuanced, fractal
grams. We examine jati-level variation in participation approach to understanding group-based inequality. In
in the MGNREGS and the State Livelihoods Programs, India, sub-castes or jati matters not just in understand-
two of rural India’s most important efforts to alleviate ing group-based inequality, but in thinking about how to
poverty. We find that after controlling for a variety of address it with anti-poverty programs.
socio-economic variables, there remains a lot of variation ACKNOWLEDGEMENT
at the jati level. Dom and Adivasi jatis are less likely to The authors are grateful to seminar participants
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
participate in the Livelihoods program, while Dobhas and at the Delhi School of Economics, Indian Statistical
Sardars are less likely to have an MGNREGS card. Institute, Delhi, and the Workshop on Human Capital
These results have several implications for research at Indian School of Business, Hyderabad, for valuable
and policy. First, jati-level data collected by the gov- comments and suggestions. The authors also thank
ernment of India as part of the 2011 Socio-Economic two anonymous referees, Martin Ravallion, Ashwini
and Caste Census should be publicly released so that Deshpande, Amrita Dhillon, Abhiroop Mukhopadhyay,
some more headway can be made on understanding the E. Somanathan, Farzana Afridi, Francisco Ferriera, Mad-
extent of jati-level inequality across India. Second, this hulika Khanna, Hemanshu Kumar, Nethra Palaniswamy,
analysis highlights the need for nationally representative Rohini Somanathan, Irfan Nooruddin and Tarun Jain, for
sample surveys, such as the NSS, and NHFS, to collect insightful comments and discussions. They are indebted
jati information from households. This would permit a to the World Bank’s South Asia Food and Nutrition Secu-
broader and more nuanced understanding of the rela- rity Initiative (SAFANSI), funded by the EU and DfID, for
tionship between jati and inequality, and its implications financial support. This paper reflects the individual views
for poverty programs. of the authors and does not in any way represent the offi-
If the degree of between jati variation holds up with cial position of the World Bank or its member countries.
more representative samples, it may call for a redesign
of anti-poverty poverty programs to be targeted not just
at the broad-caste level but at the jati level. For instance,
we would argue, consistent with several other scholars,
that Musahars and other ‘maha-dalits’ jatis are ‘truly dis- References
advantaged’ and require specially designed interventions Alatas, V. et al. (2012) ‘Targeting the Poor: Evidence From a Field
that are better able to reach and support them. In the Experiment in Indonesia’, American Economic Review, 102.
case of women’s SHG programs, it suggests a concerted Banerjee, A., and Somanathan, R. (2007) ‘The Political Economy of
effort to make inroads into Musahar hamlets with more Public Goods: Some Evidence From India’, Journal of Development
sensitive program facilitation. In the case of NREGA, Economics, 82: 287–314.
it may require an effort to monitor implementation in Banerjee, A., Duflo, E., Ghatak, M., and Lafortune, J. (2013) ‘‘Marry for
villages with Musahar groups to reduce the relatively What?’ Caste and Mate Selection in Modern India’, American Economic
Journal: Microeconomics, 5: 33–72.
higher levels of discrimination they may face in obtaining
Besley, T., Pande, R., and Rao, V. (2005) ‘Participatory Democracy in
support from village governments in accessing the pro-
Action: Survey Evidence From South India’, Journal of the European
gram.
Economic Association, 3: 648–57.
The fractal nature of inequality at the jati level also Béteille, A. (1996) ‘‘Varna and Jati’, Sociological’, Bulletin, 45: 15–27.
suggests a more local approach to poverty targeting. Borooah, V. K. (2005) ‘Caste, Inequality, and Poverty in India’, Review
While SCs and STs are more disadvantaged than upper of Development Economics, 9: 399–414.
castes, the details of jati variation within them can vary Borooah, V. K. et al. (2014) ‘‘Caste, Inequality, and Poverty in India: A
considerably at the village level. It would be difficult, Re-Assessment’, Development Studies Research’, An Open Access
given the potentially high degree of local-level hetero- Journal, 1: 279–94.
geneity that we observe in these data, to legislate a Bourguignon, F. (1979) ‘Decomposable Income Inequality Measures’,
jati-level targeting policy at the national or even the Econometrica, 47: 901–20.
Brooks, M. (1946) ‘American Class and Caste: An Appraisal’, Social
state level. What might work better is to combine the
Forces, 25: 207–11.
current system of broad-caste and proxy-means test
Cassan, G. (2015) ‘Identity-Based Policies and Identity Manipulation:
targeting, with community-based selection via the sys-
Evidence From Colonial Punjab’, American Economic Journal: Eco-
tem of constitutionally sanctioned ‘gram sabhas’ (vil- nomic Policy, 7: 103–31.
lage meetings) that exist in rural India (Sanyal and Rao Cowell, F. (1980) ‘On the Structure of Additive Inequality Measures’,
2019). Drawing on the knowledge of local-level inequality Review of Economic Studies, 47: 521–31.
that village communities possess via the gram sabhas Cowell, F. A., and Jenkins, S. P. (1995) ‘How much inequality can we
by validating the proxy-means tested list of targeted explain? A methodology and an application to the United States’
households can reduce errors of exclusion and inclusion The Economic Journal, 105: 421–30.Joshi et al. | 13
Datt, G., and Ravallion, M. (1998) ‘Why Have Some Indian States Done States’ Anderson, S., Beaman, L., and Platteau, J. P., (eds) Towards
Better Than Others at Reducing Rural Poverty?’ Economica, 65: Gender Equity in Development Oxford University Press.
17–38. Kijima, Y. (2006) ‘Caste and Tribe Inequality, Evidence From India,
Deininger, K., Jin, S., and Nagarajan, H. (2013) ‘Wage Discrimination 1883–1999’, Economic Development and Cultural Change, 54: 369–404.
in India’s Informal Labor Markets: Exploring the Impact of Caste Krugman, P. (1994) Peddling Prosperity: Economic Sense and Nonsense
and Gender’, Review of Development Economics, 17: 130–47. in an Age of Diminished Expectations, New York: WW Norton and
Desai, S., and Dubey, A. (2012) ‘Caste in 21st Century India: Compet- Company.
ing Narratives’, Economic and Political Weekly, 46: 40. Kumar, H., and Somanathan, R. (2017) Caste Connections and Govern-
Deshpande, A. (2001) ‘‘Caste at Birth?’ Redefining Disparity in India’, ment Transfers: The Mahadalits of Bihar, Working Paper 270, Delhi:
Review of Development Economics, 5: 130–44. Centre for Development Economics.
Deshpande, A. (2004) ‘Decomposing Inequality: Significance of Caste’ Lanjouw, P., and Rao, V. (2011) ‘Revisiting Between-Group Inequality
Debroy, B., and Shyam Babu, D., (eds) The Dalit Question: Reforms Measurement: An Application to the Dynamics of Caste Inequal-
Downloaded from https://academic.oup.com/ooec/article/doi/10.1093/ooec/odab004/6520734 by guest on 22 June 2022
and Social Justice. pp. 33–52Globus Books. ity in Two Indian Villages’, World Development, Elsevier, 39: 174–87.
Dreze, J., and Sen, A. K. (2002) India: Development and Participation, USA: Mandlebrot, B. (1982) The Fractal Geometry of Nature, New York: WH
Oxford University Press. Freeman and Company.
Dubey, A., Gangopadhyay, S., and Wadhwa, W. (2001) ‘Occupational Mazzocco, M. (2012) ‘Testing Efficient Risk Sharing With Hetero-
Structure and Incidence of Poverty in Indian Towns of Different geneous Risk Preferences’, The American Economic Review, 102:
Sizes’, Review of Development Economics, 5: 49–59. 428–68.
Puja, D., Murgai, R., Ravallion, M., and van de Walle, D. (2014) Right to Mukul (1999) ‘The Untouchable Present: Everyday Life of Musahars
work? Assessing India’s employment guarantee scheme in Bihar. Equity in North Bihar’, Economic and Political Weekly, 3465–70.
and Development Series, Washington DC: World Bank. Munshi, K. (2019) ‘Caste and the Indian Economy’, Journal of Economic
Elbers, C., Lanjouw, P., Mistiaen, J., and Ozler, B. (2008) ‘Re-Interpreting Literature, 57: 781–834.
Sub-Group Inequality Decompositions’, The Journal of Economic Munshi, K., and Rosenzweig, M. R. (2006) ‘Traditional Institutions
Inequality, 6: 1569–721. Meet the Modern World: Caste, Gender and Schooling Choice
Galasso, E., and Ravallion, M. (2005) ‘Decentralized Targeting of an in a Globalizing Economy’, The American Economic Review, 96:
Antipoverty Program’, Journal of Public Economics, 89: 705–27. 1225–52.
Gang, I., Sen, K., and Yun, M. S. (2008) ‘Poverty in Rural India: Caste Mutatkar, R. (2005) ‘Social Group Disparities and Poverty in India’,
and Tribe’, Review of Income and Wealth, 54: 50–70. Working Paper Series No. WP-2005-004. Mumbai: Indira Gandhi
Government of India (2014) ‘Scheduled Communities: A Social Devel- Institute of Development Research.
opment profile of SC/ST’s (Bihar, Jharkhand and W.B)’, Plan- Newman, K. S. (2009) No Shame in my Game: The Working Poor in the
ning Commission Research Paper. http://planningcommission.nic. Inner CityVintage Press.
in/reports/sereport/ser/stdy_scmnty.pdf accessed 24 July 2016. Rao, V., and Ban, R. (2007) ‘The Political Construction of Caste in South
Government of India (2017) Economic Survey of India, Department of India’, in Mimeo, Washington DC: World Bank.
Economic Affairs, Economic Division: Ministry of Finance. Sanyal, P., and Rao, V. (2019) Oral Democracy: Deliberation in Indian
Hasan, A., Rizvi, B. R., and Das, J. C. eds (2005) People of India: Village AssembliesCambridge University Press.
Uttar Pradesh, XLII, Part I, II and III: Anthropological Survey of Shorrocks, A. (1980) ‘The Class of Additively Decomposable Inequal-
India. ity Measures’, Econometrica, 48: 613–25.
Himanshu, Lanjouw, P., Murgai, R. and Stern, N. (2013) ‘Nonfarm Srinivas, M. N. (1976) The Remembered Village, No. 26. University of
Diversification, Poverty, Economic Mobility, and Income Inequal- California Press.
ity: A Case Study in Village India’, Agricultural Economics, 44: Subramanian, S., and Jayaraj, D. (2006) ‘The Distribution of Household
461–73. https://doi.org/10.1111/agec.12029. Wealth in India’, Research Paper 2016/116, Helsinki: UNU-WIDER.
Huber, J. D., and Suryanarayan, P. (2016) ‘Ethnic Inequality and the Teltumbde, A. (2007) ‘Reverting to the Original Vision of Reserva-
Ethnification of Political Parties: Evidence from India’, World Poli- tions’, Economic and Political Weekly, 42: 2383–5.
tics, 1: 149–88. Thorat, S. (2009) Dalits in India: Search for a Common DestinySAGE
Hutton, J. H. (1946) Caste in India: Its Nature, Function and Origin, Publications India.
Bombay: Indian Branch. Thorat, A., Vanneman, R., Desai, S., and Dubey, A. (2017) ‘Escaping
Jaffrelot, C. (2003) India’s Silent Revolution: The Rise of the Lower Castes and Falling Into Poverty in India Today’, World Development, 93:
in North India. pp. 210–1 Columbia University Press. 413–26.
Jaffrelot, C. (2010) Religion, Caste, and Politics in India. Primus Books. Wilson, W. J. (1987) The Truly Disadvantaged: The Inner City, the Under-
Jodhka, S. (2017) Caste in Contemporary India Routledge. class, and Public Policy University of Chicago Press.
Joshi, S., Kochhar, N., and Rao, V. (2018) ‘Are Caste Categories Mislead- Zacharias, A., and Vakulabharanam, V. (2011) ‘Caste Stratification
ing? The Relationship Between Gender and Jati in Three Indian and Wealth Inequality in India’, World Development, 39: 1820–33.You can also read
