Caste and The Indian Economy: Kaivan Munshi

Journal of Economic Literature 2019, 57(4), 781–834
https://doi.org/10.1257/jel.20171307
Caste and the Indian Economy †

Kaivan Munshi*
Caste plays a role at every stage of an Indian’s economic life, in school, university,
the labor market, and into old age. The influence of caste extends beyond private
economic activity into the public sphere, where caste politics determines access to
public resources. The aggregate evidence indicates that there has been convergence in
education, occupations, income, and access to public resources across caste groups in
the decades after independence. Some of this convergence is likely due to affirmative
action, but caste-based networks could also have played an equalizing role by exploit-
ing the opportunities that became available in a globalizing economy. Ethnic networks
were once active in many advanced economies but ceased to be salient once markets
developed. With economic development, it is possible that caste networks will cease to
be salient in India. The affirmative action programs may also be rolled back and (sta-
tistical) discrimination in urban labor markets may come to an end if and when there
is convergence across caste groups. In the interim period, however, it is important to
understand the positive and negative consequences of caste involvement across a vari-
ety of spheres in the Indian economy. ( JEL G22, J15, J71, O15, O17, Z13)
1. Introduction to, the way he is treated by his teachers,

and his interactions with his classmates. In
C aste plays a role at every stage of an

Indian’s economic life. His caste will
determine the type of school he has access
young adulthood, his caste will determine
whether or not he benefits from affirmative
action in higher education and (later) access
to government jobs. Over the course of his
* Yale University. Research support from Cambridge working life, his caste will determine how
Institute for New Economic Thinking (INET), the he is assessed by potential employers, while,
National Science Foundation through grants SES-0431827 in parallel, networks organized around his
and SES-0617847, the National Institutes of Health
through grant R01-HD046940, and the Agence Nationale caste help him find new jobs. Overlapping
de la Recherche (ANR) under the EUR Project ANR-17- caste-based networks will provide him with
EURE-0010 is gratefully acknowledged. I am grateful to credit, help him start a business if he is
my collaborators Nancy Luke and Mark Rosenzweig for
many conversations that have shaped my thinking on this endowed with entrepreneurial talent, and
topic and to Editor Steven Durlauf, five anonymous refer- provide him with insurance against income
ees, and Guilhem Cassan for comments on the paper. I am shocks and major contingencies into old age.
responsible for any errors that may remain.
†
Go to https://doi.org/10.1257/jel.20171307 to visit the The influence of the caste extends beyond
article page and view author disclosure statement(s). private economic activity. Electoral politics
781
782 Journal of Economic Literature, Vol. LVII (December 2019)
in postindependence India has had a strong action program, whose stated objective was
caste component, with political parties tar- to eliminate the historical inequities of the
geting public resources to particular castes caste system. Internal cooperation within
in return for their votes. Caste politics castes supports economic networks, which
extends down to the local level. While it is compete with each other in the market with-
well known that local politicians target their out regard to their position in the social hier-
own castes, new research shows, in addition, archy. The much more visible affirmative
that cooperation within the caste can lead to action program, in contrast, was explicitly
an increased supply of non-excludable public targeted at large groups of historically dis-
goods. advantaged castes, giving them preferred
The importance of caste in India’s econ- access to higher education, government
omy and polity should not come as a sur- jobs, and political representation. These two
prise, given the structure of its society. forces have worked in tandem to shape the
Hindu society is stratified into four hier- evolution of the modern Indian economy.
archical classes, or varnas, with a large The aggregate evidence indicates that
subpopulation of untouchables excluded
there has been convergence in education,
from this system entirely. Within each of occupations, income, and access to public
these classes, and among the untouchables, resources across caste groups in the decades
are thousands of castes or jatis. The central after independence. Some of this conver-
rule in Hindu society is that individuals must gence is likely due to affirmative action, but
marry within their own caste. Recent genetic caste-based networks could also have played
evidence indicates that this rule has been an equalizing role by exploiting the opportu-
strictly followed for over two thousand years. nities that became available in a globalizing
Each caste consists of approximately 250,000 economy. It is possible that the caste-based
members spread over a wide area covering affirmative action programs will no longer
many villages (and, sometimes, urban loca- be needed at some point in the future. The
tions). Spatial segregation within the village available evidence indicates that caste dis-
results in a high degree of local social con- crimination, at least in urban labor markets,
nectedness, with caste clusters in distant is statistical, that is, based on differences
villages linked to each other through ties of in socioeconomic characteristics between
marriage over many generations. upper and lower castes. This type of dis-
The unusual social architecture described crimination will disappear if and when there
above has given rise to economic networks is convergence between these broad caste
organized around specific castes ( jatis) that groups. Caste networks and caste politics
are exceptional in their size and scope. While will also lose their relevance once the mar-
these networks smoothed the consumption ket economy and the democratic political
of their members in the face of income fluc- system begin to function efficiently. In the
tuations for centuries, they later expanded interim period, however, it is important to
the domain of their activity to the urban understand the positive and negative conse-
labor market and into business when new quences of caste involvement across a variety
opportunities became available under colo- of spheres in the Indian economy.
nial rule. Caste networks have continued to
play an important role in the Indian econ-
2. Caste in Indian Society
omy after independence in 1947. This period
witnessed, in parallel, the establishment The caste system is a system of social
of the world’s most aggressive affirmative stratification in India that goes back as far
Munshi: Caste and the Indian Economy 783
as 1500–500 BCE. The caste system com- category rather than their specific caste.
prises four hierarchical classes, or varnas, the However, any analysis of economic networks
Brahmins, Kshatriyas, Vaisyas, and Shudras. in the Indian economy must be based on
Certain population groups, known today as the caste or jati and not on the broad caste
Dalits, were historically excluded from the group (SC, ST, OBC) or the village. As dis-
varna system and were regarded as untouch- cussed below, a number of recent studies on
ables. Within each varna, and among the networks in India have ignored this critical
Dalits, are hundreds of castes or jatis (I will requirement.
use these terms interchangeably in the dis- Unlike the varnas, the castes were explic-
cussion that follows). itly established to serve an economic func-
The view among scholars of the British tion. Gupta (2000) posits that they were
colonial era, as articulated by Dumont initially occupational guilds, tracing their
(1970), was that the hierarchical aspect of emergence to the feudal period (seventh
the caste system, characterized by the var- to twelfthcentury). Other scholars have
nas, was largely symbolic. Upper castes claimed that jati identity and the rigid bound-
maintained higher levels of ritual purity than aries between castes that exist today were a
lower castes, for example, by being vege- creation of the British colonial government
tarian, but these differences had no direct (de Zwart 2000, Dirks 2001). Although the
economic consequences. Other research has British were not the first to enumerate the
contested this view, arguing that social sta- Indian population by caste—this goes back
tus allowed the upper castes to economically to Mughal rule (Appadurai 1993)—it is true
exploit the lower castes (Gupta 2000, Dirks that they assigned government jobs to partic-
2001). In recent decades, the affirmative ular castes, possibly creating a new jati iden-
action program put in place by the Indian tity or reinforcing an existing identity (Bayly
government has given the varna system a 2001). Recent genetic evidence, however,
new role in the Indian economy. This affir- places the origins of the jati system much
mative action program reserves positions in further back than was previously assumed.
institutions of higher education, the central This research indicates that marriage within
government, and local politics for individu- castes, the lynchpin of the caste system, was
als belonging to historically disadvantaged rigidly adhered to long before the feudal
castes. These castes comprise the former period. While there was free mixing in the
untouchable castes, which are collectively Indian population in ancient times, there
classified as scheduled castes (SCs), socially was an abrupt change 1,900–4,200 years ago,
and economically marginalized indigenous after which strict endogamy within narrowly
ethnic groups that are classified as scheduled defined population groups became the norm
tribes (STs), and, more recently, another (Moorjani et al. 2013).1 The Indian popula-
group of castes that are referred to as other tion today consists of 4,000 distinct genetic
backward castes (OBCs). groups, each of which is a caste (or its
When evaluating the affirmative action non-Hindu equivalent kinship group),

program, it is appropriate to compare broad
caste categories, that is, SC, ST, OBC versus 1 Bidner and Eswaran (2015) propose an economic
the other (forward) castes. This may also be explanation for the emergence of the caste system that is
appropriate when studying discrimination in based on the complementarity between a woman’s pro-
urban labor markets, where employers may ductivity and her husband’s skill. Their theory can explain
many features of the caste system, including its hereditary
care about the socioeconomic background nature, the key rule of marital endogamy, and its hierarchi-
of job candidates, which is proxied by caste cal structure.
aking India one of the most genetically

m the caste across villages. Marriage in India
diverse countries in the world (Singh 1992).2 is endogamous (within caste) and patrilocal,
The number of genetic groups that have that is, the bride moves into her husband’s
been identified is in line with the number household. Most women leave their natal vil-
of castes (approximately 3,000) listed in the lage when they marry. A dense web of mar-
colonial censuses.3 Thus, while colonial pol- riage ties thus forms over many generations,
icies may have enhanced caste affiliations, allowing caste networks to span a wide area
they did not create them. while maintaining their connectedness.
With 4,000 castes and a national popu- Although the population census of India has
lation of one billion, each caste comprises not collected information on caste affiliation
250,000 members on average. Although since 1931, recent rounds of nationally rep-
castes are spread over a wide area, some- resentative surveys such as the 1999 REDS
times an entire state, a single caste will not and the 2005 India Human Development
have a presence in all villages in the area Survey (IHDS) report that over 95 percent of
that it covers. It will cluster, instead, in select Indians continue to marry within their caste
villages that are often located far from each (or equivalent kinship community). This is
other. This clustering shows up clearly in the not surprising, given the genetic evidence
most recent round of the Rural Economic that has recently become available, and the
and Demographic Survey (REDS) con- same patterns of strict endogamy have been
ducted in 2006. The REDS is a nation- reported by respondents in other sample
ally representative survey of rural Indian surveys conducted in rural and urban India
households that has been administered by (Munshi and Rosenzweig 2006, Munshi
the National Council of Applied Economic 2011, Luke and Munshi 2011). Figure 1 uses
Research at multiple points in time over retrospective information from the 1999
the past four decades. Each survey round REDS to report rates of outmarriage, that
includes a census of all villages that are cov- is, marriage between members of different
ered, followed by a detailed survey of a sam- castes, in rural India for the children and sib-
ple of households in each village. The census lings of household heads over the 1950–99
data from the 2006 REDS indicate that the period. Outmarriage is just around 5 percent
mean number of castes per state is sixty-four, of all marriages—this number appears with
while the mean number of castes per village remarkable consistency across all the studies
is twelve. With 340 households on average listed above—and has remained stable over
in a village, this implies that a caste will have time.4
about thirty households in the select vil- In addition to marital endogamy, the dis-
lages where it locates. This is a large enough tinctive feature of the caste system is the
number to support a local caste community
within the village, while marriage ties link
4 It has been claimed in previous research, for exam-
ple Cassan (2015), that caste identity can be manipulated
2 Muslims marry within biradaris and Christians con- when there is an economic incentive to report an affilia-
tinue to marry within their original (pre-conversion) jatis. tion with another jati. While this type of manipulation may
Other religious communities such as the Parsis and the show up in government documents and census statistics, it
Jains also practice strict endogamy. does not imply that caste identities have actually changed.
3 The first colonial census to enumerate the population The genetic evidence and the remarkably stable patterns of
by caste was in 1871 and the last was in 1931. The Indian marriage obtained across many surveys suggest that caste
government has not collected this information since inde- boundaries have remained rigid for a very long period of
pendence in 1947. The most recent census round in 2011 time. Indeed, Christians continue to marry within their
did collect this information, but it has not been made pub- original jatis in rural India, despite having changed their
licly available. religion many generations ago (Luke and Munshi 2011).
14
12
10
0
1950–59 1960–69 1970–79 1980–89 1990–99
Figure 1. Marriage outside the Caste in Rural India
Note: Percentage of marriages by women in the household head’s family (sisters and daughters) outside their
caste.
Source: REDS 1999.
restrictions it places on social intercourse is a smaller spatial unit, the average caste’s
between castes, not just between upper share increases to 14 percent, indicative of
castes and lower castes, but also between the spatial clustering along caste lines that
castes occupying the same position in the characterizes the Indian village.
social hierarchy (Ghurye 1932). These Given the strong social ties within castes
restrictions result in spatial segregation, and the weak social interactions between
with members of a caste typically living castes, we would expect informal economic
together in a particular area of the village; transactions in the village to be concentrated
indeed, streets in Indian villages are often within the caste. The 2006 REDS village
referred to by their caste names. Munshi census asked each household head to list the
and Rosenzweig (2015) use data from the names of three individuals in the village that
2006 REDS to describe the spatial distribu- he would approach for food, in the event of
tion of castes within villages, as well as social a temporary shortfall, and for a short-term
and economic interactions in the village. loan. If these individuals were approached
Focusing on the 91 percent of REDS villages without regard to caste affiliation, we would
for which information on local government expect, on average, that 6 percent of the
( panchayat) wards is available, each caste individuals providing food transfers and
makes up, on average, 6 percent of the pop- loans would belong to the respondent’s
ulation of a village. Within the ward, which caste. Based on the identity of the first listed
individual, the REDS data indicate that as the preceding discussion, we would expect
much as 50 percent of food transfers and Indians to be similarly intolerant of neigh-
loans are from individuals belonging to the bors belonging to a d ifferent caste (this is
same caste. The REDS data also tell us that certainly reflected in how they live).
more than half of caste loans (by value) The sections that follow will examine
originate outside the village. If we included the role played by caste communities in
those loans, the fraction of individuals within a wide variety of economic activities. The
the caste that would be approached for a insider–outsider dichotomy will be a com-
loan would increase well above 50 percent. mon thread running through the discussion
Informal transactions are evidently concen- in each section. While internal cooperation
trated within the caste in the Indian village. allows castes to support the economic activ-
The insider–outsider dichotomy, char- ity of their members when market institu-
acterized by close intragroup relations and tions are missing or function imperfectly, the
weak (or even adversarial) intergroup rela- disregard (and even active hostility) toward
tions, is a feature of all traditional societies.5 outsiders creates new distortions in the
This dichotomy is especially pronounced in economy. There is no perfect substitute for a
Indian society due to its unique caste-based well-functioning market, and we will see that
structure, as discussed above, and is also evi- while informal caste-based networks may
dent from c ross-country comparisons based increase economic efficiency in the short
on the World Values Survey (WVS), docu- run, they can restrict the occupational and
mented by Munshi and Rosenzweig (2015). spatial mobility of their members in the long
Figure 2 reports cross-country results from run.
wave five of the WVS (conducted between
2005–09 in different countries) on questions
3. Mutual Insurance
relating to trust and tolerance of outsiders.
Restricting the sample to countries with a
3.1 Caste-Based Insurance
population in excess of twenty million that
are classified by the World Bank as low, Consider an economy with two risk-averse
lower-middle, or upper-middle income (with individuals, A and B. Each individual earns
the addition of Russia), we see in figure 2, either 1,000 or 2,000 per period with equal
panel A, that India ranks close to the top probability. Income realizations are indepen-
of the list with regard to trust in neighbors. dent across individuals, resulting in the fol-
Given the spatial segregation that character- lowing income states: (1,000, 1,000), (2,000,
izes the Indian village, the Indian respon- 2,000), (1,000, 2,000), (2,000, 1,000), where
dents are essentially reporting that they have the first term in parentheses refers to indi-
a high degree of trust in their fellow caste vidual A’s income and the second term refers
members living nearby. However, the rank- to individual B’s income and each income
ing reverses when we look at tolerance of state occurs with probability 0.25.
outsiders—speaking a different language or Given that individuals are risk averse,
following a different religion—in figure 2, they will benefit from a mutual insurance
panels B and C, respectively. India now lies arrangement that smooths their consump-
at the very bottom of the list and, based on tion over time. Individuals who receive a
positive income shock make transfers to
5 Akerlof (1976) and Choy (2018) provide game theo-
temporarily less fortunate individuals in
retic foundations for why internal cooperation coupled with these arrangements, with the roles revers-
the exclusion of outsiders can be sustained in equilibrium. ing in later periods when the income shocks
Panel A. Trust in neighbors

100
90
80
70
60
50
40
30
20
10
0
M eru
R na
Af a
Tu ine
n d
al ia
Ch co
et ia
t
lo zil
ico
G ia
So rge sia
U rica
Th rkey
hi a
or a
Eg m
yp
h in
do n
Et esi
M aysi
in
M op
Vi Ind
b
oc
ha
na
Co Bra
In aila
ut nt
A us
a
ex
P
kr
Panel B. Tolerance of neighbors speaking a different language
100
90
80
70
60
50
40
30
20
10
0
Vi dia
do m
or a
Tu co
G ey
ai a
M land
Ru ia
Et ssia
Ch ia
U ina
So Me e
h co
Br a
il
ge ru
a
M esi
Th n
ric
in
ain
az
ys
op
In na
Ar Pe
rk
oc
ut xi
ha
In
nt
ala
Af
n
kr
hi
et
Panel C. Tolerance of neighbors following a different religion

100
90
80
70
60
50
40
30
20
10
0
a
or n
co
Vi raq
G nd
ala a
M pia
lo ne
So B ru
a
Ar Afr l
do ey
U ico
ge ica
T am
ail ia
Ch ia
R na
hi a
h zi
di
M han
in
M Ira
Et ussi
bi
Th nes
ys
Pe
oc
Co krai
ut ra
a
In urk
i
In
nt
I
ex
m
n
et
Figure 2
Notes: Panel A reports the percentage of respondents who trust people in their neighborhood. Panel B reports
the percentage of respondents who do not have a problem with their neighbors speaking a different language.
Panel C reports the percentage of respondents who do not have a problem with their neighbors following a
different religion.
Source: WVS, wave five.
are reversed, leaving everyone better off. An in the group around which mutual
alternative (equivalent) way to c haracterize insurance is organized.
mutual insurance arrangements when pref- (ii) Conditional on total income, Y  s, notice
erences are logarithmic is that the total that individual income, y  si  , has no bear-
income of all participants is pooled in each ing on consumption, c  si  . This is evident
state or period and then distributed on the from our simple example; consumption
respecified income-sharing rule,6
basis of a p is equal to 1,500 in states with unequal
incomes, regardless of whether the
λ s individual earns 1,000 or 2,000.
(1)
c si  = _
  i      Y  ,
∑ i  λi
In our example, income shares are equal.
where c  si  is the consumption of individual i More generally, individuals earning higher
in income state s , Y  sis total income (over all incomes on average will have higher income
participants in the insurance arrangement) shares. This is accounted for in the test of
in state s, and λi/∑ i  λi is individual i’s share full risk sharing by including individual fixed
of the pot, which is constant across all states. effects, which capture the last term in equa-
In our simple example, the income-generat- tion (2). Conditional on group income in a
ing process is the same for the two individu- given period, shocks to individual income,
als, and so it is reasonable to assume an equal that is, deviations from average income,
income-sharing rule: λA = λB = 1 / 2. The should have no effect on individual con-
usual assumption in models of mutual insur- sumption with full r isk sharing.
ance is that there is no storage or savings, Starting with Townsend’s (1994) semi-
which results in three consumption states: nal contribution set in village India, studies
(1,000, 1,000) with probability 0.25, (2,000, from across the developing world have docu-
2,000) with probability 0.25, and (1,500, mented extremely high levels of risk sharing
1,500) with probability 0.5. (close to full risk sharing) in rural communi-
The preceding discussion serves as the ties, for example Grimard (1997); Fafchamps
basis for a test of full r isk sharing. Taking logs and Lund (2003); and Angelucci, De Giorgi,
in equation (1), and Rasul (2015). These communities
are evidently able to smooth income risk
λ
( ∑ i  λi )
extremely effectively, despite the fact that
(2) log (c  si  ) = log (Y  s)  + log _
  i     .
private credit and government safety nets are
largely absent in the economies within which
The preceding equation can be used to they operate. While this research has shed
derive two testable predictions: new light on the important role played by
informal nonmarket institutions in smooth-
(i) Replacing states, s, with time periods, t, ing income risk in developing economies, it
individual consumption will co-move suffers from two limitations.
with total income (or total consumption) The first limitation is that the social group
within which risk sharing is assumed to take
6 More generally, with full r isk sharing, the ratio of mar-
place is often chosen on the basis of con-
ginal utilities between any two members of the arrange- venience. For example, Townsend (1994)
ment is constant in all income states. With log preferences, treats the village as the relevant social group,
this implies that the ratio of consumptions are constant. whereas our characterization of Indian society
Summing over all network partners, this implies that each
individual receives a constant fraction of the total income would suggest that the caste (within and across
that is generated in each state. villages) is the social group around which
mutual insurance arrangements are organized to establish full risk sharing at the level of
in rural India. The available evidence supports the caste but not the village.
the view that the caste or jati, rather than the Recent research continues to ignore the
village, is the relevant group around which central role played by the caste in smoothing
mutual insurance arrangements are organized consumption in rural India (among the excep-
in rural India. Caldwell, Reddy, and Caldwell tions are Mobarak and Rosenzweig 2012
(1986) surveyed nine villages in South India and Munshi and Rosenzweig 2016). Morten
after a t wo-year drought and found that nearly (2019) treats the village as the risk-sharing
half (46 percent) of the sampled households unit, completely ignoring caste, just as
had taken consumption loans during the Townsend did. Ambrus, Chandrasekhar,
drought. The sources of these loans (by value) and Elliot (2014) partition the village into
were government banks (18 percent); mon- two broad caste groups: scheduled caste/
eylenders, landlords, or employers (28 per- scheduled tribe (SC/ST) and general merit/
cent); and relatives and members of the same other backward caste (GM/OBC). As dis-
caste community (54 percent), emphasizing cussed previously, the SC, ST, OBC catego-
the importance of caste loans for smoothing ries group castes with similar social status for
consumption. the purpose of affirmative action programs
The REDS, which provides detailed implemented by the Indian government.
information on gifts and loans for a sam- Jatis within a broad caste group do not share
ple of villages covering all the major Indian a common history nor do they share a special
states, can be used to corroborate these social bond. The test of full risk sharing can
findings. Focusing on the 1982 REDS be implemented with data from any subset
round, which included a detailed classifica- of individuals belonging to the relevant social
tion of the source and purpose for each loan, group. However, it will erroneously reject full
the key findings reported in Munshi and risk sharing if it includes individuals outside
Rosenzweig (2016) are that (i) caste loans the group, such as individuals outside the
are second in overall value after bank loans, caste residing in the same village. An addi-
and (ii) caste loans are the primary source tional limitation of previous studies is that
of support for meeting major contingencies they do not collect information on network
such as illness and marriage. Gifts make up links outside the village. This limitation is
the bulk of transfers within the caste (67 per- especially relevant when the objective is
cent by value). If we included gifts, the caste to measure the household’s centrality in its
would be by far the most important source insurance network, as in Ambrus, Gao, and
of support for smoothing consumption and Milán (2017). As noted, the caste network
meeting major contingencies in rural India. spans a large number of villages.
As noted by Munshi and Rosenzweig, much The second limitation of the standard test
of this support comes from caste mem- of full risk sharing is that it does not account
bers outside the village. The same pattern for other important aspects of consumption
holds up in later REDS rounds, such as the smoothing. Returning to our simple exam-
1999 round, and in other nationally repre- ple, suppose now that there are N individ-
sentative surveys such as the 2005 IHDS. uals in the mutual insurance arrangement.
Providing the most direct evidence that the As before, income realizations are indepen-
caste, rather than village, is the relevant dent across participants, and each individual
risk-sharing unit, Mazzocco and Saini (2012) earns 1,000 with probability 0.5 and 2,000
use International Crops Research Institute with probability 0.5. The income-generating
for the Semi-Arid Tropics (ICRISAT) data process is the same for all individuals, so we
continue to assume an equal income-sharing network provides a vital and irreplaceable

rule. service (in addition to other services when
It is straightforward to show that mean markets function imperfectly that will be dis-
consumption with insurance, MI, is equal cussed in the sections that follow).7 This may
to mean income or, equivalently, mean con- explain the continuing relevance of the jati
sumption under autarky, M A, which is 1,500 in the Indian economy and society. To the
in our example: extent that marriage within the jati is a pre-
requisite for access to c aste-based networks,
( N i i ) N ( A)
  1   ∑   y  s    =  _
this also explains why most Indians continue
(3) MI  = E _ 1   NM   = M .
A
to marry within their caste.
3.2 Caste-Based Insurance and Restrictions
Although mean consumption is independent
on Mobility
of the number of participants, the variance of
consumption with insurance, VI, is decreas- The preceding view of c aste-based insur-
ing in N: ance is that it has emerged, and persisted,
because market credit and government
( N i i ) N  2( A) N
  1   ∑   y  s    =  _
VA
(4) VI  = V _ 1   NV   =  _    . safety nets are inadequate. While this view
would suggest that caste-based insurance
Extremely high levels of risk sharing have networks are efficiency enhancing, Munshi
been documented throughout the develop- and Rosenzweig (2016) show that they can
ing world. However, risk sharing does not inadvertently generate distortions in other
map one for one into consumption smooth- markets, specifically the labor market.
ing. There is full risk sharing in our exam- Their research is motivated by two stylized
ple with two individuals, but consumption facts: (i) the rural–urban wage gap in India
varies substantially across states. The exist- is exceptionally high relative to other large
ing literature has failed to recognize this dis- developing countries and has remained this
tinction, focusing exclusively on risk sharing way for many decades, and (ii) the rate of
and, more recently, on deviations from full permanent male migration has, nevertheless,
risk sharing. Given that the deviations from remained exceptionally low over this period.
full risk sharing are generally small, what This implies that there must be some fric-
makes caste-based networks exceptional in tion that prevents rural workers from mov-
their ability to smooth consumption (which ing to the city and taking advantage of what
is what we ultimately care about) is their size appears to be an arbitrage opportunity.
and scope. A caste consists of thousands of Munshi and Rosenzweig posit that the
individuals spread over a wide area, whereas source of this friction is the loss in (rural)
informal networks in other countries are
often organized around an extended family
7 Data from the 1982 REDS indicate that loan terms
residing in the same village; see, for exam-
are substantially more favorable for caste loans, on aver-
ple, Angelucci, De Giorgi, and Rasul (2015). age. Caste loans are significantly more likely to be interest
Large Nreduces the variance of consump- free than loans from banks, employers, and moneylenders.
tion, and the wide spatial scope of caste net- They are also significantly more likely to be collateral free
than loans from banks. Among the caste loans received
works diversifies income shocks, reducing in the year prior to the 1982 survey, 20 percent by value
the variance of consumption even further. required no interest payment and no collateral. The cor-
Without adequate access to private credit responding statistic for the alternative sources of credit
was close to zero, except for loans from friends where 4
and government safety nets, the exception- percent of the loans were received on similarly favorable
ally well-functioning caste-based insurance terms. Similar results are obtained with the 2005 IHDS.
network insurance when one or more mem- igration, however, it is necessary to take a
m
bers of a household migrates permanently step back and model the ex ante p articipation
to the city. Households with migrant mem- decision and the optimal design of the
bers will have reduced access to rural caste income-sharing rule. Under the assumption
networks for two reasons. First, migrants that there is full risk sharing, households can
cannot be as easily punished by the net- either remain in the village and participate in
work, and their family back home in the the insurance network or send one or more
village now has superior outside options (in of their members permanently to the city,
the event that the household is excluded increasing their income but losing the ser-
from the network). It follows that house- vices of the network. The sharing rule that
holds with migrants cannot credibly commit is chosen in equilibrium determines which
to honoring their future obligations at the households benefit more from the insurance
same level as households without migrants. network and, therefore, choose to stay.8
Second, an information problem arises if the To derive the sharing rule, and associated
migrant’s income cannot be observed. If the migration, we begin with the household’s
household is treated as a collective unit by decision rule. With log preferences, the
the network, it always has an incentive to expected utility from consumption can be
misreport its urban income so that transfers derived as an additively separable function
flow in its direction. If the resulting loss in of mean consumption and its coefficient of
network insurance from migration exceeds variation (squared). With this specification,
the income gain, then large wage gaps could a household will choose to remain in the
persist without generating a flow of workers rural insurance network and forego its urban
to higher-wage areas. This distortion is para- opportunities if the following condition is
doxically amplified when the informal insur- satisfied:
ance networks work exceptionally well, as
VI
in the Indian case for reasons listed above, (5) MI)  −  _
log( 1    _     
2 M  2I 
because rural households then have more to
lose by sending their members to the city.
VA
The simplest test of the hypothesis MA)  −  _
≥ log( 1   β  _     + ϵ.
2 M  2A  
that the potential loss in network services
restricts mobility in India would be to com-
pare migration rates in populations with and As defined above, MI, VIrefer to the mean
without c aste-based insurance. This exercise and variance of consumption with insurance.
is infeasible, given the pervasiveness of caste Here, MA , VA are the mean and variance of the
networks. What Munshi and Rosenzweig do household’s income, that is, its consumption
instead is look within the caste and theoret-
ically identify which households benefit less
8 In related research, Morten (2019) derives the rela-
(more) from caste-based insurance. They
tionship between temporary migration and mutual insur-
then proceed to test whether it is precisely ance. Her model extends Ligon, Thomas, and Worral
those households that are more (less) likely (2002) by allowing households to enhance their income
to have migrant members. in low r ural-income states by sending migrants temporar-
ily to the city. As in all models of ex post risk sharing, the
The literature on mutual insurance is size of the network is exogenously determined and there
concerned with ex post risk sharing, tak- is no (permanent) exit. Morten follows Ligon, Thomas,
ing the size of the network and the shar- and Worral by allowing the income-sharing rule to adjust
in particular income states to satisfy the enforcement con-
ing rule as given. To derive the connection straint. However, the baseline income-sharing rule (in the
between insurance networks and permanent absence of these constraints) is exogenously determined.
in autarky when all its members remain in generated by the insurance arrangement
the village. This income would increase can be increased by redistributing income
to MA (1  + ϵ̃ )  if one or more household mem- so that relatively poor households consume
bers moved to the city, where ϵ  ≡  log (1  + ϵ̃ )  more than they earn on average, that is,
measures the household’s urban opportunity MI  > MA
. This gain from redistribution
(in utility units), which is unobserved by the must be weighed against the cost to the
rest of the rural network. members of the network from the accom-
While migration will increase the house- panying decline in its size, since relatively
hold’s income, this increase must be weighed wealthy households will now be more likely
against the additional consumption risk to leave and smaller networks are less able
that a household with migrants must face. to smooth consumption. The first prediction
If a household remained in the village but of the model is that the income-sharing rule
did not have access to network insurance, will nevertheless be set so that there is some
then its consumption risk, which would amount of redistribution in equilibrium.
be the same as its income risk, would be This implies that relatively wealthy house-
(1/2)(VA /M  A2  ). The risk parameter β reflects holds within their caste benefit less from the
both the decline in income risk through the network and so will be more likely to have
diversification in household income that migrant members ceteris paribus.
accompanies migration as well as the avail- The second prediction of the model is that
ability of n onnetwork insurance. We thus households facing greater rural income risk
expect that β  < 1 . Nevertheless, the key will be less likely to have migrant members.
assumption of the model is that VI/M  I2    <  With full risk sharing, all households in the
β (VA/M  A2  ). This is the wedge that permits insurance network face the same consump-
tion risk, regardless of their income risk.
large wage gaps to exist in equilibrium.
This implies that households facing greater
Retaining the assumption that the rural
income risk benefit more from the network
income-generating process is the same
and so will be more likely to stay. This pre-
for all households, which implies an equal
diction is especially useful in distinguishing
income-sharing rule and, hence, from equa-
Munshi and Rosenzweig’s hypothesis, based
tion (3) that M I  = MA, a household will only
on the rural insurance network, from alter-
send migrants to the city if ϵ  > β (VA /M  A2 )    − native explanations for the large wage gaps
(VI/M  I2  )
. Since we have fixed the and low migration in India. One alternative
income-sharing rule, for the time being, we explanation is that a small number of castes
only need to solve for equilibrium partici- fortuitously managed to establish networks
pation, that is, the threshold ϵI above which in the urban labor market, extracting wage
households will send migrants to the city. rents for their members and excluding out-
Once we introduce heterogeneity in rural siders. There are also alternative explana-
incomes, the i ncome-sharing rule and migra- tions available for redistribution within the
tion in each income class must be simultane- caste and the increased exit from the network
ously determined. Munshi and Rosenzweig by relatively wealthy households. However,
(2016) assume that the community’s objec- none of these explanations imply that house-
tive is to maximize the total surplus generated holds facing greater rural income risk should
by the insurance arrangement (placing equal be less likely to have migrant members.
weight on all income classes), taking account Munshi and Rosenzweig begin their
of participation in each income class. With assessment of the theory by showing that
diminishing marginal utility, the total surplus there is substantial redistribution of income
within castes, using data from the most network in place, indicate that a 50 percent
recent ICRISAT panel survey and from improvement in risk sharing for households
the 2006 REDS. We know very little about with migrant members (which is still some
the distribution of resources within castes, way from full r isk sharing) would more than
and this new finding will be corroborated double the migration rate from 4.3 per-
below with data from urban India. Munshi cent to 9 percent. In contrast, doubling the
and Rosenzweig then show, using the 2006 rural–urban wage gap from 18 percent to
REDS village census, that relatively wealthy 36 percent without any change in formal
households within their caste are significantly insurance increases the migration rate by
more likely to report that one or more adult just 1.5 percent. These results emphasize the
male members have permanently left the vil- central message of this research, which is that
lage. Highlighting the role that rural income exceptionally well-functioning (caste-based)
risk plays in the migration decision, they also rural insurance networks coupled with inade-
find that households with a higher coefficient quate access to formal insurance, rather than
of variation in their (rural) income—who wage differentials as commonly assumed in
benefit more from the rural insurance net- models of migration, may explain much of
work—are less likely to have migrant mem- the low mobility in India.
bers. Additional results, using the 1982 and One way to circumvent these restrictions
1999 REDS rounds, directly support the on mobility would be for members of the
key assumption of the model, which is that rural community to move to the city as a
migration should be associated with a loss group. Members of the group could monitor
in network services (and increased marriage each other and enforce collective punish-
out of the caste). ments, solving the information and commit-
Having found evidence consistent with ment problems described above. They would
the theory, Munshi and Rosenzweig proceed also help each other find jobs at the destina-
to estimate the structural parameters of the tion. As described in the sections that follow,
model. Migration and the income-sharing the history of industrialization and urban-
rule (the λ s in equation (1)) are determined ization in India is indeed characterized by
jointly in the model. With logarithmic pref- the formation and evolution of caste-based
erences, the λ
s map directly into relative urban networks, sometimes over multiple
consumption across income classes. The generations. We will see that a limitation of
estimated λs match closely with relative con- this strategy is that a sufficiently large (com-
sumption, which is available from the REDS mon) shock is needed to jump-start the new
but not used to estimate the model. This close network at the destination, and such oppor-
match effectively validates the model, pro- tunities occur relatively infrequently. Thus,
viding empirical support for the assumptions while members of a relatively small number
that preferences are logarithmic and that the of castes with (fortuitously) well-established
community’s objective is to maximize the destination networks can move with ease,
surplus generated by the insurance arrange- most potential migrants will lack the social
ment (placing equal weight on all income support they need to move. A second strat-
classes). The substantial redistribution that egy to reduce the information and enforce-
is uncovered appears to be driven entirely by ment problems that restrict mobility is to
an attempt to equate marginal utilities across migrate temporarily. The principal limitation
income classes. C ounterfactual simulations of the temporary migration strategy is that
that quantify the effect of formal insurance it will not fill the large number of jobs in
on migration, leaving the rural insurance which there is fi rm-specific or task-specific
learning and where firms will set perma- century (Mehta 2001). However, mercantile
nent wage contracts. Nevertheless, seasonal opportunities were relatively limited until
migration has been increasing rapidly over the arrival of the British in the eighteenth
time in India (Morten 2019). The anecdotal century. Commercial activity under the
evidence, for example Dhillon, Iversen, and British was concentrated around the ports
Torsvik (2013), suggests that caste networks of Bombay and Calcutta.9 Parsis, Gujarati
are once again playing an important role in Banias, Bhatias, and Bohris dominated
this migration, and so this would appear to long-distance trade out of Bombay from

be an important and open area for future the early nineteenth century (Gupta 2014),
research. whereas the Marwaris were only able to
establish a foothold in the Calcutta trade
4. Business (against stiff competition from the British)
toward the end of that century (Timberg
4.1 Historical Background
1978). Over time, these communities moved
Business success in a developing economy from trade to manufacturing, as early as the
requires connections, know-how, and access middle of the nineteenth century in Bombay
to capital. This effectively shuts out most (Nafziger 1978) and from 1914 onward in
potential entrepreneurs unless they have Calcutta (Lamb 1955).
the good fortune to be born into a business Apart from providing new opportunities
family. The obvious question to ask, then, in trade and manufacturing, British colonial
is how first-generation entrepreneurs enter rule also encouraged the settlement of castes
business in the first place, particularly in a in overseas destinations. The Nakarattar
country like India. In Weber’s (1958) view, Chettiars, for example, served as money-
the stability of the caste hierarchy served as lenders for the British throughout Southeast
a natural barrier to occupational mobility Asia, directly providing investment capital to
and entrepreneurial activity. This view has farmers, plantation owners, and mine opera-
been contested by historians such as Rudner tors from 1870–1930 (Rudner 1994). Starting
(1994) and Chandavarkar (1994) who take from the late nineteenth century, the pati-
the position that mobility is possible even in dars (Patels) of Charotar in Gujarat moved
a caste-based society, but that this mobility to British colonies in East Africa where they
involves the movement of the entire group. set up business and, in time, dominated
We will see below that caste or jati networks commercial activity (Mehta 2001). While
have historically supported, and continue to the Chettiars were forced to return to India
support, the movement of fi rst-generation in 1930, the Patels remained in East Africa
entrepreneurs into business. Our objective till the 1970s, when they too were forcibly
will be to examine the conditions under evicted (many of them subsequently moved
which such transitions take place. to the United Kingdom where they now run
One varna, the Vaisyas, traditionally con- successful businesses).
trolled moneylending and trade in India, Although some of the castes listed above,
with castes drawn from this varna active in such as the Marwaris and Chettiars, were
different regions of the country. N on-Hindu historically engaged in business-related
communities such as the Jains, Bohris, and activities, colonial rule provided the first
Khojas were also involved in trade from
ancient times. Some of these castes (commu- 9 These cities were recently renamed as Mumbai and
nities) had migrated overseas, to East Africa Kolkata, respectively. However, we will retain the historical
and the Persian Gulf, by the seventeenth names throughout to maintain consistency.
opportunity for other communities to move statistics tell us is that the pool from which
into business. These opportunities for these successful firms are drawn is not partic-
group-level mobility have continued after
ularly socially diverse, that is, that businesses
independence in 1947, particularly after the in India are drawn from a relatively small
liberalization of the Indian economy in the number of castes and communities. This is
1990s, with the emergence of a new business because once a community-based business
class drawn from a variety of agricultural network has formed, it will strengthen rap-
castes and from castes that were histori- idly over time, bringing in hundreds of new
cally concentrated in the bureaucracy and entrepreneurs. Those entrepreneurs will
white-collar professions (Damodaran 2008). support their children, which implies that
The striking feature of Indian business his- the community network is less essential from
tory is that the narrative is entirely about the next generation. The critical challenge
castes or their n
on-Hindu equivalent kinship appears to be jump-starting the community
communities; as Gadgil (1959, p. 16) puts it, network. The model that follows, which is a
“the history of the rise and growth of a mod- simplified version of Munshi (2011), charac-
ern business class in India is largely the his- terizes conditions under which business net-
tory of the activities of members of certain works will form and grow over time.
groups.”
4.2 Caste-Based Business Networks
Although occupational mobility did occur
on occasion, the absolute number of castes Consider an economy with a continuum of
and non-Hindu communities that moved workers of unit mass entering the labor force
into business over time is small. Timberg in each period. There are two occupations in
(1978) reports that twenty-three of the this economy: a traditional occupation and
thirty-seven largest North Indian-owned business. The payoff from the traditional
industrial houses listed in the Monopolies occupation is Vfor all individuals. The pay-
Inquiry Commission Report of 1964 were off from business for individual iwho enters
Marwari or Gujarati Bania. More recently, the labor force in period tis determined by
a Times of India article (October 20, 2006) his ability, ωi, and the size of his business
estimates that Gujarati-promoted companies network, which we measure by the stock
account for 17 percent of the market capi- of community members who are already in
talization of the Bombay Stock Exchange business; ∑ τ =0   Δ ωτ, where Δ ωτ is the flow
 t−1
(BSE) BSE-500 index, followed by Marwari- of individuals into business in period τ. The
promoted companies with 11 percent and basic idea here is that the larger the existing
Parsi-promoted companies with 8 percent. network is, the greater is the support avail-
Public sector units, including banks and oil able to new entrants. Individual iwill then
companies, account for 25 percent and all select the business occupation if the follow-
other companies, including multinationals, ing condition is satisfied:
just 39 percent of the market capitalization.10
t−1
(6) ωi  + β   ∑    Δ ωτ ≥ V.
Firms that are listed on the stock exchange,
particularly those that are successful enough τ =0
to be listed on the B SE-500, are operating
outside the boundaries of their community As in Munshi and Rosenzweig’s (2016) model
network, almost by definition. What these of insurance and migration, the choice is
between the location or sector where the
10 The market capitalization of the B
SE 500 index is network is active and going it alone. In their
92 percent of the total BSE market capitalization. model, the insurance network at the rural
origin discourages individuals from moving if the initial push, Δ ω0, is sufficiently large.
independently to the city. In the current This may explain why a relatively small
model, the business network at the destina- number of c aste communities have success-
tion encourages mobility; in each period t, fully managed the transition into business
there is a threshold ability, ω   , above which historically; it is not that they lack business
individuals select into the business ¯t occupa- potential, but that they were not fortunate
tion, with the threshold declining over time enough to receive a large enough common
as the network grows in size. Note, however, shock that would have initiated subsequent
that there is no inconsistency between the group mobility.
two models; the general point is that com- Up to this point, the model has implic-
munity networks support group mobility but itly assumed that there is a single commu-
discourage the (independent) movement of nity (network) in the economy. In practice,
individual members, and this will show up multiple communities will coexist distin-
again in the subsequent section. Where the guished by their traditional occupations,
two models do differ is in the specification which determine V , and their social connect-
of preferences—we assume from now on edness, which determines β, the mapping
that individuals are risk neutral—and that from network size to business payoffs. The
network effects occur with a lag (because second implication of equation (8) is that
individuals need to have some business expe- once it does take off, the network will grow
rience before they can support others). relatively fast in socially connected commu-
To initiate the network dynamics, Δ ω0 nities (with large β) and in communities with
individuals are moved exogenously into the worse outside options (with small α and,
business occupation in period 0. Assume that hence, small V ). This last result is especially
ability is distributed uniformly on the unit interesting because it tells us that policies
interval, ω
∼ U [0, 1]. The flow of individu- that attempt to jump-start business networks
als into business in any subsequent period is could be both growth enhancing and equal-
then Δ ωt ≡ 1 − ω  t. This allows us to derive izing if they target historically disadvantaged
the relationship between¯ Δ
ωt and the his- communities.
tory of flows from equation (6), Munshi (2011) tests the implications
of the dynamic model with data from the
t−1
Δ ωt = β   ∑   Δωτ − α,
Indian diamond industry. This industry
(7)
τ = 0 occupies an important position in the Indian
economy, accounting for roughly 14 percent
where α ≡ V − 1 . The assumption here of India’s total merchandise exports, and has
is that V > 1, which ensures that no indi- competed with textiles, and more recently
vidual would choose the business occupa- with computer software, as the country’s
tion without the exogenous push in period top export industry over the past decades.
0. Equation (7) can be solved recursively to For the purpose of our analysis, the most
derive a c losed-form solution for the flow of interesting feature of the diamond industry
individuals into business in each period, is its social heterogeneity. India does not
produce rough diamonds and so the dia-
Δ ωt = (βΔ ω0   − α) (1  + β)  t−1 .
(8) monds must be imported, for the most part
from the Antwerp market, and then cut and
The first implication of the preceding equa- polished in domestic factories before being
tion is that the business network will only sold in Bombay to foreign buyers or directly
grow over time if β Δ ω0  − α > 0, that is abroad. Two traditional business castes—the
Marwaris and the Palanpuri Jains—con- threat of social sanctions ensuring that the
trolled the business end of the industry from beneficiaries of the referrals do not renege
its inception in the m id-1960s, leaving the on their commitment.11 An alternative strat-
cutting and polishing to the Kanbi Patels, a egy for sourcing rough diamonds, typically
community of lower-caste agricultural labor- followed by established (second-generation)
ers known informally in the industry as the entrepreneurs, is to set up a branch in
Kathiawaris (Engelshoven 2002). The indus- Antwerp. This allows the firm to build a rep-
try structure changed in the late 1970s with utation in the Antwerp market and operate
the discovery of massive diamond deposits independently of its caste network, both as
(one-third of the world diamond supply by an exporter and as a rough diamond supplier.
volume) in Australia’s Argyle mines. This Retrospective information from a survey
supply shock allowed Kathiawari firms to of 800 Bombay-based diamond export firms
enter business, and today all three commu- conducted in 2004–05 indicates that there was
nities account for a substantial share of the a sharp increase in the number of Palanpuri
industry. Our interest is in characterizing the firms with branches in Antwerp just around
precise conditions under which a sufficiently the time that diamonds were discovered in
large number of Kathiawaris were able to the Argyle mines. The narrative in the indus-
enter initially and then to compare the sub- try is that these Palanpuris unloaded some
sequent trajectory of the three community of the increased supply of rough diamonds
networks. that came on the Antwerp market by encour-
Because of the difficulty in enforcing aging their Kathiawari labor contractors,
legal contracts, the diamond industry is with whom they had established long-term
associated with a high degree of community bilateral relations, to enter business. A suf-
networking throughout the world (Coleman ficiently large number of Kathiawaris must
1988, Richman 2006). Over 96 percent of have entered at that time, because hundreds
Indian diamond export firms are drawn from of Kathiawaris entered business in the sub-
just three caste communities, indicating sequent decades (presumably supported by
that networks must be active in the Indian an underlying network that was growing in
industry as well. In the Indian industry, the strength over time). This discussion high-
role of the network appears to be to facil- lights the difficulty in jump-starting com-
itate supplier credit. Most exporters visit munity networks; a large supply shock was
the Antwerp market for a few days, once a needed, together with support from out-
month or every couple of months, to source siders (who were no longer reliant on their
rough diamonds. They tend to specialize in community network). Such a confluence of
stones of a particular size, and while each favorable circumstances is unlikely to occur
exporter will have a small number of regu- very often, and if this is what is needed to
lar suppliers, he will also want to buy stones jump-start a network, then it would explain
from other suppliers from one trip to the why a relatively small number of castes are
next. Given the high value of the diamonds, active in business today.
most Indian exporters must rely on supplier Once the dynamics have been initi-
credit. What the network does is allow its ated, the model tells us that networks will
members to receive credit from suppliers strengthen most rapidly in socially connected
with whom they do not have long-term con-
nections; other members of the community 11 This cooperative arrangement has many features in
who do have established relations with those common with Greif’s (1993) description of the Maghribi
suppliers stand guarantor for them, with the traders’ coalition a thousand years earlier.
c ommunities and historically disadvantaged farming to business over the course of a

communities. Given the circumstances, we single generation.13 Similar transitions have
would expect the Kathiawari network to no doubt occurred before, but the diamond
have strengthened relatively rapidly, with industry is not representative, and so we
an accompanying decline in the ability of know relatively little about the precise cir-
entrants from that caste over time. We moti- cumstances under which other transitions
vated a role for community-based business occurred. Given the importance of entre-
networks with the argument that these net- preneurship in the development process and
works substitute for a family background in the potential role played by caste networks
business. This implies that the family back- in stimulating entrepreneurship in India,
ground of the Kathiawari entrants should be further investigation along these lines could
declining particularly steeply over time, and add much value. This could involve addi-
this is precisely what we observe with data tional industry-level or caste-level case stud-
from the survey (which includes the estab- ies, or the use of comprehensive census or
lishment year of the firm and the family back- survey data to examine the evolution of pri-
ground of the entrepreneur).12 Education, vate enterprise in its entirety (subject to the
which is an alternative (and standard) mea- constraint that caste identifiers are available
sure of ability, also declines relatively steeply at the firm level).
over time for the entering Kathiawaris, after
4.3 Caste Networks and Static Inefficiency
accounting for a ge-cohort effects.
Our interpretation of the weakening abil- Once a business network has formed, we
ity of the entering entrepreneurs over time, have seen that it can transform the circum-
especially the Kathiawari entrepreneurs, is stances of an entire caste in a single gener-
that they are being supported by a strength- ation. From the perspective of the insiders,
ening caste-specific network. Administrative these informal institutions are entirely pos-
data on firm-level exports over a ten-year itive; they generate surplus for the caste by
period (1995–2004) allow us to validate this efficiently allocating resources internally, and
interpretation. The Kathiawaris keep pace they are redistributive, with less able individ-
with their Marwari and Palanpuri rivals uals benefiting from the support provided by
despite the observed decline in their abil- more capable c aste members. However, we
ity. Indeed, once this compositional change must also consider the consequences of these
is accounted for with firm fixed effects, the networks for outsiders (belonging to a differ-
Kathiawari export trajectory is significantly ent network or without a network of their
steeper than the corresponding trajectories own). One obvious inefficiency that arises in
for the other two communities (or its own an economy where business is community
trajectory without fixed effects). There is based is that outsiders are crowded out;
evidently an underlying community-level capable individuals do not receive the appro-
force that is giving a boost to the Kathiawari priate level of resources or are excluded from
firms, resulting in a dramatic shift from business activity entirely. The examples that
12 While Marwari and Palanpuri entrants are slighly less

likely to report, over time, that their father was a business-
man, this decline is particularly steep for the Kathiawaris; 13 Closer inspection of the entrepreneurs’ family back-
70 percent of the entrants in 1970 (before the supply shock grounds indicates that most of the i ntergenerational mobil-
and the establishment of the network) have a family back- ity was driven by a shift out of agriculture into business;
ground in business, but this statistic declines rapidly and 52 percent of the Kathiawaris report that their father was a
drops below 20 percent for the entrants in 2000. farmer (and this figure is much higher for recent entrants).
follow document such inefficiencies in urban analysis is motivated by two stylized facts:
and rural India, respectively. First, the Gounders hold more capital stock
Banerjee and Munshi (2004) examine than the outsiders at all levels of experience.
resource allocation within and across castes Adjusting for differences in production, the
in Tirupur’s garment export cluster in a Gounders use roughly twice as much capi-
paper that predates the misallocation litera- tal per unit of production than the outsiders.
ture in macroeconomics. Tirupur, a town in Second, exports grow faster for the outsiders
the South Indian state of Tamil Nadu, sup- than for the Gounders at all levels of expe-
plies 70 percent of India’s knitted garment rience. This is not mechanical convergence;
exports. The textile industry in Tirupur was the outsiders do start with lower export lev-
initially controlled by a local trading commu- els, but they outstrip the Gounders after five
nity. However, after a prolonged period of years of experience.
labor unrest in the 1960s, it was taken over These two stylized facts, taken together,
by the Vellala Gounders, a wealthy farming can be shown to imply that there must be
caste (Swaminathan and Jeyaranjan 1994). a misallocation of resources across commu-
Details of the circumstances under which the nities in this industry. To see this, let the
Gounders entered the industry are unavail- firm’s exports, Y , be determined by entre-
able, but as with the diamond industry, there preneurial ability, ω, and capital stock, K. We
appears to have been a fortuitous element to make the assumption that the two inputs are
the occupational transition. complements,
Tirupur continued to be dominated by the
Gounders for the next twenty years, with the Y = ω K  α ,
(9)
industry producing exclusively for the domes-
tic market. Starting from the mid-1980s, where α < 1ensures that the production
however, the export of knitted garments from function has the standard neoclassical prop-
Tirupur started to grow extremely rapidly erties. Let rbe the interest rate faced by the
(presumably as a consequence of another firm. The firm will choose K to maximize its
fortuitous shock), and by the early 1990s, the profit, which yields the following expressions,
annual growth rate of exports exceeded 50
percent. This generated an influx of entre-
K =  _
(10) ω  2   .
ω  2   , Y =  _
preneurs from other regions of the coun- 4r  2 2r
try. In 1996, when Banerjee and Munshi
conducted a survey of firms in the industry, If all firms faced the same interest rate,
about half the entrepreneurs were Gounders as they would if markets functioned effi-
and the rest belonged to traditional business ciently, then entrepreneurs with higher
castes and communities, including Gujarati ability would hold more capital stock and
Banias, Parsis, and Marwaris. have higher exports. This is evidently not
The survey collected retrospective infor- the case in Tirupur. Indeed, it can be shown
mation on investment and production over that the two stylized facts necessarily imply
a four-year period for each firm. Banerjee that the Gounders face lower interest rates,
and Munshi’s objective is not to model the < rO
rG , and have lower ability, ωG < ωO
.
entry of firms over time, as we did with the To see this, substitute the expressions for K
diamond industry, but instead to identify and Yfrom equation (10),
the mismatch between entrepreneurial tal-
ω  2G   r  2G  
ent (ability) and capital that arises because (11) KG > KO
⇒  _     >   _   ,
credit fails to cross community lines. Their ω  2O   r  2O  
ω  2G   rG

(12) YG < YO
⇒  _     <  _     . villages. Upper castes are almost entirely
ω  O  
2 rO
absent from the BAC-dominated villages. A
comparison of the two types of villages yields
Note that rG/rO > (rG/rO
)  2 from inequal- the key empirical finding, which is that lower
ities (11) and (12), which implies, in turn, castes—BACs, OBCs, and SCs—have sub-
/rO
that rG < 1. It follows, from inequality stantially higher incomes and crop yields in
(12), that ω G/ωO
< 1. BAC-dominated villages. Anderson’s expla-
The Gounders are a wealthy landowning nation for this finding is that low-caste water
caste, and garment manufacturing was their buyers in the B AC-dominated villages have
first activity outside agriculture. Given that better access to this input because they can
they have few alternative uses for their cap- trade more easily with water sellers of their
ital, it makes sense that (internal) capital own caste.
is cheaper for them than for the outsiders. One limitation of Anderson’s analysis,
Although we do not model selection into the which she acknowledges, is that all the BAC,
industry, it also makes sense that they have OBC, and SC jatis are lumped together in a
lower ability on average, simply because the monolithic group.15 Identifying the misallo-
pool of potential entrepreneurs is weaker. cation is also more challenging for Anderson
The inefficiency that Banerjee and Munshi because she does not have “conflicting” styl-
uncover is that cheap capital fails to cross ized facts, as do Banerjee and Munshi, that
community boundaries, from the Gounders can be used to uncover an underlying market
to the more productive outsiders, presum- inefficiency. Anderson notes that the caste
ably because commitment problems pre- composition of the villages in her sample has
clude such trade.14 likely been stable for centuries. What has
The inability to trade across castes is not changed is the pattern of land ownership in
restricted to industrial activity. Anderson those villages where upper-caste Brahmins
(2011) provides evidence that caste is an were absentee landlords in the colonial
impediment to trade in groundwater, a crit- period. Land reforms in the 1950s, shortly
ical agricultural input, in North Indian vil- after independence, transferred land own-
lages, despite the fact that castes have been ership to lower-caste (BAC) cultivators, and
coexisting in these villages for centuries.
it is those villages that are B AC-dominated
Anderson’s analysis exploits the fact that today. Anderson provides evidence that
there are two types of villages in her sample; the two types of villages are the same with
an upper caste is dominant, that is it owns the respect to cultivated area, population, over-
majority of the land, in 48 percent of Hindu all access to irrigation, and the functioning
villages, whereas a backward agricultural of the water market. But if this is true, then
caste (BAC) is dominant in the remaining there is a comparability problem because
the characteristics of the lower castes and
14 One alternative explanation for the two stylized facts
the quality of their land will necessarily vary
is that ability and capital are substitutes, in which case across the two types of villages.
higher-ability outsiders could grow faster even while hold-
ing less capital than the Gounders (with interest rates the
same in both communities). But if this is true, then exports
and capital stock should be negatively correlated within
each community. However, within each community, where 15 Anderson notes that in her North Indian setting, the
we expect entrepreneurs to face the same interest rates, upper castes consist primarily of Brahmins and Rajputs,
firms holding more capital have higher exports (and export while the BAC group is comprised mainly of Yadavs. The
growth) consistent with the assumption that ability and results of her analysis, based on aggregate caste groups, are
capital are complements. thus likely driven by underlying caste-specific interactions.
The lower castes were historically inde- across the two types of villages, it must be that
pendent cultivators (with absentee land- upper-caste water buyers have greater access
lords) in the BAC-dominated villages, and to irrigation in upper-caste-dominated vil-
we would expect this autonomy to have had lages (where water sellers are p redominantly
long-term effects; indeed, they have signifi- upper caste). This symmetry is reassuring,
cantly higher education than lower castes in with the general finding that trade appears to
the upper-caste-dominated villages. A sec- be smoother when buyers and sellers belong
ond source of variation is that lower castes to the same caste.
in the B AC-dominated villages will own land Taking this result as given, a natural ques-
across the quality spectrum, whereas we tion to ask is why potential buyers and sellers
would naturally expect the upper castes to belonging to different castes cannot enter
have captured the best land in the villages into long-term bilateral relationships. Social
where they dominate. The available qual- ties, and the social sanctions that go with
ity measure, based on land value, does not them, are useful in supporting cooperation
reveal systematic differences for the lower in a one-shot game. However, these farmers
castes in the two types of villages. However, have been living together in the same village
there is reason to treat this finding with cau- for centuries and, given the low mobility in
tion. If yields (and incomes) are higher for rural India, would expect to live together long
lower castes in B AC-dominated villages, into the future. With a repeated game, caste
then this should necessarily be capitalized in ties would seem to be less relevant, with the
the value of their land. The absence of such potential termination of the long-term bilat-
variation in land prices indicates that either eral relationship being sufficient to deter
the land market is functioning imperfectly, deviation from the cooperative equilibrium.
in which case the prices are not informative, One possible explanation for the failure of
or that there is measurement error in the buyers and sellers from different castes to
reported prices, which prevents a meaning- come together is that they start with negative
ful comparison. priors about the long-term stability of the
The preceding discussion indicates relationship. Just as a sufficiently large com-
that a comparison of irrigation, yields, or mon shock is needed to jump-start commu-
income for the lower castes across the two nity networks, a positive shock is needed to
types of villages will not identify a misallo- create a long-term intercaste bilateral rela-
cation unless the comparability issues are tionship. Anderson’s results would suggest
addressed. However, the direct informa- that such shocks occur fairly infrequently.
tion on the c aste-based restrictions to trade
that Anderson provides is more compel-
5. Labor
ling. The strongest empirical finding is that
lower-caste water buyers have greater access
5.1 Historical Background
to irrigation, with an accompanying increase
in their yields and income, if they reside in a Castes were originally occupational units,
BAC-dominated village where water sellers and individuals remained within their caste’s
are predominantly of the same caste group. traditional vocation from one generation to
The issue of comparability remains, but it is the next prior to the arrival of the British.
reduced because we are now comparing cul- With colonial rule came overseas trade and,
tivators on the same side of the water market subsequently, industrial manufacturing.
across villages. Given that access to irrigation The accompanying growth of cities resulted
for water buyers as a whole does not vary in the demand for labor in n ontraditional
occupations, in the mills, docks, and facto- analysis of labor recruitment and the study
ries. A second new source of labor demand of the behavior of rural recruits in an indus-
came from mines and plantations, both in trial environment … it is entirely inappro-
India and overseas. Historical accounts indi- priate to lump into larger groups because
cate that networks organized around the jati of similarity of name, function, social status,
played an important role in supplying labor or region-of-origin subcastes that are not
to cities and plantations. endogamous.”
Commercial activity under the British was The same pattern of caste-based recruit-
centered around the cities of Bombay and ment appears to have emerged in Calcutta
Calcutta. Bombay’s textile-mill and factory and in the mines and plantations. While
owners appear to have adjusted their lev- business in Bombay was controlled by out-
els of output and employment to demand siders, labor was drawn mostly from the
shocks, periodically bearing the cost of idle Maharashtrian hinterland. In contrast, labor
capacity (Chandavarkar 1994). Similar fluc- in Calcutta’s jute mills and factories was
tuations in the demand for labor have been drawn from far away, from the western dis-
documented on the docks, in the shipping tricts of Bihar, the adjoining eastern districts
industry, and in general laboring jobs. These of Uttar Pradesh, and two Orissa districts,
are precisely the conditions under which Cuttak and Balasore (Das Gupta 1981). Once
labor networks will emerge, with incumbent again, a foreman or sardar was in charge of
workers providing referrals for new hires labor recruitment, and once again he appears
from their social group. Social connections to have drawn on his caste back home to fill
within the community provide the incum- new jobs. The same source districts in Bihar
bent workers with information about the and Uttar Pradesh supplied the bulk of the
quality of new hires. The same connections labor to overseas sugar colonies—Mauritius,
ensure that new workers will not shirk, even the West Indies, and Fiji—during the latter
if they are hired temporarily, to maintain the part of the nineteenth century, and while
community’s reputation in the labor market the recruitment of indentured labor to these
(and the incumbent worker’s reputation in colonies was heavily controlled by the colo-
his firm). nial government, the concentration of par-
In Bombay, the labor networks appear to ticular castes among the migrants indicates
have been organized around the mukadam that there was, nevertheless, an endogenous
or jobber, a foreman who was in charge of a aspect to the labor supply.
work gang in the mill, factory, dockyard, or Within India, caste-based recruitment
construction site, and, more importantly, also emerged wherever labor was needed in
in charge of labor recruitment.16 In an envi- large numbers. For example, caste-based
ronment with frequent fluctuations in labor recruitment was also observed in south-
demand, we would expect the jobber to ern tea plantations. A particular feature
have recruited from his social group, which of the caste system in South India was the
in the Indian context would be the jati, and institution of agrestic slavery in which the
this is precisely what appears to have hap- members of certain castes (and their prog-
pened. As Morris (1965, p. 76) argues, “I am eny) were owned by upper-caste masters
using the term [caste] to refer to what are (Alexander 1989, Kooiman 1989). The abo-
precisely called subcastes ( jati) … for any lition of slavery in India by the colonial gov-
ernment in 1861 coincided with the opening
16 The jobber was known as the maistry in the construc- up of the plantations in South India and Sri
tion industry and the sarang in the shipping industry. Lanka. These plantations were located in
reviously uninhabited hills and so labor had

p twentieth century, and this group was pre-
to be transported from the plains in Tamil sumably large enough to jump-start a new
Nadu. Initially, the British planters sent their network that persisted and grew in size over
own agents to recruit workers, but within a the next century.
short time, workers began to migrate inde- The model of network formation and
pendently in gangs drawn from the foreman the preceding case studies emphasize the
or kangany’s jati. Our dynamic model tells importance of initial conditions. When the
us that once a network has formed, it will model is extended to multiple destination
strengthen most rapidly in communities with locations or occupations, it generates the
the weakest outside options, which in this additional prediction that the community
context would have been the former slave could get locked into those destinations
castes. Luke and Munshi (2011) surveyed where it (sometimes fortuitously) first estab-
a sample of South Indian plantation work- lished itself. Caste clusters have been doc-
ers three generations later and found that umented among Bombay’s dock workers
68 percent of the workers belonged to the (Cholia 1941), construction workers, and in
former slave castes (Pallars and Paraiyars), the railway workshops (Burnett-Hurst 1925),
highlighting the role played by caste net- the leather and dyeing industries, and in the
works in supporting mobility, not only into Bombay Municipal Corporation and the
business but also into n onagricultural labor Bombay Electric Supply and Transportation
occupations. Company (Chandavarkar 1994). Castes
Luke and Munshi show how caste net- sometimes found niches even within occupa-
works in the plantations were refreshed and tions; Gokhale’s (1957) survey of textile work-
reinforced by new arrivals from the plains, ers in the 1950s found that particular castes
through marriage, from one generation to were concentrated in particular departments
the next. Networks that formed in the city within the mills, with different castes domi-
were similarly reinforced over the course of nating a given department in different mills.
many generations by the arrival of new work- De Haan (1997) documents similar caste
ers from the rural origin. Focusing on a single segmentation in Calcutta’s jute mills. Urban
caste, Dandekar (1986) traces the evolution jobs did not necessarily relate to traditional
of a network of Jadhavs (a particular caste) rural occupations; indeed, castes that were
from a single village in interior Maharashtra traditionally weavers ended up in the spin-
to one jobber who went to Bombay in the ning section of the jute mills by historical
1930s, working first on the docks and later in accident, emphasizing the importance of
the textile mills. In 1942, 67 percent of the initial conditions in determining long-term
Jadhav migrants from that village were work- outcomes when networks are active.
ing in the textile mills and 4 percent in other
5.2 Discrimination
factories. In 1977, their descendants and
new arrivals from the village over the next One side of caste networks is that they
generation were still in the same occupa- harness social ties within the community to
tions, with 58 percent in the textile mills and solve market imperfections and increase eco-
10 percent in other manufacturing indus- nomic efficiency. The other side is that they
tries. De Haan (1997) similarly traces the exclude outsiders once they are established.
evolution of a network of Khandait migrants This discrimination is not a consequence of
in Calcutta to a single sardar. This individual social prejudice, but is simply the outcome of
brought an initial group of twenty workers economic competition between groups; the
with him from Orissa at the b eginning of the foreman would always prefer to recruit from
his own caste. In recent years, a n umber of be explained by differences in endowments

studies have examined another aspect of (mainly education and occupation).
caste discrimination in which upper-caste Madheswaran and Attewell (2007) apply
employers prefer not to hire lower-caste the Oaxaca–Blinder decomposition to
workers. This type of discrimination is not test for caste discrimination with National
targeted at specific castes, especially in Sample Survey (NSS) data. The advantage of
urban labor markets but is based on more their data is that it is nationally representa-
broadly defined social status. While the jati tive and available at multiple points in time.
may be the appropriate social unit when ana- SC and ST workers earn 5 percent less than
lyzing caste networks, the hierarchical aspect other workers in 1983, 8.4 percent less in
of the caste system, defined by the varnas, 1993–94, and 8.9 percent less in 1999–2000.
is relevant when analyzing discrimination. Corroborating Banerjee and Knight’s find-
One approach that has been used to identify ing, the endowment component explains the
caste discrimination is the Oaxaca–Blinder bulk of the earnings differential (70–87 per-
decomposition. In this method, the earnings cent) over the three time periods. If occupa-
differential between upper castes and lower tional variables are included in the earnings
castes, which is attributed to discrimination, function, the endowment effect increases
is divided into two components: differences even further. Kijima (2006) applies a vari-
in the estimated returns to endowments such ant of the Oaxaca–Blinder decomposition to
as education and occupation for upper- and explain the consumption gap between SC/ST
lower- caste groups and differences in the and non-SC/ST households over multiple
endowments themselves. NSS rounds (1983–99). In contrast with the
Banerjee and Knight (1985) apply the studies listed above, he finds that differential
Oaxaca–Blinder decomposition to a sam-
returns to endowments (education and occu-
ple of migrants that they surveyed in Delhi. pation) explain half of the consumption gap,
Earnings are 17 percent lower for the SC with the decomposition remaining stable
migrants relative to the non-SC migrants. over the sample period.
The two groups of migrants also differ in It is standard practice in this literature to
their endowments—education, the propor- interpret differences in the returns to endow-
tion who are salaried, and the proportion ments between groups as wage discrimi-
in unskilled manual jobs. The estimated nation and differences in the endowments
earnings function coefficients reveal no dif- themselves as pre-market discrimination.
ference between the two caste groups for This interpretation may, however, overstate
education, age on arrival in Delhi, occu- the extent of current discrimination. The
pation, and the intercept. However, there lower castes, particularly the SCs, were
is a greater premium for salaried workers socially disadvantaged for centuries. They
versus daily wage earners among the SCs, a were originally assigned the lowest-status
greater return to formal sector employment occupations, requiring little investment in
versus informal sector employment for the human capital, and even when they managed
non-SCs, and a steeper earnings-experience to achieve occupational mobility under the
profile for the non-SCs. Overall, differences British, they typically ended up in low-skill
in the estimated returns to individual char- industrial jobs. We have noted that caste
acteristics favor the n
on-SCs but are small. networks remain active in the Indian labor
Depending on whether SC or n on-SC earn- market. Once the upper-caste networks had
ings function estimates are used, half to established a presence in particular sections
two-thirds of the earnings differential can of the labor market, they would have ensured
that outsiders were subsequently excluded characteristics, through which intergroup

from those jobs. Persistent differences in earnings differentials are generated.18
occupations between SCs and n on-SCs Following Bertrand and Mullainathan’s
could thus be determined by historical (2004) research design that was used to doc-
circumstances rather than by discriminatory ument racial discrimination in the United
practices by upper-caste employers today. States, a field experiment in which caste
Given historically low levels of human cap- affiliation is randomly assigned to otherwise
ital, and the importance of parental human identical job applications is the cleanest way
capital in determining preschool human to identify caste discrimination in the Indian
capital, the education deficit for the SCs labor market. Thorat and Attewell (2007)
could be similarly determined by historical conducted such an experiment by replying
circumstances that are slow to change rather to job advertisements in major English daily
than by current discrimination in education. newspapers, sending three applications to
Differences in the returns to endowments, each call: an applicant with an upper-caste
particularly education, are also difficult to name, an applicant with an SC name, and an
interpret as discrimination.17 For example, applicant with a Muslim name. The appli-
well-qualified non-SC parents could transmit cants had identical educational qualifications
noncognitive skills and other traits that are and experience. No explicit mention of caste
complementary to education to their chil- or religious background was made in any
dren. If these parental inputs are observed application; the only distinguishing feature
by the market but not the researcher, then was the name of the applicant. In addition,
the estimated returns to education would be a “discordant” application was submitted to
lower for the SCs. Caste-based reservation each advertisement; for jobs requiring an
in institutions of higher education would also undergraduate degree, an applicant with an
affect the estimated returns to education. If SC name and a master’s degree was submit-
affirmative action serves its intended pur- ted, and for jobs requiring a postgraduate
pose, then SC students enrolled in a given degree, an applicant with an upper-caste
course in a given college will be less pre- name and an undergraduate degree was
pared than their non-SC classmates when submitted.19
they join. If they fail to catch up over the The main finding of the study is that
course of the program, then they will grad- equally qualified applicants with SC or
uate with the same credentials but will be Muslim names were significantly less likely
less competent. If this difference in com- to be called for an interview or for testing.
petence is accounted for by employers, but In the first discordant experimental treat-
unobserved by the researcher, then the esti- ment, the probability of success for the
mated returns to education will, once again,
be lower for the SCs. In general, it is diffi-
18 Deshpande and Newman (2007) compared the
cult to disentangle discrimination from other
job-market expectations of SC and non-SC students in
channels, based on unobserved individual their final year of university in Delhi. The SC students
expected a lower wage than the n on-SC students, but this
could be due to differences in endowments or perceived
discrimination. The SC students performed worse than
the non-SC students on the final university exam and had
weaker family backgrounds (measured by parental occupa-
17 Banerjee and Knight detect no differences in the tion and education).
returns to education for SCs and n on-SCs, whereas 19 Initially, one set of four applications was submitted
Madheswaran and Attewell’s estimates indicate that the for each advertisement. However, this was later increased
returns are consistently lower for the SCs. to three sets to increase statistical power.
o verqualified SC applicant was greater than or OBCs for software jobs. However, they do
(less than) the corresponding probability for find large and significant differences in call-
an appropriately qualified SC (upper-caste) back rates between upper castes and OBCs
applicant. However, these differences are not (and to a lesser extent SCs) for call center
significant at conventional levels. The second jobs. These results are consistent with statis-
discordant treatment had greater statisti- tical discrimination, since unobserved family
cal power, with under-qualified upper-caste characteristics that are correlated with caste
applicants being significantly less likely (as status are more likely to matter for the call
likely) to be successful as appropriately qual- center jobs. Statistical discrimination is not
ified upper-caste (SC) applicants. necessarily economically inefficient, but
These experimental results indicate that it does punish high-achieving members of
unobserved individual characteristics that underperforming groups, and it can lead

are correlated with caste identifiers may to self-fulfilling stereotypical behaviors in
not entirely explain the findings from the equilibrium.20
Oaxaca–Blinder decompositions. It appears
5.3 Caste-Based Labor Networks and
that private firms do select applicants explic-
Dynamic Inefficiency
itly by their caste status. The next question
to ask is whether this discrimination is sta- Although caste networks may reduce
tistical; that is, firms use caste identifiers imperfections in the labor market, they can
to proxy for applicants’ unobserved char- also generate inefficiencies of their own. We
acteristics that are relevant for their jobs, saw earlier that the failure of capital to cross
or taste-based, or in other words, based caste lines results in a mismatch between
on preferences for a particular group. One entrepreneurial talent and investment in
way to dig deeper and answer this question capital. Caste-based networks can similarly
is through qualitative research. Jodhka and misallocate labor inputs by excluding com-
Newman (2007) conducted in-depth inter- petent outsiders from the market. Even the
views with managers responsible for hiring gain in efficiency within the caste through
workers in twenty-five large private firms internal cooperation is not without cost.
in the New Delhi area. The central finding Munshi and Rosenzweig (2006) show that
was the view, expressed by virtually every the same caste networks that supported the
respondent, that workers should be selected movement of their members from the vil-
strictly on merit. However, merit was lage to the city historically discouraged the
defined broadly enough to include family occupational mobility of their members in
background, and while the respondents did subsequent generations. This phenomenon
not explicitly mention caste as a consider- is not unique to India. Heavily networked
ation, they did endorse regional stereotypes. blue-collar communities, originally estab-

In a situation where information on family lished by dynamic migrants, have been
background is missing, as in the field experi- observed to be especially resistant to change
ment above, they would very likely have used
caste and regional characteristics as a proxy
20 There is no direct evidence that I am aware of that
for family characteristics when making their
shows (conditional on observed credentials) that low-
hiring decision. Complementing this finding, er-caste workers are less productive. Deshpande and
Banerjee et al. (2009) conducted a resume Weisskopf (2011) use data from the Indian Railways to
audit study in Delhi that was restricted to show that an increase in the fraction of reserved workers
does not appear to lower productivity. These results must
software and call center jobs. They find no be treated with caution, however, given the endogeneity of
evidence of discrimination against SCs, STs, the inputs and the fraction of jobs that are reserved.
an early study set in Chicago, Rees (1966)

in other economies; for example, Gans found that informal sources accounted for
(1962) and Kornblum (1974). 80 percent of all hires in eight b lue-collar
To provide economic foundations for this occupations and 50 percent of all hires in
inertia, we return to the dynamic model of four white-collar
occupations. Matching
occupational choice that was described in these patterns, 68 percent of b lue-collar
the previous section. There was a continuum workers and 38  percent of w hite-collar
of ability levels in that model, which sorted workers reported having received help find-
individuals into business and the traditional ing a job in Gore’s (1970) study of migrants
non-business occupations. We now assume in Bombay. Thirty years later, Munshi and
that there are two levels of ability, high Rosenzweig (2006) document similar refer-
and low, and two occupations, skilled and ral patterns in an area close to Gore’s study,
unskilled, in the labor market. Education with 70 percent of blue-collar workers and
plays an important role in the empirical 35 percent of white-collar workers report-
application that follows by determining the ing that they received help from a member
type of job that the individual receives. We of their caste in finding their first job. The
incorporate this in the model by assum- model incorporates this feature of the labor
ing that high ( H ) education leads to the market by assuming that the skilled wage,
high-skill occupation, while individuals with WH
, is exogenously determined, whereas
low (L) education are channeled into the the unskilled wage, WL, is increasing in the
low-skill occupation. Ability sorts individuals (lagged) size of the community network in
into occupations once again, but it is now via that occupation. In particular, members of
the educational choices that they make. It the community from the previous generation
_
costs   C   for low-ability individuals to attain in the low-skill occupation provide referrals,
high education, whereas the corresponding and increase wages, for the next generation.
_ Let WL (0)be the unskilled wage when
cost for high-ability individuals is  _  C  <   C   
no one from the individual’s community
. We normalize so that the cost of attaining selected that occupation in the previous
low education is zero. period. Let WL (N)be the corresponding
Absent any other forces, high-ability indi- wage when all N members of the previous
viduals would select into the high-skill occu- generation selected the unskilled occupa-
pation if W H  −  C   > WL , and low-ability tion. If communities were concentrated
individuals would¯ select _ into the low-skill entirely in either the skilled or the unskilled
occupation if W H  −   C    < WL . Once we occupation in the initial period, then the fol-
introduce community networks, however, lowing conditions ensure that there will be
all individuals in a community could choose occupational persistence at the level of the
the same occupation, regardless of their community in all subsequent generations:
ability, with these community-level choices _
persisting over many generations. A nota- (C1) WH   −   C    > WL (0),
ble feature of the historical descriptions of
urban caste-based networks in India is that (C2)   −  _ 
WH C   < WL (N ).
they focus on working class jobs. This is a
feature of labor market networks world-
wide, possibly because the information and The first condition says that if the low-skill
commitment problems that give rise to the network was not active in the previous gen-
networks are most acute in those jobs. In eration, then individuals of both types would
invest in high education in the current gen- that identity, and values more generally, are
eration and end up in skilled jobs. The sec- purposefully instilled to further community
ond condition says that if everyone in the objectives is in line with previous work on this
community selected into the network in topic in economics; for example, Bisin and
the previous generation, then individuals of Verdier (2000) and Tabellini (2008). It also
both types would select low education and explains why heavily networked blue-collar
end up in the unskilled occupation. If con- communities tend to instill an especially
ditions (C1) and (C2) are satisfied, it fol- strong sense of identity among their mem-
lows that communities will stay in the initial bers in many economies.
equilibrium from one generation to the next, The community identity described above
with everyone either investing or not invest- aligns individual choices more closely with
ing in education. the social optimum. While this identity may
Economies will, however, restructure over thus be welfare enhancing when it is first
the course of the development process, and put in place, it can result in a dynamic inef-
one consequence of development is that the ficiency if it persists in subsequent gener-
returns to education, and the skilled wage, ations past the point where WH has grown
will start to grow. When the inequality in large enough that it is socially optimal for
condition (C2) is just reversed, high-ability the high-ability individuals to exit. Cultural
individuals from communities that tradition- norms and social values are persistent by
ally exclusively selected into the low-skill design, which explains why b lue-collar com-
occupation will invest in high education and munities often appear to stubbornly resist
earn ϵ more in the skilled occupation than change (despite the fact that the same com-
they would have if the entire community munities were extremely dynamic when they
remained in the traditional occupation. If the formed). Munshi and Rosenzweig build on
fraction of high-ability individuals in these this idea to document a dynamic inefficiency
communities is non-negligible, then their in Bombay’s labor market, which was his-
exit from the network will result in a substan- torically organized around caste-based net-
tial (first-order) decline in the wage received works. The starting point for Munshi and
by low-ability individuals. From a utilitarian Rosenzweig’s analysis is the economic and
perspective, welfare will decline, providing financial liberalization of the Indian econ-
an economic rationale for c ommunity-based omy in the 1990s. This increased white-collar
restrictions on mobility. wages (WHin the model) substantially. The
Individuals who select out of the tradi- question is whether the occupational per-
tional low-skill occupation will often move sistence associated with the blue-collar
to a new location. The conventional punish- networks prevented the next generation of
ment mechanisms that maintain cooperation workers from taking advantage of these new
within communities will then no longer be opportunities.
effective. Munshi and Rosenzweig argue that
an alternative strategy to maintain coopera-
tion in that case would be to instill a strong However, the level of these transfers would need to be
sense of community identity in childhood, agreed upon long before individuals entered the labor mar-
ket because investment in education is necessary for entry
which ensures that individuals remain tied into the skilled sector. Reaching such an agreement ex ante
to their community in adulthood.21 The idea is difficult in an economy where the (steeply increasing)
wage trajectory is uncertain. The community would also
need to credibly commit to making these transfers ex post,
21 Another possibility would be to provide compensa- thereby creating potentially substantial income inequality
tory transfers to those individuals to induce them to stay. among individuals doing the same job.
When the Indian economy restructured in schooling choices, and this is indeed what
the early 1990s, the next generation of work- they find (once parental and household con-
ers would have still been in school. Munshi trols are included in the regression). Pooling
and Rosenzweig’s strategy to test for occu- boys and girls and including caste fixed
pational persistence exploits the fact that effects, the second (complementary) result
schooling choice is often a good predictor of is that the gender gap in schooling choice is
future occupational outcomes. Schooling in increasing in male occupational choice in the
Bombay, the setting for their research, can previous generation. More importantly, this
be either in English or Marathi (the local lan- relationship remains strong and stable over
guage). Marathi schooling channels the child the course of the 1990s, even as the returns
into networked blue-collar jobs, whereas to English grew, providing m icrolevel
more expensive English schooling substan- evidence of the lock-in into specific tradi-
tially increases the probability that the child tional occupations.
will obtain a h igh-wage white-collar job in At the aggregate level, this lock-in is
the future. The economic restructuring thus reflected in the absence of convergence
effectively increased the returns to English in schooling choice between upper-caste
schooling. If networks discourage individual and lower-caste boys. Upper-caste boys
mobility, then parents in some castes could, who entered school in the 1980s are much
nevertheless, continue to make schooling more likely to be schooled in English than
choices for their children that channel them lower-caste boys, reflecting the fact that

into the traditional b lue-collar occupation. upper castes historically had access to admin-
Munshi and Rosenzweig test for this istrative and professional jobs, while lower
intergenerational occupational persistence
castes were concentrated in blue-collar
by examining schooling choices for twenty occupations. While there is a steep increase
cohorts of Maharashtrian children who in the fraction of boys enrolled in English
entered first grade between 1982–2001, in schools in the 1990s, there is no convergence
one area of Bombay city consisting of thirty between the castes. This can be explained
schools (twenty Marathi and ten English). by the stable intergenerational correlation
The key result is that the probability that a in occupational choice over the course of
boy will be schooled in English is decreas- the 1990s that we noted above. In contrast,
ing in the proportion of men in his caste while there is also a caste gap in schooling
employed in b lue-collar (working class) choice for the girls in the 1980s, this gap nar-
jobs. This result is robust to the inclusion of rows dramatically in the subsequent decade.
parental education and household income Although lower-caste networks may have
as regressors. One concern is that children historically supported the movement of their
from lower castes, who were historically members from the village to the city, with
concentrated in the b lue-collar occupations, accompanying occupational mobility, many
have lower p reschool human capital, which generations later they appear to hold their
leads their parents to favor Marathi school- descendants back.
ing. The parental and household variables As described in the previous section,
control for this to some extent, but Munshi caste-based business networks are continu-
and Rosenzweig also implement an alterna- ally forming in the dynamic Indian economy.
tive approach that relies on the fact that the We would expect labor market networks to
blue-collar networks were largely restricted be similarly forming as new opportunities
to men. Male occupational choice in the arise, with high-ability individuals leading
previous generation should not affect girls’ the way into new occupations. At the same
time, the evidence presented in this sec- e valuations of the affirmative action program
tion indicates that other caste networks will and, more generally, examinations of the
simultaneously be decaying, although social relative circumstances of the SCs and STs
frictions could substantially delay this pro- in the decades after independence. As with
cess. Munshi and Rosenzweig show that rel- labor market discrimination, it is appropri-
atively high-ability children (where ability ate to treat all castes among the SCs, and all
is proxied by the father’s years of schooling) tribes among the STs, as a single group if the
are the first to exit the traditional blue-collar objective is to evaluate the average effect of
occupation, providing a useful c ounterpoint affirmative action on these historically disad-
to the patterns of selection that are observed vantaged groups. In the most comprehensive
with caste network formation. A fruitful area study to date, Hnatkovska, Lahiri, and Paul
for future theoretical and empirical research (2012) use data from five successive rounds
would be to bring the twin processes of net- of the NSS, covering the 1983–2005 period,
work formation and network decay together. to measure differences in education, occupa-
tions, wages, and consumption between SC/
ST and n on-SC/ST households. The key find-
6. Education
ing is that there is significant convergence in
all of these outcomes over the period under
6.1 Aggregate Trends
consideration that extends well past the point
Given their historically disadvantaged in the early 1990s when the Indian economy
circumstances, the lower castes lagged far started to liberalize. Decomposition analy-
behind the upper castes with respect to edu- sis indicates that much of the convergence
cation and other socioeconomic indicators in wages and consumption is driven by con-
at independence in 1947. Article 46 of the vergence in education between the SC/STs
Indian Constitution put into place one of and n on-SC/STs. Moreover, the convergence
the largest and most aggressive affirmative is observed at all points in the respective
action programs in the world to rectify this distributions.
historical disadvantage. Seats in universities In a follow-up paper, Hnatkovska,
and central government jobs are reserved Lahiri, and Paul (2013) compare the inter-
for SCs and STs to match their share of generational mobility of SC/STs and
the population, 15 percent for the SCs and non-SC/STs using the same data. Their

7.5 percent for the STs. More recently, reser- analysis takes advantage of the fact that
vation has been extended to OBCs, bringing 60 percent of households in the NSS sample
the overall share of reserved seats close to include members from different generations.
50 percent. In an economy where university Complementing their previous findings,
seats are heavily oversubscribed and unem- intergenerational mobility of SC/STs was

ployment is high, the affirmative action pro- lower than that of n on-SC/STs in 1983 but
gram has naturally been highly controversial. rises faster over the subsequent decades and
Much of the public debate has centered on is roughly equal by 2005. For example, the
whether the SCs and STs have actually ben- probability that an SC/ST child’s education
efited materially from the reservations and, differs from his parent’s education increases
if so, whether privileged individuals within from 0.42 to 0.67 over the 1983–2005 period,
those groups have captured the bulk of the whereas the corresponding probabilities for
targeted resources. the non-SC/STs are 0.57 and 0.67. Similar
Despite the importance of these ques- convergence in intergenerational mobility is
tions, there have been few b road-based observed for income and occupation. Once
again, these gains in mobility are observed household wealth, measures of school quality,
at all points in the distribution. In particu- and teacher inputs, SC/ST children (and OBC
lar, the largest intergenerational changes in children) are significantly less likely to attend
education for SC/STs are at the bottom of school.
the distribution, out of illiteracy into primary One explanation for this finding is that the
and secondary school. school-quality and teacher-input variables do
While these results are heartening, there not fully capture the (lower) quality of edu-
are a couple of important caveats. First, the cation that lower-caste children receive. The
convergence cannot be entirely attributed government of India has made it a priority to
to affirmative action (which targets higher provide a school within walking distance of
education). Some of this c atch-up could be each rural household. Given the segregation
mechanical, since the SC/STs started behind by caste within villages, this policy has resulted
the non-SC/STs. Moreover, the government in lower-caste students being separated from
has taken steps to expand basic education upper-caste students, denying them the pos-
as described below, and caste networks, itive peer effects they would otherwise have
which are active in many economic domains, enjoyed (Kochar 2008). The additional effect
could also have played a role in this conver- of this policy is that the average lower-caste
gence. Recall that networks will strengthen student resides in a neighborhood of smaller
more rapidly, once they form, in historically size and will, therefore, be enrolled in a
disadvantaged castes with weaker outside smaller school (with worse infrastructure)
options. Second, while there may have been than the average upper-caste student (Rawal
convergence, substantial gaps in the level and Kingdon 2010).
of education between SC/STs and n on-SC/ Even if upper-caste and lower-caste stu-
STs remain. Using representative NSS data, dents were mixed in school, the available
Hnatkovska, Lahiri, and Paul show that this evidence indicates they would not be treated
gap is as large as two years of schooling. The the same. Rawal and Kingdon use data on
discussion below provides micro-foundations primary school children in rural Bihar and
for the observed convergence and the per- Uttar Pradesh to show that student per-
sistence of the c aste gap, describing the chal- formance increases substantially when the
lenges faced by SC/ST students in school, teacher belongs to the same caste. Hanna
the benefits they receive through affirmative and Linden (2012) conducted an experiment
action in university, and the role played by in which teachers were recruited to grade
caste networks in determining the demand examinations. Candidate characteristics—
for education. caste, gender, and age—were randomly
assigned to each cover sheet to ensure that
6.2 Access to Schooling
there was no relationship between these
It is well known that school participation observed characteristics and exam quality.
is particularly low among the SC/STs. Drèze The main finding is that exams assigned
and Kingdon (2001) use data from the Public to lower castes are scored lower by exam-
Report on Basic Education (PROBE) survey, iners, although the effects are extremely
covering randomly selected villages in the small in magnitude. Interestingly, this bias is
northern states of Bihar, Madhya Pradesh, accounted for entirely by low-caste graders.
Rajasthan, Uttar Pradesh, and Himachal It appears to be generated by statistical dis-
Pradesh to compare school attendance and crimination—a prior belief in that group of
grade attainment for SC/ST and n on-SC/ST graders that lower-caste students are worse
children. Controlling for parental education, on average—because the bias disappears
over time as the link between demographic action has substantially increased college
characteristics and test performance is attendance for (less qualified) lower castes
revealed (this would not be the case with in recent decades (Darity, Deshpande, and
taste-based discrimination). Weisskopf 2011).
In addition to differential treatment by
teachers, lower-caste students must also 6.3 Affirmative Action
confront negative stereotypes when they
interact with higher-caste students in the As discussed in Frisancho Robles and
classroom. A large literature in social psy- Krishna (2012), there are three issues associ-
chology finds that individuals belonging to a ated with affirmative action in higher educa-
group associated with a particular s tereotype tion: (i) targeting—whether the b eneficiaries
perform worse on tasks associated with that are truly disadvantaged or whether the
stereotype when observed by others. In the resources are captured by wealthy elites
Indian context, this implies that lower castes, within the historically disadvantaged
who were historically perceived to be intel- groups22; (ii) catch up—whether there is con-
lectually inferior, would perform worse on vergence between reserved groups, who start
cognitive tasks if the results of their efforts behind by design, and unreserved groups
were made public. Hoff and Pandey (2006) over the course of the university degree; and
test this hypothesis by conducting an exper- (iii) mismatch—whether the beneficiaries of
iment in which low-caste and high-caste affirmative action would have actually done
sixth and seventh graders were assigned the better by not taking advantage of their legally
task of solving mazes in s ix-person groups. mandated opportunity, perhaps because the
When the participants’ caste status was kept college or the major they chose turned out to
anonymous, low-castes and high-castes per- be too challenging.
formed equally well on the task. However, Bertrand, Hanna, and Mullainathan
when caste identity was revealed, in homo- (2010) address the targeting and mismatch
geneous or mixed-caste groups, there was issues with data on all individuals who
a sharp decline in low-caste performance. took the engineering entrance exam in one
This result is obtained from a stylized Indian state in 1996.23 The minimum score
lab-in-the-field experiment that may be
for admission, out of a total of 900, is 480
subject to its own biases. For example, the for the upper castes, 419 for the OBCs, and
high castes perform equally well when their 182 for the SCs, confirming that affirmative
caste status is kept private or is revealed in
mixed-caste groups. However, their per-

22 A related concern is that individuals belonging to the
formance declines substantially when their
upper castes could misreport their caste in order to take
caste status is revealed in high-caste-only advantage of affirmative action. It is commonly believed
groups. This curious result is difficult to that such identity manipulation does occur, although there
explain, and it suggests that more evidence are no statistics on the prevalence of this practice (for obvi-
ous reasons).
is needed, with more comprehensive data, 23 As discussed below, Bagde, Epple, and Taylor (2016)
in support of the experimental findings. use similar data with a different identification strategy to
Indeed, the available evidence indicates assess whether reservation improves college performance
for beneficiaries. In contrast, Kochar (2010) and Frisancho
that conditional on attendance, grade Robles and Krishna (2012) focus on catch-up by exam-
attainment does not vary by caste (Kingdon ining student progress through college. A limitation of
1998 and Drèze and Kingdon 2001). Where both studies is that they are restricted to a single college
and lack information about the entire pool of candidates
caste differences in performance do appear (differential selection by caste group into college is thus a
is in higher education, where affirmative major concern).
action does indeed bring in weaker students students above and below the entrance
at the margin from historically disadvan- exam cutoff for their group. They estimate
taged castes. To gather more information on the location of the cutoff using a standard
the socioeconomic background of the can- structural break approach; that is, they select
didates and their post-college labor market as the cutoff the score that minimizes the
outcomes, the research team conducted a sum of squared residuals in a regression of
household survey between 2004–06 in which engineering college attendance on a dummy
the sample was based on candidates on equal to one if an individual’s test score is
either side of the cutoff test score for their above a particular score threshold, zero oth-
caste group. erwise. Lower-caste students just above and
Combining the survey data with NSS data, just below the cutoff can be assumed to have
which is representative of the state popu- effectively the same intrinsic characteristics,
lation, the first finding of the study, related and so a comparison of the two groups of stu-
to targeting, is that although lower-caste dents provides an estimate of the gain from
applicants are positively selected from their affirmative action for the marginal student
population s ubgroup, affirmative action does who benefits from that policy. In practice,
redistribute educational resources to less Bertrand, Hanna, and Mullainathan sample
wealthy households. Lower-caste applicants applicants with scores in a relatively wide
to the engineering colleges have household range around the threshold. While the two
incomes that are four times larger than the groups—above and below the threshold—
average income of their c aste-group in the do have the same observed characteristics on
state. Nevertheless, the lower-caste appli- average, differences in unobserved charac-
cants who are just above the test score teristics are now a concern. Their point esti-
threshold that secures them a position in mates are also now interpreted as the effect
engineering college have household incomes of affirmative action on a group (of students)
that are just over half the income of the that is less clearly defined.
upper-caste candidates whom they displace A more recent examination of affirmative
(with test scores just below the threshold for action in higher education by Bagde, Epple,
their caste group that would have secured and Taylor (2016) uses similar data but a
them admission). different identification strategy to estimate
The second finding of the study, related to the effect of reservation in engineering col-
mismatch, is that despite the fact that low- leges in one Indian state on the performance
er-caste entrants have much weaker skills of lower-caste (SC, ST, and OBC) students
than upper-caste entrants (by virtue of their one year into the program. Their approach
substantially lower scores on the entrance takes advantage of the fact that all entrants
exam), they still gain by attending engineer- to the 215 colleges in the state must take a
ing college. However, this gain is more than common entrance exam and, once admitted,
offset by the loss in income for the high-caste sit for another common exam at the end of
candidates they displace. Thus, even if there the first year. Bagde, Epple, and Taylor also
is catch-up over the course of the engineer- obtain high school test scores for most (89
ing program, the two groups do not converge percent) of the candidates that provide addi-
with respect to their earnings potential by tional information on intrinsic ability.
the time the program is completed. Instead of focusing on the marginal
Bertrand, Hanna, and Mullainathan (2010) entrant, Bagde, Epple, and Taylor estimate
estimate the gain from affirmative action the effect of affirmative action on admitted
by comparing the income of lower-caste students over the entire ability distribution.
They do this by constructing two ranks: the One natural mechanism is that affirmative
absolute rank of each individual, based on action improves the quality of colleges to
the performance on the entrance exam, and which reserved students have access, and
the effective rank, which adjusts for the caste this improves their performance. College
reservation. Conditional on the absolute rank quality, measured by the mean entrance
or, equivalently, the score on the entrance exam rank of men in the unreserved (open)
exam, which measures intrinsic individual category, is increasing in effective rank.
ability, the effective rank determines the Instrumenting for college quality with effec-
effect of reservation on both entry into col- tive rank, first-year performance is found to
lege and fi
rst-year performance (conditional be increasing in college quality.25 Contrary
on entry). High school performance pro- to the mismatch hypothesis, access to better
vides an additional measure of ability and, colleges by less prepared lower-caste stu-
in practice, latent ability is specified to be dents significantly improves their outcomes.
a polynomial function of the score on the Bagde, Epple, and Taylor conclude their
college entrance exam and the high school analysis by quantifying these improvements.
exam. For this exercise they use the estimated coef-
The results of the analysis are as follows. ficients from the analysis described above,
First, the probability of college attendance replacing effective rank with absolute rank.
is (not surprisingly) increasing in effective The counterfactual analysis indicates that
rank. Second, conditional on latent ability affirmative action increases engineering col-
as measured above, performance on the lege attendance for men and women from
first-year exam is increasing in effective
reserved castes at all points in the entrance
rank. Note that the effective rank contains exam distribution, although the effects are
no information about individual ability strongest in the middle of the distribution.
beyond what is contained in the entrance Despite these gains, the lower castes are
exam score, that is, the absolute rank. This underrepresented in the engineering col-

result thus provides an unbiased estimate leges relative to their share of the popula-
of the effect of affirmative action on college tion. With regard to the impact of reservation
performance. An augmented specification on test scores, average gains for SC and ST
interacts effective rank with caste–gender men are 0.12 and 0.18 standard deviations
dummies. The coefficients on the interac- respectively, with similarly sized effects for
tion terms are insignificant, indicating that SC and ST women. These magnitudes are
all eligible caste groups benefit equally roughly in line with the estimated effect
from affirmative action.24 from l arge-scale education interventions
The preceding results tell us that affir- in the United States and other developing
mative action increases the college atten- countries, including India.
dance of reserved groups and improves their Although caste-based affirmative action
performance (conditional on attendance). applies to institutions of higher education
However, they do not provide a mechanism and government jobs, it can affect the choices
through which these benefits are generated. that forward-looking parents make for their
24 With a more parsimonious specification that includes 25 Conditional on the latent ability measure, which sub-
effective rank and its interaction with the gender dummy, sumes the absolute rank, the effective rank can only affect
effective rank has a positive and significant effect for both first-year performance by changing the college that a stu-
men and women, although the effect is slightly weaker for dent attends. The instrument thus satisfies the exclusion
women. restriction.
children earlier in life. Cassan (2019) exploits access to reservation, is negative and signif-
two exogenous policy changes to estimate icant for the boys but not the girls. This is
the effect of caste reservation on the educa- indeed what Cassan finds.
tional choices of SC children. The first policy
6.4 Demand for Education
change redrew the borders of a number of
Indian states in 1956. SC status was not har- The preceding studies describe how
monized within states at the time, resulting different caste groups fare in the Indian
in the new residents of a state sometimes not educational system. While the conver-

having the same reservation status as the old gence in education levels documented by
residents, despite belonging to the same jati. Hnatkovska, Lahiri, and Paul may be gener-
This changed with the second policy change ated, in part, by universal access to primary
in 1976 that ensured that all members of a schooling and affirmative action in higher
jati had the same reservation status in each education, the differential demand for edu-
state. cation at the narrower caste level may also
Cassan (2019) exploits the discrepancy in have played a role in this convergence.
reservation status within jatis between 1956– Munshi and Rosenzweig (2006) document
76 to examine the effect of affirmative action, differences in schooling and occupational
measured by SC status, on school enroll- choice, by caste and gender, in Bombay
ment. His difference-in-difference research city. In their urban setting, lower-caste
design compares the schooling choices made men, who have been engaged in heavily
for children after 1976 (the post period) with networked b lue-collar jobs for generations,
corresponding choices made before 1976 are least responsive to the new opportuni-
(the pre period) across two groups: the “new” ties that become available in the globalizing
SCs who were only reserved after 1976 and Indian economy. This inertia is reflected in
the “old” SCs who were always reserved. the schooling choices made by lower-caste
The baseline specification finds no effect of boys. Globalization will also affect schooling
affirmative action; that is, the p ost-new-SC choices in rural India, where the majority of
coefficient is close to zero and statistically the population resides, but the new oppor-
insignificant. However, stronger results are tunities that arise will mostly be low-skill
obtained when the sample is partitioned by (urban) occupations that require a different
gender. The p ost-new-SC coefficient is now type of education. Luke and Munshi (2011)
positive and significant for the boys, with exploit a unique setting—a group of tea plan-
weak and even sometimes negative effects tations in the South Indian High Range—to
for the girls, depending on the specification. assess how economic development will affect
Caste fixed effects are included in all specifi- schooling choices, by caste, in the future. The
cations, and the identifying assumption is that key requirement when estimating differen-
new SCs and old SCs within a caste have the tial effects by caste on schooling choices is to
same intrinsic characteristics. If convergence control for differences in household wealth
is immediate, then this implies that new SCs across castes. As discussed below, Luke
and old SCs will have the same educational and Munshi’s research setting satisfies this
outcomes in the p ost-1976 period. To ratio- requirement.
nalize this with the estimated effects, it must The High Range, a mountainous area
be that the new-SC coefficient, which mea- straddling the modern Indian states of Kerala
sures the difference in educational attain- and Tamil Nadu, was virgin forest until it was
ment between new SCs and old SCs in the acquired by British planters and converted
pre-1976 period when only the old SCs had into tea plantations in the last quarter of the
nineteenth century. Because the plantation of the caste system in South India was the
land was previously uninhabited, workers institution of agrestic slavery. Abolition of
were brought to the High Range from the slavery coincided with the establishment
plains in Tamil Nadu. The workers on the tea of the tea plantations, and, not surpris-
plantations today are the third-generation ingly, workers belonging to the former slave
descendants of those migrants, whose pop- castes—the Pallars and the Paraiyars—are
ulation has been supplemented by a fresh disproportionately represented in the High
influx of new workers from the plains in each Range. The slave castes were historically
subsequent generation through marriage. spread over a very wide area, with their
The workers in the High Range continue to members having minimal social interactions
be tied to their ancestral communities in rural with each other. We would expect their rural
Tamil Nadu, despite having lived on the tea networks to be relatively weak and, thus, the
plantations for many generations. Loans and expectation is that workers from those castes
transfers flow back and forth between the tea would have the greatest incentive to invest in
plantations and the origin communities, the their children’s education.
children of the workers are often sent home Using data from a survey of plantation
to study, and many workers will buy land, workers, augmented with administrative
build a house, and return to their ancestral data on income over multiple years, Luke
villages when they retire. Perhaps the most and Munshi show that workers belonging
distinctive feature of the South Indian kinship to the former slave castes have the same
structure is marriage among close relatives income and jobs in the tea plantations but
(Kapadia 1995). Reinforcing existing network are less attached to their ancestral communi-
ties, many workers continue to marry their ties and have higher levels of education than
children to relatives from the ancestral loca- the non-slave castes.26 It is worth noting that
tion in the traditional fashion. The key paren- these statistics are very similar across the
tal decision is the extent to which children will non-slave castes, despite the fact that they
be tied to their ancestral communities and the are drawn from very different positions in
traditional economy. This decision will, in the social hierarchy.
turn, determine the probability that children Looking within the household, Luke and
will marry among their relatives as well as Munshi show that part of the difference in
their (optimal) educational attainment. education across castes in the tea plantations
As in Munshi and Rosenzweig (2006), can be explained by variation in the incen-
attachment to the traditional network lowers tives of men and women to remain attached
parents’ incentives to invest in their children’s to their ancestral communities. Interactions
education but for different reasons. First, with the ancestral community and the pros-
education does not help the child (particu- pect of permanently settling there after
larly the daughter) match with a better part- retirement are especially burdensome for
ner if she is married in the customary fashion women belonging to the former slave castes.
to a relative. Second, attachment to the net- Thus, while both parents might benefit from
work reduces spatial mobility and accompa-
nying job opportunities, as in Munshi and 26 The fact that Pallar and Paraiyar workers have the
Rosenzweig (2016), with an accompanying same jobs and earn the same income but have higher edu-
decline in the returns to education. The cation is not necessarily inconsistent. Higher education for
heterogeneous caste composition in the tea someone growing up in the tea plantations increases the
probability of exit, with access to better job opportunities,
estates allows Luke and Munshi to test this in adulthood, but this outcome is not certain. What we
hypothesis. Recall that a particular feature observe are those individuals who stayed behind.
mutual insurance, old age support, and other obtained for the former slave castes alone,
forms of assistance organized around their accounting for part of the caste differences
ancestral communities, the cost of maintain- in schooling and marriage choice observed in
ing the social ties that are necessary to obtain both the parental and current generation.
this social support may be substantially larger The apparent willingness of the Pallar
for the women than the men in those castes. and Paraiyar women to act on their
The wider gender preference gap among preferences does not, however, come with-
the former slave castes is likely reinforced out a cost. Previous evaluations of female
by the greater ability among the women income-generation programs in developing
in those castes to act on their preferences. countries have documented a male backlash,
Lower-caste women historically worked as measured by the increased incidence of
outside the home, and although they were marital violence, when recently empowered
typically employed in low-paying menial women shift household decisions away from
occupations, it has been argued that the their husbands’ preferences (Schuler et al.
income and independence that this work 1996). Similar patterns of marital violence
provided gave rise to a culture in which have been documented in the international
these women were relatively assertive migration literature as well (Min 2001). Luke
within their households (Kapadia 1995, and Munshi provide evidence that a relative
Chakravarti 1993). Given that culturally increase in female income among the for-
determined beliefs and practices are often mer slave castes also leads to an increase in
slow to change, we expect these caste dif- marital violence. In contrast, female income
ferences to persist in the High Range even has no effect on household decisions and
though all women work in the tea estates, outcomes among the other castes in the tea
irrespective of their caste affiliation. estates and no effect on marital violence as
Luke and Munshi test for these gender well. Apart from providing independent
differences by caste by estimating the effect support for the cross-caste results described
of an increase in female income, net of total above, this result also links marital violence,
household income, on important decisions a serious social problem in India, to underly-
and outcomes for the children. Women have ing caste-based institutions in a way that has
permanent full-time jobs as tea-leaf pluckers not been done before.
in the plantations, and they actually earn more Munshi and Rosenzweig’s (2006) analysis
than the men on average. Exploiting exoge- of educational choices in urban India and
nous variation in male and female incomes Luke and Munshi’s (2011) analysis of these
generated by differences in rainfall across choices in the tea plantations both indicate
plantations and using a variety of indepen- that historically disadvantaged caste groups
dent estimation strategies, Luke and Munshi may be especially responsive to new eco-
find that a relative increase in female income nomic opportunities, leading them to invest in
weakens the family’s ties to the ancestral com- their children’s education because they have
munity as the woman gains bargaining power fewer ties to the traditional economy to hold
within the household. The children are less them back. D emand-side effects thus explain
likely to marry a relative, be schooled in the part of the convergence across castes that
ancestral location, and ultimately settle there. was documented by Hnatkovska, Lahiri, and
At the same time, an exogenous increase in Paul in recent decades. Luke and Munshi’s
relative female income increases the educa- analysis indicates, in addition, that disadvan-
tional attainment of the children. Most impor- taged individuals within those groups, notably
tantly, significant female income effects are the women, may be at the forefront of these
changes, providing an interesting and rela- r elationships between castes and parties, and
tively unexplored gender component to the it is thus not surprising that caste has played
role played by caste in the changing Indian a central role in Indian democratic politics.
economy.27 Yadav (1999) divides Indian politics after
independence into three electoral systems.
The first electoral system, covering the
7. Politics and Access to Public Resources
first four general elections from 1952–67,
was dominated by the Congress Party. The
7.1 State and National Politics
opposition was regionally fragmented and

Clientelism, a phenomenon by which pol- often came from within the Congress. “The
iticians deliver benefits to specific individ- voter of course did not vote as an individual
uals or groups of individuals in return for but rather as a member of a politicized jati”
their votes, is pervasive in many developing (Yadav 1999, p. 2394). For the caste to vote as
countries. For a reciprocal arrangement of a block, voters spread over a wide area must
this nature to be sustained, each side must coordinate on a candidate. One way for the
credibly commit to cooperate with the other. decentralized caste to coordinate in a given
Politicians represent political parties, which election would be with the help of a caste
have a long-term reputation to maintain with association. Rudolph and Rudolph (1960)
the electorate, and so will fulfill their prom- note that caste associations did play a role in
ises even if they are elected for a single term Indian electoral politics immediately follow-
and have personal preferences that do not ing independence. However, relatively few
align with the promises they made ex ante. Indians are members of formal religious or
While it is difficult to keep track of the vot- caste associations. For the caste to vote as a
ing behavior of individuals, well-organized block, other coordinating mechanisms are
groups can similarly credibly commit to hon- needed.
oring their reciprocal obligation to a politi- A common feature of clientelist politics
cal party (because they too have a long-term is the role played by brokers or middlemen
reputation to maintain). In India a natu- who link voting blocks to parties. Srinivas
ral social unit around which “vote banks” (1951) describes how higher caste land-
are organized is the caste or jati. We have owners and moneylenders served this func-
seen in many different settings that social tion in the Mysore village that he studied.
connectedness within the caste facilitates In other areas, representatives of the caste
private economic activity by solving informa- itself served this vital function; this is simi-
tion and commitment problems. The same lar to the role played by foremen in link-
social ties can be used to support reciprocal ing workers from their caste to firms in the
urban economy. The success of the Congress
Party in the decades after independence has
27 Given that lower-caste women face fewer restrictions
been attributed to its ability to establish a
on mobility and labor force participation, we would expect
them to be more responsive to new economic opportu- network of such middlemen (Kothari 1970,
nities than upper-caste women. However, an experiment Kohli 1990). Although most Congress lead-
providing basic financial literacy and business skills to ers came from the upper castes, the middle-
women in Ahmedabad generated the curious result that
upper-caste women but not lower-caste women responded men generated popular support for the party
to the intervention (Field, Jayachandran, and Pande 2010). that was much broader than its leadership
The argument that the returns to training were for some (Brass 1994).
reason higher from more restrictive groups in this setting
does not square with the additional finding that there was Although the Congress Party may have
no effect for the most restricted Muslim women. been successful in organizing caste-based
voting banks, many castes were left out. the state level. This reconfiguration of party
Competition within the party for scarce politics explicitly on caste lines is most visible
resources also resulted in some castes ben- in Uttar Pradesh, India’s largest state with a
efiting more than others. According to population of two hundred million, where
Yadav, the second electoral system, starting the Bharatiya Janata Party (BJP) represented
in 1971, was characterized by the entry of the upper castes, the Samajwadi Party (SP)
middle-caste (OBC) politicians and the end of represented the backward castes, and the
the complete domination of national p olitics Bahujan Samaj Party (BSP) represented the
by the Congress Party. As politics became scheduled castes. Similar social cleavages in
more competitive, there was a consolida- party politics also emerged in other states
tion of vote banks into state-wide multi-jati such as Gujarat, Maharashtra, and Bihar.
alliances, such as Kshatriya–Harijan–
The advantage of clientelist politics is
Adivasi–Muslim (KHAM) in Gujarat and that it allows reciprocal relations between
Ahir–Jat–Gurjar–Rajput (AJGAR) in Uttar political parties and population groups to
Pradesh. Nevertheless, the basic strategy of be sustained. The well-known disadvantage
creating a m ulti-caste coalition of voters at is that there will be an u nder-provision of
the local level by delivering public resources public goods in favor of transfers targeted at
to specific castes once elected continued to the elected leader’s vote bank (Bardhan and
be followed by all political parties (Chhibber Mookherjee 2012). In the Indian variant of
2002). clientelist politics, as described above, the
This changed in the late 1980s with the emergence of explicitly caste-based parties
onset of Yadav’s third electoral system, char- was accompanied by the formation of a new
acterized by the emergence of parties orga- caste-group identity (Yadav 1999). Banerjee
nized explicitly along (broad) caste lines. As and Pande (2009) examine the (negative)
discussed above, caste has always played an consequences of this identity for the qual-
important role in Indian politics. But while ity of elected leaders. Their basic idea is
the jati in its local setting constituted the that if voters care about their leader’s group
voting block in the first electoral system and identity, then the party associated with the
statewide alliances of jatis served this role in numerically dominant c aste group in a polit-
the second electoral system, the third elec- ical constituency enjoys an electoral advan-
toral system witnessed the formation of a tage. Its candidates will win even if they
new caste identity based on jati groupings are less competent than their rivals. This
such as upper caste, OBC, or SC. implies that in constituencies with a numer-
In Chhibber’s view, the event that precipi- ically dominant caste group, an increase in
tated this change in the political equilibrium caste-group identity will result in a decline
was the Mandal Commission report that in leadership quality if the elected candidate
expanded reservations in government jobs belongs to the party associated with that
and educational institutions to the OBCs. caste group. Conversely, if the elected can-
The Congress Party was forced to distance didate belongs to a party associated with the
itself from the report because forward, back- minority group in that constituency, he will
ward, and scheduled castes were part of its be of relatively high quality.
coalition. The social polarization created by Banerjee and Pande test this prediction
the report thus opened the door for other with data from one hundred constituencies
parties to step in and represent the inter- in Uttar Pradesh at two points in time, 1980
est of broad caste groupings that were large and 1996. Caste identity is believed to have
enough to influence electoral outcomes at increased substantially in Uttar Pradesh over
this period with the emergence of parties groups could have been left worse off. Apart
such as the SP, favoring the backward castes, from these distributional effects via the
and the BSP, favoring the scheduled castes. competitive democratic process, the reser-
This implies that in a constituency dominated vation for SCs and STs in state and national
by low castes, elected politicians represent- legislatures could also have shifted public
ing upper-caste parties should have relatively resources toward these historically disadvan-
high quality in 1996. Conversely, in constit- taged c aste groups. In parallel with affirma-
uencies dominated by upper castes, elected tive action in higher education and central
politicians from the same upper-caste parties government jobs, the Indian constitution also
should be of relatively low quality. Politician requires that reservation for SCs and STs in
quality is based on measures of corruption the national legislature reflect their share of
collected retrospectively from key infor- the population in each state. If political par-
mants ( journalists and former politicians) ties can credibly commit to particular policies
in each constituency. The assumption here regardless of the preferences of their elected
is presumably that more corrupt politicians representatives, as assumed in the previous
capture more resources for themselves at the discussion on clientelist politics, then reser-
expense of the electorate. vation of seats in specific constituencies (13
The first result is that the probability that percent on average for the SCs and 7 percent
a constituency dominated by the low castes on average for the STs) will have no effect.
has a leader from an upper-caste party, that However, as Pande (2003) notes, if political
is, the Congress or the BJP, fell by 38 per- parties cannot exercise complete control over
cent from 1980–96. The second result is that their representatives once they are elected
those leaders are substantially less likely to and if these representatives can influence
be corrupt in 1996 versus 1980, whereas the the allocation of resources by the legislature,
converse is true in constituencies dominated then reservation will improve outcomes for
by the upper castes (where they are more the beneficiaries of affirmative action.
likely to be corrupt in 1996). An additional Pande examines the effect of caste reser-
prediction of Banerjee and Pande’s model is vation on access to public resources using
that there should be no change in the quality state-level data from 1960–92. Political res-
of elected politicians over time in competi- ervation is measured by the fraction of juris-
tive constituencies where no caste group is dictions reserved (separately) for SCs and
dominant. This implication, which is espe- STs in a state in a given year. Targeted public
cially useful in ruling out alternative explana- resources are measured by (i) the fraction of
tions for the differential change in leadership state-government spending devoted to SC
quality across constituencies, is also borne and ST welfare programs (this is 6–8 percent
out in the data.28 of total spending on average), and (ii) the
fraction of state government jobs reserved
7.2 Affirmative Action
for SC/STs. One potential concern when
The change in Indian party politics over estimating the relationship between targeted
time would have served the interest of government spending and reservation is that
particular caste groups that succeeded in these expenditures could have been made
mobilizing politically. However, other caste in the absence of reservation. A state with a
greater share of reserved seats is, after all, a
28 Banerjee and Pande’s model also has implications for
state with a greater share of SC/STs in the
the difference in quality of the winning candidate and the population, and they would naturally have a
runner-up that are borne out in the data. say in resource allocation.
Pande’s solution to this identification goods, covering the areas of education,

problem, which is similar to Bagde, Epple, health, water, electricity, and communica-
and Taylor’s (2016) research design, takes tion, with access to each good measured by
advantage of the fact that the underlying the fraction of villages in the constituency
variable upon which the affirmative action where it is available. The 1971 data indi-
is based is not perfectly correlated with the cate that an increase in the share of SC/
fraction of SC/STs in the population at a STs in the constituency is associated with a
given point in time. In Bagde, Epple, and deficit in almost all public goods. Once the
Taylor’s application, the effective rank in 1971 and 1991 data are pooled, however,
the list of engineering college applicants is the evidence indicates that there has been
positively (but not perfectly) correlated with substantial convergence in access to public
the absolute rank, that is, the score on the goods over time.30 When this convergence is
college entrance exam. The latter variable decomposed by caste group, it turns out that
can thus be used to control for individual the gains are mostly in the SC-dominated
ability when estimating the effect of reser- areas, while the STs appear to have been left
vation, measured by the effective rank, on behind. The convergence documented by
college performance. In Pande’s application, Banerjee and Somanathan can be explained,
the slippage between the population share in part, by the political reservation policy
of the SC/STs and the fraction of reserved and its redistributive effect that Pande finds.
seats is generated by the time lag in chang- The reconfiguration of electoral politics on
ing quotas. These quotas only change in an caste lines over time, coupled with the broad
election year and are based on the preced- antipoverty programs that were pushed

ing census, whereas the population share by the Indian government, could also have
changes continuously over time.29 The main played a role in this convergence. The main
finding of the analysis is that conditional on message from the preceding analyses, com-
the SC/ST population share, an increase in plementing the findings of Hnatkovskaya,
SC reservation has a strong positive effect Lahiri, and Paul (2012) in the private sphere,
on job quotas, but not on welfare programs, is that there has been substantial equaliza-
whereas an increase in ST reservation results tion in access to public resources over time.
in a substantial increase in ST welfare pro- Did the targeting of public resources to
grams (without a significant effect on jobs or specific caste groups have negative conse-
SC welfare programs). quences for overall public good provision?
In a related study, Banerjee and Although the counterfactual experiment,
Somanathan (2007) examine the location of without affirmative action, is unavailable, the
public goods between 1971–91 in approxi- evidence indicates that overall public good
mately five hundred parliamentary constitu- provision has, if anything, increased over
encies in rural India to assess the changing time. Banerjee and Somanathan document
relationship between the caste composi- a dramatic expansion in rural infrastruc-
tion of the constituency and access to pub- ture between 1971–91. By 1991, primary
lic resources. They consider fifteen public schools had been established in 75 percent
29 The population share, which is not directly mea- 30 This convergence does not arise mechanically
sured, is derived by interpolating between census rounds. because access to public goods was close to complete in
Because the preceding census is used for this interpola- advantaged constituencies in 1971. Half the public goods
tion, the reservation variable, which is constructed from that Banerjee and Somanathan consider were available in
the preceding census alone, provides no additional infor- less than 5 percent of Indian villages in 1971 and in less
mation about the actual population share. than 10 percent of villages in 1991.
of all villages, and the share of villages with regular elections for the position of pan-
electricity increased from 18 percent to chayat president and for each ward repre-
70 percent. There was a n inefold increase sentative have been held every five years in
in access to piped water over this period. most states. At the panchayat level, seats are
Although there are many contributory fac- reserved in each election for SCs, STs, and
tors, part of this dramatic increase can be OBCs, but now in proportion to their share
directly attributed to targeted spending of the population in the district. Seats are,
through affirmative action. Pande (2003) in principle, reserved randomly across wards
shows that ST (but not SC) reservation sig- and, for the position of the president, ran-
nificantly increased per capita state gov- domly across panchayats, from one election
ernment spending, and Chin and Prakash to the next in each district. The only restric-
(2011), using a similar identification strategy, tion is that no seat can be reserved for the
show that ST (but not SC) reservation signifi- same group across consecutive elections
cantly reduced poverty between 1960–2000. within a constituency. In practice, however,
There are evidently important differences seats have historically been reserved in ear-
in the effect of SC and ST reservation, and lier election rounds in constituencies with a
these two population groups have fared very larger share of SCs, STs, and OBCs, respec-
differently in the p ostindependence period. tively (Dunning and Nilekani 2013).
Further investigation of these differences Without the discipline of a party affiliation,
would have great value for the design of and with limited possibility of reelection
development policy in India. (given the frequent exogenous turnover due
to randomized reservation), candidates in
7.3 Local Politics
panchayat elections cannot credibly commit
At the state and national level, the liter- to exert the effort that is needed to ensure
ature has devoted much attention to the an adequate supply of public goods once in
role that caste plays in party politics. At office. In addition, local political representa-
the local level, parties are less relevant. In tives are poorly compensated for their efforts;
fact, many Indian states prohibit candi- panchayat presidents are paid 50–60 dollars
dates in local elections from contesting on per month (less than the minimum wage),
party lines. Caste matters now through the while ward representatives earn even less. In
caste affiliation of the elected representa- the absence of the standard political incen-
tive, and the focus of much of the literature tive mechanisms, Munshi and Rosenzweig
on local governments in India has been to (2018) posit that social ties within castes
understand the relationship between the could be exploited to elicit effort by elected
characteristics of the leader (including representatives. As discussed above, these
caste and gender) and the supply of public ties have been used to support clientelist
resources. arrangements between political parties and
The s eventy-thirdamendment of the jatis since independence. They have also
Indian Constitution, passed in 1992, estab- been used to facilitate private economic
lished a three-tier system of local govern- activity in an economy where many markets
ments or panchayats, at the village, block, function imperfectly. In the Indian local gov-
and district level, with all seats to be filled ernments, they could be used to ensure that
by direct election. The village panchayats, the representative puts in the requisite level
which often cover multiple villages, were of effort, even when he is elected for a single
divided into 10–15 wards. Panchayats were term, while at the same time ensuring that he
given substantial power and resources, and is compensated (informally) for his efforts.
While caste networks may facilitate eco- that is optimal from caste k’s perspective,
nomic activity, their involvement in the taking as given the candidate’s ability, ω. The
economy generates particular inefficien- effort is chosen to maximize
cies, as discussed above. Many of these
inefficiencies can be traced to the fact that (13) Nk a  β  −  _
a
ω   ,
cooperation and trade cannot cross caste
lines. The use of the caste to solve political where a  βis the level of the public good,
incentive problems is similarly associated and the assumption is that the level of the
with its own limitations. In particular, the public good maps one for one into the utility
competence of the candidate put forward derived from its consumption by each resi-
by a caste and the effort he exerts, condi- dent of the constituency. The second term
tional on being elected, will be optimal from in the expression above measures the effort
the perspective of the caste rather than the cost of the caste’s chosen candidate, condi-
constituency as a whole. Because each caste tional on being elected. We make the usual
accounts for only a share of the local pop- assumption that the unit cost is decreasing
ulation, there will be a s uboptimal level of in ability. Based on the solution to the maxi-
leadership competence and effort, with an mization problem, the optimal level of effort
accompanying undersupply of public goods. from the caste’s perspective, a (ω, Nk ), is an
To derive an empirical test of c aste-specific increasing function of the candidate’s ability,
cooperation and to quantify the resulting ω, and the size of the caste in the constitu-
undersupply of public goods, Munshi and ency, Nk .
Rosenzweig develop a model in which N The next step is to determine which indi-
individuals belonging to K castes reside in a vidual will be selected by the caste as its
local constituency. Each caste k consists of candidate. Caste kwill put forward as its
Nkindividuals, with the ksubscript sorting candidate the individual with ability ω that
castes by size; N k−1 < Nk . An individual maximizes
from among the constituency’s residents
is elected to represent it in each election a(ω, Nk )
(14) Nk [a(ω, Nk )]  β  −  _ ω      − αω,
 
term. The representative’s task is to supply a
non-excludable public good to the residents
of the constituency. Representatives receive where the first term in the expression above
no monetary compensation and are elected measures the utility derived from the public
for a single term. A (long-term) reciprocal good by the N kmembers of the caste. The
relationship between the representative and second term measures the candidate’s effort
the entire constituency is thus infeasible. cost, conditional on being elected, and the
However, the representative’s caste could third term the opportunity cost to the can-
credibly commit to compensate him ex post didate of holding public office. Although the
for his efforts through informal side trans- representative could extract personal rents
fers, even if he is elected for a single term, for himself and advance his political career,
because they are connected to each other in the private benefit of holding local public
many ways outside the political system. Once office is limited. The α parameter measures
the side transfers are in place, each caste the difference in the returns to ability in
will select a candidate whose ability, and the the private sector and public office, which
effort he subsequently exerts if elected to we thus assume is positive. Substituting the
represent the constituency, are optimal from expression for a(ω, Nk )derived from the first
its perspective. We first derive the effort, a, maximization problem and then maximizing
with respect to ability, both ω and aare ulti- the supply of public goods. In contrast, the
mately derived as increasing functions of size of the representative’s caste in the village
Nk
. Intuitively, because more individuals or the number of ward residents that belong
benefit from the non-excludable public good to the representative’s broader caste group-
in a large caste, it will be in its collective ing, that is, SC, ST, or OBC, has no bearing
interest to select a more competent candi- on the supply of public goods.
date who will, in turn, exert greater effort. Munshi and Rosenzweig quantify
Cooperation within the caste results in a the undersupply of public goods due to
higher level of public good provision rela- caste-specific cooperation by estimating the
tive to the benchmark model where the rep- structural parameters of the model and then
resentative only cares about his own utility conducting counterfactual simulations. Their
(Nkwould be replaced by one in the maxi- analysis examines the supply of six major
mization problems above). However, public public goods falling within the p anchayat’s
good provision is less than first best (where jurisdiction measured by whether there were
Nkwould be replaced by N). The difference expenditures on either new construction or
between the size of the representative’s caste maintenance for each of those goods on each
and the population of the constituency, N , street in the ward in a given election term.31
determines the extent of the u ndersupply of Across the major Indian states covered by the
the public good. This model of public good REDS data, the average fraction of the ward
provision shares an important feature of pre- population receiving a given public good
vious models of insurance, migration, and across the six goods they consider is less than
occupational choice, which is that outcomes one-third in 40 percent of wards. The first
depend on the size of the relevant group, the counterfactual result is that if the ward rep-
caste network in previous applications, and resentative internalized the benefit derived
the size of the caste in the constituency in from the public goods by all residents rather
the current model. than just his caste, then the entire population
Munshi and Rosenzweig (2018) test the would receive all public goods in all wards.
hypothesis that cooperation is restricted to the The counterfactual simulations can also be
representative’s caste and then quantify the used to assess the impact of existing policies
resulting u ndersupply of public goods using on the supply of public goods. Ward repre-
data from the 2006 REDS, which covers a sentatives are responsible for the administra-
large sample of wards—the most local level of tion of welfare programs, in addition to their
government—across the major Indian states traditional role of delivering public infra-
over three election terms. The data include structure. Following Besley et al. (2004),
the caste and education of the elected rep- Munshi and Rosenzweig measure welfare
resentative, the supply of public goods at the transfers by the receipt of Below the Poverty
street level, which can be mapped to the ward Line (BPL) cards that give recipients access
level, and the receipt of welfare transfers by to subsidized food and most government
specific households in each election term. In welfare programs. BPL cards are meant to
addition, the data provide the caste of every be received by economically disadvantaged
household in each ward. Providing direct sup-
port for caste-specific cooperation, there is a 31 The six public goods—drinking water, sanitation,
positive and significant relationship between roads, electricity, street lights, and public telephones—
the size of the elected representative’s caste account for 67.5 percent of the panchayat’s discretionary
spending. Education and health are not included in the list
in the ward and both the representative’s edu- of public goods because they fall outside the purview of the
cation, which proxies for his competence, and panchayats (Bardhan and Mookherjee 2006).
households, but it is well known and well caste of the elected representative, it is sig-
documented, for example, by Besley, Pande, nificantly more likely to hold a BPL card com-
and Rao (2012), that ineligible households pared to when the representative belongs to
that are politically connected can also bene- another caste. However, there is no evidence
fit from them. If the total supply of the wel- that public goods are targeted to streets on
fare transfers is fixed and the representative which the representative’s caste is dispro-
first targets his own caste, then outsiders portionately represented. As with the results
will be worse off with respect to the welfare on the supply of public goods, the fact that a
transfers when a larger caste is in power household belongs to the broad caste group
(as documented empirically by Munshi and of the elected representative, that is, SC, ST,
Rosenzweig). This makes larger castes rela- or OBC, does not affect the probability that
tively unpopular with the electorate, reduc- it will receive a BPL card conditional on
ing the likelihood that their candidates will whether or not the household belongs to the
be elected and, thereby, lowering the supply same caste. Previous research on c aste-based
of public goods. Counterfactual simulations targeting in Indian local governments has
of the estimated structural model indicate documented that SC/ST households are
that the effect of decoupling the delivery of more likely to receive publicly provided
public infrastructure and the administration private goods (welfare transfers) when the
of welfare programs on the supply of public panchayat president’s position is reserved for
goods would be modest. However, another SC/STs; see Besley et al. (2004) and Bardhan,
policy—reservation for historically disadvan- Mookherjee, and Torrado (2010). While it
taged castes and tribes in panchayat coun- may make sense from a policy perspective to
cils—would appear to have more impact. assess whether political reservation for par-
Reservation will mechanically reduce the ticular caste groups benefits the members of
size of the representative’s caste in the ward those caste groups on average, Munshi and
by restricting the set of castes that are eligi- Rosenzweig’s results indicate that targeting,
ble to stand for election. The disadvantaged at least at the ward level, is occurring at the
minority groups will also account for a rela- finer jati level.32
tively small share of the ward population on Political clientelism is characterized by
average. Based on the counterfactual sim- the transfer of targeted public goods, jobs, or
ulations, a policy that combines the decou- services to groups of voters in return for their
pling of welfare transfers and public good political support (Stokes 2009). The caste or
provision with de-reservation would reduce jati, rather than caste groupings such as SC,
from 40 percent to 20 percent the fraction ST, or OBC, is the social unit around which
of wards in which less than o ne-third of the these reciprocal arrangements are orga-
population received each public good, aver- nized at the local level (Brass 1994). This
aged over the six goods. argument runs as a common thread through
A key assumption in Munshi and the preceding sections of the paper and is
Rosenzweig’s analysis is that the public consistent with Munshi and Rosenzweig’s
goods they consider are non-excludable at
the local—ward—level, whereas the welfare 32 Although most studies on targeting in panchayats
transfers can be targeted at specific house- focus on welfare transfers, Gille (2016) uses the 2006
holds. Exploiting changes in the caste of the REDS to show that individuals are more likely to apply for
elected representative over election terms reserved public sector jobs when the panchayat president
belongs to the same caste. Her interpretation of this result
within each ward, they are able to verify this is that the president uses his political connections at the
assumption. If the household belongs to the state level to help the applicant from his caste get a job.
finding that welfare transfers are targeted workers. The additional advantage enjoyed
to the representative’s jati. Their analysis by the Marathas is that they control local
goes beyond the standard clientelist model agricultural input and product trading net-
by allowing transfers to simultaneously flow works. Maratha patrons can offer access to
in the opposite direction, from the caste to these networks to their n on-Maratha clients
the representative. This informal compen- in return for their votes. The more workers
sation ensures that the competence of the a Maratha landlord employs, the greater is
individual selected to represent the caste, the gain from suppressing wages. It follows
and the effort he exerts conditional on being that if the Marathas are the dominant land-
elected, are optimal from the perspective of owning caste in the village, then it is in the
the caste. Anderson, Francois, and Kotwal interest of the Maratha landlords to establish
(2015) extend the standard clientelist model a clientelist political arrangement.
in a different direction by allowing the elite To test this prediction, Anderson, Francois,
group in power to transfer private rather and Kotwal estimate the following equation:
than public resources to the electorate, in
the form of insurance against income shocks, (15) Yik = β0   + β1 MLDk   + β2 MPROPk  
in return for their votes.
Anderson, Francois, and Kotwal’s analysis + β3 MLDk   ⋅ MPROPk 
is set in rural Maharashtra. An unusual fea-
ture of Maharashtrian society is that a sin- + ψ Zik  + γ Zk  + ϵik,
gle caste, the Marathas, accounts for a large
fraction of the total population. Moreover, where Yikis an outcome for household
while Marathas are often the dominant land- iresiding in village k, MLDk indicates
owning caste in the village, many Marathas whether village kis Maratha land dominated,
do not own land and supply labor to Maratha MPROPk is the proportion of Marathas in
and non-Maratha landlords. The starting the village, Zik, Zkare household- and vil-
point for Anderson, Francois, and Kotwal’s lage-level controls, and ϵik is a mean-zero
model is the assumption that landlords do disturbance term. The same specification is
not value public programs because they raise used for village-level outcomes, except that
wages and lower labor compliance. Political the individual, i, subscript is suppressed.
power allows the landlords to reduce these Based on the model, clientelism is more
programs, but this is only possible if the elec- likely to be observed when MLDk = 1.
torate votes them to power. The landlords Thus, we expect public programs and wages
achieve this objective by offering private to be decreasing in MLDk, while access
insurance to the landless in return for their to private insurance (through their land-
votes. For this reciprocal arrangement to be lords) by the landless and landlord profits
an equilibrium, each side must credibly com- should be increasing in MLDk . The sign
mit to honoring its obligation. An important of the coefficient on the interaction term,
assumption in the model is that the Marathas β3, is ambiguous. If intra-caste ties between
are more cohesive than other castes; in par- Maratha workers and landlords are more
ticular, only Maratha landlords can cooperate (less) important than the access to trading
with each other. Given that the Marathas are networks by non-Maratha workers through
the only caste on both sides of the clientelist their Maratha landlords, then clientelism
arrangement, the social cohesion within that is increasing (decreasing) in the interaction
caste can also be exploited to sustain coop- term. In contrast, β 2 = 0because if
eration between Maratha landlords and MLDk = 0 , then there is no clientelism,
irrespective of the caste composition of the (16) Yik = β0   + β1 RESERVEDk  
village population. Using data from 320
villages in rural Maharashtra, Anderson,   + γk Zk   + ϵik ,
+ ψk Xik
Francois, and Kotwal verify each of these
predictions, with the sign of the β3 coeffi- where Yikis the level of public resources
cient indicating that access to trading net- received by household iresiding in village
works dominates intra-caste cohesion as a k, RESERVEDkindicates whether the pan-
determinant of clientelism. chayat election was reserved for lower castes
Caste reservation is implicitly based on (SC, ST, OBC) in village k , Xik and Zk are
the assumption that clientelism is present, household and village-level controls, and ϵik
that is, that political representatives will is a mean-zero disturbance term. The same
target public resources to their own group. specification is used for village-level out-
The available evidence indicates that disad- comes, except that the individual, i , subscript
vantaged minorities (SCs and STs) are more is suppressed.
likely to benefit from welfare programs Based on the preceding discussion, the
when the political representative belongs to prediction of Anderson and Francois’ model
their own caste. While reservation for disad- is that β1 > 0. This is because in reserved
vantaged minorities thus has distributional elections, the next term will certainly be
benefits, Munshi and Rosenzweig’s (2017) unreserved, which implies that the ability
analysis indicates that there can be nega- of the incumbent elected representative
tive efficiency consequences because reser- to extract rents is reduced. However, this
vation reduces, on average, the size of the argument does not apply if the jati in power
representative’s caste. The other obvious accounts for a majority of the population, in
reason why the supply of public goods could which case its representative would remain
decline with reservation is that the pool of in power, with or without reservation. It
potential candidates is weaker in histori- would also not apply if the representative’s
cally disadvantaged minorities. However, caste accounts for a small minority of the
new research by Anderson and Francois population, in which case he would only be
(2017) indicates that the effect could go in elected by chance. Anderson and Francois
the opposite direction under certain cir- thus focus on panchayat elections in which
cumstances, with reservation increasing the caste in power comprises 25–50 per-
the supply of public goods. Their analysis cent of the population. In those elections,
builds on Padró i Miquel’s (2007) model the β1coefficient is positive and significant.
of clientelism in which ethnic groups will However, this coefficient is statistically indis-
tolerate the corruption of their co-ethnic tinguishable from zero in elections where the
political representative and continue to vote population share of the representative’s caste
him into power when the alternative with lies outside this range (this result is robust
another group’s representative in power is to the specific range that is specified). One
much worse. Reservation generates exog- limitation of this analysis, which also applies
enous turnover in the political leadership, to Anderson, Francois, and Kotwal (2015), is
reducing the ability of incumbent repre- that panel data are unavailable. The poten-
sentatives to extract rents. Using the same tial for omitted variable bias is always greater
data from a sample of Maharashtrian villages with a cross-sectional analysis. A second
as Anderson, Francois, and Kotwal (2015), limitation of the analysis is that the electoral
Anderson and Francois (2017) estimate the process and, hence, the caste in power, is not
following equation: endogenized. Nevertheless, these results,
as in Munshi and Rosenzweig (2017), indi- Cassan and Vandewalle find that lower-caste
cate that caste reservation interacts with the women are more likely to be elected to the
caste composition of the village in complex president’s position than upper-caste women
ways that have an important bearing on the in reserved elections (women are almost
supply of public resources in Indian local never elected in unreserved elections). They
governments. then proceed to show that goods preferred by
In parallel with the literature on caste res- women and, independently, goods preferred
ervation in local politics, there has also been by lower castes, based on preferences
substantial work on the reservation of seats obtained from the 2006 REDS, are more
in the panchayat council for women. Previous likely to be supplied in elections reserved for
research on caste networks has found that women. Research on caste that adds a gender
men and women benefit differentially from component consistently uncovers interesting
these networks, with resulting differences and novel findings. Further work in this area
in participation by gender (Munshi and will likely have high payoff.
Rosenzweig 2006, Luke and Munshi 2011).
We would similarly expect caste and gender to
8. Conclusion
interact in Indian local governments, and this
is the subject of a new paper by Cassan and Caste continues to play an important role
Vandewalle (2017). In a well-known paper, in the Indian economy. Networks organized
Chattopadhyay and Duflo (2004) examine at the level of the caste or jati provide insur-
the impact of gender reservation for the pan- ance, jobs, and credit for their members
chayat president’s position on the choice of in an economy where market institutions
public goods. The main finding is that public are inefficient. Affirmative action for large
goods that are preferred by women, based groups of historically disadvantaged castes in
on the frequency of complaints lodged by higher education and India’s representative
men versus women in public meetings, are democracy has, if anything, made caste more
more likely to be supplied in election terms salient in society and in the public discourse.
where the president’s position is randomly Newly available evidence with nationally
reserved for women. Cassan and Vandewalle representative data indicates that there has
add to this research by adding a caste com- been convergence in education, income,
ponent to the gender reservation. In particu- occupations, and consumption across caste
lar, they hypothesize that lower-caste women groups over time. If this trend continues,
will be disproportionately more likely to be then there will no longer be a rationale for
elected than upper-caste women in reserved affirmative action. The available evidence
elections. This is not because those elections indicates that caste discrimination, at least
are also more likely to be reserved for lower in urban labor markets, is statistical, that is,
castes. Within each caste-reservation category based on differences in s ocioeconomic char-
and in open elections, o ne-third of the seats acteristics between upper and lower castes.
in each election term is randomly reserved This type of discrimination will disappear
for women. However, as discussed in section once there is convergence between these
6.4, lower-caste women have historically had broad caste groups. It is important to realize,
higher rates of labor force participation and however, that these changes will not occur in
enjoyed greater autonomy than upper-caste the near future. The current gap in educa-
women. It is entirely possible that they would tional attainment between SC/STs and upper
also be more likely to participate in local gov- castes is very wide, over two years of school-
ernments. Using data from the 2006 REDS, ing. Given the strong i ntergenerational
ersistence in human capital, the key vari-

p corresponding marriage patterns in rural
able driving convergence, it will be many India in ten-year intervals from 1950–2000.
generations before income and consumption While o utmarriage remained extremely stable
are equalized across caste groups. at 5 percent over the entire fi
fty-year period in
The caste-based economic networks that rural India, o utmarriage increases over time
currently serve many functions will also in the urban data. It is even lower than 5 per-
disappear once markets begin to function
cent, somewhere between 2–4 percent, prior
efficiently. These networks continue to be to economic liberalization, but then starts to
active in the globalizing Indian economy increase steeply in the 1990s and is as high as
because information and commitment prob- 12 percent in the 1995–2000 period. Munshi
lems are exacerbated during a period of eco- and Rosenzweig’s study is set in Bombay
nomic change. In the long run, however, the city, which is at the vanguard of economic
markets will settle into place and the caste development in India. These results are thus
networks will lose their purpose. This has cer- almost certainly unrepresentative, but they
tainly been the experience in many developed do tell us what we might expect to see in the
countries. In the United States, for example, future.33 Marriage within the caste is key to
ethnic networks based on a European coun- the efficient functioning of caste-based net-
try (region) of origin supported their mem- works and the formation of a caste identity.
bers through the nineteenth century into the The changes in marriage patterns uncovered
middle of the twentieth century. Ultimately, in Bombay will likely be replicated elsewhere
however, these networks no longer served a in the future as the economy develops, and
useful role and today, outside of a few pock- this will signal the end of the caste system as
ets, European ethnic identity in the United we know it.
States is largely symbolic. We might expect These social changes, just like the end of
caste to similarly lose its salience as India affirmative action, will take many decades
develops into a modern market economy, and (even generations) to unfold. In the interim,
there is some evidence that this process may both research and policy would benefit from
have already begun. a recognition of the informal caste-based
Munshi and Rosenzweig (2006) found that institutions that are active in the economy. A
lower-caste boys in Bombay were held back counterfactual policy experiment described
by their working class networks, whereas in Munshi and Rosenzweig (2016) highlights
lower-caste girls (without that historical bag- the potential cost of ignoring these institu-
gage) were apparently willing and able to tions. Recall that Munshi and Rosenzweig’s
take advantage of the new opportunities that explanation for India’s low mobility is
became available with economic liberalization based on a combination of well-functioning
in the 1990s. This is reflected not only in their (caste-based) rural insurance networks and
schooling choices, which will presumably map the absence of formal insurance, which
into their labor market outcomes, but also in
their marriage decisions; specifically, whether
they married within their caste. Figure 3 33 Similar results are obtained by Banerjee et al. (2013),
reports the marriage choices of female family who examine the marriage advertisements that were
placed in an English-language newspaper in Calcutta and
members of the students who were sampled the matches that subsequently took place. They find that
by Munshi and Rosenzweig. The marriage 72 percent of the prospective grooms and 68 percent of the
choices are reported in five-year intervals over prospective brides who had married after a year found a
partner within their caste. These statistics are substantially
a time period (1970–2000) that substantially lower than the standard figure of 95 percent obtained from
overlaps with figure 1, which documented nationally representative surveys.
14
12
10
0
1970–75 1975–79 1980–85 1985–90 1990–95 1995–2000
Figure 3. Marriage outside the Caste in Urban India
Note: Percentage of marriages by women in the student’s family (sisters and mothers) outside their caste.
Source: Survey of school children and their families in Bombay, 2001.
includes government safety nets and pri- the size and scope of the insurance network
vate credit. Having empirically validated this is reduced by the exit of the wealthy house-
explanation and having estimated the model holds, making it function less effectively and
of insurance and migration, they proceed increasing the c onsumption variance of the
to implement a counterfactual experiment households that remain. Second, the level of
in which private credit is made available to redistribution will decline as a way of getting
wealthy households (with the resources to put some wealthy households to stay in the net-
up collateral). We see that these households work, reducing the consumption level of the
subsidize less wealthy members of their caste poor households. A well-intentioned policy
and thus are more likely to leave in the cur- that succeeded in its objective of increas-
rent equilibrium. While the absence of alter- ing migration could thus have unintended
native consumption-smoothing mechanisms distributional consequences that would go
keeps their migration currently at a low unnoticed if researchers evaluating the pro-
level, this level increases dramatically in the gram and the policy makers who designed it
counterfactual experiment when they get
were unaware of the underlying links within
access to credit and are no longer reliant on the caste. There is an increasing awareness
the rural insurance network. that spillovers must be accounted for when
The increased migration by wealthy designing and evaluating development
households leaves less-wealthy households programs. In the Indian context, these spill-
in their caste worse off in two ways. First, overs will often extend beyond the village to
cover the entire caste, spread over a wide 14–27.

area. This fact seems to have gone unnoticed Banerjee, Abhijit, Esther Duflo, Maitreesh Ghatak, and
by development economists evaluating gov- Jeanne Lafortune. 2013. “Marry for What? Caste and
Mate Selection in Modern India.” American Eco-
ernment programs and even explicitly ana- nomic Journal: Microeconomics 5 (2): 33–72.
lyzing networks in India. Banerjee, Abhijit, and Kaivan Munshi. 2004. “How
Efficiently Is Capital Allocated? Evidence from the
References Knitted Garment Industry in Tirupur.” Review of
Economic Studies 71 (1): 19–42.
Akerlof, George. 1976. “The Economics of Caste and Banerjee, Abhijit, and Rohini Pande. 2009. “Parochial
of the Rat Race and Other Woeful Tales.” Quarterly Politics: Ethnic Preferences and Politician Corrup-
Journal of Economics 90 (4): 599–617. tion.” https://epod.cid.harvard.edu/sites/default/
Alexander, K. C. 1989. “Caste Mobilization and Class files/2018-02/parochial_politics-_ethnic_prefer-
Consciousness: The Emergence of Agrarian Move- ences_and_politician_corruption.pdf.
ments in Kerala and Tamil Nadu.” In Dominance Banerjee, Abhijit, and Rohini Somanathan. 2007. “The
and State Power in Modern India: Decline of a Social Political Economy of Public Goods: Some Evidence
Order, Vol. 1, edited by Francine R. Frankel and M. from India.” Journal of Development Economics 82
S. A. Rao, 362–413. Oxford: Oxford University Press. (2): 287–314.
Ambrus, Attila, Arun G. Chandrasekhar, and Matt Banerjee, Biswajit, and J. B. Knight. 1985. “Caste Dis-
Elliot. 2014. “Social Investments, Informal Risk crimination in the Indian Urban Labour Market.”
Sharing, and Inequality.” Economic Research Initia- Journal of Development Economics 17 (3): 277–307.
tives at Duke Working Paper 179. Bardhan, Pranab, and Dilip Mookherjee. 2006. “Pro-
Ambrus, Attila, Wayne Y. Gao, and Pau Milán. 2017. poor Targeting and Accountability of Local Gov-
“Informal Risk Sharing with Local Information.” ernments in West Bengal.” Journal of Development
http://www.waynegao.com/uploads/8/1/4/6/81465138/ Economics 79 (2): 303–27.
agm170226.pdf. Bardhan, Pranab, and Dilip Mookherjee. 2012. “Polit-
Anderson, Siwan. 2011. “Caste as an Impediment to ical Clientelism and Capture: Theory and Evidence
Trade.” American Economic Journal: Applied 3 (1): from West Bengal, India.” http://thred.devecon.org/
239–63. papers/2012/2012-029_Bardhan_Political_Clientil-
Anderson, Siwan, and Patrick Francois. 2017. “Reser- ism_and.pdf.
vations and the Politics of Fear.” http://faculty.arts. Bardhan, Pranab K., Dilip Mookherjee, and Monica
ubc.ca/fpatrick/documents/reservations_Bread_ Parra Torrado. 2010. “Impact of Political Reserva-
may17.pdf. tions in West Bengal Local Governments on Anti-
Anderson, Siwan, Patrick Francois, and Ashok Kotwal. poverty Targeting.” Journal of Globalization and
2015. “Clientelism in Indian Villages.” American Development 1 (1): 1–36.
Economic Review 105 (6): 1780–816. Bayly, Susan. 2001. Caste, Society and Politics in India
Angelucci, Manuela, Giacomo De Giorgi, and from the Eighteenth Century to the Modern Age.
Imran Rasul. 2015. “Resource Pooling within Cambridge: Cambridge University Press.
Family Networks: Insurance and Investment.” Bertrand, Marianne, Rema Hanna, and Sendhil Mul-
http://citeseerx.ist.psu.edu/viewdoc/download?- lainathan. 2010. “Affirmative Action in Education:
doi=10.1.1.722.3002&rep=rep1&type=pdf. Evidence from Engineering College Admissions.”
Appadurai, Arjun. 1993. “Number in the Colonial Journal of Public Economics 94 (1–2): 16–29.
Imagination.” In Orientalism and the Postcolonial Bertrand, Marianne, and Sendhil Mullainathan. 2004.
Predicament: Perspectives on South Asia, edited by “Are Emily and Greg More Employable than Laki-
Carol A. Breckenridge and Peter van der Veer, 314– sha and Jamal? A Field Experiment on Labor Market
39. Philadelphia: University of Pennsylvania Press. Discrimination.” American Economic Review 94 (4):
Bagde, Surendrakumar, Dennis Epple, and Low- 991–1013.
ell Taylor. 2016. “Does Affirmative Action Work? Besley, Timothy, Rohini Pande, Lupin Rahman,
Caste, Gender, College Quality, and Academic Suc- and Vijayendra Rao. 2004. “The Politics of Pub-
cess in India.” American Economic Review 106 (6): lic Good Provision: Evidence from Indian Local
1495–521. Governments.” Journal of the European Economic
Banerjee, Abhijit, Marianne Bertrand, Saugato Datta, Association 2 (2–3): 416–26.
and Sendhil Mullainathan. 2009. “Labor Market Dis- Besley, Timothy, Rohini Pande, and Vijayendra Rao.
crimination in Delhi: Evidence from a Field Exper- 2007. “Political Economy of Panchayats in South
iment.” Journal of Comparative Economics 37 (1): India.” Economic and Political Weekly 42 (8): 661–67.
Besley, Timothy, Rohini Pande, and Vijayendra Rao. Bombay. New York: Longmans, Green, and Co.
2012. “Just Rewards? Local Politics and Public Choy, James P. 2018. “Social Divisions with E
ndogenous
Resource Allocation in South India.” World Bank Hierarchy.” Economic Journal 128 (615): 2711–42.
Economic Review 26 (2): 191–216. Coleman, James S. 1988. “Social Capital in the Cre-
Bidner, Chris, and Mukesh Eswaran. 2015. “A Gen- ation of Human Capital.” American Journal of Sociol-
der-Based Theory of the Origin of the Caste System ogy 94: S95–S120.
of India.” Journal of Development Economics 114: Damodaran, Harish. 2008. India’s New Capitalists:
142–58. Caste, Business, and Industry in Modern India. New
Bisin, Alberto, and Thierry Verdier. 2000. “‘Beyond the York: Palgrave Macmillan.
Melting Pot’: Cultural Transmission, Marriage, and Dandekar, Hemalata C. 1986. Men to Bombay, Women
the Evolution of Ethnic and Religious Traits.” Quar- at Home: Urban Influence on Sugao Village, Deccan
terly Journal of Economics 115 (3): 955–88. Maharashtra, India, 1943–1982. Ann Arbor: Univer-
Brass, Paul R. 1994. The Politics of India Since Inde- sity of Michigan Press.
pendence. Cambridge: Cambridge University Press. Darity, William, Jr., Ashwini Deshpande, and Thomas
Burnett-Hurst, A. R. 1925. Labour and Housing in Weisskopf. 2011. “Who Is Eligible? Should Affirma-
Bombay: A Study in the Economic Conditions of tive Action be Group- or Class-Based?” American
the Wage-Earning Classes of Bombay. London: P. S. Journal of Economics and Sociology 70 (1): 238–68.
King and Son. Das Gupta, Ranajit. 1981. “Structure of the Labour
Caldwell, John C., P. H. Reddy, and Pat Caldwell. 1986. Market in Colonial India.” Economic and Political
“Periodic High Risk as a Cause of Fertility Decline in Weekly 16 (44–46): 1781–806.
a Changing Rural Environment: Survival Strategies De Haan, Arjan. 1997. “Unsettled Settlers: Migrant
in the 1980–1983 South Indian Drought.” Economic Workers and Industrial Capitalism in Calcutta.”
Development and Cultural Change 34 (4): 677–701. Modern Asian Studies 31 (4): 919–49.
Cassan, Guilhem. 2015. “Identity-Based Policies and de Zwart, Frank. 2000. “The Logic of Affirmative
Identity Manipulation: Evidence from Colonial Pun- Action: Caste, Class and Quotas in India.” Acta
jab.” American Economic Journal: Economic Policy Sociologica 43 (3): 235–49.
7 (4): 103–31. Deshpande, Ashwini, and Katherine Newman. 2007.
Cassan, Guilhem. 2019. “Affirmative Action, Education “Where the Path Leads: The Role of Caste in
and Gender: Evidence from India.” Journal of Devel- Post-University Employment Expectations.” Eco-
opment Economics 136: 51–70. nomic and Political Weekly 42 (41): 4133–40.
Cassan, Guilhem, and Lore Vandewalle. 2017. “Identi- Deshpande, Ashwini, and Thomas E. Weisskopf. 2011.
ties and Public Policies: Unintended Effects of Polit- “Does Affirmative Action Affect Productivity in
ical Reservations for Women in India.” Graduate the Indian Railways?” https://pdfs.semanticscholar.
Institute Geneva Working Paper HEIDWP18-2017. org/0358/870b8a73d388a6547bf6ae36e3ef01a2c692.
Chakravarti, Uma. 1993. “Conceptualising Brahmani- pdf.
cal Patriarchy in Early India: Gender, Caste, Class Dhillon, Amrita, Vegard Iversen, and Gaute Tors-
and State.” Economic and Political Weekly 28 (14): vik. 2013. “Employee Referral, Social Proximity
579–85. and Worker Discipline: Theory and Evidence from
Chandavarkar, Rajnarayan. 1994. The Origins of Indus- India.” Center for Economic Studies ifo Institute
trial Capitalism in India: Business Strategies and the Working Paper 4309.
Working Classes in Bombay, 1900–1940. Cambridge: Dirks, Nicholas B. 2001. Castes of Mind: Colonialism
Cambridge University Press. and the Making of Modern India. Princeton: Prince-
Chattopadhyay, Raghabendra, and Esther Duflo. 2004. ton University Press.
“Women as Policy Makers: Evidence from a Ran- Drèze, Jean, and Geeta Gandhi Kingdon. 2001. “School
domized Policy Experiment in India.” Econometrica Participation in Rural India.” Review of Development
72 (5): 1409–43. Economics 5 (1): 1–24.
Chhibber, Pradeep K. 2002. Democracy without Asso- Dumont, Louis. 1970. Homo Hierarchicus: The Caste
ciations: Transformation of the Party System and System and Its Implications. Chicago: University of
Social Cleavages in India. Ann Arbor: University of Chicago Press.
Michigan Press. Dunning, Thad, and Janhavi Nilekani. 2013. “Ethnic
Chin, Aimee, and Nishith Prakash. 2011. “The Quotas and Political Mobilization: Caste, Parties, and
Redistributive Effects of Political Reservation for Distribution in Indian Village Councils.” American
Minorities: Evidence from India.” Journal of Devel- Political Science Review 107 (1): 35–56.
opment Economics 96 (2): 265–77. Engelshoven, Mirandu. 2002. “Rural to Urban Migra-
Cholia, Rasiklal Popatlal. 1941. Dock Labourers in tion and the Significance of Caste.” In Development
and Deprivation in Gujarat, edited by Ghanshyam Hnatkovska, Viktoria, Amartya Lahiri, and Sour-
Shah, Mario Rutten, and Hein Streefkerk. Thousand abh B. Paul. 2013. “Breaking the Caste Barrier:
Oaks: Sage Publications. Intergenerational Mobility in India.” Journal of

Fafchamps, Marcel, and Susan Lund. 2003. “Risk-Shar- Human Resources 48 (2): 435–73.
ing Networks in Rural Philippines.” Journal of Devel- Hoff, Karla, and Priyanka Pandey. 2006. “Discrimi-
opment Economics 71 (2): 261–87. nation, Social Identity, and Durable Inequalities.”
Ferraz, Claudio, and Frederico Finan. 2011. “Motivat- American Economic Review 96 (2): 206–11.
ing Politicians: The Impacts of Monetary Incentives Jodhka, Surinder S., and Katherine Newman. 2007. “In
on Quality and Performance.” https://eml.berkeley. the Name of Globalisation: Meritocracy, Productivity
edu/~ffinan/Finan_MPoliticians.pdf. and the Hidden Language of Caste.” Economic and
Field, Erica, Seema Jayachandran, and Rohini Pande. Political Weekly 42 (41): 4125–32.
2010. “Do Traditional Institutions Constrain Female Kapadia, Karin. 1995. Siva and Her Sisters: Gender,
Entrepreneurship? A Field Experiment on Business Caste, and Class in Rural South India. Boulder:
Training in India.” American Economic Review 100 Westview Press.
(2): 125–29. Kijima, Yoko. 2006. “Caste and Tribe Inequality: Evi-
Frisancho Robles, Verónica C., and Kala Krishna. 2012. dence from India, 1983–1999.” Economic Develop-
“Affirmative Action in Higher Education in India: ment and Cultural Change 54 (2): 369–404.
Targeting, Catch up, and Mismatch.” NBER Work- Kingdon, Geeta Gandhi. 1998. “Education of Females
ing Paper 17727. in India: Determinants and Consequences: A Case
Gadgil, D. R. 1959. Origins of the Modern Indian Busi- Study of Urban Uttar Pradesh.” Unpublished.
ness Class. New York: Institute of Pacific Relations. Kochar, Anjini. 2008. “Can Schooling Policies Affect
Gans, Herbert J. 1962. The Urban Villagers: Group Schooling Inequality? An Empirical Evaluation
and Class in the Life of Italian-Americans. New York: of School Location Policies in India.” India Policy
Free Press of Glencoe. Forum 4 (1): 53–99.
Ghurye, Govind S. 1932. Caste and Race in India. Kochar, Anjini. 2010. “Affirmative Action through Quo-
London: Routledge. tas: The Effect on Learning in India.” Stanford Cen-
Gille, Véronique. 2016. “The Role of Local Politics and ter for International Development Working Paper
Caste Networks in the Take-up of Affirmative Action 426.
in India.” Unpublished. Kohli, Atul. 1990. Democracy and Discontent: India’s
Gokhale, R. G. 1957. The Bombay Cotton Mill Worker. Growing Crisis of Governability. Cambridge: Cam-
Bombay Millowners’ Association. bridge University Press.
Gore, M. S. 1970. Immigrants and Neighbourhoods: Kooiman, Dick. 1989. Conversion and Social Equality
Two Aspects of Life in a Metropolitan City. Bombay: in India: The London Missionary Society in South
Tata Institute of Social Sciences. Travancore in the 19th Century. Delhi: Manohar
Greif, Avner. 1993. “Contract Enforceability and Eco- Press.
nomic Institutions in Early Trade: The Maghribi Kornblum, William. 1974. Blue Collar Community.
Traders’ Coalition.” American Economic Review 83 Chicago and London: University of Chicago Press.
(3): 525–48. Kothari, Rajni. 1970. Politics in India. Boston: Little,
Grimard, Franque. 1997. “Household Consumption Brown, and Company.
Smoothing through Ethnic Ties: Evidence from Cote Lamb, Helen B. 1955. “The Indian Business Commu-
d’Ivoire.” Journal of Development Economics 53 (2): nities and the Evolution of an Industrialist Class.”
319–422. Pacific Affairs 28 (2): 101–16.
Gupta, Bishnupriya. 2014. “Discrimination or Social Ligon, Ethan, Jonathan P. Thomas, and Tim Wor-
Networks? Industrial Investment in Colonial India.” rall. 2002. “Informal Insurance Arrangements with
Journal of Economic History 74 (1): 141–68. Limited Commitment: Theory and Evidence from
Gupta, Dipankar. 2000. Interrogating Caste: Under- Village Economies.” Review of Economic Studies 69
standing Hierarchy and Difference in Indian Society. (1): 209–44.
New Delhi: Penguin Books. Luke, Nancy, and Kaivan Munshi. 2011. “Women as
Hanna, Rema N., and Leigh L. Linden. 2012. “Dis- Agents of Change: Female Income and Mobility in
crimination in Grading.” American Economic Jour- India.” Journal of Development Economics 94 (1):
nal: Economic Policy 4 (4): 146–68. 1–17.
Hnatkovska, Viktoria, Amartya Lahiri, and Sourabh Madheswaran, S., and Paul Attewell. 2007. “Caste
Paul. 2012. “Castes and Labor Mobility.” Ameri- Discrimination in the Indian Urban Labour Market:
can Economic Journal: Applied Economics 4 (2): Evidence from the National Sample Survey.” Eco-
274–307. nomic and Political Weekly 42 (41): 4146–53.
Mazzocco, Maurizio, and Shiv Saini. 2012. “Testing India.” American Economic Review 93 (4): 1132–51.
Efficient Risk Sharing with Heterogeneous Risk Rawal, Shenila, and Geeta Kingdon. 2010. “Akin to
Preferences.” American Economic Review 102 (1): My Teacher: Does Caste, Religious, or Gender Dis-
428–68. tance between Student and Teacher Matter? Some
Mehta, Makrand. 2001. “Gujarati Business Commu- Evidence from India.” UCL Institute of Education
nities in East African Diaspora: Major Historical Department of Quantitative Social Science Working
Trends.” Economic and Political Weekly 36 (20): Paper 10-18.
1738–47. Rees, Albert. 1966. “Information Networks in Labor
Min, Pyong Gap. 2001. “Changes in Korean Immi- Markets.” American Economic Review 56 (1–2):
grants’ Gender Role and Social Status, and Their 559–66.
Marital Conflicts.” Sociological Forum 16 (2): Richman, Barak D. 2006. “How Community Institu-
301–20. tions Create Economic Advantage: Jewish Diamond
Mobarak, A. Mushfiq, and Mark Rosenzweig. 2012. Merchants in New York.” Law and Social Inquiry 31
“Selling Formal Insurance to the Informally (2): 383–420.
Insured.” https://web.stanford.edu/group/SITE/ Rudner, David West. 1994. Caste and Capitalism in
archive/SITE_2012/2012_segment_1/2012_SITE_ Colonial India: The Nattukottai Chettiars. Berkeley:
Segment_1_papers/rosenzweig.pdf. University of California Press.
Moorjani, Priya, et al. 2013. “Genetic Evidence for Rudolph, Lloyd I., and Susanne Hoeber Rudolph.
Recent Population Mixture in India.” American Jour- 1960. “The Political Role of India’s Caste Associa-
nal of Human Genetics 93 (3): 422–38. tions.” Pacific Affairs 33 (1): 5–22.
Morris, Morris David. 1965. The Emergence of an Schuler, Sidney Ruth, Syed M. Hashemi, Ann P. Riley,
Industrial Labor Force in India: A Study of the Bom- and Shireen Akhter. 1996. “Credit Programs, Patri-
bay Cotton Mills, 1854–1947. Berkeley: University of archy and Men’s Violence against Women in Rural
California Press. Bangladesh.” Social Science and Medicine 43 (12):
Morten, Melanie. 2019. “Temporary Migration and 1729–42.
Endogenous Risk Sharing in Village India.” Journal Singh, K. S, ed. 1992. An Introduction. Vol. 36, People
of Political Economy 127 (1): 1–46. of India. Calcutta: Anthropological Survey of India.
Munshi, Kaivan. 2011. “Strength in Numbers: Net- Srinivas, Mysore N. 1951. “The Social Structure of a
works as a Solution to Occupational Traps.” Review Mysore Village.” Economic Weekly. October 30,
of Economic Studies 78 (3): 1069–101. 1051–56.
Munshi, Kaivan, and Mark Rosenzweig. 2006. “Tradi- Stokes, Susan C. 2009. “Political Clientelism.” In
tional Institutions Meet the Modern World: Caste, Oxford Handbook of Comparative Politics, edited
Gender, and Schooling Choice in a Globalizing Econ- by Carles Boix and Susan C. Stokes. Oxford: Oxford
omy.” American Economic Review 96 (4): 1225–52. University Press.
Munshi, Kaivan, and Mark Rosenzweig. 2015. “Insid- Swaminathan, Padmini, and J. Jeyaranjan. 1994. “The
ers and Outsiders: Local Ethnic Politics and Public Knitwear Cluster in Tiruppur: An Indian Industrial
Goods Provision.” NBER Working Paper 21720. District in the Making?” Unpublished.
Munshi, Kaivan, and Mark Rosenzweig. 2016. “Net- Tabellini, Guido. 2008. “The Scope of Cooperation:
works and Misallocation: Insurance, Migration, and Values and Incentives.” Quarterly Journal of Eco-
the Rural-Urban Wage Gap.” American Economic nomics 123 (3): 905–50.
Review 106 (1): 46–98. Thorat, Sukhadeo, and Paul Attewell. 2007. “The Leg-
Munshi, Kaivan, and Mark Rosenzweig. 2018. “Eth- acy of Social Exclusion: A Correspondence Study of
nic Diversity and the Under-supply of Local Public Job Discrimination in India.” Economic and Political
Goods.” https://www.histecon.magd.cam.ac.uk/km/ Weekly 42 (41): 4141–45.
panchayat79.pdf. Timberg, Thomas A. 1978. The Marwaris: From Trad-
Nafziger, E. Wayne. 1978. “Class, Caste, and ers to Industrialists. New Delhi: Vikas.
Entrepreneurship: A Study of Indian Industrialists.” Townsend, Robert M. 1994. “Risk and Insurance in Vil-
Honolulu: University Press of Hawaii. lage India.” Econometrica 62 (3): 171–84.
Padró i Miquel, Gerard. 2007. “The Control of Poli- Weber, Max. 1958. The Protestant Ethic and the Spirit
ticians in Divided Societies: The Politics of Fear.” of Capitalism. New York: Charles Scribner’s Sons.
Review of Economic Studies 74 (4): 1259–74. Yadav, Yogendra. 1999. “Electoral Politics in the Time
Pande, Rohini. 2003. “Can Mandated Political of Change: India’s Third Electoral System, 1989–
Representation Increase Policy Influence for Dis- 99.” Economic and Political Weekly 34 (34–35):
advantaged Minorities? Theory and Evidence from 2393–99.
This article has been cited by:
1. Matt Lowe. 2021. Types of Contact: A Field Experiment on Collaborative and Adversarial Caste
Integration. American Economic Review 111:6, 1807-1844. [Abstract] [View PDF article] [PDF with
links]

Caste and The Indian Economy: Kaivan Munshi

Uploaded by

Copyright:

Available Formats

Caste and The Indian Economy: Kaivan Munshi

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Caste and The Indian Economy: Kaivan Munshi

Uploaded by

Copyright:

Available Formats

Journal of Economic Literature 2019, 57(4), 781–834

Caste and the Indian Economy †

1. Introduction to, the way he is treated by his teachers,

C aste plays a role at every stage of an

­ aking India one of the most genetically

Figure 1. Marriage outside the Caste in Rural India

Panel A. Trust in neighbors

Panel C. Tolerance of neighbors following a different religion

continue to assume an equal i­ncome-sharing network provides a vital and irreplaceable

c­ ommunities and historically disadvantaged farming to business over the course of a

12 While Marwari and Palanpuri entrants are slighly less

​ω​ 2G ​​ ​r​G​​

­ reviously uninhabited hills and so labor had

his own caste. In recent years, a n ­ umber of be explained by differences in endowments

that outsiders were subsequently excluded c­haracteristics, through which ­ intergroup

an early study set in Chicago, Rees (1966)

Pande’s solution to this identification goods, covering the areas of education,

­ ersistence in human capital, the key vari-

Figure 3. Marriage outside the Caste in Urban India

cover the entire caste, spread over a wide 14–27.

You might also like

aking India one of the most genetically

continue to assume an equal income-sharing network provides a vital and irreplaceable

c ommunities and historically disadvantaged farming to business over the course of a

ω  2G   rG

reviously uninhabited hills and so labor had

his own caste. In recent years, a n umber of be explained by differences in endowments

that outsiders were subsequently excluded characteristics, through which intergroup

ersistence in human capital, the key vari-