51
From Hierarchy to Ethnicity: The Institutional Origins of Ranked Identities Alexander Lee April 9, 2015 Abstract Scholars of ethnic politics have posited that in some societies ascriptive iden- tities are ranked relative to each other, but the origins of ranking are poorly understood. This paper develops a theory of ranked identities, and of why the elites of subaltern groups sometimes choose to invest in improving their status within the value systems of dominant groups, rather than in creating their own. This choice is conditioned by the fact that ranked identities are advantageous within informal patronage networks, while unranked identities are advantageous in winning elections. These theories receive some support from a panel dataset of political mobilization efforts by colonial Indian caste groups. Landed groups with few government employees tend to emphasize the ranked aspects of caste identity, while large groups in areas with elected local institutions emphasize other aspects of intergroup difference. These findings demonstrate the historical contingency of the modern concept of the ethnic group. Word Count: 9272

From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

  • Upload
    others

  • View
    2

  • Download
    0

Embed Size (px)

Citation preview

Page 1: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

From Hierarchy to Ethnicity:

The Institutional Origins of Ranked Identities

Alexander Lee

April 9, 2015

Abstract

Scholars of ethnic politics have posited that in some societies ascriptive iden-

tities are ranked relative to each other, but the origins of ranking are poorly

understood. This paper develops a theory of ranked identities, and of why the

elites of subaltern groups sometimes choose to invest in improving their status

within the value systems of dominant groups, rather than in creating their own.

This choice is conditioned by the fact that ranked identities are advantageous

within informal patronage networks, while unranked identities are advantageous

in winning elections. These theories receive some support from a panel dataset of

political mobilization efforts by colonial Indian caste groups. Landed groups with

few government employees tend to emphasize the ranked aspects of caste identity,

while large groups in areas with elected local institutions emphasize other aspects

of intergroup difference. These findings demonstrate the historical contingency of

the modern concept of the ethnic group.

Word Count: 9272

Page 2: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

1 Introduction

Classic descriptions of ethnic politics have divided ethnic identity systems into two ba-

sic types: Ranked identity systems (in which groups create relationships of superiority

and subordination to each other) and Unranked identity systems (in which groups are

conceptually autonomous)(Horowitz 1985, Weber 1958.) This distinction is based on

the fact that some identity systems, such as the caste systems of India and the racial

systems of the new world, ethnic group relations are fundamentally structured around

social inequality. Although the differences among these types of identities are widely

acknowledged among political scientists, there has been little or no theorizing on their

functions or causes. In particular, we know very little of why many subaltern groups

participate enthusiastically in ranked systems defined by others, and why over the past

century unranked systems have gradually become the dominant paradigm for identity

politics.

This paper will develop a theory of why elites, and in particular the leaders of poor

groups, participate in ranked identity systems. The choice between these strategies

is determined by the nature of the political system. Ranked identity mobilization is

motivated by the desire to gain an advantageous position in patron-client relationships,

and is thus more common among groups linked to these networks and alienated from the

formal state. Unranked identity mobilization is motivated by the desire to link elites to

ordinary group members for political purposes, and is thus common among large groups

in areas where political power is distributed democratically or quasi-democratically.

This question is examined empirically in the context of colonial India. Indian castes

are usually given as examples of ranked identities, but within India there is dramatic

variation in the degree to which groups embrace the norms of ranking. This was par-

ticularly true in the colonial period, when both patrimonial and democratic features

were present within the political system. For this reason, though the colonial period

Page 3: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

saw a massive increase in the politicization of caste identities, not all of these mobiliza-

tion efforts had identical goals. Some castes emphasized their distinctiveness from other

groups (very much as the ethnic politics literature predicts,) while other groups chose to

emphasize their hierarchical superiority to other groups, a process which often led them

to adopt the values and identities of traditionally high-status caste groups. Among Indi-

ans, this later strategy is referred to as “Sanskritization,” while the unranked alternative

has attracted names such as “Anti-brahmanism” or “Mandalization.”

To probe the plausibility of the connection between group traits, the political climate

and groups’ embrace of ranking, I use a new dataset of caste petitions to the colonial

census authorities, where group embrace of ranked norms is measured by their propensity

to adopt upper castes group names. Among petitioning groups, hierarchical rhetoric

is dominant among landed groups and groups with few state employees—the groups

most exposed to patrimonial institutions. Unranked rhetoric is dominant among large

groups in areas with high levels of political participation—the groups which stand to

gain the most from competition based upon group numbers. These findings are robust to

controlling for some of the more obvious alternative predictors of ranked mobilization,

such as a group’s position within the caste hierarchy and the policies of the census

officials, though the limitations the dataset makes them suggestive rather than probative.

Even when focusing on areas with strong traditions of social hierarchy, the contem-

porary literature on ethnic politics has tended to deemphasize ranking in favor of making

larger causal statements applicable to all forms of ethnic politics (Chandra 2004:18-19).

These accounts have thus focused on explaining variation in the mobilization of modern,

bounded ethnic categories (Posner 2004, Chandra 2004, 2012) without specifying the

economic and political conditions that made such bounded identities possible or desir-

able. More narrowly, the causes of caste ranking are at the center of traditional South

Asia studies, particularly in sociology and anthropology. However, whether these au-

thors view the caste system as an unchanging attribute of Indian civilization (Dumont

2

Page 4: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

1980[1966]) or a consequence of the policies of the colonial state (Dirks 2002, Srinavas

1966,) they tend to view hierarchical identities as being uniformly distributed across

South Asia and do not examine the variation in the degree to which castes accept these

identities. The theory in this paper builds on classic accounts of the transition from

traditional to modern caste politics (Rudolph and Rudolph 1967), though it advances

on this literature by having an articulated theory of social ranking and its causes, and

in attempting to measure changes in its incidence. It also draws on an anthropological

literature arguing that the caste system has changed in nature during the 20th century,

becoming “substantialized” (Barnett 1977) or “cultureized” (Natrajan 2011) during the

20th century, by advancing an explicit theory of how and why this transition has oc-

curred.

Section Two develops and defines what ranked and unranked identities are. Section

Three will develop a theory of why social ranking is attractive in some cases. Section

Four will discuss the quantitative data, while Section Five discusses the results of the

analysis. Section Six concludes with a discussion of the broader importance in the form

political identities take. In the online appendix, Section One provides additional tables,

Section Two discusses two historical examples of identity development, Section Three

provides additional details on the construction of the dataset, and Section Four discusses

the results of a series of robustness checks.

2 Ranked vs. Unranked Identities

2.1 Class and Ranking

The most obvious aspect of ranked identities is that ranked ascriptive identities cor-

respond, at least in theory, to differences in social status, though not necessarily to

differences political and economic power. In Max Weber’s view:

3

Page 5: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

A ’status’ segregation grown into a ’caste’ differs in its structure from a mere’ethnic’ segregation: the caste structure transforms the horizontal and un-connected coexistences of ethnically segregated groups into a vertical socialsystem of super- and subordination... Ethnic coexistences condition a mu-tual repulsion and disdain but allow each ethnic community to consider itsown honor as the highest one; the caste structure brings about a social sub-ordination and an acknowledgment of ’more honor’ in favor of the privilegedcaste and status groups (Weber 1958:185.)

In practice, this means that in ideal-type ranked systems of identity subaltern groups

have no social elite, or a small dependent one. In ideal-type unranked systems, by

contrast, each ascriptive group has an elite that considers “its own honor as the highest

one.”

In practice, of course, socio-economic status and ethnicity will rarely be perfectly

aligned. By military skill, business acumen, or luck, individuals from lower status groups

will gain wealth and power. There are numerous examples of such divergence in ranked

societies: India has many cases of impoverished Brahmans and wealthy peasants, and

the racially divided societies of the 18th century americas had noticeable populations

of both poor whites and rich blacks. If a society sees relatively little social change (as

Weber claimed that India did) such anomalies will be rare. But in fact, both ranked

social systems in general and India in particular have produced numerous examples of

low status social groups whose members attained considerable power and wealth, even

in the pre-colonial period (Bayly 1999, Bayly 1983, Kolff 1990.)

If such “anomalies” are allowed to accumulate, the basis of ascriptive ranking might

well collapse, as the nouveau-riche groups become restive at the gap between their high

economic and political position and the low social status assigned them in the ranked

system. These groups will have both the ability and the incentive to adopt forms of

identity that reject ranking. However, under certain sets of social conditions these elites

will choose to assimilate into the hierarchical system and change their position within

it—choosing not to lead their old ascriptive category but to exit it. By behavioral

4

Page 6: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

changes, advantageous marriages, and political muscle flexing, previously low status

elites often managed to “become” high status within a few generations (Bayly 1999.)

Such advancement reinforced the general correspondence of category and power rather

than weakening it. This means that studies of ranked identities must focus on the

actions of newly rich elites, whose decision to emphasize social ranking ensures ranking’s

continued relevence.

This discussion of identity “systems” would seem to assume, as Weber, and Dumont

did, that there is unity within a society on the approach that these new elites will take.

If all rising elites choose to emulate traditional ones, or ignore them, it is reasonable

to confound the ideologies of groups with the idealogical structure of the society as

a whole. Arguably, Ancient India and 21st century American fit this description, as

ranked and unranked identities (respectively) were/are present among all major social

groups. However, as we shall see, their are numerous cases in which different groups

make different sorts of ideological claims, and where ranked and unranked social theories

coexist and where a holistic ranked “system” thus reflects an idealogical ideal rather than

a practical reality. This paper will focus on how groups choose idealogical strategies

within such mixed societies.

2.2 Division and Distinction

Ranked and unranked identities differ from each other not just in their relationship

to social status, but in how they conceive of themselves. A political identity has two

elements: What I call a theory of category (who possesses particular identities and who

does not) and what I call a theory of distinction (which types of identities are normatively

desirable and which are not.) Theories of category sort human beings into groups, while

theories of distinction sort groups into the desirable and the undesirable. While the term

theory would suggest a sophisticated intellectual apparatus, in practice these theories

tend to be inchoate attitudes or schemas. While all projects of identity mobilization try

5

Page 7: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

to make identity politically important, they differ significantly on the strategic emphasis

that they give to these two elements.

Unranked identities have well-developed theories of category but weak theories of

distinction. While they erect well-articulated boundaries around the group, they make

only weak claims about the relative status of group members or the relationship of

the group to other groups. The tendency of unranked mobilization is thus absolutist.

Having taken a particular trait or identity as important, it attempts to put this trait or

identity at the center of political relations. Those that possess the mobilized attribute

are considered to have something important in common. Those that do not possess

the mobilized attribute are considered as the other, defined less by their own good

or bad qualities than by their lack of membership in the in group. The distinction

between ingroup and outgroup individuals, and the preference for the former, is thus

fundamental for unranked identities. This distinction has become the basis for a long-

running literature in psychology and behavioral economics on the causes and dynamics

of ethnic prejudice, (see for instance Allport 1954, Brewer 1999, Chen and Li 2009).

Ranked mobilization is focused on creating a theory of distinction rather than a

theory of difference. It develops a theory of what traits (both ascriptive and non-

ascriptive) are normatively good, and uses this to award social value to both coethnics

and non-coethnics.1 Frequently this project also involves a theory of proper personal

interrelationships, in which lower status individuals are expected to defer to higher

status ones. Ranked mobilization thus creates a relatively subtle pattern of political

distinction, since behavioral differences among similar groups may be quite small. As

a consequence, in unranked systems differences in identification are concentrated at

the ethnic boundary line, while in hierarchical systems these differences are scattered

throughout the system. While the differences between individual subgroups may be very

small (and hence relatively permeable,) the overall differences in identification from one

1For the classic version this argument, see Dumont (1980).

6

Page 8: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

extreme of the hierarchy to the other may be very large.

By extension, while unranked systems offers only two types of political relationships

(coethnic and non-coethnic) ranked systems offer an almost infinite menu of intergroup

relationships based on varying degrees of superiority and inferiority. In colonial Latin

America, for instance, a white person would treat a light-skinned African or Indian far

better than a dark-skinned one. In the traditional Indian caste system as imagined

by Dumont, a Brahman would have a subtly different relationship to a Jat cultivator

than to a Chamar leather worker, and be far more willing to enter a political or social

relationship with the former, due to his membership in a “clean” caste.

2.3 Internal Relations

One notable aspect of unranked identities is that they have a tendency to elide internal

cultural differences within the mobilized group. This is the result of their fixation on

the inter-group boundary— any attempt to further divide the in group would render the

group boundary irrelevant. In practice, this within-group fraternite takes the form of

both persuading poor group members to abandon local identities and behaviors in favor

of the normative behaviors of the group elite (a la Eugene Weber’s French peasants)

and of persuading elites to identify themselves closely with the interests of their poorer

coethnics, which often involve the abandonment of attempts to link themselves with a

more cosmopolitan elite identity. The result, in successful unranked identity projects like

the 19th century European nationalisms, is a world of rhetorically homogenous groups,

marked by a single set of behaviors and attributes, which may come to be interpreted

as a distinct “culture” (Natrajan 2011).

Besides assigning differential social value to other identity categories, ranked iden-

tities tend to subdivide and assign differential value to individuals within their own

category. This is a natural consequence of assigning normative worth based on actual

behavior: Inevitably, some individuals will less closely approximate the behavioral ideal

7

Page 9: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

of the group than others. This leads to subtle but important divisions in hierarchical

status, and thus in political identity, between members of the same ethnic category. As

we shall see, one important element in the development of “modern” caste identities in

India was thus the deemphasis on previously salient intra-caste differences.

2.4 Intragroup Relations

While groups in making unranked identity claims may define themselves in opposition

to other groups, they tend to view themselves as “conceptually autonomous,” (Fearon

and Laitin 2000.) This means that group identity, legitimation, and boundaries based

on a common history, culture and set of myths rather than through their relationship

to other groups, and that they are capable of existing as a social whole without either

the deference of lower groups or the patronage of higher ones. In addition, conceptual

autonomy implies that groups do not rely on myths or ideologies produced by other

groups—at least in theory, their existence is legitimated by them themselves, rather than

a group of outsiders. The search for conceptual autonomy means that the development

of hierarchical identities is closely associated with the rejection of external ideologies and

social hierarchies, such as cosmopolitan imperialism for Gellner’s Ruritanian nationalists

and Sanskritic hierarchy for Indian caste activists.

By their nature, ranked identities require the existence of other groups to make them

meaningful. The Brahman’s superiority is established by the inferiority of the groups

around him, and would be meaningless in a society comprised entirely of Brahmans.

The ethnic cleansing or exclusion so characteristic of unranked identities is entirely alien

to the hierarchical mindset, which instead would simply try to maintain other groups

in an inferior status. By the same token, ranked identities are incapable of developing

the peer relations with other groups that are common among segmentary identities.

Societies in which ranked identities are common are thus characterized by deference and

resource transfers across groups, in contrast to the unstable and rivalry-prone relations

8

Page 10: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

of segmentary groups (Horowitz 1985:29-32.)

3 Theory

The key actors in this paper are the members of “low ranked” groups who receive socioe-

conomic gains—either political, economic or educational—that make them competitive

with the traditional elite and give them the resources to engage in political action.2

When confronted by a ranked system that confined them to an inferior ritual position,

these elites faced a variety of options. As Horowitz puts it:

Theoretically, a ranked system can move in one of four directions. Subor-

dinate groups can attempt to displace superordinate groups; they can aim

at abolition of ethnic divisions altogether; they can attempt to raise their

position in the ethnic hierarchy without denying the legitimacy of that hi-

erarchy; or they can move the system from ranked to unranked. (Horowitz

1985:34)

In colonial India, as in many other historical situations, the first two of these options

were foreclosed. The economic and military dominance of the upper castes (not to

mention the colonial state) made the violent revolution necessary to overthrow the ranked

system a remote possibility. Similarly, the entrenched position of ascriptive divisions, and

their importance in the political strategies and self-images of high status groups, made

the bilateral renunciation of these divisions a remote possibility, though one occasionally

2While these elites may have a formal or informal position of leadership within their

ascriptive category, they theory does not require that this be so, and in fact many of

the groupings that promote ascriptive identities are fairly mall and unrepresentative. In

related work, I discuss why these elites often act in groups, and how they are able to

overcome problems of freeriding.

9

Page 11: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

envisaged by idealistic reformers. Realistically therefore, low status elites faced two

options: Climbing the hierarchical ladder and emphasizing more separate, unranked,

conceptions of identity. This section will describe how they might choose between these

options.

3.1 Hypothesis One

H1: Groups linked to informal and patrimonial institutions should adopt ranked iden-

tities. Political power is distributed differently in different types of societies. In some

societies, often referred to as patrimonial, power is distributed informally through net-

works of patronage and traditional obligation with little concern for the needs and rights

of individuals (see Bendix 1977, Weber 1947.) While this distinction was developed to

explain the decline of European feudalism, the concept of the patrimonial society also

fits the highly personalized politics of pre-colonial India (Cohn 1962, Fox 1971.)3

In patrimonial society, political power can be earned in two complementary ways:

by gaining the favor of one’s superiors at higher levels of the political system, and by

building up a support base of clients at lower levels. In Cohn’s case, the Talukdars

(large landowners) of Benares must at the same time maintain good relations with both

the raja and the village-level elites from whom they are collecting taxes. The raja in

turn must manage his Talukdars while at the same time propitiating supra-regional

leaders like the Nawab of Oudh. In neither case do the interests of citizens enter into

political calculations: The stock in trade of a politically ambitious man is his contacts

and reputation among other members of the elite.

Patrimonial elites have a strong incentive to politicize identities that will ease their

political relationships. An obvious response would be to adopt the same identity as one’s

3Indeed, some later Weberians have argued that pre-colonial India was a “purer”

example of patrimonialism than European feudalism. See Bendix 1977.

10

Page 12: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

political superiors, thus benefiting from coethnic favoritism. However, this course is un-

common for two reasons. Firstly, elites in patrimonial societies must think downwards as

well as upwards, and the same identity that might be an advantage with one’s superiors

might be a disadvantage with one’s inferiors. The Raja of Benares, for instance, might

gain a political advantage in his relations with the Nawab by adopting (Shia) Islam,

but such a move would create a major division between him and his (Hindu) Talukdars.

Secondly, patrons do not necessarily wish to be on equal terms with their clients. On the

contrary, they wish to use identity to emphasize their own legitimacy and the obligations

owed them by their clients4

Ranked identities fill this need. Instead of emphasizing groups’ divisions (which might

allow some inferiors to become either too familiar or too distant), ranked mobilization

emphasizes subtle distances in individual status. These differences define the patterns

of obligation in patron-client relationships and enable an ambitious politician to be at

once the humble supplicant and the gracious benefactor. A patrimonial politician with a

high status thus receives a helpful degree of legitimation in his dealings with his clients.5

For this reason, pre-colonial Indian states were deeply concerned with controlling the

hierarchical status of their subjects (Dirks 1993.) 6

Like many Weberian concepts, “patrimonial society” is an ideal type, and few soci-

eties fit it fully. In practice, most societies, particularly colonial ones, contain areas and

4On such relations in contemporary India see Chakravarti, 2001, and Shah et. al.

2006.5On the role of visible markers of social status in determining power in interpersonal

interactions see Bourdieu 1984.6It is important to note that members of the highest status groups have no reason

to mobilize to gain hierarchical status, since they have it ex ante. They will concentrate

on frustrating the attempts of other groups to gain status, which would devalue their

social position.

11

Page 13: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

groups that are primarily influenced by patrimonial institutions, while other areas and

groups are more strongly influenced by the formal state. These groups are also likely

to have a relatively traditional political culture, and a more limited set of ideological

alternatives. Within such mixed societies, we should expect groups whose elites have

substantial patrimonial ties to find hierarchical mobilization relatively attractive, and

the reverse to be true of groups whose elites are closely linked to the state and thus tend

to have a more formalized relationship to political authority. When landed or weakly

governed groups are more likely to direct their aspirations into winning a better posi-

tion in the patrimonial system than groups with stronger links to the formal political

economy.

3.2 Hypothesis Two

H2: Within areas with participatory state institutions, large groups should be more

likely to mobilize along unranked lines. Over the course of the 19th and 20th centuries,

patrimonial societies were gradually replaced by more modern, rationalized political

systems. In such systems, patrimonial politics has broken down and one’s political

position depends on one’s ability to gain the support of a majority of the population

(Bendix 1977.) In many such societies, offices are apportioned democratically, and

politicians must gain the support of voters.

If politics becomes a competition for mass support, individuals from traditionally

poor groups will be at a disadvantage relative to their more established peers, due

to their lower levels of money, contacts and political skills. One way in which they

can overcome these disadvantages is to attempt to activate politically social identities

that they share with a large segment of the population and argue that this identity is

sharply different from others. To the extent that the identity is perceived as relevant,

politicians from within the group will gain an advantage in competing with outsiders,

either because constituents expect that co-ethnics will favor them in the distribution of

12

Page 14: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

resources or because voters gain some more abstract benefit from seeing a co-ethnic in

power. By dividing some voters from others and tying them to particular elites, unranked

mobilization thus gives politicians a tool for creating constituency groups to which they

have privileged access. Ranked identities, which emphasize internal differences rather

than internal solidarity, will be much less advantageous in such a setting, since they

emphasize cultural differences with the prospective constituents of new elite members,

and reinforce the status of the incumbent elite.7

These incentives to mobilize politically are shaped by the electoral system. Most

electoral systems imposes some sort of penalty on small groups,8 and in majoritarian

systems this disadvantage may be very substantial. Since, in this theory, the ultimate

goal of ethnic constituency building is to win elections, it will not be equally attractive

to all groups. Elites, being primarily motivated by the rewards of office, are unlikely to

cultivate a constituency if it is so small that even a very successful mobilization effort

would leave them in a weak position within the electorate as a whole (Posner 2004,

7Political mobilization efforts face the generic problem that people will free ride on the

efforts of a handful of activists, potentially making activists hard to recruit. However,

decisions about status-seeking made by individuals have implications for their relatives

and coethnics in an environment where the value of one’s personal connections is an

important measure of one’s political importance. Status gains by individuals reflect well

on people in their social or kinship network, and the gains of individuals are sometimes

devalued by the low ritual position of those around them. In the colonial era, the

development of new forms of information and publicity rendered these externalities more

marked, and made group-based claims thus became a more attractive tactic.8While the relative sizes of groups may vary considerably over time, this paper will

assume that they are costly to shift, the that a priori sizes are thus still a predictor of

group strategy. In the appendix, I show that population changes do not seem to be

driving the empirical results.

13

Page 15: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

2005.) The potential opportunities opened up by mass politics are thus only attractive

to groups that are large in size. Given that changing one’s ascriptive traits is difficult or

prohibitively costly, the leaders of small groups will prefer to work along hierarchical or

non-identity lines, since the creation of hard boundaries would divide them from most of

their potential supporters. The leaders of a group with half a percent of the population,

in other words, may find the electoral rewards of ethnic mobilization so small that they

will prefer to take their chances within a ranked system.

Unranked identity development should thus be more common among larger groups in

areas where the institutional context makes numbers politically valuable. Importantly,

neither of these factors in isolation should influence mobilization strategy. In a non-

democratic context, large subaltern groups are just as marginalized as small ones, while

even in a democratic system very small groups have no-realistic chance of political power.

4 Data

Colonial India provides an excellent place to test why some groups choose to opt in to

ranked identity systems, due to the wide knowledge of Sanskritic norms of ranking and

the considerable temporal and regional variation in levels of democratic and patrimonial

institutions. This paper will thus make use of a complete panel dataset of Indian caste

groups in the 1901-1931 period. The unit of analysis is the caste-province-year, with

each caste being counted as a separate observation within each province where it occurs.

4.1 Measuring Ethnic Rhetoric

Variation in the level of importance attached to social ranking is often difficult to measure

in practice, since much of the expression of ranking norms is cultural and takes place

outside formal state structures. The role of the census in colonial India provides an

interesting exception to this pattern. The census, which asked individuals to classify

14

Page 16: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

themselves by caste, had already led to a reification of caste lines, and encouraged

some groups to emphasize their separateness from others (Dirks 2002.) The role of

the census in caste mobilization became even larger after the 1901 census, when H.H.

Risley, the census director, decided to organize the census returns by hierarchical caste

status. While Risley’s goal was to generate data supporting his own theory of the racial

origins of caste, the classification change had substantial consequences: Many lower and

middle caste elites anxious to maximize their hierarchical status organized to petition

the census authorities for a new caste name, typically one that linked them to a “higher”

caste or to one of the three higher varnas of the sanskritic caste hierarchy. 9 Other caste

elites petitioned for a new name less to claim status than to distance themselves from

the ranking system entirely, by disassociating themselves from names that suggested

subordination.

Despite its refusal to recognize most new names, the census thus became a place for

the validation of caste rhetoric, much to the annoyance of census officials:

All subsequent census officers in India must have cursed the day when it

occurred to Sir Herbert Risley, no doubt in order to test his admirable theory

of the relative nasal index,10 to attempt to draw up a list of castes according

to their rank in society. He failed, but the results of his attempt are almost as

troublesome as if he had succeeded, for every census gives rise to a pestiferous

deluge of representations, accompanied by highly problematical histories,

asking for recognition of some alleged fact or hypothesis of which the census

as a department is not legally competent to judge. (Hutton 1932: 433)

9In Sanskrit texts, society is divided into four varnas: the Brahmins (priests,) the

Kshatriyas (warriors,) the Vaishyas (traders,) and the Shudras (farmers and craftsmen.)

Individual castes (jatis) often attempt to classify themselves within one of varnas.10Risley felt that measurements of facial features were strongly associated with non-

Indian racial origins, which in turn were strongly associated with high caste status.

15

Page 17: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

While the origins of its petitioning lay in colonial racial theories, it was sustained after

Risley left his position by a vibrant culture of caste activism. The colonial period was the

golden age of the caste sabhas, or associations, groups of usually activists who sought to

advance the material and symbolic interests of their castes. The caste sabhas were led by

wealthy men, many with careers in local and provincial politics, and for this reason the

incidence of caste activism tended to coincide with rises in group wealth. Their activities

varied widely: The provision of scholarships, the encouragement of “uplifting” behavioral

changes, the adjudication of intra-caste disputes, and the production of dubious caste

histories and genealogies were all common.

Petitioning the census of authorities might seem a useless addition to these strategies,

since given the dismissive attitude of colonial officials, the chance of a petition being

granted was extremely small (only 7.1% were granted, and only 4.4% if we exclude

Madras in 1931). The decisions made by the census officials appear not to have been

based on the social political power of the groups in question, or even the historical or

scholarly merit their claims, but rather whether or not including the name would cause

confusion between different castes (Blunt 1911: 323). Perhaps for this reason, outside of

1931 Madras, rates of granting are almost identical between ranked potions (4.7%) and

unranked petitions (4.3%).11

However, the aims of groups tended to be as much rhetorical as practical. Relative

to our own time, where citizens have numerous means of publicizing their opinions—

websites, demonstrations, press releases—the public self-expression of British Indians

was severely constrained, both legally and technologically. Petitions, drafted in public

meetings and presented by large deputations, served as a way to publicly register group

claims. Unlike the other rhetorical modes employed by Sabhas, like pamphlets, petitions

have several advantages for scholarly study: They were easy to produce, even for poor

groups (who submitted oral or handwritten petitions) and were preserved by the colonial

11 Excluding this census year does not effect the empirical results.

16

Page 18: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

state rather than lost or scattered. For this reason, I use them as my main source on

the types of claims made by groups, though it should be noted that in many cases they

were part of broader strategies of identity-based activism.

The need for publicity was one reason why petitioning was an attractive strategy for

promoters of both ranked and unranked identity projects. However another factor was

also in play: The presence of a system of hierarchical norms within the census system

(implicit or explicit), that seemed to challenge their claims. For groups emphasizing

ranking, census classifications would be an refutation of their claims to be part of or

close to a high status group, all the more damaging for its ostensible objectivity. For

leaders groups attempting to emphasize their separateness from the Sanskritic system,

classification in a subordinate position or under an insulting name was an equally public

slap in the face, and an endorsement of the subordination that they denied. Petitions

were the only way these elites had of confronting and challenging colonial attempts to

impose a fixed version of the traditional hierarchy.

4.2 Coding the Dependent Variable

To test hypotheses one and two, I divided petitions that were submitted into those

making ranked and those making unranked claims, based on the name claimed in the

petition, and associated each petition with a particular caste as defined by the census

authorities. In almost all cases, the object of the petition was a new caste name, though

often these names had implications for the placement of the caste within the census or

implied an association with another caste. I classified as ranked those petitions in which

the claimed name included either the name of a specific higher status caste (e.g. “Saini

Rajput”) or the name of one of the higher varnas, which therefore made a claim to

higher status within the caste system at the expense of asserting a completely separate

identity. This could include either claiming complete identity with the varna (asking

the census authorities to use the term “Brahman” instead of the existing caste name)

17

Page 19: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

or using the varna name as an adjective modifying the caste name (e.g. “Bhumihar

Brahman.”) By tying themselves to the traditional Sanskritic hierarchy, these petitions

emphasize improving group status in the ranked system.

The petitions coded as unranked are those that make no such claim to belong to

another caste or varna. As such, they are more heterogeneous than the ranked petitions.

Most commonly, they reject a previous name thought to be insulting or associated with

a traditional occupation considered degrading. Frequently, these new names also had

positive connotations: The Goalas, for instance, claimed the name Yadava, which implied

descent from Yadu, a mythical king. What they have in common, is that they reject the

traditional social position of the group without attempting to join or associate themselves

with the traditional elite.

This does not mean that the unranked petitions reject the hierarchical worldview

entirely. While a few contain robust denunciations of brahman domination, the ma-

jority are shot through with Sanskritic rhetoric, and boasts about caste “cleanliness.”

Many of these petitioners no doubt sought to improve their own caste’s position without

destroying the system as a whole. However, by altering the traditional caste hierarchy

to envisage the possibility of prestigious groups outside the three highest varnas, the

unranked petitions make a fundamental break with the traditional ranking system, and

one highly unlikely to be accepted by the upper castes.

4.3 Independent Variables: Patrimonial Institutions

Two key concepts discussed in the theory section were the degree to which political

power was exercised through formal bureaucratic means and the degree to which people

were actively involved in politics. Both these concepts are fairly abstract and difficult to

measure by quantitative means. Both, however, exhibited considerable variance within

colonial India. The colonial state was neither a patrimonial successor of the Mughal

Empire nor a modern nation state, but a peculiar hybrid. In isolated rural areas and in

18

Page 20: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

the indirectly ruled princely states, local elites collected taxes and administered justice

very much as their ancestors had done, while in the cities and towns of the directly

ruled areas there existed local elections and mass-based political movements every bit

as organized as those in contemporary Europe. This variance gives an opportunity for

studying political identity in a context where the “pre-modern” and “modern” were

not chronological opposites but existed in uneasy proximity. No other period in Indian

history exhibits an equal degree of variation in political participation and institutional

form.

To measure the relative exposure of castes to patrimonial and formal state institu-

tions, I use two measures drawn from the occupational section of the census. The first is

the percentage of males in a caste cultivating or owning land, which should be associated

with close links to patronage systems and weaker links to formal institutions. This is

in part a function of the spatial distribution of rural groups, since even strong states

have difficulty projecting themselves in rural areas. More generally, rural areas tend

to be affected more slowly by social changes than urban areas, and thus more likely to

preserve older patterns of patronage, political culture and obligation. Finally, control

over land in colonial India involved a cultivator in an intricate network of social obliga-

tions, both to the laborers who did the actual work and to the network of intermediaries

(bureaucratic, aristocratic, or village-level) who collected taxes and administered justice

in rural areas. The management of these obligations involved linking oneself to patrons

and the development of clients, and descriptive accounts of this period emphasizes the

importance of patron-client ties among rural Indians (Stokes 1978.)

To measure the direct exposure of the caste to formal state institutions, I use the

proportion of male workers employed in the army and police, taken from the census.

This was chosen rather than public employment as a whole because the majority of

“public employees” in colonial India were informally employed village servants. Since,

in traditional theories (eg. Weber 1947,) the spread of rationalized bureaucracies is

19

Page 21: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

associated with the decline of traditional political linkages, we should expect groups with

strong links to the bureaucracy to be the first to find patrimonialism irrelevant. Put

another way, in the context of a modern state groups with many bureaucrats possess a

much more viable alternative to traditional social networks than landed and uneducated

groups do.

4.4 Independent Variables: Participatory Institutions

Politics under the colonial state was always a limited business, as the most important

decisions were made by European bureaucrats. There was however, a small but expand-

ing level of autonomy granted to local electoral institutions and an expanding sphere

of protest against colonial rule. I will use two measures of participation, the spread

of democratic elections and the spread of the Congress party. Though the franchise in

colonial India was limited to property owners, and the nationalist movement confined to

even wealthier elites, both these changes helped create the beginnings of a mass political

awareness. To measure electoral institutions, I use the percentage of elected members

of district and subdistrict boards. These were local councils, primarily concerned with

sanitation, roadbuilding and public education. Originally, district board members had

been appointed by the district collector, who also chaired the meetings. Gradually,

however, the bodies expanded in size and began to include elected members, first as a

minority and then as a majority. Despite their modest powers, membership of these

bodies, each the only political body in an area with a population of several hundred

thousand, conferred considerable prestige, and many future national politicians served

at this level.

The timing and extent of the change from elected to appointed boards varied greatly

across provinces, and appears to have been driven in part by local-level demand. The

British were confident that granting elected seats would both diffuse nationalist demands

and coopt their leaders, and thus tended to expand elections in areas where there existed

20

Page 22: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

a politically aware Indian elite. Due to this complex causal pattern, the effect of electoral

institutions should be interpreted very broadly, as an indicator of participatory politics.

Data on the proportion of local elected officials was collected from the government’s

Report on the working of the district boards for each province-census year.

In addition to measuring electoral institutions, I use a more direct measure of political

activity, the province’s average per capita attendance at the annual sessions of the Indian

National Congress, collected from the congress’s annual reports. At this stage of Indian

history, the Congress was less a political party than a discussion forum for politically

aware Indians, and it admitted virtually any well-dressed individual who wanted to

attend. Despite its elite character, Congress attendance thus should be a reasonable

proxy for the number of people in a province committed enough to the nationalist

movement to invest a week of their time and a long train journey. In fact, the annual

attendance figures seem to tally relatively well with trends in political activity described

in the secondary literature. Bengal, for instance, sees a peak in attendance around 1911

(during the political turmoil surrounding the partition of the province).

The political changes that occurred at the province level were obviously correlated

with the wide variety of social and economic changes taking place in colonial India.

To account for such time-varying confounders, I include three control variables often

associated with the level of economic and social modernization: the provincial literacy

rate, the proportion of the population living in urban areas, and the proportion of the

population engaged in public administration, all taken from the census tables.

4.5 Control Variables: Ritual Status

We have good reason to expect that differences in a priori status should influence the

mobilization strategies of groups: As groups with a higher ritual status have a more

advantageous position within the existing hierarchy, they are more likely to desire its

preservation. Table One shows the rates of petitioning by caste categories, the details

21

Page 23: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

of whose coding are discussed in the appendix. Aside from the one Brahman petition,

ranked petitioning is universal among high caste groups, and less common (though still

predominant) among lower caste groups. The exception is in the “Lower Unclean Shu-

dra” category, which has higher levels of hierarchical petitioning than the untouchable

or “Upper Unclean Shudra” categories. This is probably a result of the small size of

many of the groups in this category which gave them little political incentive to mobilize

on an unranked basis.

Table 1: Number of Petitions by Caste Status

Caste Status Ranked Unranked Total HierarchicalNo Petition Petition Petition Percent

Untouchable/Dalit 200 12 20 245 62.5Lower Unclean Shudra 322 23 90 430 79.6Upper Unclean Shudra 143 21 47 204 69.1Intermediate/Dominant 101 4 26 131 86Upper/“Twice Born” 62 0 17 79 100Brahman 39 1 0 40 0Total 868 61 201 1130

The numbers in table represent caste-province-years. For details on the coding of thecaste categories, see the data appendix.

5 Analysis

The leaders of Indian castes faced primary choices in interacting with the census author-

ities: Whether to petition for a name change at all and whether or not to choose a new

name that emphasized social ranking. This decision structure is modeled as a sequential

logit model with the choice of ranking language nested within petitioning cases. This

means that Pr, the probability of a province-caste-year recording a ranked caste petition,

can be estimated as the product of two probabilities:

22

Page 24: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Pr = Pp ∗ Pr|p (1)

Where Pp is the probability of submitting a petition, and Pr|p is the probability

of submitting a ranked petition given that a petition is already submitted. In related

work, I examine the causes of identity mobilization in general, Pp, which appears to be

strongly related to group socio-economic status. This paper will focus instead on Pr|p,

the probability of ranked mobilization among mobilizing groups. This can be estimated

by using a nested logit model, with the choice of ranked or unranked petitioning being

nested within the choice to petition at all:

Pr|p =eYr∑j∈n e

Yr(2)

Where Yr are the independent and control variables of interest.12

It is important to note that this model is not estimating the probability the groups

will be involved in politics, or even the probability that their political involvement will

emphasize caste. Rather, it estimates the probability that already mobilized caste iden-

tities will emphasize ranking in their rhetoric. 13

All reported models control for the caste status of the caste at the time of petition.

The inclusion of these variable account for the possibility that the results might by

12Dropping cases of no petitioning and running a simple binary model of the incidence

of ranked produces substantially identical results.13Elites may be involved in politics even if their identity remains unmobilized—the

Indian National Congress, for instance, built a durable political coalition with signifi-

cant lower caste support without using explicit caste rhetoric or conceding significant

authority to lower caste elites. This paper focuses on the rhetoric of those activists who

remain outside such “vertical” coalitions (Rudolph and Rudolph 1967), and choose to

emphasize jati as a politically relevant identity.

23

Page 25: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

driven by lower incentives to social repositioning among higher castes, since groups are

compared to thus with similar ex ante social status. All the models also include fixed

effects for province and year, to account for the possibility of geographical or time-

varying confounds, such as the common hypothesis that hierarchical norms are stronger

in the North of India than in the South.

In Table A.2, we see the changes that occurred in petitioning over time. Overall, the

level of both ranked and unranked petitioning rose over time. This first order result is

largely the result of changes in Pp. Since the overall incidence of petitioning is related

to the socioeconomic status of groups, the gradual expansion of wealth and literacy in

early 20th century India led to steep increases in petitioning, even among poorer and

more rural groups. In the raw data, this effect swamps the changes in the incidence of

ranking traceable to the rise of cities and elected local institutions. To study the rhetoric

of petitioning independent of the changing incidence of petitioning, it is necessary to use

the sequential model. 14

5.1 Hypothesis One

The effect of caste landedness on ranked mobilization is examined in Table Two, which

includes the two independent variables of interest, group ownership of land and employ-

ment in the army and police. Landownership has a strong positive relationship with

ranked petitioning, supporting the idea that that ranking is more attractive to groups

enmeshed in rural-patron client relationships. The effect of landownership is fairly large

substantively: Holding all other variables at their means, increasing a caste’s percentage

of landowners from 10% to 20% would increase the probability that a submitted petition

is ranked from 46.7% to 57.1%.15

14Readers interested in predictors of Pp should refer to Table A.3, which reports the

results of the first equation of the sequential model.15Neither variable appears to affect the overall incidence of petitions.

24

Page 26: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Table 2: Traditional Insitutions and Social Ranking: Sequential Logistic Regression

(1) (2) (3) (4) (5) (6)VARIABLES Ranking Ranking Ranking Ranking Ranking Ranking

Caste Land 3.517*** 4.359*** 4.796*** 4.698**Cultivation Rate (1.186) (1.557) (1.546) (1.889)Caste Army/Police Rate -68.43 -76.73** -82.99** -52.18

(36.55) (33.78) (34.44) (40.37)Prop. Local Boards -2.782 -11.59Elected (3.378) (10.88)Prop. Local Boards -91.44** -219.2***Elected*Population (43.44) (77.56)Prov. Congress Delegates 188.3**

(87.87)Prov. Congress -3,136**Delegates*Population (1,243)Population as Proportion -15.04 -14.52 11.99 -49.90* 40.13

(14.24) (15.26) (30.05) (28.72) (36.84)Population 00000s 0.0244 0.0351 0.236* 0.352 0.621**

(0.0699) (0.0724) (0.126) (0.196) (0.263)Male Literacy Rate -10.51 -12.17 2.219 -0.611 -0.855

(7.926) (8.458) (7.935) (9.721) (11.93)Male Literacy Rate Sq. 32.34 33.70 10.63 6.512 19.43

(17.64) (19.78) (19.63) (23.79) (31.48)Caste Ag. Labor Rate -4.792

(2.770)Caste Traditional -0.743Occupation Rate (1.186)Constant -0.642 -4.383 0.0453 3.718 6.915 16.93

(0.873) (20.74) (21.22) (4.375) (8.091) (28.08)

Observations 412 390 390 587 401 300Province-Year Controls NO YES YES YES YES YESProvince FE YES YES YES YES YES YESYear FE YES YES YES YES YES YESCaste Status FE YES YES YES YES YES YES

Robust standard errors in parentheses*** p¡0.01, ** p¡0.05

These are pooled sequential regression models, with fixed effects for caste status, year andprovince. This table only reports the results of the second equation, where ranked petitioningis the dependent variable. The corresponding results of the first equation, with petitioningas the dependent variable, are reported as Table A-3. The province-year controls, proportionurban, proportion in public employment, provincial population, and the provincial literacyrate, are not reported for reasons of space.

25

Page 27: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

As expected, employment in the army and police has a negative effect on ranked

petitioning.Though the effect of police employment is statistically significant in most

models (and close to significance in the others), it is much smaller than the effect of

landedness, with a similar 10% increase decreasing the probability of ranked petitioning

by only four percentage points. These results seem to indicate that, consistent with

Hypothesis One, investment in traditional social institutions leads to an emphasis on

social ranking, while exposure to the more modern institutions of the colonial state

reduces this emphasis.

These effects remain substantial in Model Two, which includes control for group pop-

ulation and literacy, along with measures of provincial social change and fixed effects for

caste status. Squared literacy has a marginal positive effect on social ranking, indicating

that when very wealthy groups petition, they tend emphasize social hierarchy.

Model Three adds a set of measures of occupational structure of each caste, including

the percentage of caste members working as agricultural laborers and the percentage

practicing the group’s traditional occupation. Neither of these variables changes the

effect of land ownership and paramilitary employment.

Another interesting aspect of Model Three is that the percentage of group members

employed in the group’s traditional occupation is unrelated to their choice of ranked

ideological forms. This tells against a well-known argument in the literature that tradi-

tional caste arrangements are the product of economic exchange (Frietas 2010, Dubois

1806.) Since the degree to which groups are reliant on out-of-group economic exchange

to provide their material needs is unrelated to their emphasis on hierarchy, it seems that

either economic ties or economic intimidation can explain identity ranking among lower

castes.

Ranked petitioning is thus most common among landowning groups with weak links

to the state, while unranked petitioning is most common among land-poor groups with

strong links to the state, providing some support for the hypothesis that this variation

26

Page 28: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

is predicted by the distinction between patrimonial and formal politics. These effects

can be seen graphically in Figure One, which present kernel density plots for ranked

and unranked petitions relative to land ownership. Figure One shows that, relative to

unranked petitions, a substantially higher proportion of ranked petitions are made by

land-rich groups, and a correspondingly smaller proportion by land-poor groups. Since

only a small proportion of groups have a majority of landowners, the majority of ranked

petitions, like unranked ones, are made by groups with less than 50% landownership.

Figure 1: Kernel Density Plot of Ranked and Unranked Petitioning Groups by LandOwnership

0.5

11.

5 D

ensi

ty

0 .5 1 Landownership

UnrankedRanked

kernel = epanechnikov, bandwidth = 0.1150

Landownership by Type of Petition

5.2 Hypothesis Two

Hypothesis Two, by contrast, posits that unranked petitioning should occur for only a

particular kind of mobilization (large) under a particular set of circumstances (institu-

tions that value mass political participation.) In practice, this means that increasing

political participation should not affect mobilization among small castes (since they will

tend to be politically weak even when mobilized) and that increasing levels of popula-

tion should be unrelated to unranked mobilization in areas with low participation (since

27

Page 29: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

they cannot be brought to bear politically.) However, the interaction of the two variables

(corresponding to the effect of a large group population in areas with high participation)

should be significant and positive.

Model Four of Table Two adds the crucial variable of interest for Hypothesis Two,

the interaction between group population and local elections. While local elections has

a little direct effect on ranked petitioning, the interaction of elections with group size

is statistically significant and negative in Model Four. This indicates that when large

groups in areas with democratic elections enter politics, their involvement tends not

to emphasize social ranking. This effect is substantively massive: Holding all other

variables at their means, increasing the proportion of local board seats elected from 20%

to 50% (a typical increase for the 1920s) would cause the predicted probability that a

group with 10% of the population would emphasize ranking in their petition to decrease

from 75.4% to 24.9%.

Model Five of Table Two reports the results of the same basic model, with the per

capita number of Indian National Congress attendees substituted for the proportion of

local officials. Despite a dramatic drop in sample size (Congress attendance data is not

available for the princely states, or for the year 1931,) the interaction of population

with participation is positive and statistically significant. This supports the hypothesis

that group size and regional politicization affect ethnicity only in association with each

other. In Model Six this effect is shown to be robust to the inclusion of the occupational

variables from Model One, though the inclusion of the election variables causes the effect

of police employment to become statistically insignificant, though still negative.

Section Four of the online appendix examines the effects of a variety of potentially

confounding variables: The probability of granting petitions, different census classifica-

tion systems, individual movement between castes, the spread of the nationalist move-

ment, and the influence of the Arya Samaj. None of these variables seem to be driving

the results, and few have a direct impact on petition language at all.

28

Page 30: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

6 Conclusion

There is no single way for a social identity to be expressed politically. On the contrary, its

expression is highly dependent upon the local institutional context, and a group’s place

within that context. The decision to mobilize on a unranked basis seems dependent on

the electoral advantages that can be gained from such a move: This type of mobilization

strategy is common among groups with a large share of the population situated in areas

with participatory political institutions. The decision to invest in a ranked identity is

guided by a group’s position within more traditional power systems: Landed and non-

bureaucratic groups are likely to find this option more attractive then groups less tied

to the traditional patronage system.

This set of theories offers a fresh perspective on the politics of modern India, es-

tablishing the scope conditions for modern ideas of caste and caste politics. In the

pre-colonial period, low levels of literacy and patrimonial polities kept caste mobiliza-

tion at relatively modest levels and within hierarchical norms. The spread of literacy in

the colonial period, in the context of a state that was still pre-modern, led to an explo-

sion of ranked mobilization (“Sanskritization”) as groups attempted to use new wealth

to achieve a traditional set of social goals. In the late colonial and post-independence

periods, primary education became the norm and democratic elections became perva-

sive. The result was an expansion of unranked mobilization: Lower caste elites switched

from the cultivation of their social superiors to that of their coethnics, leading to the

activation of politicized caste groups with well-marked external barriers and diminishing

internal barriers. However, the persistence of patrimonial institutions in many parts of

rural India means that ranked identities have not died out entirely, and it is notable

that those areas of India (such as rural Bihar) with very weak state institutions also

tend to be the areas in which traditional conceptions of caste are most entrenched. The

persistence of the colonial relationship between hierarchical norms and state weakness

29

Page 31: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

in the contemporary world is a promising topic for future study.

The trend away from social ranking is mirrored in the politics of the developing

world more generally. As democratic norms have spread and patrimonial states have

declined, ranked notions of identity have become increasingly unusual, replaced by the

standardized norm of the differentiated, internally homogenous, self-legitimating ethnic

group. As it has done so, is has become less common for the elites of subaltern groups to

assimilate into groups above them. Instead, like the elites of India’s lower caste groups,

they have embraced their identity and sought to lead the groups they were born into.

This is a development with substantial, and poorly understood, implications for ethnic

conflict and cooperation. Future work on this topic may explore how the decline of

ranked identities has effected the incidence of ethnic conflict, the formation of ethnic

parties, and patterns of intermarriage and everyday social interaction.

30

Page 32: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Bibliography

Allport, G.W. The Nature of Prejudice Cambridge: Addison-Wesley. (1954)

Barnett, Steve. ”Identity choice and caste ideology in contemporary South India.” In

The new wind: Changing identities in South Asia, edited by Kenneth David (1977):

393-414.

Bayly, C. A. Rulers, Townsmen, and Bazaars : North Indian Society In the Age of

British Expansion, 1770-1870. Cambridge [Cambridgeshire]: Cambridge University

Press, 1983.

Bayly, Susan. Caste, Society and Politics in India: From the Eighteenth Century to the

Modern Age. Cambridge UP, Cambridge, 1999.

Bendix, Reinhard. Nation-building and Citizenship Berkeley: University of California

Press, 1977.

Blunt, Edward. 1912. Census of India. United Provinces. Vol. XVI Part I, Report.

Lucknow: Census of India.

Bourdieu, Pierre. 1984. Distinction Trans. Richard Nice. Cambridge: Harvard UP.

Brewer, M. B. The psychology of prejudice: Ingroup love and outgroup hate?. Journal

of social issues, 55(3), 429-444.(1999).

Cassan, Guilhem. “British law and caste identity manipulation in colonial India” Delhi

School of Economics Development conference. 2010.

Chakravarti, Anand. Social power and everyday class relations : agrarian transformation

in North Bihar. New Delhi ; Thousand Oaks, Calif. : Sage Publications, 2001.

Chandra, Kanchan. Why Ethnic Parties Succeed Cambridge: Cambridge UP, 2004.

Chen, Yan, and Sherry Xin Li. ”Group identity and social preferences.” The American

Economic Review (2009): 431-457.

Cohn, Bernard S. “The Initial British Impact on India: A Case Study of the Benares

Region.” The Journal of Asian Studies , Vol. 19, No. 4 (Aug., 1960), pp. 418-431.

31

Page 33: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Degler, Carl N. Neither black nor white: Slavery and race relations in Brazil and the

United States. Univ of Wisconsin Press, 1971.

Dirks, Nicholas B. The hollow crown: ethnohistory of an Indian kingdom. University of

Michigan Press, Ann Arbor, 2 edn. 1993.

Dirks, Nicholas. Castes of Mind Princeton University Press, 2002.

Dubois, Abbe J.A. , Hindu Manners, Customs, and Ceremonies, (trans. Henry Beuchamp),

Clarendon Press, Oxford, 1906.

Dumont, Louis. Homo hierarchicus Chicago : University of Chicago Press, c1980 [1966.]

Dunning, Thad. 2009. “The Salience of Ethnic Categories“Society for Political Method-

ology, conference paper,

Eifert, Benn, Edward Miguel, and Daniel N. Posner. “Political competition and ethnic

identification in Africa.” American Journal of Political Science 54.2 (2010): 494-

510.

Fearon, James D., and David D. Laitin. “Violence and the social construction of ethnic

identity.” International Organization 54.4 (2000): 845-877.

Fox, Richard Gabriel. Kin, Clan, Raja, and Rule: State-hinterland Relations In Prein-

dustrial India. Berkeley, University of California Press, 1971.

Freitas, Kripa,“The Indian Caste System as a Means of Contract Enforcement,” Stanford

Institute for Theoretical Economics seminar paper, 2010.

Freyre, Gilberto, and Samuel Putnam. ”Masters and the Slaves.” (1946).

Hardgrave, Robert L. The Nadars of Tamilnad: the Political Culture of a Community

In Change. Berkeley: University of California Press, 1969.

Horowitz, David. Ethnic Groups in Conflict. Cambridge: Cambridge UP, 1985.

Hutton, J.H. Census of India. Vol. I Part I, Report. Calcutta: Census, 1932.

Htun, Mala. “From “racial democracy” to affirmative action: changing state policy on

race in Brazil.” Latin American Research Review 39.1 (2004): 60-89.

32

Page 34: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Kolff, D. H. A. Naukar, Rajput, and Sepoy : the Ethnohistory of the Military Labour

Market of Hindustan, 1450-1850. Cambridge: Cambridge University Press, 1990.

Marx, Anthony W. Making Race and Nation. A Comparison of the United States, South

Africa, and Brazil. Cambridge: Cambridge University Press, 1998.

Natrajan, Balmurli. The culturalization of caste in India: identity and inequality in a

multicultural age. Vol. 47. Routledge, 2011.

Posner, Daniel. 2004. “The Political Salience of Cultural Difference” American Political

Science Review, 98(4): 529-45.

Posner, Daniel N. Institutions and Ethnic Politics In Africa. New York: Cambridge

University Press, 2005.

Rudolph, Lloyd. and Susanne Rudolph. The Modernity of Tradition Chicago, University

of Chicago Press [1967]

Shah, Ghanshyam, Harsh Mander, Sukhadeo Thorat, Satish Deshpande, Amita Baviskar,

G. Shah, H. Mander, S. Thorat, S. Deshpande, and A. Baviskar. Untouchability in

rural India. New Delhi ; Thousand Oaks, Calif. : Sage Publications, 2006.

Srinivas. M.N. Social Change in Modern India. Berkeley: University of California Press,

1966.

Stokes, Eric. The Peasant and the Raj : Studies In Agrarian Society and Peasant

Rebellion. In Colonial India. Cambridge: Cambridge University Press, 1978.

Telles, Edward Eric. Race in another America: the significance of skin color in Brazil.

Princeton University Press, 2004.

Templeman, Dennis. The Northern Nadars of Tamil Nadu : an Indian Caste In the

Process of Change. Delhi: Oxford University Press, 1996.

Weber, Max. The theory of social and economic organization. Tr. by A.M. Henderson

and Talcott Parsons. New York, Oxford University Press,c1947.

Weber, Max, and Hans Heinrich Gerth. From Max Weber : Essays In Sociology. New

York: Oxford University Press, 1958.

33

Page 35: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Online Appendix

1 Additional Tables

Table A.1: Summary statistics

Variable Mean Std. Dev. Min. Max. NProp. Local Boards Elected 0.525 0.228 0.057 0.933 828Prov. Congress Delegates 0.007 0.014 0 0.067 616Per Capita Land Revenue 23.207 9.047 7.83 40.7 823Population as Proportion 0.029 0.036 0 0.277 1428Population 000s 242.634 348.608 1.5 3226 1670Male Literacy Rate 0.121 0.162 0 0.936 1300Male Literacy Rate Sq. 0.041 0.097 0 0.877 1300Prov. Proportion Urban 0.06 0.025 0.018 0.134 1428Prov. Literacy Rate 0.057 0.034 0.023 0.174 1634Prov. Public Employment Rate 0.013 0.008 0.001 0.03 1428Prov. Population 000s 23705.766 17214.272 1108 68059 1648Prov. Agricultural Employment Rate 0.66 0.105 0.389 0.839 1428Land Cultivation Rate 0.381 0.269 0 0.991 703Caste Public Employment Rate 0.02 0.054 0 0.592 703Caste Traditional Occupation Rate 0.468 0.279 0 0.985 703Caste Industry Owners Rate 0.001 0.003 0 0.035 730Local Social Status 0.054 0.114 0 0.476 1441Change in Population 000s 0.213 1.026 0 29.885 1279Hierarchical Census Classification 0.699 0.459 0 1 1670Hierarchical Census Class.* Caste Status 3.906 2.796 0 9 1670

34

Page 36: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Table A.2: Disaggregated Types of Caste Petition by Year

1901 1911 1921 1931 TotalNo Petition 134 239 267 228 861Ranked PetitionsRanked Petition 17 33 50 65 165Rankedl Petition by Subgroup 5 10 2 3 19Unranked PetitionsUnranked Petition 2 9 9 17 37Unranked Petition by Subgroup 3 2 1 2 8Untouchable Caste Seeking Merger 0 0 2 10 12Other PetitionsMerged Untouchable Caste (“Adi-”) 0 0 1 3 4Conflicting Petitions 0 2 2 12 16Previously Granted Petition 0 0 0 1 1

Total 161 294 334 341 1130

35

Page 37: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Table A.3: Traditional Insitutions and Social Ranking: Sequential Logistic Regression

(1) (2) (3) (4) (5) (6)VARIABLES Petitioning Petitioning Petitioning Petitioning Petitioning Petitioning

Caste Army/Police Rate -14.27 -16.99 -10.64 -13.04(10.91) (17.91) (16.74) (37.45)

Caste Land -0.130 -0.330 -0.388 -1.019Cultivation Rate (0.638) (0.739) (0.751) (0.777)Prop. Local Boards 0.387 -0.611Elected (1.796) (3.281)Prop. Local Boards 9.450 3.980Elected*Population (19.20) (29.37)Prov. Congress Delegates -38.58

(32.09)Prov. Congress 392.4Delegates*Population (468.1)Population as Proportion 8.313 8.876 -5.369 5.832 3.295

(5.520) (5.499) (11.83) (12.25) (13.72)Population 00000s 0.0856* 0.0927** 0.104 0.0648 0.0950

(0.0479) (0.0450) (0.0737) (0.0712) (0.0853)Male Literacy Rate 12.56*** 13.85*** 10.81*** 10.74*** 20.01***

(4.223) (4.314) (3.119) (3.468) (6.901)Male Literacy Rate Sq. -24.93*** -26.47*** -16.60** -19.48** -52.04**

(9.089) (8.794) (6.713) (7.766) (20.29)Caste Ag. Labor Rate 1.232

(1.239)Caste Traditional 1.516***Occupation Rate (0.551)Constant -2.112*** 7.273 5.147 0.434 -1.656 1.747

(0.494) (5.734) (5.766) (2.440) (5.077) (5.576)

Observations 412 390 390 587 401 300Province-Year Controls NO YES YES YES YES YESProvince FE YES YES YES YES YES YESYear FE YES YES YES YES YES YESCaste Status FE YES YES YES YES YES YES

Robust standard errors in parentheses*** p¡0.01, ** p¡0.05, * p¡0.1

These are pooled sequential regression models, with fixed effects for caste status, year andprovince. This table only reports the results of the first equation, where petitioning is thedependent variable. The corresponding results of the second equation, with ranked petitioningas the dependent variable, are reported as Table Two. The province-year controls, proportionurban, proportion in public employment, provincial population, and the provincial literacyrate, are not reported for reasons of space.

36

Page 38: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

2 Historical Examples

2.1 From Ranked to Unranked: The Shanan/Nadars

The Shanans of Tamil Nadu were a large caste traditionally associated with tenant labor

and the harvesting of coconuts and the production of toddy liquor. The caste was not

considered a prestigious one in Tamil society, and Shanans shared with other lower castes

a set of humiliating markers of their low ritual status, most notably a ban on Shanan

women to cover their upper body. Despite the general poverty of the community, there

existed a literate Shanan elite associated with tax collection and small trade. There

even existed a wealthy subcaste, the Nadars (or Nandans), who owned whole villages

and adopted the manners of local aristocrats (Templeman 1996: 18-29.) This elite

grew in size in the 19th century, as the growth of the South Indian economy improved

the position of tenants and Christian missionaries expanded access to education among

traditionally poor groups (Hardgrave 1969: 43-55.)

In the late 19th century, the Shanan elite began to organize politically against its low

social status in the late 19th century. Two very notable aspects of this early mobilization

were its constant reference to traditional Hindu norms and its refusal to engage with poor

Shanans. In the minds of the Shanan elite, they were a class of warriors who had been

degraded by contact with toddy cultivation, and their primary goal was to assimilate

themselves to the practices of elite Hindus. The cultural changes they adopted included,

among others, vegetarianism, cremation, dowries, Sanskritic child names, the wearing of

the sacred thread and the hiring and feeding of Brahmans. To formalize these changes,

they petitioned to the census authorities to have Shanans listed as Kshatriyas, echoing

earlier attempts by wealthy Shanans to use the name of their subcaste instead of the

caste name overall (Hardgrave 1969: 105-136.) The activist’s main contact with poor

Nadars was adversarial, as their attempts (enforced by violence) to ban the sale of toddy

deprived many poorer Shanans of their livelihood (Hardgrave 1969: 136-140.)

37

Page 39: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

However, this Sanskritic and ranked concept of Shanan identity eventually became

obsolete, mostly through the efforts of an organization founded in 1910, the Nadar Ma-

hajana Sangam. The Sangam leadership was dedicated to the creation and social uplift

of a united Shanan community under the name of Nadar, which had formerly described

only the Shanan upper class. To this end, they gradually abandoned their pretentions

to Kshatriya status and demanded that the census bureau classify them as Nadar, a

demand that was granted in Madras in 1921 and Travancore in 1931. Much of the

Sangam’s program was designed to include the Nadar lower classes, setting up media-

tion boards to resolve inter-caste disputes and energetically lobbying for social welfare

measures to help the workers hurt by the toddy prohibition measures they had earlier fa-

vored. The organization used the wealth of the Nadar traders to set up a large network

of Nadar social service organizations, including elementary schools, scholarships, and

the Nadar Bank (Templeman 1996: 82-90.) But the Sangam’s most publicized activities

(and the reasons for its prominence in the secondary literature) were in the political

sphere, where it campaigned on behalf of the lower caste Justice Party and against a

Congress Party associated with the Brahman elite and opposition to caste-based affir-

mative action (Hardgrave 1969, Rudolph and Rudolph 1967.) With the support of their

poorer coethnics and the organizational muscle of the Sangam, many elite Nadars, such

as Sangam leader W.P.A. Soundrapandian, gained political office in the colonial period

and after.

The early efforts of Shanan elites thus emphasized many of the common features of

ranked identity systems, and the usual policy of lower status elites within such systems.

Instead of trying to become the leaders of their ethnic category, they sought to exit it

entirely and assimilate into the high status group. This involved an increased emphasis

on behavioral status distinctions, a deterioration of internal relations within the group,

and a wholesale adoption of external cultural legitimation. More modern Nadar elites, by

contrast, sought to lead an unranked “pillarized” group, a change which was associated

38

Page 40: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

with a harder drawing of group boundaries, a de-emphasis on internal social distinctions,

and a rejection of other groups as a cultural model.16

2.2 Racial Identities in the Americas

Both Brazil and the Southern US were shaped by plantation agriculture and black

slavery, and both thus developed into societies where skin color was an essential aspect

of social identity. These societies differed, however, in the way in which these identities

were expressed politically. Racial identity in Brazil developed in a context where national

political power was exercised in an unapologetically authoritarian manner, and the power

of local political bosses, usually wealthy planters, remained virtually unchecked well into

the 20th century. In these circumstances, racial identities in Brazil were recognizably

hierarchical. While there was an elite consensus on the social value of white skin (and

the set of “respectable” social behaviors associated with it,) society was not divided into

groups of whites and non-whites but organized along a spectrum, with large groups of

racially mixed individuals striving to emulate the values and behavior of the planter

16The gradual transition of the Shanan community from a ranked to an unranked

conception of identity, and the central role of the Sangam in this transition, is a major

example in Lloyd and Susanne Rudolph’s classic account of caste politics (Rudolph

and Rudolph 1967:36-49). However, the Rudolphs do not attempt to explain why this

change occurred, though they do note that it coincided with the gradual expansion of

democratic participation in Southern India. Rather, they use the Shanans as an example

of “horizontal mobilization” (mobilization by a group’s own leaders) which they oppose

to “vertical mobilization” (mobilization by external elites). While this distinction may

seem similar that between ranked and unranked identities, it is in fact quite distinct: As

the Shanan case shows, horizontal mobilization may embrace by ranked and unranked

strategies.

39

Page 41: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

class, the only source of political power (Degler 1971.) As in the Indian case, many

low-status individuals succeeded in improving their status through strategic changes

of behavior and association, and were able to enjoy improved social status due their

higher levels of whiteness, often through the use of intermediate social categories such

as “mulatto.” Ambitious black men usually sought to become mulatto, and mulattos

sought to be considered white, rather than attempting to be the elites of the lower

(black) status category (Marx 1998).17

In the United States, by contrast, political identities developed in the context of

a robustly democratic political culture, both at the local and state levels. African

Americans could thus not be counted out as potential competitors, particularly if they

were light skinned and free. The result was the development of one of the world’s must

rigidly bounded identity systems, with a very small amount of African ancestry necessary

to place a person in the African American racial category. This enabled white politicians

to appeal successfully to white voters against a perceived racial threat. Interestingly,

this segmentary tendency became even more developed in the aftermath of the Civil

War, which by formally enfranchising blacks made the system far more participatory,

and the need to exclude black elites more immediate. While the US census(and local

opinion) recognized the third ethnic category “mulatto” well into the 20th century,

the possibility of racial ambiguity gradually faded with the patrimonial power of the

17While race in Brazil has many characteristics of a ranked identity system, the dom-

inant way to characterize Brazil’s racial politics has been negative and comparative,

emphasizing the lack of entrenched racial division in Brazil relative to other parts of

the Americas (Marx 1998, Freyre and Putnam 1946), rather than examining the in-

ternal logic of the ranked system. Marx’s account also differs from my own in tracing

inter-American differences in identity to state policies, and the incentives of the ruling

coalitions shaping those policies. This emphasis makes it difficult to access the causes

in variation in identity strategies across groups exposed to similar state policies.

40

Page 42: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

planters. In the Brazilian case, the gradual expansion of political participation in the

late 20th century has been associated with some blacks becoming interested in unranked

ideas racial identity, and pressing for policies dependent on hard racial divisions, such

as race-based affirmative action(Telles 2004, Htun 2004.)

Racial status in colonial Brazil illustrated the main characteristics of a ranked sys-

tem. It was very strongly associated with social class, it emphasized subtle social distinc-

tions over hard and vast differences, it created elaborate internal differences within eth-

nic categories, and tended to lead to the subordinate group defining themselves through

the ideological tools of the super-ordinate group. In America, especially since the civil

war, racial divisions are not perfectly predictive of elite status, are relatively rigid, and

are attended by rhetorical pledges of racial solidarity and racial independence. As in

the Nadar case, this transition occurred at the same time as the growth of democratic

politics.

3 Data Appendix

3.1 Data Sources

The primary data used here are drawn from the published reports of the Census of India

for the years 1891, 1901, 1911, 1921, and 1931. Before 1891, the census did not cover

all of India, while after 1931 it did not include information about jati. Since petitioning

began in 1901 in response to Risley’s policy change, the 1891 caste-provinces are not

included in the main results.

For each census year, I used the individual tables volume for each province to collect

information on every caste with over .5 percent of a province’s population, using a

variety of tables within each volume. While population data is available for nearly every

caste-year, some provincial census superintendents collected occupational data on castes

that they considered important. For this reason, models that include the occupational

41

Page 43: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

variables lose a number of observations.

The data on petitioning was based on the report volume for each province year,

where the petitions was usually noted by the superintendent. Usually, only the caste

and the name claimed was recorded, though in some cases the superintendent also noted

the group making the claim, and some details on their justification.

While most petitions appear not to have been preserved, I was also able to review

some petitions (either from the Central Indian Agency or Bihar) in the National Archives

of India.18 The detail of the petitions varied widely, from handwritten notes to masses of

supporting documentation, some of which is discussed more extensively in related work.

3.2 The unit of analysis

The unit of analysis on is caste-province-year, but the “provinces” require some expla-

nation. Colonial India was divided into areas ruled by the British government, and

areas in which sovereignty was delegated to native princes, usually descendants of early

British allies. The directly rules areas were divided into fairly large provinces, of which

the most important were: The United Provinces, Bengal, Bihar and Orissa, Bombay,

Madras, Punjab, the Central Provinces and Berar, and Assam. Four very small provinces

(two of which are overwhelmingly Muslim border areas) have been ignored.

The indirectly ruled areas were divided into three main groups: Four large states

(Jammu and Kashmir, Hyderabad, Mysore and Baroda) that enjoyed a direct relation-

ship with the central government, two large groups of states (the Rajputana Agency

and the Central Indian Agency) which reported to a political agent appointed by the

central government, and a large number of states who reported to the political depart-

ment of the local British province. The policy followed here has been to report the large

states and agencies separately, as the equivalent of British provinces, and include the

18For examples, see: Home Department. Public Branch. F. 45/6/31-Pub., and Home

Department. Census Branch. A Progs. December 1900. Nos. 24-27.

42

Page 44: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

other states with the province that controlled them. The only exception to this was the

large princely state of Travancore, which published its own census report, and has been

treated as separate from Madras.

There were several changes to provincial boundaries during this period, so I have

consolidated and divided provinces to keep the unit of analysis consistent. For the

1891 and 1901 census years Berar has been treated as part of the Central Provinces

even though it was in fact independent, and for the same years, Bihar has been treated

separately even though it was in fact part of Bengal. In 1931 the recently created

Western Indian States Agency is treated as part of Bombay, and for 1911, 1921 and

1931 Gwalior state is treated as part of the Central Indian Agency.

I use the caste classifications used in the census tables at the 1891, which ignore

subcaste divisions. In a few cases where later census year clubbed castes together, I used

the relative proportions of the component groups in 1891 to separate these categories

into their constituent groups.

3.3 Independent Variables

The primary independent variables are drawn from the census tables for each year. For

many, (though not all) castes, we also have the male literacy rate of the caste, which the

census authorities calculated on the fairly generous basis that anyone capable of writing

his name in any language was literate. The square of the literacy rate is be used to

index very high levels of literacy. I also include controls the male population for each

caste, which I use to calculate their proportion of the provincial population and their

absolute male population. In some province years the literacy data was only reported

for areas in which the caste was especially prevalent. The reported results are robust to

the exclusion of these cases.

Per capita Congress attendance is calculated by taking the number of attendees

from a given province present at the Indian National Congress’s annual session in the

43

Page 45: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

year of the census and in the two years immediately before and after, excluding years in

which the province hosted the Congress, and dividing by the province’s population.

The proportion of elected officials within each province was calculated using the

annual Report on the working of the district boards for each census year. These reports,

like the annual reports of the congress, are rare and difficult to find outside of India and

the British Library.

3.4 Petitioning

The main procedures for coding petitions are explained in the text. However, some

special cases require additional explanation.

1. A complicated chain of events occurred in South India at the 1931 census. In

Madras, Mysore and Hyderabad, activists from untouchable groups demanded that their

groups be not counted separately, but combined under a common name (depending on

the region, Adi-Hindu, Adi-dravid, Adi-Andhra and Adi-Karnataka.) The Adi prefix

emphasized the claim that these groups were in fact the original inhabitants of South

India, and had been enslaved and degraded by later invaders. The census policy towards

these claims varied by state: In Hyderabad, the castes were still listed separately, but

in Mysore they were combined in the tables, while in Madras both the old and the new

names were given as options, which respondents were free to choose between. The policy

I followed in these cases was to treat all the old untouchable castes in these provinces

as having submitted petitions, while ignoring the new castes. The major results are

robust to the exclusion of these province-years. Madras in 1931 also contains the only

two cases of jatis not petitioning because their demands were fully granted in a previous

year. They were dropped from the analysis.

2. The only instances in which all cases in a province year were coded as zero were

when the superintendent specifically noted receiving no petitions. Other province-years,

which failed to mention petitions at all, are coded as missing.

44

Page 46: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

3. In 1931, some sabhas began mailing petitions for reclassification to every provin-

cial census officer in India, even in areas where they had no organizational presence.

These mailed petitions are also coded as a zero, unless there is some indication they

were supported by a local organization.

4. For this purpose I exclude castes that were below .5% of the provincial population,

non-Hindu castes and tribal groups. There are in fact two instances of petitions by tribal

groups, though the overall rate is very low and in one instance there is evidence that

they were encouraged to petition by an outside group of upper caste Hindus. Due to

marginal spatial position of these groups, and their complicated relationship with the

caste hierarchy, I have excluded them from the main analysis. Including them would

slightly strengthen the reported results.

5. A few petitions were submitted on behalf of sub-groups of a caste group recognized

by the census, requesting a name change for themselves rather than the caste as a whole.

These are treated as being submitted by the whole caste. Excluding these cases does

not affect the reported results.

3.5 Ritual Status

This variable is coded using the infamous “tables of precedence” compiled for the 1901

census and recorded in the census report of each province, supplemented with informa-

tion from the Castes and Tribes series for each province that were published in the same

period. In all but a few cases the classifications agree with those made by the post-

independence government for the purpose of affirmative action. The status variable is

constructed as an ordinal variable four to nine, with each jati being assigned to a cate-

gory. From highest to lowest, the categories are Brahmans, other clean twice born castes

(“upper castes,”) high status cultivating castes (“Intermediate castes,” “middle castes,”

“dominant castes”), low status cultivating castes (“Upper OBCs,” “unclean shudras”),

low status occupational castes (“lower OBCs,” “artisan castes”) and former untouch-

45

Page 47: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

ables (“dalits,” “harijans.”) While the terminology varies, this six-fold classification is

familiar to India scholars, as it lies at the heart of most previous work on caste politics

(Jaffrelot 2003, Jaffrelot and Kumar 2009, Frankel and Rao 1989) and is the format used

in most contemporary surveys. While there is much blurring at the edges, particularly

among the shudra categories, this scheme seems to capture certain important aspects of

status hierarchy in India.

4 Robustness Checks

The conclusions we can draw from the results in Table Two are limited by the size of

the dataset. It is important to note, however, that the inclusions of state and caste

status fixed effects and literacy controls accounts for some of the more obvious alterna-

tive hypotheses: That differences in the importance identity ranking can be traced to

some unobserved cultural aspect of areas, to positions within existing ranking systems,

or group socioeconomic status. Tables A-4 and A-5, explores a few more alternative

hypotheses, including changing census policies, Christian and Hindu missionary work,

and individual movement between groups, with Table A-4 examines the robustness of

the results of Hypothesis One, while Table A-5 examines the robustness of the results

of Hypothesis Two.

Model One of both tables shows that the results are not being driven by the proba-

bility of rejection or acceptance of petitions. The rate of petition granting has no effect

on the incidence of ranking, and have no effect on the main results. Model Two of both

Tables examines the effect of the classifications used in the census table. While in some

province-years the census authorities classified castes by ritual “precedence” in other

years castes were organized alphabetically. The use of ranked census norms might well

encourage the use of ranked language in caste petitions. However, while ranked cen-

sus classifications do seem to have a positive effect on ranked petitioning (particularly

46

Page 48: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

for higher castes) this effect does not reduce the effect of caste employment, or of the

interaction of caste population and elections.

Model Three of both tables tests the hypothesis that intergroup movement could

serve as a substitute for collective action. Instead of petitioning collectively to raise their

caste status, individuals could merely reclassify themselves as members of higher castes.

Such individual movement among castes is well attested in colonial India, and is readily

apparent in census figures (see Casson 2010.) Model Three includes a control for the

percentage change in caste membership since the last census. While caste population

changes do appear to be positively correlated with petitioning, their inclusion as a control

does not affect the effects of the key independent variables.

Are the differences in petitioning driven by differences in politicization and the

spread of the nationalist movement? Models Four of Table A.4 and Model Five on

Table A.5. Test this idea, by controlling for the per capita attendance at Indian National

Congress sessions. This variable has no effect on the effects of landedness and the size-

election interaction, but it does eliminate the effect of state employment, primarily by

halving the size of the sample.

It is also possible that the effects of a cultural movements in early 20th century

India that might affect the adoption of norms of social ranking: the Hindu revivalist

Arya Samaj. The Arya Samaj sought to recreate an idealized form of the original Hindu

caste hierarchy. Model Five of Table A.4 and Model Four of Table A.5 includes the

provincial levels of these variables, and their interactions with caste status. While the

Arya Samaj does appear to have an impact on petitioning behavior, it has no effect on

petition language. Also, the inclusion of these variables does not alter the effects of the

key independent variables.

One final possibility is that the adoption of ranking is especially common among the

so-called “martial races.” These were ethnic groups considered by the British especially

loyal and suitable for military service, and were preferential access to the armed forces

47

Page 49: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

and other government benefits. Given the sophisticated discourse of ethnic superiority

that developed around these groups, it is reasonable to think that their political identities

might develop differently than other groups, a major concern given that military and

police recruitment is one of my measures of state access. However, it is certain that

these groups are not driving the results, because no groups in this category are included

in the dataset. The majority of the major martial groups were either from outside

India (Gurkhas) or members of religious minorities (Sikhs, Baluchis, Punjabi Muslims,

Hindustani Muslims) and are thus are outside the sample frame. The only Indian Hindu

groups generally considered martial were the Dogras of Kashmir and Punjab, the Rajputs

of the Himalayan regions and Rajasthan, and the Marathas. Since none of these groups

even submitted a petition, they have no effect on the results in Table Two.19

19Removing these groups also has no effect on hypothesis testing about the causes of

petitioning.

48

Page 50: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Table A.4: Robustness Checks: Sequential Logistic Regression

(1) (2) (3) (4) (5)VARIABLES Ranking Ranking Ranking Ranking Ranking

Population as Proportion -28.99** -26.07 -22.31 -60.32*** -22.69(14.61) (14.10) (13.37) (19.69) (12.35)

Population 00000s 0.0680 0.0555 0.0586 0.161 0.00485(0.0701) (0.0679) (0.0658) (0.101) (0.0649)

Male Literacy Rate -13.55 -11.61 -10.82 5.917 -7.646(8.905) (9.061) (8.817) (12.17) (7.858)

Male Literacy Rate Sq. 41.61** 35.47 37.11 -27.70 26.08(20.43) (20.19) (20.40) (42.92) (17.68)

Caste Land Cultivation Rate 5.058*** 4.933*** 5.436*** 5.433*** 3.197***(1.409) (1.386) (1.651) (1.843) (1.199)

Caste Army/Police Rate -75.18** -72.95 -80.13** 141.3 -75.98**(37.77) (41.51) (37.77) (107.1) (36.08)

Petition Grant Rate -7.612(5.697)

Hierarchical Census -0.995Classification (3.887)Hierarchical Census 0.398Classification* Caste Status (0.654)Population Change 1.478

(1.734)Prov. Congress Del. -143.2**

(55.65)Prov. Prop. Arya Samaj -22.18

(359.6)Prov. Prop. Arya Samaj -15.78* Caste Status (69.68)Constant -4.023 23.89 9.211 -2.743 -2.593

(16.90) (35.82) (28.61) (2.440) (24.70)

Observations 365 391 389 208 391Province-Year Controls YES YES YES YES YESProvince FE YES YES YES YES YESYear FE YES YES YES YES YESCaste Status FE YES YES YES YES YES

Robust standard errors in parentheses*** p¡0.01, ** p¡0.05

These are pooled sequential regression models, with fixed effects for caste status, year andprovince. This table only reports the results of the second equation, where ranked petitioningis the dependent variable. The corresponding results of the first equation, with petitioningas the dependent variable, are not reported. The province-year controls, proportion urban,proportion in public employment, provincial population, and the provincial literacy rate, arenot reported for reasons of space.

49

Page 51: From Hierarchy to Ethnicity: The Institutional Origins of ... · Section Two develops and de nes what ranked and unranked identities are. Section Three will develop a theory of why

Table A.5: Robustness Checks: Sequential Logistic Regression

(1) (2) (3) (4) (5)VARIABLES Ranking Ranking Ranking Ranking Ranking

Population as Proportion 11.15 24.80 20.18 7.765 11.10(30.25) (30.70) (32.05) (25.96) (28.57)

Population 00000s 0.239 0.283** 0.235 0.250** 0.224**(0.126) (0.135) (0.137) (0.122) (0.103)

Male Literacy Rate 1.359 3.421 -0.359 -0.299 -0.742(8.454) (7.265) (8.303) (8.800) (9.615)

Male Literacy Rate Sq. 12.56 9.582 19.24 15.16 9.417(21.41) (16.01) (20.82) (22.81) (23.14)

Prop. Local Boards Elected -1.995 -1.675 -3.145 -2.944 -0.468(3.377) (3.336) (3.527) (3.438) (2.148)

Prop. Local Boards Elected*Population -91.52** -123.9** -101.7** -94.08** -100.1**(43.34) (48.47) (47.07) (42.31) (45.28)

Petition Grant Rate -2.774(3.338)

Hierarchical Census -5.188Classification (3.289)Hierarchical Census 1.188**Classification* Caste Status (0.592)Population Change -1.412

(1.588)Prov. Prop. Arya Samaj 506.5

(421.0)Prov. Prop. Arya Samaj* -77.85Caste Status (83.87)Prov. Congress Delegates 38.03

(28.84)Constant 0.871 2.705 1.998 4.790 3.715

(4.736) (4.696) (4.678) (4.683) (2.022)

Observations 561 587 556 587 401Province-Year Controls YES YES YES YES YESProvince FE YES YES YES NO NOYear FE YES YES YES NO NOCaste Status FE YES YES YES YES YES

Robust standard errors in parentheses*** p¡0.01, ** p¡0.05,

These are pooled sequential regression models, with fixed effects for caste status, year andprovince. This table only reports the results of the second equation, where ranked petitioningis the dependent variable. The corresponding results of the first equation, with petitioningas the dependent variable, are not reported. The province-year controls, proportion urban,proportion in public employment, provincial population, and the provincial literacy rate, arenot reported for reasons of space.

50