WHAT IS ETHNIC IDENTITY AND DOES IT MATTER?

Forthcoming in the Annual Review of Political Science Kanchan Chandra New York University Department of Politics 726 Broadway, Rm 718 New York NY 10003 Tel: 212-998-8525 Kanchan.chandra@gmail.com

WHAT IS ETHNIC IDENTITY AND DOES IT MATTER? Keywords: Ethnicity, Culture, Violence, Democracy, Patronage Abstract Since the publication of Donald Horowitzs Ethnic Groups in Conflict, there has been an increasing convergence on the classification of ethnic identities among comparative political scientists. But there is no agreement on the definition that justifies this classification and the definitions that individual scholars propose do not match their classifications. This article proposes a definition that captures the conventional classification of ethnic identities in comparative political science to a greater degree than the alternatives. According to this definition, ethnic identities are a subset of identity categories in which membership is determined by attributes associated with, or believed to be associated with, descent (described in the article simply as descent-based attributes). I argue, on the basis of the definition proposed here, that ethnicity either does not matter, or has not been shown to matter, in explaining most outcomes to which it has been causally linked by comparative political scientists, including violence, democratic stability and patronage.

WHAT IS ETHNIC IDENTITY AND DOES IT MATTER? What is ethnic identity? Since the publication of Donald Horowitzs Ethnic Groups in Conflict in 1985, there has been a convergence among comparative political scientists on which identities we classify as ethnic. For Horowitz, ethnicity is an umbrella concept that easily embraces groups differentiated by color, language, and religion; it covers tribes, races, nationalities, and castes. (Horowitz 1985, 53). Much of the recent theoretical literature on ethnic politics explicitly follows this umbrella classification (e.g. Varshney 2001, Chandra 2004, Wilkinson 2004, Htun 2004, Posner 2005). Even more importantly, the four principal datasets on ethnic groups that constitute the foundation for cross-national empirical studies of the effect of ethnic identity in comparative politics the Atlas Narodov Mira (Miklukho-Maklai Ethnological Institute 1964), a dataset on ethnic groups in 190 countries published by Alesina et al in 2003 (Alesina et al 2003), a comparable count of ethnic groups in 160 countries published by James Fearon in the same year (Fearon 2003) and the Minorities at Risk (MAR) project (http://www.cidcm.umd.edu/inscr/mar/) -- also by and large employ this classification. There remain only some quibbles on the margin about whether castes should be excluded (e.g. Fearon 2003) or retained (e.g. Atlas Narodov Mira 1964, Varshney 2001, Chandra 2004, Wilkinson 2004, Sambanis 2004, Htun 2004, Posner 2005) and whether region and clan should be included in the list . But we do not have a definition which matches this classification. Many comparative political scientists do not define the term before using it. And those that do often classify identities as ethnic even when they do not correspond to their own definitions. Horowitz, for instance, counts Hindus and Muslims in India, Christians and Muslims in Lebanon, and Creoles and Indians in Guyana and Trinidad as ethnic categories even though they do not possess his primary defining characteristic of a myth of common ancestry (Horowitz 1985). Fearon counts Hindi-speakers as an ethnic group even though individuals who either speak Hindi or have Hindi as their mother tongue do not meet his definitional criterion of having a distinct history as a group or a shared culture valued by the majority of members (Fearon 2003). And Chandra often counts categories based on region as ethnic, even though it is not clear whether these groups meet her definition of ethnic groups as ascriptive groups (Chandra 2004, Chandra 2005). Why is a definition necessary if we are approaching a consensus on classification? Because a definition tells us how to evaluate and build theories about ethnic identity and concepts based on ethnic identity, such as ethnic diversity, ethnic riots, ethnic parties, ethnic voting and so on as an independent variable. We now have a large body of work in comparative political science that argues that ethnicity matters for violence, for democratic stability, for institutional design, for economic growth, for individual well-being and so on and makes general, cross-country predictions about its effects (e.g. Horowitz 1985, Rabushka and Shepsle 1972, Przeworski et al 2000, Posen 1993, Petersen 2002, Geertz 1973, Dahl 1971, Bates 1974, Landa 1994, Caselli and Coleman 2001, Fearon 1999, Chua 2003, Cox 1997, Kaufmann 1996, Reilly 2001, Toft 2003). Claims that tell us why these variables matter for some outcome are always based on the assumption that ethnic identities have particular properties that explain the outcome. If we are to assess these claims and build new ones, we need some basis on which to judge which properties can reasonably be associated with ethnic identities. A definition provides an analytical basis for making this judgment.

This article proposes a definition of the term ethnic identity which captures the conventional classification to a greater degree than previous definitions. According to this definition, ethnic identities are a subset of identity categories in which eligibility for membership is determined by attributes associated with, or believed to be associated with, descent (described in the article simply as descent-based attributes). This definition differs from previous ones in two ways. First, it introduces a distinction between categories of membership and the attributes that qualify individuals for membership in that category. The two concepts have been conflated in previous work, but making a consistent distinction between them has large consequences for how we think about ethnic identity. Second, it empties the definition of characteristics such as a common culture, a common history, a common territory and a common language, which are only sometimes associated with the identities that we think of as ethnic and thus cannot be thought of as defining characteristics. My approach to building this definition is to identify the principles underlying the classification of the term ethnic identity in the specialized community of comparative political scientists not to define ethnicity according to some objective criteria, or to capture broader social scientific usage or everyday usage. The virtue of this approach lies in the opportunities it provides for cumulation of research on ethnic identity as an independent variable. In order to theorize about the effect of ethnic identity in a cumulative fashion, we must evaluate the previous body of work in comparative political science that argues that ethnicity matters, retain the insights that survive an evaluation and discard those that do not. And in order to evaluate whether and how ethnicity matters in the way these works propose, we need a definition that tells us what the properties of ethnic identities, as classified by this body of work, are. It would be an added advantage if this definition also captured the way in which the term is understood among other communities but that is not my main concern. I show here that many of the properties commonly associated with ethnic identities in our explanatory theories do not characterize the identities that we classify as ethnic in general, although they may well apply to particular subsets of these identities. These properties include common ancestry, a common myth of ancestry, a common language, a common culture, and a common homeland. Based on the definition proposed here, I also identify two properties that are indeed intrinsic to ethnic identities, on average: constrained change and visibility. By the property of constrained change, I mean that while an individuals ethnic identities do change in the short term, the pattern of change and the mechanisms driving it are constrained by the underlying set of attributes. By the property of visibility, I mean that some information about an individuals ethnic identity categories and the categories to which she does not belong -- can be obtained through superficial observation. But these two properties are not uniquely associated with ethnic identities -- they characterize all identity categories in which membership is determined by descent-based attributes. And they are only associated with descent-based identity categories on average. Particular descent-based identities may well resemble particular non-descent- based identities in particular contexts. This forces us to conclude that ethnic identity and concepts related to ethnic identity such as ethnic diversity, ethnic riots, ethnic parties, ethnic violence, ethnic conflict and so on either do not matter or have not been shown to matter by most previous theoretical work on ethnic identity as an independent variable. Only a small subset of our previous claims about why and how ethnic identity matters rely on properties that have been shown to be intrinsic to ethnic identities. In most 3

instances, the mechanisms driving our explanatory theories about the effect of ethnic identity assume properties such as fixedness of identity, cultural homogeneity, and a shared history that are not associated with ethnic identities even as classified by this body of work. The cause of the outcome they seek to explain, thus, must be some other variables that act independently or in interaction with ethnic identity. Thus, these theories must either be reformulated by taking into account the role of one or more omitted variables or read as theories that are not about the effect of ethnic identities at all. As we seen, then, defining the concept of ethnic identity in a way that accords with the conventional classification in comparative political science is an enterprise distinct from defending or perpetuating the use of the concept in our explanatory theories. Rather, it provides a basis on which to question the findings of previous theories and the continued use of the concept. But if we discard the findings of previous theories or discontinue the use of the concept, we must have a clear idea of what we are rejecting and why and what to replace it with. Thus, just as we need a definition of ethnic identity to defend the use of the concept, we need a definition of ethnic identity to justify why it should be discarded. Section 1 elaborates on the definition proposed here. Section 2 situates this definition in the lineage of past definitions, eliminating in the process some of the properties routinely associated with ethnic identity. Section 3 identifies two of the properties that can indeed be taken to be intrinsic to ethnic identity constrained change and visibility. Section 4 uses the discussion of the properties that can and cannot be associated with ethnic identity to evaluate theories about how ethnicity matters. 1. Definition of Ethnic Identity By identity, I mean any social category in which an individual is eligible to be a member. Ethnic identity categories, I propose, are a subset of identity categories in which eligibility for membership is determined by descent-based attributes. By attributes that determine eligibility for membership I mean either those that qualify an individual for membership in a category or those that signal such membership. By descent-based attributes, I mean attributes associated with, or believed to be associated with descent. By attributes associated with descent I mean attributes that are acquired genetically (e.g. skin colour, gender, hair type, eye colour, height, and physical features), or through cultural and historical inheritance (e.g. the names, languages, places of birth and origin of ones parents and ancestors), or acquired in the course of ones lifetime as markers of such an inheritance (e.g. last name, or tribal markings). By attributes believed to be associated with descent, I mean attributes around which a credible myth of association with descent has been woven, whether or not such an association exists in fact. The definition thus includes both a subjective and an objective element. The set of identity categories in which membership is determined by descent-based attributes is large. Ethnic identity categories are a subset of this larger set defined by the following restrictions: (1) They are impersonal. (2)They constitute a section of a countrys population rather than the whole. (3) If one sibling is eligible for membership in a category at any given place, then all other siblings would also be eligible in that place. (4) The qualifying attributes for membership are restricted only to ones own genetically transmitted features or to the language, religion, place of origin, tribe, region, caste, clan, nationality, or race of ones parents and ancestors. 4

If some of the restrictions that constitute the subset of ethnic identities appear somewhat arbitrary, they are. Why impose a rule that requires siblings to be equally eligible for membership before a category can be called ethnic? Why allow this particular set of descent-based attributes and not others? I do not offer analytical justifications for these restrictions here indeed, I argue later that there is not so far a good reason to wall off ethnic identities from other types of descent-based identities. But my purpose here is simply to identity those restrictions that we must impose in order to approximate the conventional classification of ethnic identities. Once we have identified these features, we are in a position also to recognize their arbitrariness and discard them where necessary. Let me illustrate this definition using the fictionalized example of Helen, imagined from a mlange of characters in Mary Waters study of West-Indian immigrants in New York (Waters 1999). Born in the English-speaking island of Trinidad to parents of African origin, she has dark skin, dark brown eyes and straight hair. She moved to the US after obtaining her high school diploma in Trinidad and works there as a food service employee, earning $25,000 a year. She belongs to, and votes for, the Democratic party there. She married a Haitian man in New York, and learned to speak French, which is now her primary language of communication with her children and husband. Her brother Derek, who has lighter skin, light brown eyes and otherwise similar features, remained behind in Trinidad. Their parents are well-educated professionals who belong to the Peoples National Movement (PNM) in Trinidad. They are Presbyterians, but Helen herself converted to Catholicism after meeting her husband. According to the definition above, Helens ethnic identity categories include Black (in which the qualifying attribute, according to current norms, is descent from African parents, signaled by attributes such as the colour of her skin and physical features) and West-Indian (in which the qualifying attribute is descent from parents who lived in Trinidad, signaled by her accent among other attributes). Both these categories are determined by attributes associated, or believed to be associated with descent, and both of which place Helen and Derek in the same categories. Her ethnic identity categories also include "African-American" (in which membership is determined by the attribute of descent from African parents, skin colour and physical features, in the US). Derek, because he stayed behind in Trinidad, cannot call himself "African-American." But if he were to move to the US, he would be eligible for membership in this category just like his sibling. Helens ethnic identity categories according to this definition do not include several other identity categories also based on descent-based attributes such as descendant of PNM supporters (excluded because it is not based on either on physical features or on the language, religion, race, tribe, caste, nationality and place of origin of her parents), people with dark brown eyes (excluded because, while it is based on her physical features, it excludes her sibling, Derek), and female (also excluded because it excludes Derek). They also do not include Catholic (determined not either by descent or a myth of descent, but by conversion, openly acknowledged, during her lifetime), French-speaker, (determined by a language learned during her lifetime rather than her ancestral language), working class (determined by attributes acquired during her lifetime such as her high school diploma and her job as a food service worker), and Democrat (determined by her joining the Democratic party during her lifetime).

This definition captures most, but not all, of the categories that comparative political scientists include in the list of ethnic categories. For instance, although that list includes all categories based on language, this definition makes a distinction between categories in which the attribute for membership is an ancestral language (or a language presented as an ancestral language), which it takes to be ethnic and categories in which the attribute for membership is the language openly acquired during a lifetime, which it does not take to be ethnic. It also provides a decision rule for whether and when to classify ambiguous categories such as those based on clan and region as ethnic categories. A regional category is an ethnic category according to this definition only if the membership rule takes into consideration the region of origin of an individuals parents and ancestors, rather than the region in which an individual currently resides. 2. Comparison with Other Definitions The most widely used definitions of ethnic identity proposed in previous literature include the following: According to Max Weber ethnic groups are those human groups that entertain a subjective belief in their common descent because of similarities of physical type or of customs or both, or because of memories of colonization or migration; this belief must be important for the propagation of group formation; conversely, it does not matter whether or not an objective blood relationship exists. (Hutchinson and Smith 1996, 35). According to Donald Horowitz: Ethnicity is based on a myth of collective ancestry, which usually carries with it traits believed to be innate. Some notion of ascription, however diluted, and affinity deriving from it are inseparable from the concept of ethnicity. (Horowitz 1985, 52). According to Fearon and Laitin, an ethnic group is a group larger than a family for which membership is reckoned primarily by descent, is conceptually autonomous, and has a conventionally recognized natural history as a group. (Fearon and Laitin 2000, 20). In a subsequent refinement, Fearon defines a prototypical ethnic group as one that has several of the following features as possible: (1) Membership is reckoned primarily by descent (2) Members are conscious of group membership (3) Members share distinguishing cultural features (4) These cultural features are valued by a majority of members (5) The group has or remembers a homeland (6) The group has a shared history as a group that is not wholly manufactured but has some basis in fact. (Fearon 2003, 7) According to Anthony Smith, an ethnic group is, a named human population with myths of common ancestry, shared historical memories, one or more elements of a common culture, a link with a homeland and a sense of solidarity. (Hutchinson and Smith 1996, 6)

Virtually all definitions in this inventory agree that descent is somehow important in defining an ethnic group. The differences are over how precisely to specify the role of descent, and whether and how other features should be combined with it in defining ethnic groups. The role of descent is specified in four different ways: (1) a common ancestry (2) a myth of common ancestry (3) a myth of a common place of origin and (4) a descent rule for membership. The features combined with

descent include: (5) a common culture or language (6) a common history and (7) conceptual autonomy. Below, I consider whether any of these characteristics, separately considered, can adequately define ethnic identity, and then evaluate definitions based on combinations of them. Taking examples of sets of categories commonly classified by comparative political scientists as ethnic, I ask three questions of any single defining characteristic: (1) Do the membership rules for inclusion in any single category within a set require individuals to have this characteristic in common? (2) Do the membership rules for inclusion in different ethnic categories within a comparable set require individuals to have different values on this characteristic? (3) Do the sets of categories that comparative political scientists classify as ethnic uniquely possess these characteristics? An ideal definition should fulfill all three conditions. But the first two are most consequential for the purpose of this project. A definition that covers most of the categories that we call ethnic allows us to evaluate claims about how ethnic identity matters even if it includes some extra categories, since it describes, at a minimum, the entire sample from which the inferences are drawn. But a definition that excludes categories that we routinely classify as ethnic makes us less able to evaluate these claims because it describes a truncated sample. a. Common ancestry A common ancestry definition must stipulate the nearness of the connection required to call a group an ethnic group does common ancestry mean a shared ancestor one branch ago, a hundred branches ago, or a million branches ago? Without such a stipulation, we can find a point of intersection in the family trees of any two individuals by going back far enough, thus eliminating group differentiation altogether (Cavalli-Sforza 2000). Lets stipulate that by common ancestry, we mean that individuals in the same ethnic group share a more proximate ancestor than individuals in different ethnic groups. This minimally reasonable stipulation excludes many of the categories that we classify as ethnic. Take the categories Black and White. As a set of categories based on race, they fit the conventional classification in comparative politics, and are explicitly counted as ethnic in at least Fearon 2003, Alesina et al 2003, and the MAR dataset (which uses the term African-American rather than Black). We know that many white Americans from former slave-owning families share proximate ancestors with black Americans. Consider the example of E.C. Hart, classified as a Louisiana white, who had several children with Cornelia, a woman of colour (Dominguez 1997, 2627). Harts children with a white wife would have shared common ancestry at the most proximate level -- the same father -- as his children with Cornelia. But the two sets of children were not classified in Louisiana as members of the same group Harts children with Cornelia were then classified as colored and would now be classified as Black, while his children with a white wife would then and now have been classified as White. Thus, in classifying these categories as ethnic, comparative political scientists appear to be following a rule that does not rely on common ancestry. At the same time, individuals whose ties of ancestry that are far more distant, if they exist at all, such as Irish-Americans and Lithuanian-Americans, or Jamaican and Nigerian immigrants, are classified as members of the same ethnic categories. In the case of other categories, such as Hindus and Muslims in India, Punjabis and Sindhis in Pakistan, Flemish and Walloon in Belgium, 7

the question of common ancestry does not even arise. We cannot, thus, take a common ancestry to be a defining feature of an ethnic group indeed, individuals often belong to different ethnic groups despite the objective fact of common ancestry. b. A Myth of Common Ancestry Let us turn now to a myth of common ancestry. There are certainly many cases of ethnic groups with a myth of common ancestry, especially among tribes in Africa. For example, Yorubas in Yorubaland trace their descent to the mythical ancestor Oduduwa, and Yorubas in different ancestral cities trace their descent even more specifically to particular sons of Oduduwa (Laitin 1986, 110). Zulus in South Africa claim direct descent from the patriarch Zulu who was born to a Nguni chief in the Congo basin area. Kikuyus in Kenya claim descent from the single ancestor Gikuyu. Indeed, the great influence that scholars who study Africa have had on the study of ethnicity may be one reason why a myth of ancestry is so often proposed as a defining characteristic of ethnic groups. But this criterion excludes a good number of groups that we also classify as ethnic, in Africa but also in other regions, which do not claim an ancestor in common and do not differentiate themselves from others in a comparable set on the basis of myths of ancestry. This is the case with Blacks and Whites in the U.S., Punjabis, Sindhis, and Pathans in Pakistan, Hindus and Muslims in India, Serbs and Croats in the former Yugoslavia, Blacks, Whites, Coloreds and Indians in South Africa, among others, none of which claim an ancestor in common (whether or not such an ancestor can be exists in fact), or differentiate themselves from each other on the basis of myths of ancestry. The irrelevance of a myth of common ancestry to membership rules in groups that we commonly think of as ethnic can be illustrated especially clearly when we consider the process by which new ethnic groups form as a result of fissures from old ones. Consider one example of this process of fission:

Pakistani

Bangladeshi Pakistani Punjabi Sindhi Pathan Baluch

The anti-colonial struggle produced the initial ethnic category of Pakistani a An early fissure in this category, itself a product of a fissure from the larger category Indian, between Bengali Muslims and others, resulting in the separate state of Bangladesh for Bengali Muslims. Other new ethnic groups to emerge within Pakistan included Punjabis, Sindhis, Pathans and Baluchs. Myths of common ancestry were not part of the process by which entrepreneurs and masses within these groups distinguished themselves from each other. Yet these groups are routinely classified by

comparative political scientists as ethnic (e.g. Horowitz 1985, 281, Fearon 2003, Alesina et al 2003). Perhaps, we might object, what is important here is not a myth of common ancestry but a myth of common origin, with which it is often conflated. In the case of Blacks and Whites, for instance, we might concede that while neither group is defined or distinguished by a common ancestor, both categories are associated with distinct points of origin: Blacks in Africa and Whites mainly in Eurasia. Similarly, while Serbs and Croats are not distinguished by common ancestry, they can be associated with distinct territories in central and southern Europe. And in the example of group fission that I raised above, each fissure is associated with a distinct territory. I will turn to this point next. But first, let me relate the critique so far to the definition that I propose in this article. To argue that the fact or myth of common ancestry does not capture the classification of ethnic groups used by comparative political scientists does not imply that ancestry, real or imagined is not important quite the opposite. According to the definition I propose in this article, ancestry, or a myth of ancestry, is critical to the definition of an ethnic group, but common ancestry, or a myth of common ancestry is not. For instance, the membership rule for classification of individuals as Black or White in the US separates individuals who have, or are believed to have, some African blood from individuals who do not, or are not believed to, have a drop of African blood. (Nobles 2000). This membership rule renders the categories Black and White ethnic categories because it requires the possession of attributes based on ancestry (the proportion of African blood) even though it does not require common ancestry. By the same logic, the category Mohajir in Pakistan would be classified as ethnic because the qualifying attribute is descent, or fabricated descent, from parents who immigrated from North India. The fact of common ancestry, even if it were verifiable, is irrelevant to this membership rule. c. A Common Region of Origin, or Myth of a Common Region of Origin. Suppose, instead of common ancestry, we define an ethnic group as a collection of individuals with a common point of origin, or a myth of common origin. We can certainly identify a subset of ethnic groups that are associated with a common point of origin. But the perception of a common homeland, presupposes the existence of a group, it does not define it. Given a group with this perception, we can reason backward, proposing that definition of a region and choosing that time span that allows us to discover a common region of origin. But we cannot, from the point of origin, predict ethnic group boundaries. To illustrate, lets return to the case of Blacks and Whites in the U.S. We might associate Blacks in the US with a common origin in Africa, and Whites with a common origin in Eurasia. But these associations depend on (1) the definition of region that we employ and (2) the time period we choose to start at. Some definitions of region and choices of time period would reveal distinct common homelands for both groups, while others would reveal a shared one. If we categorize Africa and Eurasia as distinct regions, then we can associate Blacks and Whites with distinct regions of origin and thus as distinct ethnic groups. But there is no analytical reason why we should think of Africa and Eurasia as distinct regions instead of disaggregating further within both categories. If we thought of regions in Africa, including present-day Liberia, 9

Ghana, Cote dIvoire, Togo, Benin and Cameroon, as distinct, then we should have not one but several ethnic groups corresponding to origin in these regions. Similarly, if we thought of regions in Eurasia, including France, Ireland Germany, Lithuania, Poland and Italy as distinct, then we would also have several ethnic groups. The identification of Africa and Eurasia as regions of origin requires us first to define Blacks and Whites by some other criteria other than region and then employ that definition of region of origin that includes members of these pre-defined groups and excludes non-members. Further, even if we take Africa and Eurasia to be distinct regions, arguing that Blacks and Whites originate in these two regions makes sense only when based on an arbitrary choice of time period. We can identify distinct points of origin for Blacks, for instance, if we start with the beginning of the slave trade in the sixteenth century. But if we continued going backwards from this point, we would find a point at which both Blacks and Whites originated in present-day Africa (Cavalli Sforza 2000). And if we went forward, we could just as easily say that both groups share a common homeland the US. The argument that ethnic groups are defined by association with a homeland, thus, holds up only when we take the groups in question as given and go backward in time, stopping at exactly the point at which the ancestors of the two groups are found in separate geographical regions. Serbs and Croats and Bosnian Muslims in the former Yugoslavia, for instance are now associated with distinct homelands, in Serbia, Croatia and Bosnia. But going back in time simply to the 9th century, we might have thought of all three groups as possessing a common homeland in the South Slav regions, and so to be one, not three ethnic groups. Punjabis, Sindhis, Pathans and Baluchs can be seen equally as belonging to distinct territories or as belonging to the same one. Similarly, Sinhalas and Tamils in Sri Lanka have distinct homelands only if we define northern Sri Lanka as a distinct region, and the rest of Sri Lanka as another distinct region or, going even further back, we define present-day Tamil Nadu as a distinct region and trace the migration of Tamils from there to Sri Lanka. But why should we define the region in this way? Why not think of all of Sri Lanka, or the northern tip of Sri Lanka and the southern tip of India as a single region, and thus think of both the Sinhalese and Tamils as having a common homeland? To the extent that the definition of group comes before the definition of a distinct region of origin, a fact or myth of a common region of origin cannot be seen as a defining characteristic of ethnic groups. Further, there are also many sets of ethnic groups, defined often by religion and race, in which the distinctions between categories in the same set do not depend upon the fact or myth of a distinct homeland. Some examples include Hindus and Muslims in India, Christians and Muslims in Nigeria, Blanco, Pardo and Preto in Brazil and so on. d. The Descent Rule The descent rule, identified by Fearon and Laitin, is the most recent attempt to clarify the role of descent in the definition of an ethnic group. As they point out: In deciding a persons ethnicity, we do not need to know anything about his or her cultural habits, mother tongue, religion or beliefs of any sort. Rather, we simply need to know about parentage. In ordinary usage, to ask What is her ethnicity? is to ask about what ethnic group her parents (or other close ancestors) were assigned to All that is necessary to be counted as a member of an ethnic group is to be able to have accepted the claim to be immediately descended from other members of the group (Fearon and Laitin 2000, 13). 10

There is an important ambiguity in this definition: Should we define a group as ethnic if the criterion for membership in the group is that the parents of all members were coded as members of the same group whether or not that group was also considered ethnic? (hereafter Descent Rule 1). Or should we define a group as ethnic if the criterion for membership in the group is that the parents of all members were also coded as members of the same group and that group was also considered ethnic? (hereafter Descent Rule 2) To illustrate the difference between the two versions, take the example of American Jews. According to Fearon and Laitin, many Americans who cannot perform a Jewish ritual and dont speak Hebrew still consider themselves and are considered by others as ethnically Jewish because that is the way their parents and grandparents were coded. (Fearon and Laitin 2000, 14). For the analyst to code American Jews as an ethnic group, is it sufficient simply that the criterion for membership in the category American Jew is that ones parents were coded as Jewish whether or not Jews were considered an ethnic group in the lifetime of ones parents (Descent Rule 1)? Or must the criterion for membership in the category American Jew be that ones parents were also coded as ethnically Jewish, requiring ones grandparents to also be coded as ethnically Jewish, which in turn would require ones great-grandparents to be coded as ethnically Jewish and so on into infinity (Descent Rule 2). Descent Rule 1 covers a larger set of cases that we commonly understand to be ethnic than the rules that ethnic groups must have a myth of common ancestry or common origin. But it excludes several cases in which individuals routinely consider themselves, and are considered as members, of a group that we classify as ethnic even when their parents were not coded as members of this group. Take for example the category Yoruba when it was invented in Nigeria in the nineteenth century. At this time period, the parents of those who were classified as Yoruba were not themselves classified as Yoruba for the reason that this category did not exist during their lifetimes. According to Descent Rule 1, then, the category Yoruba in the nineteenth century would not be coded as ethnic. But the category Yoruba is universally coded as an ethnic category by all comparative political scientists, without making a distinction between time periods. As another example, consider the category Backward Caste in India, which included as members individuals who possessed a given set of last names and/or ancestral occupations. The category was introduced by the Indian Central Government in 1990. Within a few years, 52% of the Indian population classified itself and was classified as backward. Yet the parents of those who termed themselves Backward Caste were not coded by themselves or by others as Backward Caste because, as in the case of the first generation of Yorubas, this category did not exist during their lifetimes. A straightforward application of Descent Rule 1 would lead us to code the first generation of Backward Castes as a non-ethnic group to the extent that membership in the category Backward Caste did not depend upon ones parents also being classified as backward caste. As a third example, take the identity category Mohajir, routinely classified as an ethnic group in our datasets (Fearon 2003, Alesina et al 2003). As noted above, this category refers to those who migrated to Pakistan from north-India and their descendants. The parents of many of those classified as Mohajir were not classified as Mohajir themselves. Applying Descent Rule 1 would 11

lead us to code the first generation of Mohajirs as a non-ethnic group, to the extent that membership in this category did not depend upon being able to code ones parents as Mohajirs. As a fourth example, take the case of the Creoles in Louisiana (Dominguez 1986): individuals were initially classified as Creole in Louisiana if they were born in colonial Louisiana and descended from French and or Spanish parents the parents were in many cases not themselves classified as Creole. Again, Descent Rule 1 would lead us to code Creoles as a non-ethnic group, since you could be Creole even if your parents were not coded as Creole. Finally, take the case of categories such as Colored. Individuals were classified as Coloured in the U.S. not only when their parents were also classified as Coloured, but also when one of their parents was classified as White and another Black. (Nobles 2000, Dominguez 1986). An application of the descent rule would treat Coloureds as non-ethnic categories. In a partial attempt to address such cases, Fearon and Laitin note that in the case of mixed marriages, arbitrary (and political) conventions that may vary from place to place are employed. (Fearon and Laitin 2000, 13). But while arbitrary conventions might be employed about which group the children of mixed marriages should be assigned to black, white, colored, mulatto, quadroon, mestizo, multiracial and so on), the conventional classification of all such groups, at least within political science, is that they are ethnic. A definition of the term ethnic, therefore, should include such cases. The discussion above suggests that although Descent Rule 1 covers a large number of categories classified as ethnic, it needs to be modified to take these additional cases into account. The definition here is one such modification. According to this definition, Yorubas, Mohajirs, OBCs, Creoles and Coloreds would all be ethnic categories because membership in each depended upon the possession of descent-based attributes that refer to language, religion, tribe, race, caste, nationality, region or physical features: descent from parents who belonged to a set of ancestral cities in the case of Yorubas, descent from parents who were born in North India in the case of the Mohajirs, descent from parents who belonged to a set of caste categories in the case of the OBCs, descent from parents who were born in Europe in the case of Creoles, and descent from parents and ancestors who included both Whites and Blacks in the case of Coloreds. This definition does not require that children share in the same descent-based categories as their parents just that the attributes that qualify them for membership in ethnic categories are acquired through descent. Descent Rule 2 is circular: for a present day American Jew to be qualified as ethnic, every preceding generation of his ancestors must also have been descended from ethnically American Jews, who themselves should have been descended from ethnically American Jews and so on into infinity there is no logical way to identify an original set of ethnic American Jewish parents. We could apply Descent Rule 2 only if we arbitrarily stipulated that some set of identities were ethnic at some starting point, and then coded other identities as ethnic in relation to this initial set. And this arbitrary application of Descent Rule 2 would eliminate even more of the categories that are routinely classified as ethnic than Descent Rule 1. e. A common culture Suppose we define a common culture broadly as a shared set of symbols, values, codes and norms (consistent with Rogowski, cited in Wedeen 2002). Broad definitions of this sort do not specify which symbols, values, codes and norms it is important to share. If we require group members to share all of them, that would probably disqualify almost all the groups that we count as 12

ethnic. Take the category Yoruba. Those classified as Yoruba share some aspects of culture so defined for instance, a common myth of origin, the worship of a common set of deities, and a common language. But there are also many aspects of culture that they do not share. Yorubas who trace their origin to different ancestral cities in Yorubaland, for instance, speak different dialects of the Yoruba language, have localized festivals, and distinct myths of origin, institutions and rituals. Christian and Muslim Yorubas not only have different modes of religious practice and observance but different sets of value orientations (Laitin 1986). The symbols, codes and norms of those classified as Black, similarly differ a great deal: there are actual or perceived differences between Jamaican blacks and American blacks based on language, work ethic, and family values (Waters 1999), between blacks from the north and the south, (Lemann 1992 ), between middle class and poor blacks and urban blacks and rural blacks (Malcolm X 1964). Consider, finally, the category Hindu. There are many symbols, codes and norms that those classified as Hindu do not share, or do not believe they share: Adharmis, for instance, speak different languages, have different rules for interdining and intermarriage, and practice different rituals than Satnamis the same goes for Bengali-speaking Hindus and Malyalam speaking Hindus, for rich and poor Hindus, for urban and rural Hindus, for Hindus who are academics and Hindus who are businessmen and so on. Suppose we interpret the common culture definition to mean, not that individuals should share all or most of the same values, symbols, codes and norms to qualify as members, but that they should share more with each other than they do with out-group members. In other words, a pair of individuals from the same ethnic group should have more in common than a pair of individuals from different ethnic groups. Although it appears initially appealing, this condition is logically unsustainable in a world of cross-cutting ethnic identities. Lets return, first, to the case of the Yorubas. On one dimension of ethnic identity, we classify the Yorubas, the Ibos and the Hausa-Fulanis as distinct ethnic groups. In order to satisfy this condition, we would need to show that a pair of Yorubas might have more in common with each other than a pair consisting of a Yoruba and an Ibo, or a pair consisting of a Yoruba and a Hausa-Fulani. But we know that there is at least one other dimension of identity on the basis of which we can also organize the same population that of religion which we also classify as an ethnic identity. On the dimension of religion, we classify the same population as Christian and Muslim, and we also think of Christian and Muslim as ethnic identity categories. If we define ethnic categories by a common culture, this means that any two Christians must have more in common with each other than a Christian and a Muslim. As long as these two dimensions cross-cut to some extent, this second claim contradicts the first. If all Yorubas have more in common with other Yoruba than with Ibos and Hausa-Fulanis, for instance, then all Christians cannot have more in common with each other than with Muslims, because there will be at least some Christians and Muslims who are both Yoruba. The same logic also applies to our other running examples Hindu and Muslim, and Black and White. Individuals in many groups that we classify as ethnic have cross-cutting memberships in other ethnic groups. So, if we say that ethnic groups are defined by a common culture as defined above, we would be ruling out an unreasonably large number of cases. Consider a third conceptualization of culture. Suppose when we say that a group has a common culture, we mean simply that they inhabit the same framework of meaning -- they use the same concepts, and can understand each other, whether or not they subscribe to an identical set of symbols, values, codes and norms, and whether or not they speak the same language. This is akin 13

to Wedeens conceptualization of culture as a semiotic practice (Wedeen 2002). Individuals who share a common culture, then, must, to paraphrase Geertz, be able to agree that something is a wink rather than a twitch, whether or not they wink at the same time and whether or not they value a wink in the same way. Even with this very reasonable definition of culture, we cannot define ethnic groups as cultural groups. Many ethnic groups are composed of still smaller groups with specialized vocabularies that are not mutually intelligible. In New York in the 1960s, the street vocabulary of sections of Harlem was the equivalent of Sanskrit for many middle class Blacks (Malcolm X 1964, 317). But both sections would classify themselves, and be classified by others, as members of the same ethnic category Black. And although a middle class Black may have been able to comprehend perfectly a white neighbour from her own class and educational background, this shared conceptual vocabulary would hardly lead her, or others, to classify these two individuals as part of the same ethnic group on that basis. Given some prior basis for defining ethnic groups, we can then probe the extent to which they share common frameworks of meaning. But if we were to define ethnic groups as groups that shared such common frameworks, we would lose many of the groups that we routinely classify as ethnic. Consider a fourth definition perhaps by a common culture, we mean not that group members share all symbols, values, codes and norms, but that they share some key symbols, values, codes and norms that distinguish them from members of other groups. This is a restatement of Barths claim that ethnic groups are defined by the cultural markers that differentiate the boundaries between them, not by the cultural stuff that is contained within these boundaries (Barth 1969). But in many of the groups that we classify as ethnic, cultural markers need to be backed up by descent-based attributes, or the myth of having descent-based attributes in order to delimit group boundaries. Sikhs in Punjab can acquire the cultural markers associated with being Hindu, for instance, by cutting off their hair and ceasing to wear a turban. But as long as they own descent from a Sikh family, they count themselves, and are counted as Sikhs rather than Hindus. Further, cultural markers that coincide with descent-based attributes are often the product of ethnic group mobilization rather than a pre-existing characteristic that defines the group itself. Finally, suppose we define culture simply as language. An ethnic group, then, is simply a group that has a common language. Different ethnic groups should have different languages. A survey of even a handful of examples illustrates the lack of fit between this definition and the groups that we often recognize as ethnic. Some ethnic groups certainly have a common language, which distinguishes them from other groups in the same partition of the population: the Yoruba are an example, as are Tamils and Sinhalese in Sri Lanka, Zulus and Xhosas in South Africa, and French and German speakers in Switzerland. But many groups that we classify as ethnic do not have a common language. Hindus include speakers of at least 17 major languages, not to mention hundreds of dialects. Christians and Muslims in Nigeria do not speak the same language, but we think of them as an ethnic category. And Blacks and Whites in the US speak the same language for instance but are classified as different ethnic groups. f. A common history Consider now another way of defining an ethnic group an ethnic group is a group that has a common history. Suppose we define history to mean simply a shared past. What does it mean to 14

say that individuals share a past? Does it mean having lived through the same key events in the course of a lifetime? In that case, all individuals of the same age group throughout the world would share a common history, in that they all lived through the same events whether or not they were aware of them. Suppose we try a definition that is less absurd, defining history to mean events that occurred at least one generation previously, and which were claimed to have been part of the particular experience of some group. For instance, although the potato famine may in some sense be part of the history of the world, it is particularly part of the history of the Irish living at that time, who experienced it most directly. Having a common history then means sharing a connection to events that that marked the lives of the generations which preceded us. But how would an individual know which generations of people to affiliate herself to when looking for a common history? Should someone of Irish descent born in the US affiliate herself to generations who were born in or lived in the US? Should she affiliate herself to generations born in Ireland? Should she affiliate herself to subsets of generations born in the US for instance, the ancestors of University professors? Or the ancestors of black men? In order to identify her history, she needs a rule to tell her which group she belongs to. A common history, then, cannot be the defining characteristic of a group, ethnic or otherwise. To have a common history already presupposes the existence of a group based on other criteria. Anthony Appiah puts it best: Sharing a common group history cannot be a criterion for being members of the same group, for we would have to be able to identify the group in order to identify its history. Someone in the fourteenth century could share a common history with me in a historically extended race only if something accounts for their membership in the race in the fourteenth century and mine in the twentieth. That something cannot, on pain of circularity, be the history of the race. (Appiah 1992 32) g. Conceptual Autonomy By conceptual autonomy, Fearon and Laitin mean that the existence of an ethnic category does not depend conceptually on the existence of any particular ethnic category. (Fearon and Laitin 2000, 16). But we know from a large literature that the definition of any ethnic group presumes and depends upon the existence ethnic others. Indeed, membership rules exist for the sole purpose of distinguishing insiders from outsiders. This criterion, if applied, would eliminate virtually all ethnic groups from the definition. h. A Combination of Characteristics. Although I considered each characteristic individually, most definitions of ethnic identity incorporate several secondary characteristics in addition to the primary characteristic of descent, although the precise combination they employ varies. But any definition that requires a combination of characteristics captures only a subset of most of the identities that comparative political scientists classify as ethnic. This is because, as I have tried to show above, each characteristic, taken singly, captures only a subset of these identities at best. Since any single defining characteristic discussed above captures at best a subset of the classification that we started with, any combination of characteristics will capture a still smaller subset. The figure below illustrates this point.

15

Non-Ethnic Identities

Set of Identities Classified as Ethnic Descent-based attributes Descent Rule 1 Language/ Myth of Common Ancestry/ Descent Rule 2/ Fact of Common Ancestry Combination of Characteristics

The bold circle represents the set of identities conventionally classified as ethnic by comparative political scientists that is, identities based on race, language, caste, tribe, religion, nationality, and, sometimes, region and clan. Each circle within it represents a subset captured by a particular definition. The largest subset is captured by definition proposed in this article (labeled DescentBased attributes), which eliminates only categories based on religion, or language and so on if the attributes for membership are acquired voluntarily over a lifetime rather than by descent. This is followed by Descent-Rule 1, which would eliminate several additional categories such as Yorubas in Nigeria in the 1900s, OBCs in India in the 1990s, Mohajirs in present-day Pakistan, the first generation of Irish Americans in the US, the first generation of Creoles and Coloreds and so on. These two subsets are followed, in some unspecified order, by language, a myth of common ancestry, the fact of common ancestry and Descent Rule 2, each of which cover some smaller subset of identities. I do not include independent subsets defined by characteristics such as a common culture, a common history, a common territory and conceptual autonomy since they are secondary rather than primary criteria that may be associated with a pre-existing group but do not define it. The smallest subset in the diagram represents a definition based on some combination of characteristics. If we define an ethnic group as a group in which membership is determined by Descent Rule 1 and by a distinct history, for example, we would eliminate still more sets of categories, including Hindus and Muslims in India, Blacks and Coloreds in Louisiana (Dominguez 1994), and Christians and Muslims in Yorubaland, since categories in each set share a history with others in the same set. With every additional characteristic that we include in the combination, we would exclude still more. As such, these definitions cannot be used to evaluate causal claims associated with ethnic identity in general, because they would truncate the sample of categories that the claims refer to. 3. Properties of Ethnic Identity Categories I identify two intrinsic properties of such descent-based attributes: stickiness and visibility. From these properties of descent-based attributes, I infer two properties of ethnic categories constrained change and visibility.

16

(a) Constrained Change Attributes associated with, or believed to be associated with, descent are, on average, difficult to change in the short term. In contrast, attributes not associated, or not believed to be associated with descent, can, on average, be easily changed even in the short term. This claim, illustrated in the figure below, refers to the average tendency of distributions of attributes, rather than any single one. Imagine a scale that orders all attributes according to the degree of difficulty associated with changing them in the short term. Descent-based attributes are distributed in the upper half of this scale, while others are distributed in the lower half of this scale, with an area of overlap in between.

Figure 1

Degree of Stickiness

To illustrate, lets go back to Helen. One of the attributes given to her at birth skin colour lies at one extreme on the scale of stickiness. Skin colour can, with medical help, be changed within a single generation, and there are a handful of cases where it has been (Griffin 1996, Larsen 1997). Skin colour can also be changed through a change in the interpretive framework within which it is viewed: if a sufficiently large number of external observers adjust their frameworks, Helens particular shade of skin colour might be reinterpreted as light rather than dark. But such a change is likely to be made with a great deal of difficulty, and over a period of time. There are degrees of difficulty. Changes in skin colour and gender are likely to be most difficult. Changes in physical features are likely to lie in the middle, in part because of the constraint of making change in any one feature appear consistent with others. Helen might easily change her hair type, for instance, by having it curled, or her eye colour, by wearing contact lenses. But in order to make the myth that these new attributes were given at birth credible, she would also need to change other attributes, such as skin colour and features, to make them consistent with that myth. Effecting changes in the place of birth of ones parents and ancestors, or last name or ancestral religion may be least difficult, especially in the context of relocation or migration, when new histories can be more easily invented.

17

In contrast, attributes not based on descent, or at believed to be based on descent, are, on average, easier to change. Helen might change her educational qualifications, for instance, simply by returning to school. She might alter her income by asking for a raise or both her occupation and income by getting a higher paying job. She can bring about these changes without having to erase her previous attributes. While one cannot claim birth in the US without renouncing birth in Trinidad, getting a masters degree does not require her to disguise the fact that she has a high school diploma. Here as well, there are degrees of difficulty. Changing your initial endowment of economic resources acquiring more land for instance can in some contexts be almost as difficult as changing your skin colour. Changing your educational credentials is often less difficult but it requires a significant investment in time and money. And changing jobs or party membership perhaps easier. But the degree of difficulty here depends upon context rather than on some intrinsic property of such attributes. There is also an area of overlap in the degree of stickiness of individual attributes that fall into the two types. Some attributes that are not descent-based such as land ownership or accent may lie towards the high end of the scale of stickiness. And some descent-based attributes, such as last name or the religion of ones parents, may lie towards the low end to the extent that the myths required by these changes may be easily invented. Such overlap is consistent with differences in the average tendency of both types of attributes. It is common to assume that because the attributes defining them are fixed in the short term, ethnic identity categories are also fixed in the short term. Individuals can change between identity categories, often quite rapidly, by combining and recombining elements from their set of attributes differently. But the pattern of change should be constrained by the underlying distribution of attributes. Thus, the property of constrained change can be legitimately associated with ethnic identities in the short term, while the property of fixedness can not. Consider Helen again. Helens stickier attributes consist of her skin colour, her birth in Trinidad and her descent from African American parents. While she cannot change these attributes in the short term, she can, by selecting different attributes, change the categories in which she activates her membership. If she emphasizes the attribute of birth in the English-speaking island of Trinidad, she can activate membership in the category West-Indian. But if she de-emphasises her foreign birth, and highlights instead the attributes of skin colour and descent, she can activate membership in the category Black. Her ability to change the categories that she activates is constrained: while she can change to categories defined by a different selection of sticky attributes that she possesses, she cannot change to categories based on sticky attributes that she does not possess (e.g. German, if the category German is defined on the basis of birth). (b) Visibility A second property of descent-based attributes is that they are, on average, more visible than attributes acquired in the course of a lifetime. By visible attributes, I mean attributes which can be ascertained through superficial data sources such as the name, speech, features and dress of individuals. The larger the number of sources that contain information about an attribute, the more visible it is. By invisible attributes, I mean attributes which can be ascertained only by more careful

18

observation and background research (Chandra 2004). As above, this claim refers to distributions rather than particular attributes. Once again, lets go back to Helen to illustrate. Some of Helens descent-based attributes, such as her gender, skin colour, hair type, physical features are immediately displayed on her person. Others, such as the language of her parents (English), are contained in her name. Less precise information on the religion of her parents is also available in her name, which, even though it does not allow the observer to pinpoint it precisely, allows the observer to rule out religions such as Islam or Hinduism. The way in which observers code such information is the result of social construction. Helens shade of skin colour, for instance, might be interpreted in the U.S. as black, but in Brazil as brown. This difference is the result of differences in historical and institutional contexts that have taught individuals to instinctively code the same data in different ways (see Chandra 2004). But the main point is that some raw data about some of the attributes given at birth are always visible, often in multiple data sources, regardless of how the observer interprets them. Information about attributes not associated with descent can also be visible. Perhaps the most visible of non-descent-based given attributes is age, always visible in speech and features. Education, income and place of residence (rural or urban) are sometimes visible in accent and appearance, but typically not in the name or features (unless they are correlated with descent). Profession or organizational membership can be visible in dress (the suits that lawyers and bankers wear, the uniforms of policemen and firemen, badges and rosettes signaling party membership etc) but not when an individual is off-duty and also not in the name or features. But on average, attributes not based on birth are less visible and less frequently visible. For instance, Helens occupation as a food-service worker is signaled during working hours by her uniform. And when she is wearing her uniform, or when she is at work, her income and educational background may be inferred from her occupation. But when she is dressed differently and/or off duty her non-birth given attributes immediately become less visible. In contrast, some information about her descent-based attributes is always visible, no matter what the context of observation, in her name and features. Here, as above, there is an overlap in the degree of visibility associated with both types of attributes, probably larger than the area of overlap associated with the stickiness of both types of attributes. Gang membership is not given at birth, but, when proclaimed in tattoos, haircut and other marks on the body, can sometimes be more visible, and more frequently visible, than many descent-based attributes. And the religion of ones parents, while given at birth, can often be less visible than ones educational qualifications or age. The link between the property of attributes and categories is straightforward in this case. When attributes are visible, membership in the categories based on them will also be visible. But note that visibility can coexist comfortably with multiple identities and with error (Chandra 2004). The information a person displays on her person can make several of your identity categories visible, not just a single one. And even though information on an individuals declared and potential ethnic memberships is visible, observers can often get it wrong. Lets return to Helen to illustrate. Two of the ethnic categories in which she is eligible for membership are West-Indian, and Black, made visible by her skin colour, hair and features. But observers might well miscode her as Nigerian or

19

Brazilian. But the main point is that our ethnic categories are visible enough, on average, to permit such guesses. Note, further, that the properties of constrained change and visibility apply to all descent-based identities rather than ethnic identities specifically -- and apply to all descent-based identities only on average. Since ethnic categories are only a subset of categories based on descent-based attributes, we should also see constrained change and visibility in some categories which are based on descent-based attributes but which we do not think of as ethnic e.g. gender. But because some attributes associated with descent are either not sticky or not visible, those ethnic categories based on such attributes may not exhibit the properties of constrained change and/or visibility. And finally, since some non-descent based attributes may also be sticky and/or visible, we might also see the properties of constrained change and visibility associated with some categories that are neither ethnic nor descent-based. The two properties identified here need not be exhaustive -- there may well be others that apply precisely and uniquely to ethnic identities as identified here. One promising direction in which to proceed in identifying others is to explore the fact that offspring of the same parents typically share the same set of ethnic identity options. It may well be that because individuals are more emotionally connected to siblings than non-siblings, membership in ethnic categories should arouse greater emotional attachments, or create a greater tendency to demonize ethnic others, or increase the stakes of conflict, than membership in other types of categories, even those based on descent (e.g. gender). By the same logic, if we assume that individuals care more about the wellbeing of siblings than more distant connections, we may see a greater degree of within-group altruism among members of ethnic categories than members of other types of categories.1 This may be a critical difference between ethnic identity categories such as Black, other descentbased identities such as women and non-descent based identities such as Republicans. Even when these other identities are also visible (e.g. women) and sticky (e.g. Republican in the US), they often cut across siblings, and this may be a consequential fact. Such properties are routinely associated with ethnic groups by comparative political scientists but without justification.2 The definition proposed here provides an analytical foundation on which to infer such an association, or establish its absence. 4. Evaluating Theories About the Effect of Ethnic Identity It is startling how few of our previous claims about the effect of ethnic identity are sustainable in light of the intrinsic properties of ethnic identity identified here. These include Caselli and Coleman and Fearons recent work on patronage and ethnic mobilization (Fearon 1999, Caselli and Coleman 2001), Janet Tai Landas work on trading networks (Landa 1994), and my own work on patronage (Chandra 2004). Caselli and Coleman and Fearon argue that there should be an association between patronage politics and ethnic politics because patronage politics favours coalitions based on identities that are hard to change, and ethnic identities are hard to change. This argument is consistent with the property of constrained change identified here. Ethnic
1 2

I was introduced to this idea by Will Le Blanc, a doctoral candidate in the Department of Political Science at MIT. Indeed, it is surprising, given how important descent has been in the definitions of ethnic identity in comparative political science, how few of our explanatory theories actually explore fields such as sociobiology which can tell us something about the properties of groups which keep offspring together. The work of Pierre Van Den Berghe is a prominent exception.

20

identities, I suggest, are not quite as hard to change as Caselli and Coleman and Fearon argue it can be easy for individuals to switch between ethnic identities within the constraint of the underlying set of attributes. But to the extent that change in ethnic identities is more constrained than change in non-ethnic identities, the argument is upheld. Landa (1994) argues that under conditions of uncertainty, the visibility of ethnic cues allows individuals to select trading partners and enforce contracts. Chandra (2004) argues that given the information constrained environment of elections in a patronage-democracy, the visibility of ethnic identities makes them more likely to be activated in voter and party behaviour than non-ethnic identities. While these arguments remain to be tested empirically, their reliance on properties that can be taken to be intrinsic to ethnic identities at least makes them logically sustainable. However, the arguments made even in this handful of works should be read as applying to descent-based categories in general, not ethnic identities in particular. There are constraints not only on changing identity categories based on ancestral language and physical features but also on changing identity categories based on gender, or the party membership or class of ones parents (e.g. descendants of landowners, children of Communists). And, while ethnic identities are visible, on average, so are identities based on gender or age. These theories, therefore, must be read as theories about a larger class of identities than simply ethnic identities. By far the largest number of explanatory claims about ethnicity rest on properties that I have argued above are not intrinsic to ethnic identities in general, such as fixedness, a common culture, and territorial concentration. As such, they cannot be taken as claims about the effect of ethnic identities in general they should apply either to a specific subset of ethnic identities, or claims about the effect of ethnic identities combined with some additional variable. Rather than reading them to mean that Ethnicity is associated with some dependent variable Y, we should read them as claims meaning either that A particular subset of ethnic identities are associated with dependent variable Y or that Ethnicity, along with some other variable X, is associated with the dependent variable Y. Consider, first, a family of theories which link the politicization of ethnic divisions with democratic destabilization. With some variations, the reasoning underlying such arguments is as follows: Democracies are destabilized by the permanent exclusion of some minority groups from power. Ethnic divisions are fixed. Elections based on ethnic divisions, therefore, produce permanent winners and permanent losers based on ethnic demography. Consequently, the politicization of ethnic divisions threatens democratic stability (Horowitz 1985, Rabushka and Shepsle 1972, Chua 2003). The key property of ethnic identity that this argument rests on is fixedness: if ethnic identities are fluid, not fixed, then the other propositions fall through. But we know, based on the definition proposed in this article that fixedness is not an intrinsic property of ethnic identities constrained change is. If we find that ethnic identities consistently acquire fixedness in a democratic context and that fixedness in turn threatens democratic stability, it must be due to some extrinsic variable that interacts with ethnic identity which has not been theorized. Consider, further, the body of work on institutional prescriptions for multi-ethnic democracies. Among the most influential of these prescriptions is the proposition that democratic governments should accord ethnic minority groups collective cultural autonomy (Kymlicka 1995, Lijphart 1977). The reasoning goes as follows: Cultural recognition is an important right. Ethnic minority groups have distinct cultures. Therefore, democratic governments should assign minority groups cultural 21

rights (Kymlicka 1995). Opponents of this proposal take the position that economic rights are more important than cultural rights (Barry 2001). But all sides assume that ethnic groups possess the property of a common culture. As I argued above, no matter how we define a common culture, it is not a distinguishing principle of ethnic groups in general. Rather than thinking of culture as a constant property of ethnic groups, we should reconceptualize it as a variable associated with all types of groups to varying degrees (Wedeen 2002). In their current form, then, these institutional prescriptions cannot be taken to apply to multi-ethnic democracies in general. They should be read and evaluated as prescriptions for democracies in which social groups have acquired a high degree of cultural cohesiveness, whether or not such groups are ethnic. Take, next, the family of arguments linking ethnic identities with various forms of violence. According to one such argument, one cause of civil war between ethnic groups is the security dilemma introduced by the collapse of the state (Posen 1993). The reasoning is as follows: the collapse of the state governing a multi-ethnic society creates an environment analogous to anarchy in the international environment, with ethnic groups analogous to states. In an anarchic environment, ethnic groups, like states, arm themselves out of fear for their own security. But this makes other groups, like states, more fearful and gives them an incentive to arm themselves also. The result is an increased threat of war, reducing everyones security in the long run. War is especially likely the groups in question have a history of rivalry, since this gives each more reason to assume the worst of the other. This argument assumes that ethnic identity categories, like states, are fixed entities for if individuals could change their ethnic identities, then one response to the collapse of the state might be simply to switch to less threatening identities rather than go to war. Second, it implies that ethnic groups are more likely than other types of groups to have a common history. Otherwise, the security dilemma should be an explanation for inter-group conflict in the wake of state collapse in general, rather than ethnic conflict in particular. But we cannot take either fixity or a common history to be intrinsic properties of ethnic identities. Ethnic identities can change even in the short term as individuals combine and recombine elements from their fixed set of attributes differently. Further, a common history does not distinguish ethnic groups, but is a product of some process by which ethnic groups are mobilized. This argument, thus, cannot be read as an argument about the effect of ethnic identities per se. The effect of ethnic identities here is contingent upon some extrinsic, omitted, variable that produces fixity in ethnic identities and a perception of a common history, which should be included in the analysis. According to another argument in this family, ethnic violence is motivated by emotions such as fear, hatred or resentment (Petersen 2002). The initial trigger for such violence, according to this argument, lies also in the collapse of the state. But given this trigger, the target of violence will depend upon the specific emotional response aroused among ethnic groups. Groups motivated by fear will choose as their target those ethnic others who are the greatest threat. Groups motivated by resentment will choose as their target those ethnic others who are farthest up the status hierarchy whether or not they are the greatest threat. And groups motivated by hatred will target those ethnic others with whom they have battled in the past, regardless of their threat potential and their position on the status hierarchy.

22

If this argument is to be read, as it is intended, as an argument about ethnic violence rather than violence in general, then ethnic groups must have the property of experiencing stronger, or distinct, emotional reactions in the wake of state collapse than other types of groups. Neither the definition of an ethnic group that I have proposed here, nor the alternative definitions, justify this assumption. This does not mean that this argument is wrong but it does mean that t is incomplete. In order to sustain the argument, we would have to demonstrate why we can reasonably associate individuals to have distinct emotional associations with their ethnic identities which they do not with other types of identities. Another argument in this family theorizes that intra-group policing can prevent inter-ethnic violence. One of the assumptions underlying this argument is that ethnic groups are distinguished by the property of dense network ties, which make within-group cooperation easier than across-group cooperation. As Fearon and Laitin put it: Why are inter-ethnic relations frequently characterized by a tension that is relatively absent in intra-ethnic relations, giving interethnic relations, even when peaceful, an ominous quality? Because social networks are better developed and interactions more frequent within ethnic groups, individuals have easier access to information about their coethnics than they do about ethnic others. (Fearon and Laitin 1996, 730). But we cannot infer this property based on the definition that I have proposed here and the classification of the ethnic categories that it justifies. In order for this argument to be plausible, thus, we would have to demonstrate either that such ties are indeed intrinsic to the groups we classify as ethnic or identify some omitted variable that creates social networks within ethnic groups and not otherwise. Consider a final argument, which explains the association between modernization and ethnic politics by invoking the property of territorial concentration (Bates 1974). The reasoning is as follows: The goods of modernity are distributed on a spatial basis. It makes sense for individuals desiring access to these goods, therefore, also to organize on a spatial basis. According to the author, there is no denying that the members of an ethnic group tend to cluster in space (Bates 1974, 464). The struggle for access to these goods, therefore, is organized on an ethnic basis. The argument is informed by the study of ethnic politics in Africa, and it certainly appears to be the case that several ethnic categories in Africa are territorially concentrated. But we cannot take either territorial concentration or the memory of a common territorial homeland to be an intrinsic property of ethnic categories in general. And we can certainly imagine non-ethnic categories that are territorially concentrated. This argument, thus, should be read, not as a general argument about ethnic identities but as an argument about territorially concentrated identities, a concept which has an imperfect overlap with a subset of ethnic identities. This is only a quick survey of a few influential examples of the body of work that theorizes about the effect of ethnic identities, or concepts related to ethnic identities. A more extensive and more detailed discussion of several such arguments can be found elsewhere (Chandra 2005, Chandra 2001a, Chandra 2001b). The broad point I wish to make here is not that these arguments are wrong but that we do not so far have reason to believe that they are right. These families of theories have not demonstrated on analytical grounds that ethnic identity categories, as they classify them, should have an explanatory effect on the outcomes of interest. In order for these theories to be sustainable, we must show at a minimum that the properties they take to be intrinsic to ethnic identity can actually be inferred from the definition proposed here, or a better one.

23

Ultimately, if we are not able to identify any further properties that are unique to ethnic identity, we would be better off substituting the concept of ethnic identity in our theories with concepts such as descent-based identities or identities based on sticky or visible attributes. But if we do, does this mean that this attempt at definition was wasted? To the contrary. The negative claim, that ethnicity does not matter, is a discovery of great magnitude. It should have far-reaching consequences for research and data collection, suggesting that we should abandon the large number of theories and datasets that we have on ethnicity and start again on an entirely different foundation. A claim of this magnitude cannot be made lightly. It must be justified perhaps even more strongly than the claim that ethnicity matters. And, just as we need to define the concept of ethnic identity to establish that it matters, we also need a definition in order to establish that it does not matter.

24

LITERATURE CITED Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. (2003). Fractionalization Journal of Economic Growth. 8.2: 55-94. Appiah, Anthony (1992). In my Fathers House: Africa in the Philosophy of Culture. New York: Oxford University Press. Appiah, Anthony and Amy Gutmann (1996). Color Conscious. Princeton NJ: Princeton University. Atlas Narodov Mira (1964). Moscow: Miklukho-Maklai Ethnological Institute of the Department of Geodesy and Cartography of the State Geological Committee of the Soviet Union. Barth, Frederik (1969). Ethnic Groups and Boundaries. Prospect Heights, IL: Waveland Press. Bates, Robert (1974). Ethnic Competition and Modernization in Contemporary Africa. Comparative Political Studies. Bhabha, Homi (1994) The Location of Culture. London: Routledge. Bhabha, Homi, ed. (1990), Nation and Narration. London: Routledge. Bringa, Tone (1995). Being Muslim the Bosnian Way. New Jersey: Princeton University Press. Brubaker, Rogers, Mara Loveman and Peter Stamatov (2003), Reviving Constructivist Theory: The Case for a Cognitive Approach to Race, Ethnicity and Nationalism. Manuscript. Carr, Jack L. and Janet T. Landa (1994), The Economics of Symbols, Clan Names and Religion. In Janet Tai Landa, Trust, Ethnicity and Identity. Ann Arbor: University of Michigan Press: 115140. Caselli, Francesco and Wilbur John Coleman (2001), On the Theory of Ethnic Conflict. Manuscript, Harvard University Press. Cavalli Sforza, Luigi (2000), Genes, Peoples and Languages. New York: North Point Press. Chandra, Kanchan (2005). Ethnic Parties and Democratic Stability. Perspectives on Politics, June 2005, Vol 3(2): 235-252. Chandra, Kanchan (2004). Why Ethnic Parties Succeed. Cambridge: Cambridge University Press. Chandra, Kanchan (2001a). Ethnic Bargains, Group Instability and Social Choice Theory. UPolitics and SocietyU. Volume 29 (3): 337-362. Chandra, Kanchan. (2001b). Cumulative Findings in the Study of Ethnic Politics. UAPSA-CPU. Vol. 12 (1): 7-11. 25

. Chua, Amy (2003). World on Fire. New York: Doubleday. Cox, Gary (1997). Making Votes Count. Cambridge University Press. Dahl, Robert A (1971). Polyarchy. New Haven: Yale University Press. Dirks, Nicholas (2001). Castes of Mind. Princeton NJ; Princeton University Press. Dominguez, Virginia (1997). White by Definition. New Brunswick: Rutgers University Press. Fearon, James (2003). Ethnic Structure and Cultural Diversity by Country Journal of Economic Growth 8.2: 195-222. Fearon, James and David D. Laitin (1996), Explaining Interethnic Cooperation. American Political Science Review, Volume 90, Issue 4: 715-735. Fearon, James. (1999). Why Ethnic Politics and Pork Tend to Go Together. Paper presented at SSRC-Macarthur Workshop on Ethnic Politics and Democratic Stability, University of Chicago. Fearon, James and David D. Laitin (2000), Ordinary Language and External Validity: Specifying Concepts in the Study of Ethnicity. Paper presented at the October 2000 meeting of LICEP, University of Pennsylvania. Geertz, Clifford (1973). The Interpretation of Cultures. Basic Books. Griffin, John Howard (1996). Black Like Me. New York: Signet. Kaufmann, Chaim (1996). Possible and Impossible Solutions to Ethnic Civil Wars. International Security, Vol. 20, No. 4): 136-175. Hale, Henry (2004). Explaining Ethnicity. Comparative Political Studies, Vol. 37, No. 4: 458-485. Hall, Stuart and Paul du Gay, eds (1996). Questions of Cultural Identity. London, Sage Publications. Hall, Stuart (1996). Who Needs Identity? in Stuart Hall and Paul du Gay, eds. Questions of Cultural Identity. London, Sage Publications: 1-17. Horowitz, Donald (1985). Ethnic Groups in Conflict. Berkeley: University of California Press. Htun, Mala (2004). The Political Representation of Identity Groups. Perspectives on Politics, September. Jung, Courtney (2000). When I was Black. New Haven: Yale University Press. Hutchinson, John and Anthony D. Smith (1996), Ethnicity. Oxford: Oxford University Press. 26

Kymlicka, Will, ed (1995).,. The Rights of Minority Cultures. Oxford: Oxford University Press. Kymlicka, Will (1995). Multicultural Citizenship. Oxford: Oxford University Press. Laitin, David (1986). Hegemony and Culture. Chicago: University of Chicago Press. Laitin, David (1998). Identity in Formation. Ithaca: Cornell University Press. Landa, Janet Tai (1994). Trust, Ethnicity, and Identity. Ann Arbor: University of Michigan Press. Larsen, Nella (1997). Passing. Penguin. Lemann, Nicholas (1982). The Promised Land: The Great Black Migration and How it Changed America. New York: Vintage Books. Malcolm X (1964) The Autobiography of Malcolm X. New York, Ballantine Books. Malouf, Amin (2000). In the Name of Identity: Violence and the Need to Belong. New York: Penguin. Minorities at Risk Database (http://www.cidcm.umd.edu/inscr/mar/) Nobles, Melissa (2000). Shades of Citizenship. Stanford CA: Stanford University Press. Ordeshook, Peter and Olga Shvetsova (1994), Ethnic Heterogeneity, District Magnitude, and the Number of Parties. American Journal of Political Science, 38, 1: 100-123. Petersen, Roger (2002). Understanding Ethnic Violence. Cambridge University Press. Posen, Barry (1993). The Security Dilemma and Ethnic Conflict. In Michael Brown ed., Ethnic Conflict and International Security, Princeton NJ: Princeton University Press. Posner, Daniel (2005). Institutions and Ethnic Politics in Africa. Cambridge University Press. Przeworski, Adam et al (2000). Democracy and Development. Cambridge University Press. Rabushka, Alvin and Kenneth Shepsle (1972), Politics in Plural Societies. Columbus, Ohio: Charles E. Merrill. Reilly, Benjamin (2001). Democracy in Divided Societies. Cambridge University Press. Sambanis, Nicholas (2004). What is an Ethnic War? Manuscript. Toft, Monica (2003). The Geography of Ethnic Violence. Princeton NJ: Princeton University Press. Varshney, Ashutosh (2002). Ethnic Conflict and Civic Life. New Haven: Yale University Press. 27

Waters, Mary (1999). Black Identities. New York: Russell Sage Foundation. Weber, Max (1996). The Origins of Ethnic Groups. In John Hutchinson and Anthony D. Smith eds., Ethnicity. Oxford: Oxford University Press: 35-40. Wedeen, Lisa (2002). Conceptualizing Culture: Possibilities for Political Science. American Political Science Review, Vol. 96, No. 4. Wilkinson, Steven (2004). Votes and Violence: Electoral Competition and Ethnic Riots in India. Cambridge University Press.

28