Sources of Ethnic Identification in Africa*

Edward Miguel University of California, Berkeley and NBER Daniel N. Posner University of California, Los Angeles

26 January 2006

This paper draws on data from over 24,000 respondents in twelve African countries to investigate the sources of ethnic identification in Africa. Contrary to popular assumptions that Africans are intrinsically and uniformly “ethnic” people, we find that just 41 percent of respondents rank their ethnic group as their most important associational membership, and that this share varies considerably across countries. Within countries, we find the sources of ethnic identification to lie in exposure to competition for jobs and political power. In particular, we find the salience of ethnic identities to be positively related to employment in non-traditional economic sectors and to the proximity of the survey to a competitive national election. We also find the salience of ethnicity to be negatively related to ethnic diversity – a result that challenges widely held assumptions in the literature. The results offer some of the first comprehensive cross-national evidence for the proposition that ethnic identification is a product of “modernity” rather than tradition.

*

This paper builds on earlier work co-authored with Alicia Bannon.. The authors thank Daniel Young and Elizabeth Carlson for their research assistance; the U.S. National Science Foundation (Miguel, SGER-#0213652) and the Academic Senate at UCLA for their support; members of the Working Group in African Political Economy; seminar participants at the Leitner Political Economy Seminar at Yale University; and the editors of the Afrobarometer Working Paper Series for their extremely helpful comments on an earlier version of the paper.

Introduction In the popular imagination, Africans are deeply and uniformly ethnic people. Ask and African “who she is,” most people assume, and you will get an ethnic response: “I am Yoruba,” “I am Kikuyu,” “I am Baganda.” Moreover, ask most people why ethnicity is so salient in Africa and they will tell you that it is because Africans are so “backward.” Once Africans become more educated and urbanized – in short, more “modern” – it is assumed that ethnicity will cease to cause so much conflict, distort so many elections, and pervert so many public policies. The notion that ethnic identities are vestiges of a pre-modern past that are destined to wither away in the face of modernization dates to the writings of Marx, Weber, Durkheim, and Parsons, all of whom predicted that ethnicity would eventually be replaced by class as the organizing principle of social life. In these authors’ views, industrialization and universalism would lead to the displacement of ethnicity, “which could hardly be expected to survive the great tidal wave of bureaucratic rationality sweeping over the Western world” (Weber, quoted in Parkin 1978). As late as the 1950s and 1960s, “it was accepted that…parochial ethnic loyalties were mere cultural ghosts lingering on into the present, weakened anomalies from a fast receding past” and that these loyalties “were destined to disappear in the face of the social, economic and political changes that were everywhere at work” (Vail 1989: 1). Notwithstanding Vail’s claim that “people from all sectors of the political spectrum believed in this vision,” a wave of scholarship began to emerge in the 1960s that disputed the expectations of this early body of modernization theory. This revisionist wave was spearheaded by researchers like Crawford Young (1965), P.C. Lloyd (1967), Robert Melson and Howard Wolpe (1970), P.T. Gulliver (1971) and Robert Bates (1983), whose views were shaped by their first hand experiences in Africa. Their research showed that, rather than cause ethnic identifications to “disappear into museums” (Davidson 1992: 100), the processes of urbanization, industrialization, education, political mobilization, and competition for jobs would deepen ethnic identities as individuals exploited their

1

ethnic group memberships as tools for political, economic, and social advancement. Few serious scholars today would dispute this (what might be termed) “second wave” modernization view. Yet it is still, by and large, not reflected in either media representations of or popular assumptions about Africa. This paper draws on survey data from over 24,000 respondents in sixteen surveys in twelve African countries to test systematically these two different understandings of the origins of ethnic identifications in Africa. Our results provide strong support for the “second wave” modernization perspective. We show first that ethnicity is not nearly as central to Africans’ conceptions of who they are as is frequently assumed. We then investigate the factors that predispose individuals to identify themselves in ethnic terms. We examine both individual- and country-level characteristics and find support for the proposition that the sources of ethnic identification lie not in economic or political backwardness, but rather in “modernity.” Specifically, we find that exposure to political mobilization and working in non-traditional occupations that expose people to competition for employment increases the likelihood that individuals will see themselves primarily in ethnic rather than non-ethnic terms. Strikingly, we find that the closer a country is to a competitive national election, the greater the likelihood that respondents from that country will say that they identify themselves ethnically first and foremost. In addition, contrary to a central assumption in much of the literature on ethnic diversity, we also find a robust negative relationship between country ethnic fractionalization and ethnic salience. Apart from the support these findings provide for the association between modernization and deepening ethnic identification, they also provide strong empirical confirmation for well established situational and instrumental approaches to ethnicity. The situational nature of ethnic identifications is borne out by the finding that the salience of ethnicity varies both across and within African countries, and that it does so in predictable ways. The instrumentality of ethnic identifications is suggested by the finding that competition for political representation (as evidenced by proximity to a

2

competitive election) and for jobs (as evidenced by participation in the “modern” economy) tend to increase the likelihood that a person will identify him or herself in ethnic terms. Though perhaps startling to consumers of popular and journalistic accounts of African affairs, our findings will not be surprising to most professional Africanists or most students of ethnicity. This paper’s contribution thus lies not in the novelty of its theoretical arguments but in its presentation of systematic, crossnational evidence to support a set of claims that, while well known and widely accepted in contemporary academic circles, have never been convincingly established empirically.

Data and Measurement To investigate the sources of ethnic identification in Africa, we employ data collected in rounds 1 and 1.5 of the Afrobarometer, a multi-country survey project that employs standardized questionnaires to probe citizens’ attitudes in new African democracies. The surveys we employ were administered between mid-1999 and late-2002. Nationally representative samples were drawn through a multi-stage stratified, clustered sampling procedure, with sample sizes sufficient to yield a margin of sampling error of ±3 percentage points at the 95 percent confidence level.1 The Afrobarometer surveys include a question designed to gauge the salience for respondents of different group identifications. The question is worded as follows:

We have spoken to many [people in this country, country X] and they have all described themselves in different ways. Some people describe themselves in terms of their language, religion, race, and others describe themselves in economic terms, such as working class, middle class, or a farmer. Besides being [a citizen of X], which specific group do you feel you belong to first and foremost?

1

Further details of the Afrobarometer project, including the sampling procedures used in collecting

the data, are available at the project’s web site: www.afrobarometer.org.

3

The main dependent variable used in the analyses in the paper – the measure of “ethnic salience” – is constructed from the responses to this question by 24,815 respondents in sixteen separate Afrobarometer survey rounds conducted in twelve countries that span the continent geographically: Botswana, Ghana, Malawi, Mali, Mozambique, Namibia, Nigeria, South Africa, Tanzania, Uganda, Zambia, and Zimbabwe.2 We group respondents’ answers to the “which group do you feel you belong to first and foremost” question into four categories: ethnic, religion, class/occupation, and other. We define ethnic responses in two ways. A first, narrow, definition equates ethnicity with tribe (or sub-tribe) and language. However, recognizing that other non-tribal and non-linguistic identities – e.g., race in South Africa, region in Malawi, religion in Nigeria – are sometimes also understood in ethnic terms, we prefer coding our dependent variable in a more encompassing, way to include these other categories as ethnic responses as well.3 We employ this second, broader, definition in the analyses reported in the main paper (though our findings are robust to using the narrower definition). 2

We use data from two different survey rounds in Namibia, Nigeria, South Africa, and Uganda.

Lesotho, for which an Afrobarometer survey from 2000 is available, was excluded because the name of the country nationality is identical to the name of the country’s dominant ethnic group, the Sotho (who comprise 99.7 percent of the population, CIA 2003). Thus the survey question (which asks, “besides being Mosotho, which group do you feel you belong to first and foremost?”) rules out an ethnic response by definition. Cape Verde, for which a 2002 Afrobarometer survey is available, was excluded because of missing data on ethnic fractionalization in the cross-country datasets. 3

In addition to language and (sub-) tribe, we code race as an ethnic response in the former settler

colonies of Mozambique, Namibia, South Africa, Zambia, and Zimbabwe; region as an ethnic response in Malawi and Nigeria; and religion as an ethnic response in Nigeria. In Table 2, race is grouped under “other” in the non-former settler colonies.

4

Before turning to the findings, several qualifications of the analysis bear mention. First, the salience of any social identification – be it ethnic or otherwise – is necessarily context specific, so we must be clear from the outset that our goal is not to pinpoint the factors that would lead respondents to define themselves in ethnic terms at all moments or in all settings. The Afrobarometer data only permits us to ascertain the way respondents identified themselves in the specific context in which they were being surveyed. Our task is to use what we know about that context (in particular, when and where the survey was administered) to make inferences about the factors that determine when ethnic group memberships become most salient. The context-specificity of respondents’ answers is not something we ignore, but is built right into the project design.4 Second, quite apart from the issue of the reliability of responses across contexts, the use of self-reported identities introduces the possibility of bias. Respondents in societies where the social norm is not to talk openly about ethnicity might be less likely to confess that their most important social affiliation is with their ethnic community, and this would generate a downward bias in measured ethnic salience in that society. This may be particularly likely if the survey enumerator is not a co-ethnic or in a context where open confessions of ethnic solidarity are frowned upon by the regime and where the enumerator is suspected of being affiliated with the government. While this concern cannot be ruled out, it is dampened by the way the Afrobarometer survey was conducted – confidentially and in private by enumerators who are not affiliated with the government or any political party. Also, and quite importantly, the Afrobarometer survey is not primarily about 4

Of course, there are aspects of the context in which respondents were surveyed for which we cannot

control, such as enumerator ethnic background (about which we have no information) or the proximity of the survey interview to religious festivals, harvest times, and other events that might have caused some identities to become momentarily more salient. In any case, such idiosyncratic situational factors should make it harder for us to find statistically significant relationships.

5

ethnicity. The question we use to construct our measure of ethnic salience is just one of more than 175 questions asked in the standard Afrobarometer questionnaire, only a handful of which make any mention of ethnicity. Respondents are thus likely to have treated the “with which group do you identify” question as a background query rather than as the central issue around which the survey revolved. We therefore expect that respondents were probably less guarded in their responses about their ethnic identities than might otherwise have been the case. In addition, to the extent that social norms against confessing the strength of one’s ethnic identification vary by country, the country controls we include in our regressions may allow us to partially control for these differences. Two additional concerns stem from the way the ethnic salience question was structured. A first issue is that the question explicitly bars respondents from describing themselves in terms of nationality: it asks “besides being your nationality [e.g., Namibian, Zambian, etc], which specific group do you feel you belong to first and foremost?” We therefore cannot rule out the possibility that respondents might consider national identity as more important to them than the identity categories recorded in our data. A second issue is that the question provides information about the salience of ethnicity in relative, not absolute, terms. All we are able to infer from respondents’ answers is the identity that they rank first among those identity categories explicitly mentioned in the survey question. We have no way of knowing how much importance respondents attach to their first- (or second- or third-) ranked group memberships. Thus to conclude on the basis of our data that ethnicity is more salient in country A than country B because a larger share of survey respondents in country A ranked ethnicity first is not quite right. It is conceivable, though we think unlikely, that ethnicity might be more salient in absolute terms to people in country B, even though a larger share of them rank some other category of identity as even more important than ethnicity. Finally, legitimate concerns can be raised about the generalizability of our findings. Although broadly representative of Africa as a whole, the twelve countries included in our study are still not a perfect substitute for a continent-wide sample. Our sample includes just one Francophone

6

and one Lusophone country (Mali and Mozambique, respectively), no countries that have failed to introduce at least some democratic or market reforms over the last decade, and, with the exception of Uganda, no countries currently involved in civil wars. As Table 1 indicates, per capita income in the twelve countries are nearly double the African average (though this is mainly driven by the cases of Botswana, Namibia, and South Africa – the other nine countries are actually poorer than the African average). Along the same lines, rates of under-5 child mortality in our sample are slightly lower than in Africa as a whole. Both urbanization and ethnic fractionalization – a measure of the likelihood that two people chosen at random from the country will be from different ethnolinguistic groups – are roughly comparable to what is found elsewhere on the continent. But citizens in the twelve sample countries enjoy slightly more extensive political rights than the average African country (note that on the Freedom House scale, which runs from 1 to 7, lower numbers indicate greater rights). Our findings therefore must be interpreted with the caveat that they may not be entirely representative of Africa as a whole. TABLE 1 HERE This caveat notwithstanding, the measure of ethnic salience adopted in this paper represents an advance over those employed in earlier studies, almost none of which measure ethnic salience directly.5 Most studies that deal with this issue rely on inferences based on the presumed effects of ethnic salience. In effect, they reason that, because there is ethnic violence in the country in question or because voting patterns or the distribution of patronage appears to follow ethnic lines, ethnicity must be a salient motivating factor in people’s behavior. Others rely on assumptions about what the diversity of ethnic groups in a country implies about the salience of ethnicity in the country’s politics – a relationship that we demonstrate below works opposite from the often hypothesized direction. 5

Mattes (2004), who also uses Afrobarometer data and adopts a methodology similar to our own, is

an exception.

7

Neither approach is as defensible as the one pursued here, which bases its assessment of ethnic salience on the self-reported identities of individuals as collected in nationally representative sample surveys.

The Salience of Ethnic Identities Table 2 reports the frequency distribution of responses to the “which specific group do you feel you belong to first and foremost” question for all respondents taken together, and then broken down for each of the sixteen surveys in our sample. Contrary to the stereotype that Africans are intrinsically “ethnic” people, a minority of 41 percent of the respondents in the twelve countries identified themselves first and foremost in ethnic terms.6 Indeed, only slightly more respondents chose “ethnic” identities than “class/occupation” identities, which were chosen by 37 percent of respondents. This pattern is even stronger when we employ our narrower definition of ethnicity, which limits “ethnic” groups to language groups and tribes (see Table A1 in the Appendix). In addition, responses vary tremendously across countries and (to a lesser degree) within countries over time. Thus, whereas 92 percent of respondents in Botswana identified themselves “first and foremost” in ethnic terms, only 3 percent of respondents did so in Tanzania.7 And whereas 41 percent of South African respondents identified themselves in ethnic terms in the 2000

6

Note that the “all respondents” row weights each survey round equally, so that respondents from

countries with larger sample sizes are weighted less. The raw (unweighted) share of respondents identifying in ethnic terms is 40 percent and the share when weighting each survey round by country population is 44 percent. 7

In Botswana, where approximately 80 percent of the country’s population is Setswana, ethnic

responses were in terms of sub-tribes (i.e., Mongwato, Mokweme, Mokgatla, and so forth).

8

Afrobarometer survey, just 21 percent did so in the 2002 survey.8 This diversity of responses across countries and over time suggests that both the national context and the events of the moment matter for the salience of ethnicity, points we return to shortly. TABLE 2 HERE If Africans are not uniformly “ethnic” people, what makes some Africans more likely to identify in ethnic terms? As we have noted, there is a curious gulf between popular and scholarly answers to this question. While journalists regularly invoke the “ancient” nature of ethnic divisions in Africa, forty years of scholarship has shown that ethnic identities are products of modernization rather than an atavism that modernization will cause to disappear. We investigate the sources of ethnic identification using the Afrobarometer dataset in two stages. First, we address individual-level determinants. Then we turn to the country-level factors that predispose individuals to identify themselves in ethnic terms. Descriptive statistics for the data used in these analyses are reported in Table 3. TABLE 3 HERE

Individual-Level Sources of Ethnic Identification

8

The surveys are repeated cross-sections rather than panels, so we cannot rule out the possibility that

sampling variation is behind some of the change between 2000 and 2002. However, both South Africa survey rounds (as well as both rounds in Namibia, Nigeria, and Uganda) employed the same sampling methodology and are nationally representative in terms of age, gender, and urban-rural location, so given the large sample of individuals we can reject (at high levels of confidence) the hypothesis that sampling variation is behind the shift. We control for individual-level factors in the regression analyses below to further control for any changes in sample composition over time.

9

The key individual-level determinants of ethnic identification that we investigate are gender, age, education level, occupation, urban/rural location, and media exposure (as proxied by how often a respondent gets his or her news from the radio or from newspapers).9 Given the high degree of variation in reported ethnic salience across countries, we first employ country fixed effects in Table 4, column 1 (our preferred specification for analyzing the impact of individual-level variables) and later use extensive country controls when we move from individual- to country-level factors in subsequent analyses. In all specifications, we cluster regression disturbance terms at the country level and weight each observation by 1/(number of observations from that country) in order to weight each country survey round equally. The main effect of including the country population weights is to reduce the influence of Nigeria and South Africa, which together account for 38 percent of the 24,815 respondents our analysis.10 TABLE 4 HERE Gender and age have no statistically significant effects on ethnic identification. Education, however, has a discernible effect, even when controlling for other variables like occupation. Compared to people with at least some primary education (the omitted education category), respondents with no formal education are roughly 4 percentage points less likely to say that they feel they belong to an ethnic group first and foremost. Even a small amount of formal education (that is, 9

Questions about respondents’ household incomes are only included in some of the Afrobarometer

surveys, so the effect of wealth can only be explored directly at the country level, as we do in the next section. However, some of the controls we employ (e.g., for education and occupation) are correlated with income, so some indirect inferences about the impact of income can be made. 10

The main results reported below are robust to the exclusion of the country population weights.

They are also robust to replicating the analysis using the use of the narrower definition of ethnicity (regressions not shown).

10

moving from the “no education” to the “some primary schooling” category) significantly increases the likelihood that a respondent will identify him or herself in ethnic terms. Beyond this minimal education level, additional schooling has no statistically significant impact on ethnic identification – at least until a respondent has finished secondary school, at which point the relationship reverses. Respondents with at least some post-secondary education are more than 7 percentage points less likely to identify themselves in ethnic terms than those with just some primary schooling. The data thus suggests a nonlinear relationship between education and the salience of ethnic identity: those at the lowest and highest levels of the educational attainment continuum are least likely to identify themselves in ethnic terms, whereas those in the middle ranges are most likely to do so, with ethnic identification effects strongest for those who have had some post-secondary schooling. Arguably consistent with the curvilinear relationship between education and ethnic salience, we also find that respondents at very high levels of media exposure (those who get their news from the radio daily and/or from a newspaper at least weekly) were less likely than those at low and moderate levels of media exposure to identify themselves in ethnic terms. Running somewhat against the grain of the other results, and contrary to the expectations of “second wave” modernization theory, our indicator variable for whether or not a respondent is located in a rural location is small and not statistically significant.11 We speculate that this may be due to the fact that rural location is only a rough proxy for participation in non-traditional economic sectors, i.e., teachers, factory workers, and government officials, and people with a range of educational attainment, are well-represented in both rural and urban areas. We thus focus on more precise individual-level characteristics.

11

The variable remains insignificant if we drop the occupation indicator variables, so this finding

does not appear to be due to the collinearity of occupation and urbanization.

11

The salience of ethnicity does vary strongly with individual occupation. Compared to farmers and fishermen (the omitted occupational category in the regression), workers in the modern sector – be they white collar workers, blue collar workers, miners, students, business people, or the unemployed – are significantly more likely to identity themselves in ethnic terms. Respondents in all of these occupational categories were between 6 and 11 percentage points more likely than those in the traditional sector (farmers and fishermen) to volunteer an ethnic membership when asked to specify the group with which they identify first and foremost. Our interpretation of this pattern is that stronger ethnic identification among respondents in the modern sector stems from the competition that such individuals face for scarce jobs and contracts and the role that ethnic connections commonly play in securing advantages in this competition. This linkage between job competition and patronage linkages is corroborated by Bratton et al (2004: 158), who find in their own cross-national analysis of the Afrobarometer data that people in trading occupations are significantly more likely than others to make use of informal patronage ties. Taken together, the findings are broadly consistent with the hypothesis that individuals entering the “modern” sector – that is, individuals who have at least some education and who have left behind livelihoods of farming or fishing to work in non-traditional occupations – are more prone to identify themselves in ethnic terms. We hypothesize that this is because such individuals are more subject than their counterparts in the traditional sector to competition for employment and income, and thus more likely to view their fate as bound up with the ethnic affiliations that so frequently determine who gets jobs, contracts and other forms of government patronage in Africa. However, we also find suggestive evidence that the effect of exposure to modern currents and competitive pressures changes sign at the very highest levels (notably, among those with post-secondary education). Ethnicity would appear to be least salient for those at the bottom and top of the “modernity” spectrum and most salient for those moving through its ranks.

12

Country-level Sources of Ethnic Identification Our country-level findings broadly reinforce and extend the individual results presented thus far. First, we find strong support for the thesis that political competition makes ethnicity more salient: the proximity to and competitiveness of national elections are associated with a higher likelihood that respondents will identify in ethnic terms. Second, and more provocatively, we find that a higher degree of country ethnic fractionalization is associated with less ethnic salience – a result that runs counter to the assumptions of scholars who use ethnic diversity as a proxy for the importance of ethnicity in political and social life. The key limitation of the findings presented in this section is that they are based on an analysis of only twelve countries (or, for the electoral proximity analysis, sixteen survey rounds). Data limitations make this shortcoming impossible to overcome until further rounds of Afrobarometer are made publicly available. We again cluster regression disturbance terms at the country level to ensure that standard errors reflect the true sample size rather than the much larger number of individual observations in our regressions. While this prevents us from having false confidence in our findings, it of course does not solve the fundamental small sample size problem. We present three different country-level regression specifications in Table 4. In column 2, we focus on political factors. In column 3, we focus on ethnic diversity. In column 4, we combine all the country-level variables in a specification that has the virtue of being the most controlled but the drawback of pushing the limits of the number of explanatory variables that our small country sample will allow, due to limited degrees of freedom. We thus present column 4 mainly for the purpose of completeness and as a robustness check. All of our country-level regressions include a measure of country wealth (log of per capita income) as a basic control. The variable is always positive, but never statistically significant at conventional levels. The interpretation of the coefficient estimate in column 2 is that a doubling of a country’s per capita income is associated with a 12.5 percent increase in the share of the population

13

identifying itself in ethnic terms. Given the moderate size (and lack of statistical significance) of this effect, we hesitate to read too much into this result. We do, however, note that it is in keeping with the hypothesis that economic development will make ethnicity more salient rather than less. A much stronger finding, and one of the two key results in the country-level analysis, is that the salience of ethnicity is closely related to the intensity of political competition. We find that the proximity of the timing of the Afrobarometer survey to a national election (defined as the absolute value of the number of months since or until the closest presidential contest12) is a highly significant explanatory variable: the more proximate the survey is to an election (i.e., the smaller the time gap), the more likely respondents are to respond that they view themselves first and foremost in ethnic terms (Table 4, column 2).13 The interpretation of the electoral proximity coefficient is that a respondent in a country that is holding an election at the time of the survey will be more than 30 percentage points more likely to identify him or herself in ethnic terms than a respondent in a country whose election took place a year ago or whose election is scheduled to take place a year hence – a truly massive effect. Figure 1 plots electoral proximity against average ethnic salience for our sixteen survey round cases. FIGURE 1 HERE 12

In Botswana and Zimbabwe, the electoral proximity variable is calculated in terms of the number

of months before/after the most proximate parliamentary election. In the case of Botswana this is because the country does not hold presidential contests; in the case of Zimbabwe it is because presidential and parliamentary elections are not held concurrently, and the most proximate national election to the Afrobarometer survey we use was the parliamentary contest of June 2000. 13

We find no statistically significant difference between the effect of time since an election versus

time before an election on the extent of ethnic identification (regression not shown), so we focus on a measure that treats time before versus after an election symmetrically.

14

These cross-sectional findings are corroborated by a closer look at variation over time within the Niger Delta of Nigeria. Drawing on the results of three different Afrobarometer survey rounds, Lewis (2004) finds marked volatility in the share of respondents who identify themselves in ethnic terms in the Niger Delta. In the January 2000 Afrobarometer survey, 67 percent of respondents in this region said that they belonged to an ethnic or linguistic group first and foremost. A year and a half later, in the August 2001 survey, this share dropped to 42 percent. But in the next survey round, in October 2003, it jumped back up to 90 percent. How might we account for these wide swings in ethnic salience? Lewis surmises that the answer may lie in political circumstances. This hunch is borne out by Figure 2, which reveals a strong positive association between the salience of ethnicity and the proximity of an election. The slope of the line in the Figure suggests that for every month closer that the survey was to an election, the share of respondents who identified themselves in ethnic terms rose by roughly 3 percentage points. Of course, with just three data points, this example and estimate can only be illustrative. But the association is consistent with the cross-country result that electoral proximity may play a key role in determining ethnic salience. FIGURE 2 HERE Also consistent with the broad finding that ethnic identities are products of political mobilization is the finding (see Table 4, column 2) that respondents are more likely to identify themselves in ethnic terms when elections are competitive, which we define in terms of the margin of victory between the winning presidential candidate and the candidate with second highest vote share in the most proximate election.14 The logic here is that the more competitive the election, the more likely politicians are to play the ethnic card to secure a marginal advantage in the competition for votes, and the more likely citizens are, as a consequence, to identify themselves in ethnic terms. The 14

In Botswana and Zimbabwe, the margin of victory is calculated in terms of the gap between the

parties with the highest and second highest vote shares in the parliamentary election.

15

results bear this out: the margin of victory coefficient suggests that every one percentage point increase in the gap between the winning candidate and the runner up is associated with more than a full percentage point decrease in the likelihood that a respondent from that country will identify him or herself in ethnic terms. This is both a substantively large and a statistically significant effect, and provides strong empirical support for the thesis that ethnic salience is a product of increased political competition.15 These findings also have clear implications for the design and implementation of survey research on ethnicity in Sub-Saharan Africa, since surveys conducted near election time appear to generate systematically different response patterns. Strikingly, neither the electoral proximity nor the margin of victory findings are as strong when these variables are entered separately without the other: both coefficients decrease by about half in magnitude and drop below traditional levels of statistical significance (regressions not shown) when entered singly. This raises some concerns about the robustness of these results, although note that the magnitudes remain large even in that case and the loss of statistical significance is in part a result of the small country sample.16 15

Both the margin of victory and electoral proximity findings are robust to substituting the narrow

ethnic salience measure and to dropping countries one at a time (regressions not shown). 16

We also include an interaction term between electoral proximity and the margin of victory, the

coefficient estimate on which is positive and significant. While a positive coefficient on this interaction variable might seem to undermine the claim that ethnic salience is affected by the proximity to a competitive election, it turns out that, even with this positive interaction term, the marginal effect of an increase in the electoral proximity variable (i.e., a greater time gap between the survey and an election) is always negative for all values in our data. The same is true of the margin of victory. The interpretation is simply that the negative marginal impact of electoral proximity and margin of victory diminish somewhat as they both approach zero.

16

A second key finding from our country-level analysis is that ethnic fractionalization is negatively related to the salience of ethnicity in the country in question.17 In the large literature that employs indices of ethnic fractionalization to account for outcomes such as civil war (Collier 2001; Elbadawi and Sambanis 2002; Garcia-Montalvo and Reyna-Querol 2002), economic growth (Easterly and Levine 1997; Collier and Gunning 1999; Alesina et al 2003), and the quality of governance (Mauro 1995; La Porta et al 1999), ethnic diversity is frequently interpreted as a proxy for the salience of ethnic identity per se. Our results suggest that the assumption underlying this approach has it backwards. It turns out that the more diverse a country is, the less salient ethnicity is for its citizens in our African sample (Table 4, column 3). For a sense of the magnitude of this relationship, an increase in ethnic fractionalization of 0.17, or one standard deviation in our sample, is associated with a reduction in expressed ethnic identification of 15 percentage points – a large effect. Figure 3 makes clear graphically that no single country drives the result – a finding we confirm when we re-run the specification in column 3, dropping countries one at a time.18 The result is also robust to multiple measures of ethnic diversity (see Table 5), when individual characteristics and country population weights are excluded (Table 5, regression 2), and when using the narrower measure of ethnicity (regression not shown). The finding drops about one third in magnitude and loses statistical significance, however, when included jointly with the full set of country economic and political controls (Table 4, regression 4). FIGURE 3 AND TABLE 5 HERE

17

We use Fearon’s (2003) measure, which we feel to be the most reliable.

18

When countries are dropped one at a time, the coefficient estimate on ethnic fractionalization

remains large, negative and significant in all cases (regression not shown). This holds even for Botswana and Zambia, the two apparent outliers in Figure 3.

17

The empirical finding that ethnic diversity is negatively related to ethnic salience is broadly consistent with theoretical claims advanced by Collier (2001), Bates (2000), and Horowitz (1985) regarding the relationship between ethnic diversity and politics. They argue that ethnic rivalries are likely to be muted in the most highly diverse societies, since no single ethnic group will be strong enough to attain power on its own and incentives will arise for cooperation across ethnic lines. At low levels of diversity, ethnicity will also not be salient for the simple reason that everyone is a member of the same group. But as ethnic diversity increases from very low levels to the middle of the range, ethnicity becomes increasingly salient, as minority groups begin to challenge the dominant ethnic group for power. In these theories, ethnicity becomes most salient in a situation where two more or less equally sized groups are competing for power (corresponding to fractionalization equal to 0.5). Our empirical results, while consistent with this curvilinear hypothesis, do not allow us to test it completely, since we do not have country observations in the relevant range: even the most ethnically homogeneous societies in our twelve-country sample (Botswana and Zimbabwe) are reasonably ethnically diverse, with fractionalization measures in the 0.4 range (see Figure 3). So our sample only allows us to capture the right (downward sloping) part of the curve.

Conclusion The findings of this study challenge two persistent conventional wisdoms about Africa: that Africans are uniformly and unidimensionally ethnic, and that the salience of ethnicity is a product of the region’s low levels of political and economic development. The study’s central result is that exposure to education, non-traditional occupations, and political competition powerfully affects whether or not people identify themselves in ethnic terms. Taken together the findings provide strong confirmation for what we term “second wave” modernization approaches to ethnicity, and for theories that link identity choices with context and instrumentality. Beyond their relevance for these

18

academic literatures, the paper’s results also have important implications for policymakers and researchers interested in ethnicity’s effects. Political scientists and economists use the concept of ethnic salience to help explain everything from economic growth to civil conflict and the effectiveness of foreign aid. When they do so, they frequently employ measures of ethnic diversity as indicators of ethnic salience, the assumption being that greater diversity implies greater ethnic salience. Perhaps surprisingly, then, we find that high levels of country ethnic fractionalization actually reduce the likelihood that individuals will identify themselves first and foremost in ethnic terms in our African sample. While unfortunately based on only 12 country (and 16 country survey round) cases, the finding is sufficiently robust to call into question a central assumption on which many studies are based. We also find evidence that the salience of ethnicity can change – not just over the course of years, but even over the course of a few months, particularly near election time. This result, which is consistent with situational approaches to ethnicity, challenges empirical work that takes ethnic identities as static and historically determined. Particularly for researchers undertaking survey work, it provides a caution that the timing of data collection – particularly the proximity of the survey exercise to large-scale political events such as national elections – can have significant effect on the answers respondents provide about their ethnic identifications. The strong relationship we find between the intensity of political and economic competition on the one hand and the salience of ethnicity on the other also makes it clear that as African countries institute democratic and market reforms it will become more urgent – not less – for African governments to develop policies and institutional mechanisms that are capable of dealing with ethnic divisions. Kenya’s recent political developments are informative. After the reintroduction of competitive multi-party politics in the early 1990s, Kenya’s reform efforts have increasingly become mired in tribal politics, including violent ethnic clashes that have left hundreds dead. Policies and institutions such as those in place in neighboring Tanzania – a country known for its efforts at nation-

19

building through the promotion of Swahili as a national language, civic education, and institutional reforms like the abolition of chiefs, as described by Miguel (2004) – might serve as a model for how Kenya and other African countries might dampen destructive ethnic divisions. Perhaps due in part to these reforms, Tanzania has the lowest degree of ethnic identity salience in our Afrobarometer sample, at just 3 percent.19 Finally, our work brings new evidence to bear on the stubbornly persistent popular misconception that ethnicity in Africa is an atavism that can be “solved” by political and economic development. Scholarly consensus has long disputed this position, but the popular view remains firmly entrenched. Part of this disconnect may lie in lingering racism, which leads some to uncritically accept representations of Africans as backward and tribe-bound. But another part of the answer may lie in the fact that nearly all of the research that documents the positive association between modernization and deepening ethnic identification is either anecdotal or based on analyses of single countries. Absent systematic, cross-national analyses of the sort presented in this paper, old stereotypes and media-reinforced misperceptions are frustratingly difficult to break. The results of this paper, based on precisely the kind of cross-national data that has hitherto been lacking, provide new support for the claim that ethnicity is salient in Africa because people are becoming more modern, not less, and because political competition on the continent is increasing, not diminishing.

19

Tanzania’s outlier status in this regard is reflected in Figure 1, where it is clear that the close proximity between the country’s 2001 Afrobarometer survey and its 2000 presidential election has little impact on the share of the population that identifies itself in ethnic terms.

20

Sources Cited Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003. “Fractionalization.” Journal of Economic Growth 8 (June): 155–94. Bates, Robert H. 1983. “Modernization, Ethnic Competition and the Rationality of Politics in Contemporary Africa.” In State versus Ethnic Claims: African Policy Dilemmas, ed. Donald Rothchild and Victor A. Olorunsola. Boulder, CO: Westview, 152-171. Bates, Robert H. 2000. “Ethnicity and Development in Africa: A Reappraisal.” AEA Papers and Proceedings 90 (May): 131-34. Bratton, Michael, Robert Mattes, and E. Gyimah-Boadi. 2004. Public Opinion, Democracy, and Market Reform in Africa. New York: Cambridge University Press. CIA, 2003, World Factbook. Collier, Paul. 2001. “Implications of Ethnic Diversity.” Economic Policy 16 (April): 129-66. Collier, Paul, and Jan Willem Gunning. 1999. “Why Has Africa Grown Slowly?” Journal of Economic Perspectives 13 (Summer): 3–22. Davidson, Basil. 1992. The Black Man’s Burden: Africa and the Curse of the Nation State. New York: Times Books. Deutsch, Karl. 1961. “Social Mobilization and Political Development.” American Political Science Review 55 (September): 493-514. Easterly, William, and Ross Levine. 1997. “Africa’s Growth Tragedy: Policies and Ethnic Divisions.” Quarterly Journal of Economics 112 (November): 1203–50. Elbadawi, Ibrahim, and Nicholas Sambanis. 2002. “How Much War Will We See? Estimating the Incidence of Civil War in 161 Countries, 1960–1999.” Journal of Conflict Resolution 46 (June): 307-34. Fearon, James D. 2003. “Ethnic Structure and Cultural Diversity by Country.” Journal of Economic Growth 8 (June): 195-222.

21

Freedom House. 2001. Freedom in the World, 2000-2001. Garcia-Montalvo, Jose, and Marta Reynal-Querol. 2002. “Why Ethnic Fractionalization? Polarization, Ethnic Conflict, and Growth.” Typescript, Universitat Pompeu Fabra. Gulliver, P. H. 1971. Tradition and Transition in East Africa: Studies of the Tribal Element in the Modern Era. Berkeley and Los Angeles: University of California Press. Horowitz, Donald. 1985. Ethnic Groups in Conflict. Berkeley and Los Angeles: University of California Press. La Porta, Rafael, Florencio Lopez-de-Silanes, Andrei Shleifer, and Robert Vishny. 1999. “The Quality of Government.” Journal of Law, Economics and Organization 15 (March): 222–79. Lewis, Peter M. 2004. “Identity and Conflict in Nigeria’s Niger Delta: New Evidence from Attitude Surveys.” Paper presented at the Annual Meeting of the American Political Science Association, Chicago (September). Lloyd, Peter C. 1967. Africa in Social Change: Changing Traditional Societies in the Modern World. Baltimore: Penguin Books. Mattes, Robert. 2004. “Understanding Identity in Africa: A First Cut.” Afrobarometer Working Paper No. 38. Mauro, Pablo. 1995. “Corruption and Growth.” Quarterly Journal of Economics 110 (August): 681– 712. Melson, Robert and Howard Wolpe. 1970. “Modernization and the Politics of Communalism: A Theoretical Perspective.” American Political Science Review 64 (December): 1112-1130. Miguel, Edward. 2004. “Tribe or Nation? Nation-building and Public Goods in Kenya versus Tanzania.” World Politics 56 (April): 327-62. Morrison, Donald, Robert Mitchell and John Paden. 1989. Black Africa: A Comparative Handbook. New York: Paragon House.

22

Parkin, Frank. 1978. Marxism and Class Theory: A Bourgeois Critique. (New York: Columbia University Press). Posner, Daniel N. 2004. “Measuring Ethnic Fractionalization in Africa.” American Journal of Political Science 48 (October): 849-63. Vail, Leroy. 1989. The Creation of Tribalism in Southern Africa. Berkeley and Los Angeles: University of California Press. World Bank. 2004. African Development Indicators. Washington: The World Bank. Young, Crawford. 1965. Politics in the Congo. Princeton: Princeton University Press.

23

Table 1: How Representative is Our Sample of African Countries?

All Sub-Saharan Africa Twelve Sample Countries Botswana Ghana Malawi Mali Mozambique Namibia Nigeria South Africa Tanzania Uganda Zambia Zimbabwe

Per capita GNI in US$ (2001) 468 888 3,170 300 160 230 200 2,150 300 2,840 270 250 320 460

Under-5 Mortality per 1,000 (2001) 169 146 110 100 183 231 197 67 183 71 165 124 202 123

Percent Urban (2002) 35.5 35.8 49.9 36.7 15.5 31.5 34.3 31.9 45.7 58.4 34.2 14.9 40.1 36.7

Ethnic Fractionalization (Fearon) .71 .74 .35 .85 .83 .75 .77 .72 .80 .88 .95 .93 .73 .37

Political Rights Score (2000-01) 4.5 3.2 2 2 3 2 3 2 4 1 4 6 5 6

Table 1 Notes: GNI, under-5 mortality, and urbanization figures are from World Bank (2004). Ethnic Fractionalization figures are from Fearon (2003). Political rights score is from Freedom House (2001).

24

Table 2: Respondent Self-Identifications “…which specific group do you feel you belong to first and foremost?” (from Afrobarometer)

All Respondents Botswana, 1999 Ghana, 2002 Malawi, 1999 Mali, 2001 Mozambique, 2002 Namibia, 1999 Namibia, 2002 Nigeria, 2000a Nigeria, 2001a South Africa, 2000 South Africa, 2002 Tanzania, 2001 Uganda, 2000 Uganda, 2002 Zambia, 1999 Zimbabwe. 1999

Ethnic 0.41

Religion 0.11

0.92 0.39 0.63 0.39 0.30 0.48 0.69 0.70 0.52 0.41 0.21 0.03 0.12 0.18 0.11 0.53

0.02 0.33 0.11 0.23 0.06 0.06 0.06

0.16 0.06 0.05 0.09 0.08 0.32 0.09

Class/ Occupation 0.37

Other 0.11

Obs. 24815

0.03 0.22 0.24 0.25 0.33 0.42 0.20 0.28 0.41 0.16 0.45 0.79 0.67 0.61 0.54 0.33

0.01 0.05 0.02 0.13 0.28 0.04 0.05 0.02 0.07 0.27 0.29 0.12 0.12 0.13 0.02 0.05

839 928 1080 1938 778 823 988 3487 2083 2132 1733 2146 1927 2153 901 879

Table 2 Notes: The “ethnic” category includes tribe, language, race (in the former settler colonies of Mozambique, Namibia, South Africa, Zambia and Zimbabwe), region (in Malawi and Nigeria) and religion (in Nigeria). The “other” category includes gender, region (except in Nigeria and Malawi), race (except in the former settler colonies), and other responses. The rows may not sum to 100 percent because of rounding errors. The first row is weighted such that each survey round is weighted equally, the same weighting system used in Tables 3, 4, and 5. a The empty cells in the “religion” column for Nigeria are due to the fact that religious identities are defined as ethnic in this country; hence the share of respondents who identified themselves in terms of religion are included in the ethnic tally.

25

Table 3: Descriptive Statistics Variable Panel A: Country-level characteristics Log per capita income (2000, 2000 US$ – source: World Bank 2004) Proximity to closest next or previous election, in months (source: authors) Margin between highest and second-highest vote-receiving presidential candidates in most proximate election (source: authors) Ethnic fractionalization – Fearon (source: Fearon 2003) Ethnic fractionalization – Alesina (source: Alesina et al. 2003) Ethno-linguistic fractionalization (source: Easterly and Levine 1997) Size of largest ethnic group (source: Morrison et al. 1989) Politically relevant ethnic groups score (source: Posner 2004) Panel B: Individual-level characteristics Female Age (years) No formal education Some primary education Completed primary education Some secondary education Completed secondary education At least some post-secondary education Occupation: Farming or fishing Occupation: White collar, teacher, or government employee Occupation: Blue collar or miner Occupation: Student Occupation: Business, shop keeper, or petty trader Occupation: Other (e.g., unemployed, housewife, don’t know) Lives in rural area Gets news from radio daily Gets news from newspaper at least weekly

Mean

Std. dev.

Obs.

6.3 13.3

1.1 7.8

24815 24815

0.33 0.77 0.72 0.76 0.40 0.49

0.23 0.17 0.15 0.13 0.19 0.19

24815 24815 24815 24815 24815 24815

0.45 35.6 0.15 0.19 0.17 0.21 0.16 0.11 0.28 0.17 0.10 0.08 0.14 0.23 0.57 0.61 0.32

0.55 14.1 0.36 0.39 0.37 0.41 0.37 0.32 0.45 0.37 0.30 0.27 0.35 0.42 0.50 0.49 0.47

24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815 24815

Table 3 Notes: The education categories are mutually exclusive. The occupation categories are mutually exclusive. Reported figures are weighted such that each survey round is weighted equally, the same weighting system used in Tables 2, 4, and 5.

26

Table 4: Sources of Ethnic Identification: Individual and Country Characteristics Dependent variable: “Belongs to an ethnic group first and foremost” (1) (2) (3) (4) Female

0.007 (0.010) 0.0006 (0.0005) -0.038* (0.020) -0.002 (0.016) 0.007 (0.023) -0.002 (0.026) -0.074** (0.031) -0.021** (0.011) -0.048* (0.026) 0.026 (0.030) 0.060** (0.028) 0.098*** (0.029) 0.090*** (0.033) 0.085*** (0.021) 0.114*** (0.033)

0.009 (0.022) Age (years) -0.00003 (0.0005) No formal education 0.024 (0.026) Completed primary education -0.027 (0.042) Some secondary education -0.038 (0.031) Completed secondary education -0.0007 (0.060) -0.065 At least some post-secondary education (0.070) -0.022 Gets news from radio daily (0.014) -0.082*** Gets news from newspaper at least weekly (0.030) Lives in rural area -0.011 (0.041) 0.069** Occupation: White collar, teacher, or gov’t employee (0.030) 0.074* Occupation: Blue collar or miner (0.041) Occupation: Student 0.175*** (0.036) 0.119*** Occupation: Business, shop keeper, or petty trader (0.038) 0.111*** Occupation: Other (e.g., unemployed, housewife, DK) (0.038) Log per capita income (in 2000, US$) 0.050 (0.077) -0.021** Proximity to closest election, in months (0.009) Margin of victory in most proximate election -0.680 (0.434) 0.045** Proximity * Margin (0.023) Ethnic fractionalization – Fearon measure -0.86*** -0.68 (0.32) (0.44) Country fixed effects Yes No No No Observations (respondents) 24815 24815 24815 24815 Table 4 Notes: Probit estimation, with marginal coefficient estimates (at mean values for the explanatory variables). Huber robust standard errors in parentheses. Significantly different than zero at 90% (*), 95% (**), 99% (***) confidence. Regression disturbance terms are clustered at the country level. The omitted education category is “Some primary education”. The omitted occupation category is “Occupation: farming or fishing”. The F-test on the hypothesis that all of the country fixed effects (in regression 1) equal zero has p-value<0.001. Regressions weight each observation by 1 / (Number of Afrobarometer observations for the survey round in which the respondent is located), thus effectively weighting each survey round equally, as in Tables 2, 3, and 5.

27

0.041 (0.029) -0.0004 (0.0007) 0.034 (0.025) -0.022 (0.039) -0.035 (0.031) -0.006 (0.063) -0.064 (0.070) -0.026** (0.013) -0.077*** (0.029) -0.015 (0.044) 0.089*** (0.032) 0.089** (0.037) 0.185*** (0.032) 0.126*** (0.040) 0.144*** (0.034) 0.125 (0.080) -0.026*** (0.008) -1.11*** (0.382) 0.047** (0.022)

-0.007 (0.020) -0.0001 (0.0005) 0.004 (0.034) -0.045 (0.047) -0.043 (0.039) -0.006 (0.066) -0.077 (0.080) -0.015 (0.019) -0.069* (0.036) 0.019 (0.042) 0.082** (0.033) 0.103** (0.050) 0.184*** (0.037) 0.129*** (0.040) 0.115*** (0.036) 0.038 (0.048)

Table 5: Sources of Ethnic Identification: Ethnic Diversity Measures

(1) Ethnic fractionalization – Fearon measure

***

-0.86 (0.32)

Dependent variable: “Belongs to an ethnic group first and foremost” (2) (3) (4) (5) -0.95 (0.37)

-0.85** (0.38)

Ethnic fractionalization – Alesina measure

-1.20*** (0.36)

Ethno-linguistic fractionalization

0.93*** (0.19)

Size of largest ethnic group Politically relevant ethnic groups score Individual and country characteristics Observations (respondents)

(6)

***

Yes 24815

No 24815

Yes 24815

Yes 24815

Yes 24815

-0.51 (0.32) Yes 24815

Table 5 Notes: Probit estimation, with marginal coefficient estimates (at mean values for the explanatory variables). Huber robust standard errors in parentheses. Significantly different than zero at 90% (*), 95% (**), 99% (***) confidence. Regression disturbance terms are clustered at the country level. Regression 1 reproduces the result from Table 4, regression 3. The individual and country characteristics in columns 3-6 are as in Table 4, regression 3. Regressions weight each observation by 1 / (Number of Afrobarometer observations for the survey round in which the respondent is located), thus effectively weighting each survey round equally, as in Tables 2, 3, and 4.

28

1

Figure 1: “Belongs to an Ethnic Group First and Foremost” and Proximity to Closest Election, in Months (with fitted regression line, unconditional)

.8

BOT-1999

NIG-2000

NAM-2002

.6

MWI-1999 ZIM-1999

NIG-2001

NAM-1999

.4

SA-2000 MLI-2001

GHA-2002 MOZ-2002

.2

UG-2002

SA-2002

UG-2000

ZAM-1999

0

TZ-2001

0

10 20 | Months pre/post election | Ethnic identification

29

Fitted values

30

1

Figure 2: “Belongs to an Ethnic Group First and Foremost” and Proximity to Closest Election, in Months, in the Niger Delta of Nigeria (with fitted regression line and date of survey round indicated)

.9

Oct 2003

.5

.6

.7

.8

Jan 2000

Aug 2001

5

10 15 | Months pre/post election | Ethnic identification

20

Fitted values Data Source: Lewis (2004)

30

1

Figure 3: “Belongs to an Ethnic Group First and Foremost” and Ethnic Fractionalization, Fearon Measure (with fitted regression line, unconditional)

.8

BOT-1999

NIG-2000

NAM-2002

.6

MWI-1999 ZIM-1999

NIG-2001 NAM-1999 SA-2000 GHA-2002

.4

MLI-2001 MOZ-2002

.2

SA-2002 UG-2002 UG-2000

ZAM-1999

0

TZ-2001

.4

.6 Ethnic fractionalization Ethnic identification

31

.8

Fitted values

1

Appendix Table A1: Respondent Self-Identifications (narrow ethnic definition) “…which specific group do you feel you belong to first and foremost?” (from Afrobarometer)

All Respondents Botswana, 1999 Ghana, 2002 Malawi, 1999 Mali, 2001 Mozambique, 2002 Namibia, 1999 Namibia, 2002 Nigeria, 2000 Nigeria, 2001 South Africa, 2000 South Africa, 2002 Tanzania, 2001 Uganda, 2000 Uganda, 2002 Zambia, 1999 Zimbabwe. 1999

Ethnic 0.34

Race 0.05

Religion 0.13

Class/ Occupation 0.37

Other 0.11

Obs. 24815

0.92 0.39 0.61 0.39 0.29 0.34 0.67 0.47 0.29 0.18 0.09 0.03 0.12 0.18 0.07 0.39

0.03 0.01 0.01 0.00 0.01 0.14 0.02 0.00 0.00 0.23 0.11 0.00 0.00 0.00 0.04 0.14

0.02 0.33 0.11 0.23 0.06 0.06 0.06 0.21 0.21 0.16 0.06 0.05 0.09 0.08 0.32 0.09

0.03 0.22 0.24 0.25 0.33 0.42 0.20 0.28 0.41 0.16 0.45 0.79 0.67 0.61 0.54 0.33

0.01 0.05 0.05 0.13 0.28 0.04 0.05 0.04 0.09 0.27 0.29 0.12 0.12 0.13 0.02 0.05

839 928 1080 1938 778 823 988 3487 2083 2132 1733 2146 1927 2153 901 879

Table A1 Notes: The “other” category includes gender, region, and other responses. The rows may not sum to 100 percent because of rounding errors. The first row is weighted such that each survey round is weighted equally, the same weighting system used in Tables 3, 4, and 5.

32