Regime Change and Ethnic Cleavages in Africa

 

 

 

 

 

 

 

 

 

 

 

 

Daniel N. Posner

Department of Political Science

University of California, Los Angeles

Los Angeles, CA  90095-1472

dposner@polisci.ucla.edu

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Paper prepared for the Workshop on Democracy in Africa in Comparative Perspective, Stanford University, 27 April 2001.

 


 

 

In this paper, I explore the impact of the (re)introduction of competitive multi-party elections on ethnic conflict in Africa.  In doing so, however, I examine an overlooked consequence of regime change.  Rather than focus on the effects of democratization on the intensity of inter-group conflict, I focus instead on how the shift from one-party to multi-party rule altered the relevant dimension of ethnic cleavage around which political competition and mobilization takes place.  Most studies to date seek to measure and/or explain the impact of democratization on the level of ethnic conflict in the region.[1]  In this paper, I seek to document and account for its effects on the relative political salience of alternative, potentially mobilizable ethnic cleavages.

 

The starting point for such an analysis is the observation that African political systems, like political systems everywhere, possess multiple lines of social cleavage, and that the relative salience of each cleavage varies over time and across political contexts.  Although ordinarily lumped under the umbrella term “ethnic,” communal conflict can take many forms.  Sometimes competition takes place along religious lines.  At other times competing groups are distinguished from one another by language.  At still other times in-group/out-group distinctions are made on the basis of tribal affiliation, clan membership, geographic region of origin, or race.  Within a single country, each of these distinctions may serve, in different situations, as a potential axis of social differentiation and conflict.  Describing the multi-dimensional nature of ethnic politics in Ghana, for example, Naomi Chazan (1982: 467-68) observes that

 

sometimes ethnic solidarity was expressed in cultural and linguistic terms.  At other times ethnicity was presented in regional or geographic terms.  At still other points, ethnicity was manifested in local-communal – traditional, political or kin – terms…All possible ethnic-political presentations, either separately or in conjunction, could be brought to bear on the political situation depending on particular conditions.

 

            The question begged by such an analysis, however, is:  what “conditions” generate which sorts of “presentations?”  Can we identify the contexts in which one dimension of ethnic cleavage will emerge as politically salient instead of another?  The purpose of this paper is to present and test an argument that does this.  I do so by showing how the different institutional logics of one-party and multi-party political competition cause different dimensions of ethnic cleavage to become politically salient in each context.  I show that, under conditions of multi-party competition, ethnic cleavages that define large blocks of people tend to emerge as the axis of political conflict, coalition-building and voting, whereas under conditions of one-party political competition, ethnic cleavages that define smaller, more localized groups of people tend to play this role.  The salience of ethnicity per se may not change, but the salience of the particular dimension of ethnic cleavage that structures politics is transformed by the shift in regime type.

 

            The argument that I present draws on Posner (1998), which explores the relationship between one-party and multi-party political institutions in Zambia and the relative political salience of linguistic and tribal divisions.  In that work, I show that linguistic cleavages structured ethnic politics during the two periods when Zambia was a multi-party regime (1964-1972 and 1991-present), whereas tribal divisions served as the basis for political mobilization during the long one-party era (1973-1991).  In the first part of the paper, I review the logic behind this finding.  In the second part, I present some of the empirical evidence that supports it.

 

            In the third part of the paper, I shift attention from the general relationship between regime type and ethnic cleavage salience to the effects of the recent wave of democratic opening in Africa.  I also expand my emphasis from the specific case of Zambia to two other countries in the region:  Kenya and Tanzania.  If my Zambia-inspired argument is generalizable, we would expect the move from one-party to multi-party rule in these countries to have caused a parallel shift from localized, narrowly-defined ethnic cleavages to social divisions that define more encompassing groups.  The particular dimension of ethnic cleavage that will emerge as salient in each country – language, religion, geographic region – will depend on the constellations of cleavage dimensions from which political actors can choose in each country.  But, whatever the specific dimension of social division that emerges as salient, we should find the narrower cleavages that structured political competition during the one-party era being displaced by cleavages that define broader coalitions of members.

 

            I begin the third part of the paper by reviewing the Zambian evidence, but restricting my analysis to a comparison of ethnic voting patterns in the two elections on either side of the recent democratic transition, those of 1988 and 1991.  I then present parallel analyses of the last one-party and first multi-party elections in Kenya (1988 and 1992) and Tanzania (1990 and 1995).  In all three cases, I find evidence that the change from one-party to multi-party political competition altered the dimensions of ethnic identity that voters emphasized in choosing how to allocate their electoral support.

 

 

Regime Type and Ethnic Cleavages:  a Simple Model

 

            The explanation I provide for why regime type matters for the kinds of ethnic cleavages that emerge as politically salient rests on two claims about the nature of African politics.  The first is that African voters seek to maximize the amount of resources they can secure from the state and that politicians, knowing this, seek to attract and maintain their political followings by promising resources to those who support them.  The second is that voters believe that having a member of their own ethnic group in a position of power will increase their access to such resources.[2]  They discount the election promises made by candidates who are not their ethnic kin and find credible only those promises made by candidates who share their ethnic background.  Both of these rather uncontroversial claims are supported by the vast literature on neo-patrimonialism in Africa (Bayart 1993, Joseph 1984, Chabal and Daloz 1999), and require little elaboration.

 

            The critical complication with the strategy of voting for members of one’s own ethnic group, however, is that both candidates and voters belong to many groups.  This means that the question arises as to which group membership is relevant in deciding which candidate to support.  If I am a Muslim Yoruba from Ibadan, do I vote for the fellow Muslim, who may be a Hausa-speaker?  the fellow Yoruba, who may come from a different ancestral city-state?  or the fellow Ibadan resident, who may be a Christian?  Of course, if the goal is to put someone from my own group in power, then any of these candidates would be preferable to a candidate who shared none of my religious, linguistic or geographic background.  But given at least some shared group membership with all three, which would be best in terms of maximizing the resources that are likely to flow to me?

 

            The answer is that I will do best by supporting the one who puts me in the winning coalition.  Throwing my support behind a candidate from my own group will not do me any good if that candidate does not win, so a first criterion is whether the group is large enough relative to other groups to win.  So long as everyone else is also voting exclusively along group lines, this boils down to an issue of which of my group memberships puts me in the largest group vis-à-vis the other groups that are defined by that cleavage dimension.[3]  But if more than one category of group membership would put me in a winning coalition, then I would prefer the one that puts me in the smaller of the two coalitions.  This way I will get to enjoy a larger share of the patronage that the coalition, by virtue of winning, is able to control.  For the same reason, if the population is homogeneous with respect to a particular line of cleavage (for example, everyone is a Swahili-speaker), then voting for a candidate based on our shared membership in that group would be disadvantageous, since such a coalition is, by definition, not minimum-winning.

 

            Thus, as a voter, my optimal strategy is, first, to think about all the principles of group division (that is, all the potentially mobilizable social cleavages) that divide the political community – religion, language, region of origin, tribe, clan, etc. – and, for each of these, compare the relative sizes of my own group with the others that the cleavage defines.  Thus, if the cleavage is “religion,” I will need to compare the size of my own religious group (say, Muslims) with the sizes of all the other religious groups (say, Christians, Hindus, Jews) in the political arena.  Then, I should select the principle of group division that puts me in the most advantageous group vis-a-vis the other groups, and throw my support behind the candidate from that group.

 

            This is almost certainly a too-precise (even too-rational) account of how African voters determine, in practice, which group membership should guide their electoral decision-making.  In many cases, they will lack the exact demographic information necessary for parsing the relative sizes of all the groups at each cleavage dimension.  And in some instances personal relationships with particular candidates or independent knowledge about candidates’ relative abilities to extract resources from the state will trump the considerations I have discussed.  But as a general set of principles, the logic I have outlined probably does guide voters’ choices, particularly in situations where the relative sizes of the groups in which the voter can claim membership are very different or where the boundaries of the political arena are such that every voter is a member of the same group with respect to one or more dimensions of social cleavage.

 

            How does the one-party or multi-party nature of the political system affect these calculations?  First, it shapes the kinds of information that voters have about the patronage commitments of each candidate.  In one-party elections, voters make inferences about how candidates will channel patronage by focusing their attention on the candidates’ own ethnic backgrounds.  In multi-party settings, making such inferences is complicated by the fact that candidates are the bearers of two different kinds of information about their patronage loyalties:  that conveyed by their own ethnic background and that suggested by their party affiliation.  Voters must decide which clues to focus on in deciding who to support.  Again and again in multi-party elections, they look beyond the backgrounds of the candidates and focus their attention on the information conveyed by their party labels.[4]

           

            The reason why voters focus their attention on party affiliations rather than candidate’s backgrounds in multi-party elections is two-fold.  First, the centrality of “big men” in African politics means that voters know that the ethnic orientation of the party’s leader will set the agenda for how that party will distribute patronage.  Under such a system, even the most well-meaning locally-oriented politician will be unable to bring development resources to the locality if his party leaders make it their priority to channel their funds to a different part of the country.  Second, any candidate lucky enough to be standing on the ticket of a party that is associated in voters’ minds with the dominant ethnic group in the constituency (for example, a candidate standing in the Western region on the ticket of a party whose president is a Westerner) will be sure to claim that his rivals will sell out the interests of the local group to those of the groups associated with their own parties.  Individual candidates in multi-party elections will thus be tarred, identity-wise, with the same brush as the parties with which they are affiliated.  And electoral competition will come to be structured by the ethnic cleavage in terms of which these party affiliations are defined.

 

            How then do the ethnic affiliations of parties tend to be defined?  Because parties are competing at the national level, the sorts of cleavages that it makes sense for them to mobilize – and in terms of which they tend to be categorized by both opponents and supporters – tend to be national in scope.  For a party to portray itself as representing the interests of a tiny group (say one of forty tribes in the country) is not a particularly good strategy for trying to capture national power.  The party might successfully win the support of all the members of that small group, but it would probably make few inroads in the rest of the country.  It would be far better for the party to identify itself with a group that comprises a larger share of the national population (say one of the three broad linguistic or regional groups in the country).  The national arena in which parties are competing thus pushes them to identify themselves with large, national-scale groups.  This tendency is reinforced by the incentives for rival parties, whose ability to mobilize their own supporters often depends on invoking the threat posed by their opponents, to highlight the linkages between particular parties and particular large (and thus threatening) ethnic groups. 

 

            Individual candidates, on the other hand, are competing at the level of the electoral constituency and find it advantageous to focus voters’ attention on a dimension of ethnic identity that is narrow enough to allow them to distinguish themselves from their rivals.  A broad regional identity shared by all the candidates in the race will be of little use in this regard.  But a local identity that allows a candidate to portray himself as a champion of the interests of a particular segment of the community will be advantageous.  So candidates, like the voters themselves, have an interest in emphasizing a dimension of social cleavage that provides a basis for dividing the electoral constituency into multiple groups.  And this means local-scale cleavages rather than national-scale cleavages.

 

            Thus, political parties and individual candidates will tend to define themselves and to be seen by others in terms of different social cleavages:  the former in terms of cleavages that define large-scale identity groups and the latter in terms of social cleavages that define smaller-scale, more localized identity groups.  In one-party elections, where candidates’ own backgrounds are all voters have to guide them, competition will revolve around local ethnic cleavages.  In multi-party elections, where voters are guided by candidates’ backgrounds political competition will revolve around national-scale cleavages.  The result is that candidates’ local ethnic backgrounds should thus be a good predictor of their electoral prospects in one-party races but a poor predictor of their performance in multi-party contests.  In the latter, the match between the broad ethnic group associated with the party on whose ticket the candidate is standing and that of the population in the constituency in which they are located should be the key to their electoral success. 

            Before testing these theoretical predictions, it should be stressed that the extent to which this model of political choice can be expected to hold will vary across political systems.  Single-member plurality systems, in which individual candidate characteristics should be particularly salient in one-party races and in which party labels are especially important in multi-party races, are the institutional setting in which the model should perform best.[5]  Former British colonies, then, provide the clearest and purest test of the predictions of the model.  The cases of Zambia, Kenya and Tanzania are thus well suited for the task at hand.  In the following section, I present confirmatory evidence from the model drawn from Zambia’s experience with one-party and multi-party rule.  Then, in the next section, I test the model’s predictions in all three countries, focusing on the differences in ethnic voting patterns in the two elections that took place on either side of the democratic transition.

 

 

Regime Type and Ethnic Cleavages in Zambia

 

            In Zambia, two different ethnic cleavages provide potential bases for political mobilization:  tribe and language.[6]  The former divides the country into roughly seventy highly localized groups, whereas the latter divides the country into four broad regional coalitions.  Given this menu of options, the model I have presented generates a relatively simple expectation about which cleavages should have emerged as the axis of political competition during periods of one-party and multi-party rule.  In one-party settings, electoral contests should have revolved around tribal divisions, whereas in multi-party settings, they should have revolved around linguistic/regional divisions.  There is no reason to expect voters to have voted more or less “ethnically” in either institutional context, since expectations about patronage networks should have been unchanged.  But their ethnic choices should have been guided by different yardsticks in each case:  by the tribal backgrounds of the competing candidates in one-party elections, and by the language-group orientations of the parties on whose tickets the candidates were standing in multi-party elections. 

 

            If this is so, then, in one-party elections, we should find the share of votes cast for candidates from a particular tribe to be roughly equal to the share of voters in the constituency that are members of that tribe.  In multi-party elections, the match between the tribal demographics of the constituency and the candidates’ vote shares should be less good.  A better predictor of electoral success in multi-party contests should be the match between the presumed language-group orientation of the party with which the candidate is affiliated and the language group of the bulk of the population in the constituency.

 

            To test these predictions, I collected information about the tribal backgrounds of every one of the more than 1,200 parliamentary candidates that ran for seats in the Zambian national assembly between 1968 and 1998.[7]  I also used data from the 1990 census to calculate the exact tribal demographics of every electoral constituency for both the one-party and multi-party delimitation schemes.[8]  Taken together, these two unique data sets made it possible for me to compare patterns of tribal voting in 791 electoral constituencies:  364 from the multi-party elections of 1968, 1991, 1996 and the by-elections that took place after 1991, and 445 from the one-party elections of 1973, 1978, 1983 and 1988. 

 

            Table 1 presents regression results comparing, across all one-party and multi-party elections ever held in Zambia, the extent to which the share of voters in each constituency belonging to the dominant tribe predicts the share of the vote won by candidates from that tribe.  If the model is right, we would expect this relationship to be stronger in the one-party contests than in the multi-party contests.  Both controlling for province-level effects and not, we find that it is.

 

TABLE 1 HERE

 

            This analysis, while suggestive, contains a critical weakness.  The hypothesis being tested is that voters will cast their votes for candidates who come from their own tribes.  But for them to do so there must be a candidate from their tribe in the race for whom they can vote.  The problem is that this is not always the case.  And when it is not, we are left without a “decision-rule” to account for the behavior of such voters.

 

            Take the example of the hypothetical constituency depicted in Table 2.  The constituency contains 6 different tribes (tribes A-F).  Of that number, only tribes A, B and C have candidates from their own groups in the race.  If the ethnic voting hypothesis is correct, we would expect voters from these three tribes to cast their ballots in the manner suggested by the dark-shaded cells:  voters from tribe A will support candidates from tribe A, voters from tribe B will support candidates from tribe B, etc.  The problem is that our ethnic voting hypothesis generates no predictions about which candidates voters from tribes D, E or F will support.  When the share of the total population of the constituency made up of such “free agent” voters is anything but tiny, our ability to make inferences about ethnic voting patterns from the aggregate data will be severely undermined.[9]

 

TABLE 2 HERE

 

It is possible, of course, that because they do not have a candidate from their own tribe to vote for, voters from tribes D, E and F will simply abstain from voting.  It is also possible that they will vote for candidates from tribes A, B and C, but divide their support among them in proportion to the population shares of those groups.  If either of these were the case, it would make our job much easier.  But it is unlikely that either is.  And we are thus left with a problem:  the inferences we would like to make are based on the total vote shares won by the candidates from tribes A, B and C, but those vote shares will necessarily be affected – perhaps quite significantly – by the theoretically unpredictable electoral choices of the “free agent” voters from groups D, E and F.

 

            My solution is to re-do the analysis presented in Table 1 but drop all constituencies where fewer than 85 percent of the voters had a candidate from their tribe in the race.  Although this results in a significantly smaller sample size (we are left with 53 one-party cases and 41 multi-party cases), the dramatic increase in the reliability of the inferences I am able to draw from the analysis more than justify the loss of data.[10]  The results of the re-analysis are presented in Table 3.  This time the model’s predictions are borne out even more clearly.  Not only does gap between the one-party and multi-party coefficients increase relative to those reported in Table 1, but the relationship between the share of voters belonging to the dominant tribe and the share of the vote won by dominant tribe candidates ceases to be statistically significant in the multi-party sample.  Moreover, the coefficient on the share of voters belonging to the dominant tribe approaches 1 in the one-party sample, suggesting a nearly perfect correlation between it and the share of the vote won by dominant tribe candidates.  The results strongly suggest that voters are casting their ballots along tribal lines in one-party elections but not in multi-party elections.

 

TABLE 3 HERE

 

            Thus far, I have sought to test the expectations of the model by looking for greater evidence of tribal voting in one-party than multi-party elections.  But the model predicts not just that tribal voting will decline after a transition to multi-party rule (or increase following a transition from it) but that it will be replaced by party-oriented language group voting.  Tables 4 and 5 present evidence from all multi-party elections held in Zambia since Independence that tests this expectation.

 

TABLES 4 AND 5 HERE

 

            As we see in Table 4, candidates that are members of the dominant tribe in the constituency in which they are standing win the seat 51 percent of the time (213 times out of 414) in multi-party elections.  Those that are not from the dominant tribe win 47 percent of the time (86 times out of 183).  Being a member of the dominant tribe helps a little bit, but not very much.  Turning to Table 5, and leaving aside the match between the candidate’s own tribal affiliation and that of the dominant group in the constituency, we see that candidates standing on the tickets of parties that are affiliated with the dominant language groups in the constituencies in which they are standing win the seat 87 percent of the time (152 times out of 174).  If they are standing on the tickets of parties that are not affiliated with the dominant language groups in the constituencies, the likelihood of winning drops to just below 35 percent (147 times out of 423).  In multi-party elections, the data suggest that standing on the ticket of a party affiliated with the dominant language group matters a lot.  A candidate’s party affiliation would appear to be a much better predictor of electoral success than his/her tribal affiliation.

 

We can take a closer look at these patterns by comparing the performance of four different kinds of candidates (see Table 6).  Type A candidates are members of the dominant tribe in the constituency and standing on the ticket of a party that is affiliated with the constituency’s dominant language group.  We would expect such candidates to perform very well, and they do:  of the 131 type A candidates that I was able to identify in all multi-party elections ever held in Zambia, 118 (90 percent) won the seat that they were contesting.  Type D candidates, by contrast are neither members of dominant tribe nor standing on the ticket of a party that is associated with the dominant language group.  To the extent that voters allocate their support along ethnic lines, we would expect type D candidates to fare poorly.  Again, the data bear out this expectation:  of the 140 such candidates that was able to identify, just 52 (37 percent) won the race they were contesting. 

 

TABLE 6 HERE

 

One implication of these findings is that ethnicity matters but is not determinative of electoral outcomes.  The fact that ninety percent of those candidates whose ethnic backgrounds (both tribally and, through their party affiliations, linguistically) matched those of the plurality of the voters whose support they were seeking were able to win their election suggests that ethnicity does matter in the voting calculus in Zambia.  But the fact that thirty-seven percent of those candidates whose ethnic background did not match that of the dominant group of voters were also able to win the seats that they were contesting suggests that factors other than ethnicity also motivate voters decision-making.[11] 

 

Recall, however, that my purpose in this paper is not to demonstrate the salience of ethnicity per se but to show that, to the extent that ethnic considerations shape electoral behavior, different kinds of ethnic considerations do so in one-party and multi-party elections.  Indeed, from the standpoint of testing my model, comparing the success rates of type A and D candidates is of little use, since we have no way of knowing whether their victory or loss was because of their individual tribal affiliations or because of the language group identities ascribed to them by virtue of their party affiliations.  The key candidate types to look at for this purpose are B and C. 

 

Type B candidates are standing on the ticket of a party that is affiliated with the constituency’s dominant language group but do not belong to the dominant tribe.  Type C candidates, on the other hand, belong to the dominant tribe but are standing on the ticket of a party that is not affiliated with the constituency’s dominant language group.  If voters in multi-party contests were behaving exactly as they did in one-party elections (that is, allocating their support based on the match between their own tribal affiliations and those of the candidates in the race, and paying no attention to party affiliations), then we would expect type C candidates to outperform type B candidates, and success rates among type A and C candidates (and also among type B and D candidates) to be roughly the same.  If, on the other hand, as the model predicts, voters in multi-party elections ignore candidates’ tribal affiliations and instead cast their votes based on the language group affiliation communicated by each candidate’s party label, then we would expect type B candidates to outperform type C candidates, and type A and B (and type C and D) candidates to perform roughly equally.  As Table 7 shows, the latter is exactly what we find.  Moreover, in the twenty instances where head-to-head contests took place between candidates of type B and type C, the type B candidate won fifteen times.

 

TABLE 7 HERE

           

Another implication of the model is that type B candidates will not only outperform type C candidates, but that type B candidates will win a larger share of the vote than their tribe’s share of the population would lead us to expect.[12]  Analogously, the model would also imply that type C candidates should win smaller shares of the vote than their tribe’s population share would lead us to expect had they voted along purely tribal lines.  I was able to test this expectation in fifteen of the twenty head-to-head contests between type B and type C candidates.[13]  Thirty-three candidates stood in these fifteen contests, 16 of type B and 17 of type C.  Of the 16 type B candidates, 14 (or 87.5 percent) won more votes than their tribe’s share of the pop would have led us to expect; of the 17 type C candidates, 14 (or 82.3 percent) won fewer votes than their tribal share would have led us to expect.  Again, the findings are in keeping with the theoretical expectations of the model:  tribal voting would seem not to be taking place in the way it did in the one-party contests.  Ethnicity still matters, but a different dimension of ethnicity.

 

 

 

Democratization and Ethnic Cleavage Change in Zambia, Kenya and Tanzania

 

            The model’s predictions would seem to be borne out by the Zambian evidence.  But is the pattern of variation I have documented a generalizable effect of regime change or simply a quirk of the Zambian case?  To find out, I collected data on two additional African countries, Kenya and Tanzania, that would allow me to compare patterns of ethnic voting in one-party and multi-party contexts.  I chose these cases because the elections I hoped to examine – those that took place immediately before and after their recent transitions to multi-party rule – were competitive, and because, like Zambia, these countries also possess single-member plurality electoral rules.[14]  Although data limitations make the tests I present below less comprehensive than those I was able to undertake in Zambia, the tests are nonetheless suggestive.  And what they suggest is that the shift from one-party to multi-party rule in these countries did affect the kinds of ethnic cleavages that voters considered when they decided how to cast their ballots.

 

            Before turning to the Kenyan and Tanzanian evidence, it will be useful to replicate the analysis presented in Table 5 with the Zambian data from just the 1988 and 1991 elections.  This will allow me to confirm that the patterns discernible in one-party and multi-party elections generally also obtain in the two elections that fall on either side of the country’s recent political transition.  It will also allow me to re-select the cases I include in the analysis to match the case selection rules I employed in Kenya and Tanzania, where I only look at candidates who stood in both elections, and in the same constituency.  Limiting the data set in this way has the benefit of automatically controlling for candidate- and constituency-specific factors that might otherwise compromise my results by varying across the pools of candidates standing in each election.  It also makes it possible to compare directly the performance of each candidate in each institutional setting.[15]

 

            Table 8 breaks down the election outcomes for the 73 candidates that ran for parliament in Zambia in both 1988 and 1991 and stood in the same constituencies in both elections.  Among the sixteen candidates who were standing on the tickets of parties affiliated with the dominant language group in the constituency, fifteen (or 94 percent) won the seat they were contesting.  Among the fifty-seven who were standing on the tickets of parties affiliated with other language groups, forty-six (or 81 percent) lost.  This is precisely the pattern the model would lead us to expect in a multi-party election.

 

TABLE 8 HERE

 

            Since I lack reliable data on both constituency-level ethnic demographics and the tribal backgrounds of Kenyan and Tanzanian parliamentary candidates, I cannot test directly for different degrees of tribal voting in one-party and multi-party elections, as I did in Zambia in Tables 1, 3 and 4.  But I can make some, admittedly weaker, inferences about the changes that took place between one-party and multi-party elections by comparing the performance of candidates in each election.  Since neither the candidates’ own tribal backgrounds nor the tribal demographic of the constituencies in which they are standing changed between the two elections, a continuation of tribal voting into the multi-party era would be reflected in an unchanged vote share for each candidate.  But if the importance of the candidate’s tribal background was displaced by some other factor, like the regional orientation ascribed to them by their party affiliation, then we would expect their vote share to change across the two contests.  Thus, if the model is right, we would expect candidates standing on the tickets of parties affiliated with the dominant regional group in their constituencies not just to win but to out-perform their one-party election results.  Similarly, we would expect candidates standing on the tickets of parties not affiliated with the dominant regional group in their constituencies not just to lose but also to under-perform their one-party election results.  In Zambia, as Table 9 shows, this is the general pattern we find.[16]

 

TABLE 9 HERE

 

            In Kenya and Tanzania, I understand the choice facing voters to be between the broad regional cleavages associated with parties (e.g., FORD-Asili in Kenya was generally associated with the Central Province in 1992; CUF in Tanzania was identified with non-mainlanders and Muslims in 1995) and the local tribal identities of the candidates themselves.[17]  Thus, for example, a hypothetical Kikuyu voter in Kenya’s Central Province would have to decide between supporting the non-Kikuyu candidates standing on the FORD-Asili ticket or the Kikuyu candidate standing for KANU.  The theory predicts that the voter would choose the former. 

 

            Tables 10 and 12 report the results for Kenya and Tanzania that parallel those reported for Zambia in Table 8.   In both cases, as we found in Zambia, candidates standing for parties associated with the region in which the constituency is located tended to win and candidates standing for parties not associated with the region tended to lose.  This pattern is particularly strong in Tanzania, where only nine of the 95 cases fail to fall in the predicted cell in Table 12.  In Kenya, although the general pattern is what we would expect, the 36 percent failure rate of “insider parties” (in the lower left cell of Table 10) and 23 percent success rate of “outsider” parties (in the upper right cell) departs somewhat from the predictions of the model.  This is probably due to a combination of incumbency advantage and outright vote-buying by KANU candidates.[18]  In the 34 cases where a candidate’s “insider” party affiliation failed to win them the seat, they were standing against a KANU candidate, who ultimately won, thirteen times.  Similarly, in the 19 cases where the regional orientation of the candidate’s party would not have predicted victory but the candidate won anyway, 14 of the victors were KANU candidates. 

 

            More powerful evidence for the shifting salience of tribal and regional orientations is provided in Tables 11 and 13, which compare how the candidates fared in the two elections.  As in the Zambian results presented in Table 9, the strong diagonal effect (i.e., almost all the cases falling in the upper left and bottom right cells) suggests that the additional information provided by candidates’ party affiliations shifted the degree of support that voters were willing to give them, and did so in precisely the ways the model would lead us to expect.  Holding tribal backgrounds constant, standing for a party with the “correct” regional orientation systematically increased candidates’ levels of electoral support, and standing for a party with the “wrong” regional orientation systematically decreased them. 

 

TABLES 10-13 HERE

 

            These comparative results can only be suggestive.  More definitive findings must await both better data and more fine-grained analysis of electoral contests in key constituencies.  But the evidence does suggest that regime change may affect more than just the level of ethnic conflict in African countries.  It may also alter the kinds of ethnic cleavages in terms of which political competition and ethnic mobilization takes place.

 

 

 


Bibliography

 

 

Bayart, Jean-Francois, 1993, The State in Africa: The Politics of the Belly (London:  Longman).

 

Carey, John and Matthew Shugart, 1995, “Incentives to Cultivate a Personal Vote: A Rank Ordering of Electoral Formulas,” Electoral Studies 14 (December), pp. 417-39.

 

Chabal, Patrick and Jean-Pascal Daloz, 1999, Africa Works: Disorder as Political Instrument (Oxford: James Currey).

 

Chazan, Naomi, 1982, “Ethnicity and Politics in Ghana,” Political Science Quarterly 97 (Fall), pp. 461-85.

 

de Nevers, Renee, 1993, “Democratization and Ethnic Conflict,” in Michael E. Brown, Ethnic Conflict and International Security (Princeton: Princeton University Press), pp. 61-78,

 

Glickman, Harvey, 1995, Ethnic Conflict and Democratization in Africa (Atlanta:  African Studies Association Press).

 

Gurr, Ted Robert, 2000, People Versus States: Minorities at Risk in the New Century (Washington, DC: United States Institute of Peace Press).

 

Horowitz, Donald, 1985, Ethnic Groups in Conflict (Berkeley and Los Angeles: University of California Press).

 

Joseph, Richard, 1984, “Class, State and Prebendal Politics in Nigeria,” in Nelson Kasfir, State and Class in Africa (London: Frank Cass), pp. 21-38.

 

Ottaway, Maria, 1994, Democratization and Ethnic Nationalism: African and Eastern European Experiences (Washington, DC: Overseas Development Council).

 

Posner, Daniel N., 1988, The Institutional Origins of Ethnic Politics in Zambia, Ph.D. Dissertation, Department of Government, Harvard University.

 

Smith, Zeric Kay, 2000, “The Impact of Political Liberalization and Democratization on Ethnic Conflict in Africa: An Empirical Test of Common Assumptions,” Journal of Modern African Studies 38, 1, pp. 21-39.

 

Snyder, Jack, 2000, From Voting to Violence: Democratization and Nationalist Conflict (New York: W.W. Norton and Company).

 

Throup, David and Charles Hornsby, 1998, Multiparty Politics in Kenya: The Kenyatta and Moi State and the Triumph of the System in the 1992 Elections (London: James Currey),

 

Young, Crawford and Thomas Turner, 1985, The Rise and Decline of the Zairian State (Madison: University of Wisconsin Press).


 

 

Table 1:

Does the Share of the Dominant Tribe in the Constituency Predict the Share

of the Vote Won by Dominant Tribe Candidates?  Are the Results

Different in Multi-Party and One-Party Contests?

 

(all cases)

 

 

 


Dependent Variable is Share of Vote Won by

Dominant Tribe Candidates

(standard errors in parentheses)

 

 

multi-party

one-party

multi-party

one-party

share of voters belonging to dominant tribe

 

0.377**

(0.105)

 

0.639**

(0.065)

 

0.331**

(0.156)

 

0.643**

(0.090)

Central

Province

 

 

-0.015

(0.092)

0.032

(0.050)

Eastern

Province

 

 

0.001

(0.102)

-0.039

(0.054)

Luapula

Province

 

 

0.171

(0.127)

0.107

(0.062)

Lusaka

Province

 

 

-0.056

(0.093)

-0.042

(0.096)

Northern

Province

 

 

0.111

(0.114)

0.012

(0.052)

Northwestern

Province

 

 

0.028

(0.099)

0.015

(0.066)

Southern

Province

 

 

-0.096

(0.102)

-0.004

(0.058)

Western

Province

 

 

-0.081

(0.093)

-0.031

(0.044)

N

155

239

155

239

R2

 

0.08

0.29

0.13

0.31

 

** significant at the 0.01 level


 

 

 

Table 2:

Voters Whose Behavior Cannot Be Accounted for

in the Tribal Voting Analysis

 

 

 

 

 

Candidates in the race

are members of

 

 

 

Tribe A

 

Tribe B

 

Tribe C

Voters in the constituency are members of

 

Tribe A

 

 

 

 

 

Tribe B

 

 

 

 

 

Tribe C

 

 

 

 

 

Tribe D

 

 

?

 

?

 

?

 

Tribe E

 

 

?

 

?

 

?

 

Tribe F

 

 

?

 

?

 

?

 


 

 

Table 3:

Does the Share of the Dominant Tribe in the Constituency Predict the Share

of the Vote Won by Dominant Tribe Candidates?  Are the Results

Different in Multi-Party and One-Party Contests?

 

(including only cases in which more than 85% of voters

had a candidate from their tribe in the race)

 

 

 

 


Dependent Variable is Share of Vote Won by

Dominant Tribe Candidates

(standard errors in parentheses)

 

 

multi-party

one-party

multi-party

one-party

share of voters belonging to dominant tribe

 

0.203

(0.339)

 

0.801**

(0.175)

 

0.506

(0.359)

 

0.943**

(0.219)

Eastern

Province

 

 

 

-0.043

(0.070)

Luapula

Province

 

 

-0.218

(0.174)

0.404

(0.218)

Northern

Province

 

 

0.129

(0.122)

 

Northwestern

Province

 

 

0.054

(0.174)

0.005

(0.135)

Southern

Province

 

 

-0.256

(0.113)

-0.100

(0.100)

Western

Province

 

 

-0.235

(0.294)

0.042

(0.119)

N

41

53

41

53

R2

 

0.01

0.29

0.26

0.36

 

** significant at the 0.01 level


 

 

Table 4:

Tribal Voting in Multi-Party Elections in Zambia

 

(includes all electoral constituencies in 1968, 1991, 1996 and all 3rd Republic by-elections)

 

 

 

 

Is the Candidate from the Dominant Tribe in the Constituency?

 

 

Yes

 

No

 

Candidate Won Seat

 

 

213

 

86

 

Candidate Did Not

Win Seat

 

201

 

97

 

 

 

 

 

 

 

 

Table 5:

Language-Group Voting in Multi-Party Elections in Zambia

 

(includes all electoral constituencies in 1968, 1991, 1996 and all 3rd Republic by-elections)

 

 

 

 

Is the Party on Whose Ticket the Candidate is Standing Affiliated with the Dominant Language Group in the Constituency?

 

 

Yes

No

 

Candidate Won Seat

 

 

152

 

147

 

Candidate Did Not

Win Seat

 

22

 

276

 


 

 

 

 

Table 6:

Four Types of Candidates in Multi-Party Elections

 

 

 

 

 

Is the Candidate from the Dominant Tribe in the Constituency?

 

 

 

Yes

 

No

 

Is the Party on Whose Ticket the Candidate is Standing Affiliated with the Dominant Language Group in the Constituency?

 

Yes

 

 

A

 

B

 

No

 

C

 

D

 

 

 

 

 

 

 

 

Table 7:

Percent of the Time that Each Type of Candidate

Wins the Seat He/She is Contesting

 

 

 

 

 

Is the Candidate from the Dominant Tribe in the Constituency?

 

 

 

Yes

 

No

 

Is the Party on Whose Ticket the Candidate is Standing Affiliated with the Dominant Language Group in the Constituency?

 

Yes

 

 

90%

 

79%

 

No

 

34%

 

37%

 

 


 

 

Table 8:

Language-Group Voting in Zambia’s 1991 Multi-party Election

 

(includes only candidates who stood in both 1988 and 1991, in the same constituencies)

 

 

 

 

Is the Party on Whose Ticket the Candidate is Standing Affiliated with the Dominant Language Group in the Constituency?

 

 

Yes

No

 

Candidate Won Seat

 

 

15

 

11

 

Candidate Did Not

Win Seat

 

1

 

46

 

 

 

 

 

 

 

Table 9:

How Did Electoral Performance Vary Between 1988 and 1991

Among Candidates Who Stood in Both Zambian Elections?

 

(includes only candidates who stood in both 1988 and 1991, in the same constituencies)

 

 

 

Candidates Standing on Tickets of Parties Affiliated with the Dominant Language Group in the Constituency Who Won the Seat

(upper left cell in Table 8)

Candidates Standing on Tickets of Parties Not Affiliated with the Dominant Language

Group in the Constituency

Who Lost the Seat

(lower right cell in Table 8)*

 

Candidate Out-Performed His/Her 1988 Results

 

13

 

8

 

Candidate Under-performed His/Her 1988 Results

 

2

 

32

 

 

 

* Six candidates were unopposed in 1988 and were thus dropped from the comparison


 

Table 10:

Regional Voting in Kenya’s 1992 Multi-party Election

 

(includes only candidates who stood in both 1988 and 1992, in the same constituencies)

 

 

 

 

Is the Party on Whose Ticket the Candidate is Standing Associated with the Region in which the Constituency is Located?

 

 

Yes

No

 

Candidate Won Seat

 

 

61

 

19

 

Candidate Did Not

Win Seat

 

34

 

64

 

 

 

 

 

 

Table 11:

How Did Electoral Performance Vary Between 1988 and 1992

Among Candidates Who Stood in Both Kenyan Elections?

 

(includes only candidates who stood in both 1988 and 1992, in the same constituencies)

 

 

 

Candidates Standing on Tickets of Parties Affiliated with the Region in which the Constituency is Located Who Won the Seat

(upper left cell in Table 10)*

Candidates Standing on Tickets of Parties Not Affiliated with the Region in which the Constituency is Located

Who Lost the Seat

(lower right cell in Table 10)**

 

Candidate Out-Performed His/Her 1988 Results

 

42

 

9

 

Candidate Under-performed His/Her 1988 Results

 

0

 

44

 

 

* Nineteen candidates were unopposed in either 1988 or 1992 and were thus dropped from the comparison.

** Eleven candidates were unopposed in 1988 and were thus dropped from the comparison.


 

Table 12:

Regional Voting in Tanzania’s 1995 Multi-party Election

 

(includes only candidates who stood in both 1990 and 1995, in the same constituencies)

 

 

 

 

Is the Party on Whose Ticket the Candidate is Standing Associated with the Region in which the Constituency is Located?

 

 

 

Yes

No

 

Candidate Won Seat

 

 

48

 

7

 

Candidate Did Not

Win Seat

 

2

 

19

 

 

 

 

 

 

Table 13:

How Did Electoral Performance Vary Between 1990 and 1995

Among Candidates Who Stood in Both Tanzanian Elections?

 

(includes only candidates who stood in both 1990 and 1995, in the same constituencies)

 

 

 

Candidates Standing on Tickets of Parties Affiliated with the Region in which the Constituency is Located Who Won the Seat

(upper left cell in Table 12)*

Candidates Standing on Tickets of Parties Not Affiliated with the Region in which the Constituency is Located

Who Lost the Seat

(lower right cell in Table 12)**

 

Candidate Out-Performed His/Her 1990 Results

 

48

 

2

 

Candidate Under-performed His/Her 1990 Results

 

0

 

17

 

 

* Two candidates were unopposed in 1990 and were thus dropped from the comparison.

** Seven candidates were unopposed in 1990 and were thus dropped from the comparison.

 



[1] See, for example, Ottaway (1994), Glickman (1995) and Smith (2000).  Outside of Africa, De Nevers (1993), Gurr (2000) and Snyder (2000) treat the effects of democratic change on ethnic conflict in broader comparative perspective.

[2] For evidence that this is the case in Zambia, see Posner (1998: ch. 3).  For similar evidence in Zaire, see Young and Turner (1985).

[3] Regarding the assumption that others are also voting along ethnic lines, Horowitz (1985) stresses that even voters who might be inclined not to vote ethnically often do so because they assume that voters from other groups will do so:  “The incentives toward reactive ethnic voting are strong.  When voters of one group choose, in effect, not to choose but to give their vote predictably on an ethnic basis to an ethnically defined party, they put voters of the other group who do choose among parties at a collective disadvantage.  All else being equal, such voters will seek to reduce their disadvantage by concentrating their votes in a comparable ethnic party.  In such a situation, ethnic voters tend to drive out non-ethnic votes” (323).

[4] Horowitz (1995: 321) points out that “ethnically aware voters have understood that presenting a multiethnic slate is an exigency of political life, even for an ethnic party, and have accordingly voted for the ethnic party rather than for or against the ethnic identity of the individual candidates.  When voters elect minority members of their ethnic party, it is wrong to regard this as non-ethnic voting.  Quite the contrary:  it is party and not candidate ethnic identification that counts.”

[5] Carey and Shugart (1995).  Although their analysis only deals with multi-party systems, one of the systems they discuss ( SNTV with open endorsements and district magnitude = 1) exactly captures the one-party single-member plurality system that I describe in this paper.  This is the system in which, according to their typology, candidates’ personal reputations are most important.   

[6] For an explanation of why tribe and language, but not other potential bases of ethnic identity, play this role, see Posner (1998: ch. 2).

[7] I did this by compiling a list, for each of the country’s fifty-seven administrative districts, of every candidate that had ever stood for parliament in that district.  I then independently asked at least two “experts” from each district (ordinarily long-time residents of the district who had recently migrated to Lusaka, where I collected these data) to help me identify the tribal backgrounds of each candidate based on their names.  In cases where “experts” disagreed, a third person from the district was consulted.

[8] Remarkably for a developing country, Zambia’s 1990 census collected information on respondents’ tribal backgrounds.  By “building up” electoral constituencies from census supervisory area-level data, I was able to use the census data to calculate the exact tribal demographics of every constituency. 

[9] Making inferences about individual voting behavior from aggregate data risks committing an error of ecological inference.  To make certain this was not the case, I re-analyzed the data using Gary King’s Ecological Inference program and was able to confirm the findings I report in this paper.  For details of these and other tests and analyses, see Posner (1998).

[10] An additional benefit of narrowing the analysis to cases where more than 85 percent of the voters have a candidate from their tribe in the race is that it solves the problem caused by the variation across elections in the average number of candidates that are competing (which ranges from 6.1 candidates per seat in 1983 to 1.7 in 1968).  All things equal, the more candidates per seat, the greater the likelihood that voters will have candidates from their tribe to vote for.  This means that, in the full sample, the accuracy of our inferences about tribal voting varied across elections (higher in the elections with more candidates; lower in those with fewer).  What makes this important for our purposes here is that the average number of candidates in the race differed somewhat between multi-party elections (2.7) and one-party elections (4.0), so our inferences about tribal voting in one-party contests were more reliable than those for multi-party contests.  Looking only at cases where more than 85 percent of the voters have a candidate from their tribe in the race minimizes this source of potential bias.

[11] It is possible, of course, that these results occur in cases where dominant tribe and language group voters split their support between two candidates and a third candidate, not of the dominant tribe or language group, emerges victorious.  Such an outcome would “look” in the data like a type D winner, but would, in fact, be an instance of strict ethnic voting.

[12] Although ecological inference problems plague a number of the analyses I have presented, a finding that a given candidate won a larger share of votes than his own tribe’s share of the voting population is definitive evidence of cross-tribal voting. 

[13] These were the fifteen that were held during the 3rd Republic, where it was thus possible to draw on census data to identify the exact share of the population that belonged to each candidate’s tribe.  I was unable to do this for the five head-to-head contests that took place during the 1968 election because of limitations in my data on constituency-level ethnic demographics in that election.

[14] I have plans to conduct parallel analyses in Malawi, but was unable to complete the necessary data collection in time for this draft of the paper.

[15] It also reduces the number of candidates for whom I will ultimately needed to collect information about ethnic group memberships to fully replicate the tests I perform on the Zambian data.

[16] To determine whether a candidate fared better or worse in 1991 than in 1988, I used the following coding rules:  If the candidate won in 1988 but lost in 1991, then the answer is “no” (i.e., fared worse).  If the candidate won in 1991 but lost in 1988, then the answer is “yes” (i.e., fared better).  If the candidate either lost or won in both 1988 and 1991, then I compared the candidate’s vote share in each contest against 100/total number of candidates in the race (i.e., against a 20 percent “expected vote” if there were five candidates in the race), and determined whether, relative to this benchmark, the candidate did better or worse in 1991.

[17] I am indebted to Timothy Ayieko and Bernadeta Killian for help in coding the match between parties’ regional orientations and the regional locations of each constituency. 

[18] For evidence of the latter, see Throup and Hornsby (1998).