Regime Change and Ethnic Cleavages in Africa

1. Because I focus on the impact of transitions from one-party to multiparty political competition, I only treat countries that previously held reasonably competitive single-party elections and now hold reasonably competitive multiparty contests (or have held at least one such election since 1989). This is a more limited understanding of regime change than is found in much of the literature, as it excludes transitions from regimes in which elections were never held, such as military regimes and settler oligarchies.

2. All of these identities are “ethnic,” in the broad sense that they all designate communities whose members are assumed to be related by descent (see Fearon & Laitin, 2000). Here, and throughout the article, I use the term tribe to refer to an ethnic community that is (or was historically) organized under the authority of a traditional chief. Membership in a tribe is determined by the answer to the question, “Are you (or were your parents) subjects of Chief X?”

3. The argument assumes a single member plurality or majoritarian system. The logic becomes less clear, and the predictions of the argument weaker, under conditions of proportional representation (PR). In a PR system, given a low enough threshold, the criterion of winnability may apply to multiple groups.

4. Regarding the presumption that everyone else is voting along ethnic group 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 . . . 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” (p. 323).

5. Note, however, that even where voters have only hazy understandings of relative group sizes, or of how new electoral institutions will aggregate their votes, vote-seeking politicians and parties—especially those that can claim to represent the minimum-winning group—will go out of their way to explain to voters why a vote for them is in the voters' best interest. So a voter's job may be less to deduce from scratch which strategy is optimal than the much simpler task of weighing the different logics presented by the competing candidates and parties.

6. The stipulation that ethnic coalitions not be close in size may also be important for a second reason. The literature on oversized parliamentary coalitions (e.g., Groseclose & Snyder, 1996) suggests that voters may find it advantageous to choose memberships in coalitions that are larger then minimum-winning. Although the logic of parliamentary coalitions (where defections are a constant threat) is somewhat different from the logic of voting coalitions (where defection is only meaningful at election time), the general message of this literature may nonetheless be relevant. The implication would be, once again, that the clear decision rule I ascribe to voters will hold best when voters are choosing between two potential coalitions where the larger of the two is more than just marginally larger than minimum-winning.

7. Political parties are also strategic actors in the story, at least in multiparty elections. They must decide which candidates to run and what issues or ethnic affiliations to stress during the campaign. I do not discuss these decisions explicitly because I assume that they follow the same logic as that ascribed to the politicians/candidates.

8. One-party and multiparty regimes often differ on other dimensions as well—for example, in the freedom they provide for civil society groups and the press, the opportunities they offer for incumbent legislators to be displaced by challengers, and the degree of governmental control they require over campaigning practices and electoral appeals. However, it would be difficult to link variation on these dimensions to variation in the kinds of ethnic cleavages that emerge as salient in one-party and multiparty settings.

9. Scheiner's (2005) work on Japan provides a parallel argument about how the centralization of political power in a patronage-oriented system causes politics—even local politics— to become nationalized.

10. The argument about the localization of cleavages under one-party politics owes inspiration to V. O. Key (1949). Bates (1989, p. 92) provides a slightly different argument that leads to the same result. He argues that national issues, and the national frame, will be salient in multiparty elections because voters will view candidates as potential members of coalitions that might conceivably form the government and shape national policy. In single-party elections, however, voters know that each candidate will have only a negligible impact on national policy, because even if they are successful, they will only be a single member of Parliament. This calculation, Bates argues, shifts voters' attention from national policy issues (in multiparty contests) to patronage concerns (in single-party contests). Thus, national issues and cleavages will animate multiparty politics and local-level rivalries will structure one-party politics. Although deductively sound, there are two problems with applying this argument to most African countries. First, it overestimates the extent to which voters ever view candidates as shapers of national policy agendas. Politics in contemporary Africa is so president-centered that legislatures play very little role in shaping policy, and voters know this. Indeed, Wantchekon's (2003) findings show clearly that policy considerations are of only minor concern to African voters. Second, Bates's account underestimates the role of patronage concerns in competitive party settings. Targetable benefits—be they individualized benefits such as money or jobs or local public goods such as roads or clinics—play a central role in both one-party and multiparty elections in Africa (though in the latter case, they are used both for individual vote buying and for organization building). The account provided here reaches the same conclusions as Bates does without making assumptions about either the extent to which voters see candidates as policy makers or the relative salience of patronage in one-party and multiparty regimes.

11. Although Carey and Shugart's (1995) analysis deals only with multiparty systems, one of the systems they discuss (single nontransferable vote [SNTV] with open endorsements and district magnitude = 1) exactly captures the one-party single-member plurality system described in this article. This is the system in which, according to their typology, candidates' personal reputations are most important.

12. Urban constituencies should reflect a different pattern. High rates of in-migration to Zambia's urban areas during the colonial era transformed urban districts into demographic microcosms of the country as a whole. Thus, the structure of national- and constituency-level cleavages in urban areas of Zambia are nearly identical, and we would expect the shift from one-party to multiparty rule to generate no change in the kinds of identities that emerge as salient. Because 80% of Zambia's electoral constituencies are located in rural areas, the rural outcome can reasonably be taken to be the outcome writ large. Nonetheless, in the analyses that follow, I limit my treatment to rural constituencies only.

13. Of course, such an analysis is only possible if constituencies are ethnically heterogeneous, which they are in Zambia (but, unfortunately, are not in Kenya).

14. I did this by compiling a list for each of the country's 57 administrative districts of every candidate who ran for Parliament in that district in these elections. I then independently asked at least two long-time residents of each district to help me identify the tribal backgrounds of the candidates on my list based on the candidates' names. In cases where my informants disagreed on the ethnic background of a candidate, I consulted at least two additional people from that district. For a fuller discussion of these procedures, see Posner (2005, Appendix C).

15. Zambia's 1990 census included the question, “What is your Zambian tribe?” By building up electoral constituencies from local census units, I was able to calculate the exact tribal demographics of every constituency. For details, see Posner (2005, Appendix D).

16. There are well-known dangers in making inferences about individual-level behavior (such as voting decisions) from aggregate data. But as long as there is no reason to believe that the potential bias from the ecological inference problem is greater in one-party than multiparty elections, this should not present a major difficulty for the analysis I am undertaking here.

17. I exclude candidates that ran unopposed in either the one-party or the multiparty contest.

18. Dominant language groups were identified from constituency-level 1990 census figures. Mambwe- and Namwanga-speakers were included in the Bemba-speaking group, and Tumbuka-speakers were included in the Nyanja-speaking group. I coded the Movement for Multiparty Democracy (MMD) as a “Bemba” party because its president in 1991 was from the Bemba language group. I coded the United National Independence Party (UNIP) as a “Nyanja” party because its president at the time was popularly viewed as a Nyanja speaker. MMD candidates were coded as running on the ticket of the “right” party when they were running in constituencies in the Bemba-speaking Northern Province, Copperbelt Province, and Luapula Province; UNIP candidates were coded as running on the ticket of the “right” party when they were running in constituencies in Nyanja-speaking Eastern Province. I excluded from the analysis candidates that ran as independents (of which there were eight) and candidates where the match between their party label and the dominant language group in the constituency was ambiguous (of which there were, again, eight).

19. All of the candidates were incumbents; hence the absence of a control for incumbency.

20. Note that whereas the analysis permits claims to be made about the relative salience of national-scale and local cleavages in multiparty contests, I do not have a measure of the salience of national-level cleavages during one-party elections and thus cannot make claims about the impact of regime change on the absolute importance attached to national-level cleavages.

21. The 1969, 1974, and 1979 contests were just de facto one-party contests, as a one-party state was not formally declared until 1982.

22. In a 2002 interview (Quist-Arcton, 2002), Kenyatta confirmed this interpretation. Asked why he lost the 1997 parliamentary election even though he was running in a heavily Kikuyu district, Kenyatta replied that “my problem was that I was in the wrong party. I was in KANU. . . . The main push then was `join DP, join any other party and we will vote for you.' . . . People were voting largely on the basis of ethnicity and it depended on which party you were in and, as far as they were concerned, [the DP] was the party that they wanted the Kikuyu community to be in and that was it.”

23. I coded parties' home areas as follows: Democratic Party of Kenya (DP; Central and Eastern Provinces); Forum for the Restoration of Democracy—Asili (FORD-Asili; Central Province and Nairobi); Forum for the Restoration of Democracy—Kenya (FORD-K; Nyanza Province); Kenya African National Union (KANU; Northeastern, Coast, Rift Valley, Western Provinces, and parts of Eastern Province). I am indebted to the late Judy Geist for her generous help with this coding effort. Where the match between a candidate's party label and the dominant language group in the constituency was ambiguous, I dropped them from the analysis. There were nine such cases.

24. Only two parties competed in the 1991 election in Zambia (MMD and UNIP, the incumbent), so the “ran on MMD ticket” dummy variable in the Zambia regression also picks up the effect of running on a UNIP ticket (and with it, any advantage that UNIP candidates may have obtained from their party's control of the state apparatus).

25. Note that, unlike in the Zambia analysis, it is impossible to compare the relative salience of local- and national-scale ethnic attachments in Kenya's multiparty races. This is because, as noted, all Kenyan candidates are members of the dominant tribal groups in their areas, which makes it impossible to test for the impact of this factor on election results. Also, as in the Zambia analysis, I do not have a direct measure of the strength of national-level cleavages in one-party elections and thus cannot make claims about changes in the absolute salience of national-level cleavages.

26. For a good summary, see Gibson (2002).