Chapter Five—
Electoral Systems for a Divided Society

The electoral system is by far the most powerful lever of constitutional engineering for accommodation and harmony in severely divided societies, as indeed it is a powerful tool for many other purposes.^[1] Unfortunately, one would hardly sense the potential of electoral innovation for conflict or accommodation from reading the standard literature on electoral systems. Most of that literature is concerned with the impact of various electoral systems on the number of political parties, the proportionality of seats to votes, the strength or weakness of party organizations, and the relationship (or its absence) between legislators and their constituents. Ethnic and racial relations are a decidedly secondary theme.^[2]

Only when severely divided societies actually come up against the

[1] See Giovanni Sartori, "Political Development and Political Engineering," Public Policy 17 (1968): 261–98.

[2] For a sampling of the literature, see Arend Lijphart and Bernard Grofman, eds., Choosing an Electoral System (New York: Praeger, 1984); Vernon Bogdanor, ed., Representatives of the People? (London: Gower, 1985); Maurice Duverger, Political Parties: Their Organization and Activity in the Modern State (London: Methuen, 1965); David Butler, "Electoral Systems," in David Butler et al., eds., Democracy at the Polls: A Comparative Study of Competitive National Elections (Washington, D.C.: American Enterprise Institute, 1981), pp. 7–25; Rein Taagepera and Mathew S. Shugart, "Designing Electoral Systems," Electoral Studies 8, no. 1 (April 1989): 45–58. For a South African example of how the ethnic and racial implications of various electoral systems are hardly noticed, see W. H. Olivier, "Party Systems and Electoral Systems," in D. J. van Vuuren and D. J. Kriek, eds., Political Alternatives for Southern Africa (Durban: Butterworths, 1983), pp. 334–55.

need to innovate, the need to choose an electoral system, do they survey the alternatives. Because the literature does not respond directly to their concerns or even help them to think about the question,^[3] they tend to have arbitrary or half-formed preferences. In South Africa, for example, many people continue to think the Anglo-American plurality system is the most democratic. Some others think that some form of proportional representation might be more appropriate for a divided society and sometimes rather casually recommend a version of PR for a future South Africa.^[4] They rightly fear the frequent tendency of plurality systems to underrepresent minorities and to produce legislative majorities from mere pluralities—or even less than pluralities—of voters.^[5] This is a phenomenon with which South Africans have had repeated experience. In 1948 and 1953, the National Party gained a majority of seats on less than half the vote. In 1948, the defeated United Party actually received more than half the vote; and in 1953, the National Party won 60 percent of the seats on under 50 percent of the vote, while the United Party won only 37 percent of the seats on nearly the same vote.^[6] In 1989, suffering erosion on both liberal and conservative flanks, the Nationalists again benefited from a seat bonus attributable to the plurality system, winning a 56 percent majority of seats on only 48 percent of the vote.

"The surest way to kill the idea of democracy in a plural society," wrote Sir Arthur Lewis a quarter-century ago, "is to adopt the Anglo-American electoral system of first-part-the-post."^[7] Lewis's sage discus-

[3] For one of the very few notable exceptions, see J. A. Laponce, "The Protection of Minorities by the Electoral System," Western Political Quarterly 10, no. 2 (June 1957): 318–39. Cf. S. M. Lipset, "Party Systems and the Representation of Groups," European Journal of Sociology 1, no. 1 (1960): 50–85.

[4] Often, the single transferable vote (discussed below) is recommended. See, e.g., Charles Simkins, Reconstructing South African Liberalism (Johannesburg: South African Institute of Race Relations, 1986; mimeo.), p. 79; Donald B. Molteno et al., Final Report of the Commission Set Up by the Progressive Party to Make Recommendations on a Revised Constitution for South Africa , vol. 1 (N.p. 1960), p. 16.

[5] See M. Wiechers, "The Franchise and Alternative Electoral Systems," in L. J. Boulle and L. G. Baxter, eds., Natal and KwaZulu: Constitutional and Political Options (Cape Town: Juta, 1981), p. 186. See also L. Schlemmer, "An Overview," in ibid., p. 209.

[6] For a table of seats and votes, see Gwendolen M. Carter, The Politics of Inequality: South Africa since 1948 , 2d ed. (New York: Praeger, 1959), pp. 448–49. Compare the figures in Richard Hodder-Williams, "South Africa: Democratic-Centralism versus Elite-based Parties," in Alan Ware, ed., Political Parties: Electoral Change and Structural Response (Oxford: Blackwell, 1987), p. 25.

[7] W. Arthur Lewis, Politics in West Africa (London: Oxford University Press, 1965), p. 71. The full discussion of electoral systems appears in ibid., pp. 69–74. Its main omission is the tradeoff between influence and representation—values assumed to be compatible. Ibid., p. 73. I shall discuss this tradeoff shortly.

sion of elections and group identity has largely been forgotten. Today, debate rarely proceeds beyond the relative fairness of PR versus plurality elections and the overall suitability of the two systems for divided societies.^[8] Yet it is necessary to go much further than where the literature and most constitution makers in divided societies leave off, in order to assess the appropriateness of alternative electoral systems. The test of a good electoral system is not to be found merely in the ratio of seats to votes or in the number of parties that emerge. The test lies instead in the posture adopted by the parties with respect to other parties and with respect to voters. Concretely, does the electoral system dispose the parties to ethnic and racial inclusion or exclusion? Will one system, rather than another, encourage parties to seek intergroup compromise and accommodation? The answers to such questions are a complicated matter, because they cannot be given a priori; competing electoral systems require evaluation in context. But this much is clear. The choices are far more extensive than those implied by the simple dichotomy of proportional or plurality systems.

Not only are the choices more extensive; they are also clear-cut. There are two different goals for promoting the inclusion of any ethnic or racial group in a divided polity. One is ethnic representation, in the tangible but narrow sense of legislative office holding. The other goal also relates to representation, but in the broader sense of incorporating the concerns and interests of a given ethnic or racial group in the calculations of politicians belonging to a variety of groups. Measures that will guarantee seats to a minority group may not foster the inclusion of that group's interests more broadly in the political process—indeed, may have the opposite effect. And vice versa: measures that encourage the accommodation of group claims do not necessarily result in proportionate office holding for all groups. Those who aim at ethnic office holding tend to prefer electoral systems that make it easy for minorities to obtain their own seats.^[9] If, however, the aim is moderation on divisive issues and the amelioration of intergroup conflicts, a quite different electoral system might be preferred. Lack of clarity about the tradeoff

[8] See, e.g., F. van Zyl Slabbert and David Welsh, South Africa's Options: Strategies for Sharing Power (New York: St. Martin's Press, 1979), pp. 152–53; Political Commission of the Study Project on Christianity in Apartheid Society, South Africa's Political Alternatives , SPRO-CAS Publication no. 10 (Johannesburg: Ravan Press, 1973), p. 206.

[9] See, e.g., Arend Lijphart, "The Power-sharing Approach," in Joseph V. Montville, ed., Conflict and Peacemaking in Multiethnic Societies (Lexington, Mass.: Lexington Books, 1990), p. 508; Simkins, Reconstructing South African Liberalism , p. 79.

between influence and office holding is responsible for much of the confusion in thinking about electoral systems for South Africa and other divided societies.

Systems and Parties:
A Glossary

Since I intend to discuss the impact of electoral systems on party systems and on intergroup conflict and accommodation, perhaps it is best to insure that the terminology is clear from the outset. So I begin with a brief narrative glossary.

First of all, there are several electoral systems. As indicated earlier, first-past-the-post refers to plurality elections. Whichever candidate receives the largest number of votes is elected, even if that number of votes is less than a majority. First-past-the-post or plurality systems are generally used in single-member constituencies, those that consist of one representative per area. Proportional representation refers to a variety of electoral systems designed to insure that parties receive approximately the same percentage of the total number of seats as the percentage they received of the total number of votes.

Two of the three systems to be discussed—the single transferable vote (STV) and the alternative vote (AV)—I shall describe at the time I introduce them into the discussion. The alternative vote is perhaps better described as a majority system than as a PR system, since it declares elected only those candidates who receive a majority, rather than a plurality, of votes. But, like PR systems, AV mitigates the winner-take-all aspects of plurality systems and generally achieves better proportionality of seats to votes than plurality systems do.^[10] AV can also be described as a preferential system , since it enables or requires voters to list their choice of candidates in order of preference. The third PR system—list-system proportional representation —which we shall encounter very soon, permits each party to put up a list of candidates. Each voter then votes for one or another party list. Each list is numbered in order of the party's preference for its own candidates. Candidates at the top of a list are elected first. So a party with, say, 20 percent of the vote secures election of the top 20 percent of its list.^[11] List-system PR can operate

[10] See, e.g., the rerun of the 1987 British general election under a hypothetical AV system in John L. Old, "The Alternative Vote and the 1987 General Election," Politics 8, no. 1 (April 1988): 8–13.

[11] There are some variants of list-system PR that permit voters to alter the party's rank order of candidates. One of these variations will be discussed later in this chapter. A variation that permits voters to choose across party lists I mention in note 27, below.

either using territorially demarcated constituencies, each with its own competing lists of candidates to represent that constituency, or using a whole country as a single constituency, each party putting up just one national list. The former I shall refer to as constituency-list-system PR; the latter, as national-list-system PR.

I shall also refer to several types of party formation. Ethnically based parties are parties that draw their support from a single ethnic or racial group. Any ethnic or racial group, however, may give its support to one or more than one such party. Multiethnic coalitions are temporary or permanent alliances formed by two or more than two ethnically based parties. Flank parties are ethnically based parties surrounding a multiethnic coalition and typically espousing ethnically more extreme positions than the coalition, with its mixed support, is able to do.

Finally, there are a few unfamiliar terms that I shall use regarding electoral demography, numbers of parties, coalitions, and the appeals of parties to voters. Constituencies may be either homogeneous or heterogeneous , depending on whether they are composed of significant numbers of two or more ethnic or racial groups in conflict. Party proliferation is the growth of parties to form a multiparty system , usually making coalition governments necessary, as opposed to a two-party system, in which one party is able to form a government by itself, leaving the other in opposition. Coalitions are made possible as parties pool seats to achieve a legislative majority. Vote pooling , on the other hand, refers to an exchange of votes of their respective supporters by two parties. The vote pooling to which I refer takes place across ethnic or racial lines and is generally the result of agreements between parties for the exchange of electoral support. I shall say much more about vote pooling as we proceed, because—unlike the mere pooling of seats—it lies at the heart of intergroup compromise in severely divided societies.

Electoral Systems:
Beyond Proportionality

The most careful work on the South African electoral future has been done by a non—South African, Arend Lijphart.^[12] Lijphart has advanced a full-blown consociational plan for South Africa, which I have commented on in Chapter 4. Here I want to focus on the electoral system he recommends, because it is assuredly the best-thought-through elec-

[12] Arend Lijphart, Power-Sharing in South Africa (Berkeley: University of California Institute of International Studies, 1985).

toral scheme for South Africa and because his plan has had considerable attraction for South Africans in quest of answers to their problems. Above all, it is addressed to the problems that will likely haunt the South African future; the plan does not wish those problems away. As I shall show, however, the comparative evidence is heavily against it.

Lijphart argues for proportional representation as an integral part of a consociational constitution, according the segments of a divided society a proportional say in the running of that society. In Power-Sharing in South Africa , the principal justification he provides for proportional representation on specifically electoral grounds is that it is fairer than first-past-the-post.^[13] Whereas plurality systems typically exaggerate the number of seats won by the largest contenders, artificially diminishing minority representation, PR can be designed so that the number of seats reflects faithfully the number of votes each party has obtained. This is true as far as it goes, but it seems a curious rationale, if that is all there is to it, for Lijphart's plan is also to give minorities—even "relatively small groups"—an "absolute veto on the fundamental issues, such as cultural autonomy, and a suspensive veto on non-fundamental questions."^[14] Moreover, he wants chairs of the executive to rotate, that is, a rotating prime ministership.^[15] In the face of these arrangements, small disparities in representation would seem less important than they might otherwise be.

But Lijphart has a further reason for advocating PR: it allows the constituent elements of a plural society to form into political parties. "The beauty of PR is not just that it yields proportional results and permits minority representation—two important advantages from a consociational perspective—but also that it permits the segments to define themselves ."^[16] This is an important argument in South Africa, where the government has decided what the categories of group membership are and who belongs to each, but it should not obscure the fact that first-past-the-post and most other electoral systems also permit self-identification and self-organization. PR is not superior on these grounds.

In a later paper, Lijphart expands upon his justification for PR and argues for a specific variant: party-list-system PR, which he juxtaposes to the single transferable vote version of PR.^[17] There he prefers PR be-

[13] Ibid., p. 8.

[14] Ibid., p. 81.

[15] Ibid., p. 80.

[16] Ibid., pp. 68–69 (emphasis in the original).

[17] Arend Lijphart, "Choosing an Electoral System for Democratic Elections in SouthAfrica: An Evaluation of the Principal Options," Institute for the Study of Public Policy, University of Cape Town, Critical Choices for South Africa series (September 1987). As indicated above, I shall say more about these varieties of PR later in this chapter.

cause of its tendency to encourage many parties, so that each group can have its own party if it wishes. List-system PR is better than the single transferable vote, because it generally provides closer proportionality of seats to votes and obviates the need for unduly large constituencies to avoid ethnic gerrymandering. All of this is generally true,^[18] but what does it mean for a severely divided society like South Africa?

On this point, Lijphart is clear. He assumes that, with a receptive electoral system, there will be a plurality of parties in South Africa and that they will need to form a coalition in order to govern,^[19] hopefully, in his view, a "grand" coalition representing all the groups.^[20] This is made clear in his book on South Africa. Commenting on Zimbabwe, he says: " . . . when one ethnic group comprises a majority of the population, PR can obviously not prevent this majority from gaining a majority of seats."^[21] So the assumption is that Black voters in South Africa will not be cohesive enough to form a Zimbabwe-like majority and a coalition will be necessary, once each party has gathered the support of its own group at the polls.

Even under PR, I am not certain that this would be the outcome in South Africa after the first universal suffrage elections. Given the history of White domination and the likely magnitude of the immediate post-election White threat, it is possible that Blacks, even if somewhat divided, might nonetheless line up behind a party sufficiently to give that party more than half the seats so it could form a government alone. This is what Robert Mugabe and the Shona managed in Zimbabwe, despite opposition from the disproportionately represented Whites and from Joshua Nkomo and the Ndebele. For reasons I shall explain in Chapter 7, a plurality rather than a majority of votes seems likely to me as well, but this is not a foregone conclusion.

More to the present point, there is already in existence an example of list-system PR imposed to induce fragmentation of a majority into its

[18] But see Michael Gallagher, "The Political Consequences of the Electoral System in the Republic of Ireland," Electoral Studies 5, no. 3 (December 1986): 253–75, at 258.

[19] "Here we have another reason to prefer proportional representation to plurality for segmented societies: proportional representation encourages multi-partyism which in turn encourages coalition government." Lijphart, "Choosing an Electoral System for Democratic Elections in South Africa," p. 5.

[20] See Arend Lijphart, Democracy in Plural Societies (New Haven: Yale University Press, 1977), pp. 25–31.

[21] Lijphart, Power-Sharing in South Africa , p. 20n.

component groups so as to break its electoral hold and facilitate coalitions. The example comes from the Guyana elections of 1964, shortly before independence. The British imposed list-system PR in the hope, among others, that the majority East Indians would divide into Hindus and Muslims. British expectations were utterly disappointed. Confronted with a strong challenge from other groups, Hindus and Muslims stayed together in the party with which they had previously aligned, and the parties induced to proliferate because of PR all made such a poor showing that together they obtained 1 percent of the vote and no seats. List-system PR does assure a good measure of proportionality, but it does not guarantee party proliferation or the fragmentation of majority groups.^[22]

But let us pass this point and suppose Lijphart is right that list-system PR will produce party proliferation in South Africa so that no party has a majority of seats. What motivates the formation of the coalition? Will it produce the democratic stability postulated by the consociational model? Will the need for the coalition induce the compromise Lijphart envisions as a result of the need to coalesce?

We already have clear, discouraging answers to these questions. Political parties in newly independent countries often had to form coalitions after elections, even though the elections were frequently conducted under first-past-the-post. The reasons lie in ethnic pluralism, which produced several ethnically based parties and a division of seats that left no party with a majority. The elections in Nigeria in 1959, in Uganda in 1962, in Benin (then Dahomey) several times between 1957 and 1965, and in the Indian Punjab in 1967 and 1969 all involved three main parties, and all resulted in the need to form coalitions. I have examined these elections in detail elsewhere.^[23] I have called the resulting coalitions "coalitions of convenience," because they were arranged merely to aggregate the number of parliamentary seats to more than 50 percent, in order to form a government—and for no other purpose, such as interethnic compromise. No compromise took place in any of these cases.^[24] The coalitions were short-lived, they fell apart over a divisive

[22] I have discussed the Guyanese election of 1964 in Donald L. Horowitz, Ethnic Groups in Conflict (Berkeley and Los Angeles: University of California Press, 1985), pp. 643–46. For a fuller treatment of electoral engineering in conflict-prone societies, see ibid., pp. 628–51.

[23] Ibid., pp. 369–78. See also the discussion of the dissolution of the Guyanese coalition that had been fostered by list-system PR. Ibid., p. 318.

[24] These were not coalitions of the moderate middle. Proximity of policy positionshad, if anything, a negative role in their formation. Parties that had some ethnic policy positions in common and that had competed with each other for some part of the same voting clientele tended not to be part of these coalitions. After all, it would be difficult to justify to party activists the need to form a coalition today with yesterday's electoral competitors. The result is that lack of electoral overlap was more likely to induce formation of a coalition. As a consequence, coalition partners had opposite views on the most serious ethnic issues that divided the society. Accordingly, there was no prospect for compromise but plenty of opportunity for issues to split the coalitions apart.

ethnic issue, and they left ethnic conflict worse than it was when they took office. Unless there is an incentive to compromise over ethnic issues, the mere need to form a coalition will not produce compromise. The incentive to compromise, and not merely the incentive to coalesce, is the key to accommodation.

As a matter of fact, the need to pool seats will not alone produce interethnic compromise in a severely divided society, even under conditions much more favorable to compromise. Northern Ireland has been down this road and shown where it leads. The election to which I refer took place in March 1973, after the British parliament provided for a return of self-government to Northern Ireland. It is worth recounting in some detail.^[25]

It is well known that Northern Ireland contains a Protestant majority fearful of the Catholic minority. Both of these groups were represented by more than one party, so that the party proliferation envisioned by Lijphart for South Africa was already present in Northern Ireland by 1973. More than that, Northern Ireland had something there is no guarantee South Africa will have—a small but significant party of the middle, drawing votes from both major groups, polling just under 10 percent of the total vote, and expressing an aspiration for accommodation. This party was called the Alliance.

The 1973 elections were scheduled by the British parliament at a time when Britain was actively seeking compromise and even making mutual concessions on Northern Ireland with the Irish Republic. More important, the British specified that no post-election government would be permitted to take office unless it were committed to "power sharing," defined, not merely as an aspiration to interethnic accommodation, but in terms of the actual participation in government of Catholic-based parties as well as Protestant-based parties. To complete the list of favorable conditions, the election was held under the single transferable vote system of proportional representation. Like other PR systems, this

[25] The account here is based on Richard Rose, Northern Ireland: Time of Choice (Washington, D.C.: American Enterprise Institute, 1976).

was thought conducive to party proliferation, so that coalition government would be necessary.^[26]

The STV system, however, has one other advantage, not possessed by list-system PR. Voters cast their ballots for candidates, not parties, in order of preference. In a multimember constituency with four seats but ten candidates, a voter might cast up to ten preference votes. This at least opens up the possibility that some of a given voter's votes might be cast for candidates across the ethnic divide. By contrast, list-system PR requires that votes be cast for a single party list. Where parties are ethnically based, there is no way to transfer votes across ethnic lines. A voter is locked wholly within his ethnic party.^[27]

Under the Northern Ireland system, a quota of votes is computed by dividing the number of seats plus one into the total number of votes cast and adding one to the quotient. Any candidate who attains more than this quota is deemed elected. It should be noted that the quota for election may be far below a majority of votes cast, depending on how many seats are at stake. If there are four seats in a constituency, a candidate could win with about a fourth of the vote. That is why STV is so different from a winner-take-all system and so conducive to proportionality. Once a candidate receives a quota and is elected, his "surplus vote"—that is, his vote above the quota—is "transferred in proportion to its size to the candidates who stand next in preference among his supporters. Once this is done, the candidate with the lowest total of votes is eliminated, and his votes are transferred to those ranking next in the preference of his supporters."^[28] Second and third preferences are therefore likely to count. Protestant voters may allocate them to Protestant candidates or to Catholic candidates, as they see fit, and vice versa.

In this respect, STV permits a measure of interethnic vote pooling

[26] Ibid., pp. 77–78.

[27] This statement is not always literally true, but it is true as a practical matter. Under the Swiss version of list-system PR, a voter may make choices across party lists. In a severely divided society, however, where parties are ethnically based, this option is entirely hypothetical. The lists of different parties compete in every way, so that a vote for Party A is a vote against Party B. Voters of ethnic group B will not cast such ballots, and the two parties will not be able to reach an agreement for such vote transfers. As I shall explain below, the situation is different where voting is preferential (that is, under STV or AV), for then voters are casting ballots for second and subsequent choices, to be counted only if their first choice is eliminated (or, under STV, if their first choice does not need their vote to make his quota). It is entirely possible for parties of different groups to agree to exchange such lower-order preferential ballots, and voters can be induced to cast them.

[28] Rose, Northern Ireland: Time of Choice , pp. 76–77.

that list-system PR completely precludes. Voters of one group could provide the margin of victory for a candidate of another group, who might then be responsive to their concerns. If vote pooling of this kind occurred as a result of agreements between parties, the basis would be laid for interethnic policy compromise.

Under STV, moderate Catholic parties or candidates might have made interethnic vote-pooling agreements for the second and third preferences of supporters of moderate Protestant parties or candidates; and, of course, such agreements would be reciprocal. Presumably, such agreements could only be consummated between interethnic moderates, and therefore such agreements should also foster moderation. If the choice for a divided society is between list-system PR and the single transferable vote, STV is a far better choice than list-system PR.

Notwithstanding these conditions—which were certainly more favorable to accommodation than those envisioned by Lijphart for South Africa's first universal suffrage election—the results were a defeat for accommodation. In the first place, a new Protestant party sprung up to oppose power sharing; it and other Protestant parties that were opposed to an interethnic coalition together secured 35 percent of the vote, compared to 27 percent for the Protestant party committed to power sharing.^[29] When the power-sharing cabinet was inaugurated, even the Protestant party committed to power sharing, fearful of erosion of its support, rejected the arrangement, dooming it to a short life.

A United Kingdom election, coincidentally held the next month, confirmed the erosion. In Northern Irish constituencies for the British parliament, Protestant parties opposed to power sharing gained an outright majority of votes and seats. A survey of Protestant opinion showed only 28 percent strongly in favor of power sharing. Under Protestant protest, the Northern Ireland government fell.^[30] The accommodating Protestant party was vulnerable to accusations of sellout by the other Protestant parties on its flank.^[31] By 1975, the accommodating party was down to 7.7 percent of the vote.^[32]

Significant vote pooling across ethnic lines did not take place, either in 1973 or in subsequent PR elections.^[33] The reason, explained by Richard Rose, is straightforward: "The electoral system offered parties no

[29] Ibid., pp. 29–30.

[30] Ibid., pp. 30–31.

[31] Ibid., p. 82.

[32] Ibid., p. 97.

[33] Ibid., pp. 78, 93.

incentive to seek votes across the religious divide, because the chances of winning an extra seat by adding a few votes from the other community were much less than the chances of losing votes by appearing 'soft' on the issues that were of central concern within the party's home community."^[34] No vote-pooling agreements were made in advance by the parties because of the same fear of a negative net result.^[35] Without such agreements, the parties were never forced to confront and compromise their differences. Even with the possibility of vote pooling and a certain amount of good will at the top, plus the participation of the small Alliance party, committed to accommodation, the coalition ended up as a coalition of convenience, just like the Nigerian, Ugandan, Beninese, and Punjabi coalitions described above.

None of this is an indictment of vote-pooling incentives as a strategy of fostering accommodation. On the contrary, that is the way to do it. The Northern Ireland incentives, however, were insufficient because they permitted parties to take a share of seats and even enter government, whether or not they had pooled votes. Since a fraction of the total vote was enough to reach a candidate's quota, the incentives were weak. As we shall see, there are stronger incentives to vote pooling that are possible and that promise better results.

The Northern Ireland elections are a refutation of the notion that PR can simply deliver coalitions and compromise in severely divided societies. If there are examples of this phenomenon from severely divided societies, they have yet to emerge. As the African and Asian examples show, coalition alone (based on pooling seats alone) is no assurance of compromise. As the Northern Ireland case shows, even weak vote-pooling incentives in a favorable milieu may not be enough to induce accommodation.

Lijphart, who is very clearheaded about these matters, does not say that coalition formation alone is enough. He wants coalitions in tandem with other consociational arrangements.^[36] In 1986, the KwaZulu Natal Indaba, a standing conference of citizens of the two territories, took his advice and proposed a consociational scheme to govern the two regions if they should merge into a single administration.^[37] The proposals em-

[34] Ibid., p. 78.

[35] Ibid., p. 93.

[36] "Choosing an Electoral System for Democratic Elections in South Africa," pp. 3–4.

[37] KwaZulu Natal Indaba, Constitutional Proposals Agreed To on 28 November 1986 (Durban: n.p., n.d.; pamphlet).

body mutual vetoes, mandatory group representation in the cabinet, consensus decision making in the executive, cultural councils, and a bicameral legislature. The upper chamber is to be elected largely on the basis of racially reserved seats, with overrepresentation of the minorities. Both chambers are to be elected by list-system PR, the lower chamber by constituency lists and the upper chamber by treating the entire merged province as a single constituency.

This is, of course, a bold attempt to meet the problems of ethnic diversity and to foster conciliation. It is almost unkind to criticize such a plan. Nevertheless, many components of the proposals boil down to guarantees of group rights and would be vulnerable to the objections I have already stated with regard to such schemes. The electoral system would do nothing to foster vote pooling and accommodative coalitions; it would certainly not provide electoral disincentives for the strongest party or parties to dismantle the minority protections. Even if one were to favor a group rights approach, it would be necessary to design an electoral system that gives politicians electoral reasons to be supportive of their entrenchment.

Nothing I have said, of course, vitiates the importance of coalitions. Coalitions should be the centerpiece of accommodative arrangements. But not any coalition will do, only a coalition likely to produce compromise rather than perpetuate conflict. Absent incentives, politicians will not engage in intergroup compromise in a severely divided society. List-system PR may or may not produce enough party proliferation to require a coalition in South African conditions. Even if it does, list-system PR contains no incentives to vote pooling or compromise and will produce only coalitions of convenience that will dissolve. If politicians can, by inducing aisle-crossings or other methods, elbow coalition partners out of the coalition and govern alone, sharing the spoils with no other parties and having no longer to justify the coalition with ethnic opponents to their own supporters, politicians will prefer to govern alone.

Pooling seats, then, is not enough to provide the appropriate incentives. If, however, popular votes are actually pooled, compromise may follow, as I suggested in Chapter 4 with respect to Malaysia. Unfortunately, so far as I know, no one working on or in South Africa has emphasized the distinction between mere seat pooling and vote pooling. Vote pooling, which is the engine of compromise, has been left out of account altogether. Lijphart, for example, is opposed to presidential systems of government, because, he says, the president, being one per-

son, "is bound to represent one particular segment to the exclusion of other segments."^[38] Never does he pause to ask how a president is to be elected. Never does he note that the president of the Nigerian Second Republic, although a rather parochial Northern politician, was elected under a formula that provided him with an incentive he could not resist (if he wished to win election) to accommodate members of groups other than his own. As a candidate, he responded to the incentive to pool votes; later, as an officeholder, he responded to the desire to be re-elected on pooled votes.

The emphasis here is not on statesmanship alone, but on incentives. In that respect, there is a fundamental difference of approach from consociationalism. Politicians, consociational theorists suggest, are motivated by notions of the common good. "If these leaders were exclusively or mainly motivated by a desire to survive politically, it would be much more to their advantage to fan the flames of intersegmental antagonism than to engage in the risky enterprise of striking compromises with rival segmental leaders."^[39] Exactly so. Consociationalism and its electoral component come down to statesmanship, not electoral incentives. If the experience of severely divided societies shows anything at all, it is that statesmanship alone, statesmanship without tangible reasons, statesmanship without rewards, will not reduce conflict. Without incentives, statesmanship will be in short supply.

In such societies, it has been argued,^[40] there is a need to choose between majoritarian democracy and consociational democracy. I do not believe this to be an accurate statement of the choices. The need is merely to choose between two kinds of majoritarian democracy: a majoritarian democracy that will produce racially or ethnically defined majorities and minorities or a majoritarian democracy that will produce more fluid, shifting majorities that do not lock ascriptive minorities firmly out of power. That is why Lijphart is right to oppose first-past-the-post in single-member constituencies. The winner-take-all assumptions of such electoral systems generally (though not always) degenerate into ascriptive majority rule. But PR alone—PR without vote pooling across group lines—will not prevent that result.

[38] Power-Sharing in South Africa , p. 60. See also ibid., p. 80; Lijphart, Democracy in Plural Societies , pp. 178, 187, 210.

[39] Lijphart, Power-Sharing in South Africa , p. 107.

[40] Ibid., p. 109.

Vote Pooling:
Statesmanship with Rewards

To promote intergroup accommodation, the need to form a coalition is, then, a necessary but not a sufficient condition. Not all coalitions in severely divided societies are inclined to accommodation. It follows that party proliferation, which produces the need to form a coalition, is also not enough. The proliferation of parties must be accompanied by the rewards to moderation that accrue when parties are dependent, in part, on vote transfers from members of groups other than the groups they principally represent. Only coalitions that rest on intergroup vote pooling, as well as seat pooling, have reason to be accommodative.

Vote pooling makes statesmanship rewarding. Now, we have already seen that the mere theoretical possibility of transferring votes across ethnic lines is not, by itself, sufficient to induce party agreements for vote transfers or the transfers themselves. And for exactly the reason specified earlier by Rose in the Northern Ireland case: there will not be intergroup vote-pooling agreements if the number of own-group votes likely to be lost by parties contemplating such agreements exceeds the number of votes likely to be gained by transfers across the ethnic divide. The mechanism that underlies this calculation is that, to obtain votes across ethnic or racial lines by agreements with other parties to trade second or third or fourth preferences, reciprocal moderation on ethnic or racial issues is required.^[41] In Northern Ireland, that moderation was likely to repel more of each group's own voters than the number of voters it was likely to attract from the other ethnic group.

Some clear conclusions emerge. Since vote pooling can only be accomplished if there is a showing of intergroup moderation by the parties seeking to exchange votes across group lines, the prospect of vote pooling with profit is the key to making parties moderate and producing coalitions with compromise in severely divided societies. Therefore, it becomes essential to reverse the calculations that prevailed in Northern Ireland in 1973—that is, to make intergroup vote pooling a positive-sum transaction for parties that participate in it. The point, in short, is to use the electoral system to induce changes in the behavior of ethnically based or racially based parties.^[42]

[41] On vote transferability across parties and the role of party distance, see Giovanni Sartori, Parties and Party Systems (Cambridge: Cambridge University Press, 1976), p. 343.

[42] It is well accepted, if rarely emphasized, that electoral systems can change the behavior and posture of parties. See, e.g., Sartori, "Political Development and Political Engineering," p. 287.

The negative-sum Northern Ireland calculations can indeed be reversed. There is an example of this, the logic of which I have explicated earlier,^[43] in the case of Malaysia. There an interethnic coalition was formed and benefited from pooling votes across ethnic lines in heterogeneous constituencies so as to defeat ethnically based competitors on both flanks. The ethnically based flank parties held issue positions that were ethnically extreme and thus differentiated them from the coalition and its compromises. Those flank competitors therefore could not make any agreements to pool votes across ethnic lines, even if they had wanted to. But, it should be stressed, the moderation of the coalition parties did not predate formation of the coalition—it came into being simultaneously, precisely because it was electorally rewarding at that moment. Nor did their moderation ever produce a grand coalition of the consociational type; it produced instead an ethnically inclusive but majoritarian government, flanked by ethnic parties in opposition.

The Malaysian coalition illustrates something already implicit in my formulation of the problem of incentives for vote transfers. There are two separate components: prospective vote losses resulting from appeals to voters across the ethnic divide and prospective vote gains from such appeals. I call these separate, because they obviously come from different groups of voters, and the net is the sum of the behavior of both. The losses come from voters of one's own group who believe the party's moderation on ethnic issues constitutes a sellout of their group interests. They therefore choose to give their votes to another party representing their group. The gains come from voters of another ethnic group who are attracted by interethnic moderation and who do not have a better choice (or, in the case of preferential voting, do not have a better second or third choice) from among parties representing their own group. So the question then becomes, What choices are available to these groups of voters at the moment vote pooling could occur?

In the Malaysian case, there were no good alternatives on the flanks. The Malaysian coalition was formed at a moment when there was substantially no intraethnic competition on the Malay or on the non-Malay (Chinese and later Indian) side, so the potential losses from vote pooling were minimal on both sides. This is because the party to be defeated in the first elections that induced formation of the interethnic coalition was not an ethnically based party but a multiethnic party led by a prominent and charismatic figure, whose popularity, it was thought, made

[43] See Chapter 4, pp. 154–55, above.

his party a dangerous competitor. So the Malaysian coalition was created at a rare, and never repeated, moment of exclusively centripetal—that is, multiethnic—competition. For supporters of the Malay and non-Malay parties that agreed to pool votes, the competitor multiethnic party did not hold much appeal, so the agreement could be made without the prospect of losing one's own supporters.

Thereafter the agreement, which resulted in the coalition's consistently running a single slate—that is, only one candidate (whether Malay, Chinese, or Indian) for each office—was sustained, even in the face of first-past-the-post elections, by some other favorable conditions. The first of these was the prevalence of ethnically heterogeneous constituencies. The second was the emergence of ethnically based flank parties that locked themselves into extreme positions. The electorate in most constituencies was thereafter split several ways, and the coalition, using vote transfers, could win seats on pluralities in three- or four-way contest. Even in a straight contest with one ethnic flank party, the coalition could afford to lose some votes to that flank party, if, depending on the ethnic composition of the constituency, it could compensate for those losses with support from the group whose flank party did not enter a candidate in that constituency. Most of the time, although not always, more net votes were to be had by a coalition candidate espousing interethnic moderation and practicing vote pooling than by a candidate practicing ethnic extremism in behalf of his own group.^[44]

It will be noticed immediately that, under admittedly unusual conditions, even first-past-the-post can be made to serve the ends of vote pooling, intergroup coalitions, and accommodation. The mere plurality requirement means that, in a badly split constituency, only a small increment of pooled votes may be required to come out ahead of competitors. But this is not easy to engineer. In a conflict-prone society, voters will generally not cross ethnic lines. The purpose of incentives is to create floating voters at some level of preference, in a situation where they would ordinarily be lacking. The main point of the Malaysian illustration is simply to show the obverse of the Northern Ireland calculation, in a society divided at least as severely as Northern Ireland, as

[44] For some estimates of vote pooling, even in the election of 1969, in which the ethnic flank parties did unusually well, see Karl von Vorys, Democracy without Consensus: Communalism and Political Stability in Malaysia (Princeton: Princeton University Press, 1975), pp. 302-03, 305. For the calculations of the coalition partners, see Horowitz, Ethnic Groups in Conflict, pp. 396–416, 424–26; Donald L. Horowitz, "Incentives and Behaviour in the Ethnic Politics of Sri Lanka and Malaysia," Third World Quarterly 11, no. 4 (October 1989): 18–35, at 24–29.

well as to expose the logic that supports the calculation. At a minimum, that logic requires at least some heterogeneous constituencies and party proliferation.

The example also shows the importance of sequencing. Once the Malaysian coalition and its opponents on the flanks divided up the party spectrum, the coalition's vote-pooling arrangements proved to embody the most profitable vote-maximizing strategy, certainly more profitable than going it alone on an ethnic-party basis. In other words, disintegrative options were quickly closed off. But it was sheer fortuity that vote pooling looked attractive in the specific conditions of the first election. Had the first elections simply been a contest between ethnically based parties, nothing in later elections would have produced vote pooling.^[45] Constitutional engineers cannot count on fortuity. They need to build in vote-pooling incentives at the outset.

Before I turn to other ways of maximizing the chance of vote pooling, some further observations are in order. It has been argued that representing whole groups united under single leaderships, rather than fractions of groups, is superior for intergroup accommodation.^[46] We can see from Malaysia the initial advantage of monopoly support for reducing the losses and maximizing the gains of vote transfers across group lines. But several caveats need to be entered. First, monopoly position is not always necessary, and in any case monopoly support is not always present at the crucial moment when an intergroup coalition is to be formed. As I said, the Malaysians were lucky. Second, there is a tradeoff between monopoly position, which provides extra latitude to make compromises, and party proliferation, which is necessary to produce the need to form any coalition in the first place. If each group is represented by only one party, there may be a danger that one party may secure a majority of seats by itself, and thus a coalition will not be necessary or at least, before the election, will not appear to be necessary. This fact may inhibit the disposition to make a vote-pooling agreement in advance. Finally, after such an agreement is made in a severely divided society, ethnically based parties will in any event position them-

[45] This is the Sri Lankan case. See Horowitz, "Incentives and Behaviour in the Ethnic Politics of Sri Lanka and Malaysia," pp. 21–26.

[46] Lijphart, Democracy in Plural Societies, pp. 25–31; Robert A. Dahl, Polyarchy: Participation and Opposition (New Haven: Yale University Press, 1971), pp. 114–21; cf. Eric A. Nordlinger, Conflict Regulation in Divided Societies, Harvard University Center for International Affairs, Occasional Paper no. 29 (Cambridge: Harvard University Center for International Affairs, January 1972), p. 118.

selves on the flanks. Therefore, intraethnic monopoly is not possible to maintain once accommodative agreements are in place.

I want particularly to underscore the first of these points—that monopoly position is not always necessary for vote pooling—by looking a bit more closely at the Northern Ireland case. The reason it is worth examining is that the Northern Ireland experience could plausibly, but I think wrongly, be read as precluding vote pooling in the absence of intraethnic party monopolies. One thing we know is that Catholics and Protestants are each represented by more than one party. What we do not know intuitively—but what turns out to be true—is that there are also significant barriers to vote transfers even between parties of the same ethnic group.

A careful study of Catholic voters aligned with the moderate Social Democratic and Labor Party (SDLP), on the one hand, and the Sinn Fein (aligned with the Irish Republican Army), on the other, shows very limited "willingness of voters of one party either to vote for the other party when their own party is not standing [in a given constituency], or to give lower-preference votes to the other party in PR elections. . . ."^[47] Aptly called "Not Many Floating Voters Here," the study makes clear that the two Catholic parties are relying on quite different pools of voters, in terms of age, class, employment status, and education. Only to a small degree are the two "fishing for votes in the same pond."^[48] So what looks like an intraethnic choice that might preclude interethnic vote transfers is not necessarily a choice in the eyes of the voters. As they survey the existing party landscape, voters discontented with interparty conciliation across ethnic lines may find they have nowhere better to go. Very few SDLP voters would move to Sinn Fein if the SDLP agreed to pool votes with a moderate Protestant party. Rather, the opposite: in the 1985 local government elections, more SDLP vote transfers appear to have gone to the intersectarian Alliance than to Sinn Fein.^[49] Although Protestant voters were more willing than Catholic voters to transfer votes among Protestant parties,^[50] nothing we know about

[47] Cynthia Irvin and Eddie Moxon-Browne, "Not Many Floating Voters Here," Fortnight (Belfast), May 1989, pp. 7–9.

[48] Ibid., p. 7.

[49] Ibid., p. 9.

[50] Nearly three times more willing, in fact. In the 1985 local council elections, 20 percent of all Protestant vote transfers went from the Official Unionists to the Paisleyites, compared to 7 percent of Catholic vote transfers from the SDLP to Sinn Fein. I am indebted to Cynthia Irvin for these data.

Northern Ireland supports the conclusion that the existence of more than one party per ethnic group necessarily dooms interethnic vote pooling. The point is, rather, that, in the face of intraethnic party proliferation and the possibility that some voters will shift to more extreme parties, the incentives to vote pooling need to be strong.

So far, all I have said may be interesting to analysts but is not yet useful to engineers. It comes down to this: (1) The exchange of votes by parties across group lines is essential to the emergence of coalitions that have a tangible interest in moderation and compromise on intergroup issues that can provoke severe conflict. (2) Vote exchange of this sort seems hard to arrange. Under propitious conditions, it did not occur in Northern Ireland, but it did occur in Malaysia. (3) The difference between the two cases is explicable, but the explanation rests on fortuitous, indeed nonrecurring conditions, especially the momentary fluke of entirely centripetal competition in Malaysia. (4) To underscore the idiosyncratic character of the two outcomes, it should be stressed that, because of the different party-competitive configuration in the two cases, the electoral system more likely to induce vote pooling (STV) did not do so in Northern Ireland, whereas the electoral system less likely to induce vote pooling (first-past-the-post) did do so in Malaysia. (5) To induce interethnic vote exchange, three elements are needed: (a) party proliferation, (b) heterogeneous constituencies, and (c) electoral incentives that make vote pooling politically profitable.

One could, of course, read this comparative lesson as simply discouraging for South Africa. Instead, since constitutional engineers, like all engineers, cannot quit when discouraged, constructive use can be made of the discouragement. What it argues for is a stronger dose of incentives for pooling votes than might otherwise have been thought necessary.

There is, in any case, a compelling reason for a strong dose of incentives for interethnic vote pooling, apart from the fact that an ordinary dose did not work in Northern Ireland. The compelling reason is that, whatever arrangements are ultimately agreed upon, electoral and related institutions are likely to yield, as I suggested earlier, a mixed system of incentives to accommodation and to ethnic extremism.

Wherever the system of incentives is mixed, there are always dangers that one set of incentives can outweigh the other. As I hinted with respect to the Nigerian Second Republic, such tensions existed in its arrangements. The president was elected by a system that rewarded him

for interethnic accommodation, but the national legislature was elected by a system that built in no such rewards. Legislators were simply representatives of their own ethnic groups and proponents of group claims; they perpetuated conflict, even as the president damped it down.

Where the system is mixed, the incentives to accommodation can also be undone. In Malaysia, for example, periodic constituency delimitation has tended to reduce the number and impact of significantly heterogeneous constituencies. Specifically, Malay candidates of the coalition have become less dependent than they formerly were on Chinese votes for their margin of victory, although Chinese coalition candidates are still frequently dependent on Malay votes. The inequality of the exchange is a source of pro-Malay bias in policy outcomes.

It cannot be assumed, therefore, that a mixed system of incentives will always be sufficient to avert severe conflict or that it will remain in its original equilibrium in the face of pulls to change it. So there are, in the end, several reasons for strong incentives: to be sure they are adequate to induce vote pooling; to be sure the extent of vote pooling is sufficient to repel conflict-prone claims that will emerge; and to shield the vote-pooling incentives themselves from erosion.

If a strong dose can be adopted, however, it has an enormous advantage for South Africa. It will be recalled that South Africans of all racial groups agree on the need to avoid the domination of any group by another, but they do not tend to agree once various means to these ends are proposed. The attraction of working on ameliorating conflict through a system of incentives lies in their self-renewing character. If the incentives are strongly built in, the politicians will generally behave in accordance with the rewards structured by those incentives.^[51] To put the point more sharply, in a divided society like South Africa, the polity that adopts the strong incentives approach need only agree on one set of means at the outset. It can then let political interest take its course. This is no small advantage where consensus is hard to find once the question shifts from abstract goals to concrete means.

[51] There are no guarantees, of course, that politicians will play by these rules, but that argument proves too much—it is true of all rules. The point about these is that they probably need to be put in place only once. Of course, all of this assumes that South Africa can, in the first instance, surmount the hegemonic aspirations attributed to its main participants. This problem I have described in Chapter 1 and shall return to in Chapter 7.

Electoral Paths to Accommodation

If South Africans decide that the promise of intergroup coalitions committed to accommodation is worth pursuing, as I obviously hope they will, South Africans will have to work out an appropriate institutional design. Since such coalitions are dependent on vote pooling by parties across ethnic lines, it is time to provide ideas about how a strong dose of incentives can be built in to induce the exchange of votes.

There are at least three approaches. The first is to enact distribution requirements for electoral victory, over and above plurality or majority requirements. The second is to opt for preferential voting, and here there are two main varieties, the alternative vote and the single transferable vote. The third is to require or nearly require ethnically mixed slates, along the lines of the Malaysian or Lebanese practice.

The Nigerian formula for the election of the president, put into effect for the 1979 and 1983 elections, had a double set of requirements. The victorious candidate needed to obtain the largest number of votes (a majority if there were only two candidates, a plurality if more than two). He also needed geographic distribution: at least 25 percent of the vote in no fewer than two-thirds of the then-nineteen states in the Nigerian federation. To reach the 25 percent threshold, given the geographic distribution of the various ethnic groups, it would not be enough for any two of the three largest groups to unite behind a single candidate, even though the three largest groups together formed nearly two-thirds of the population. This was a system put in place deliberately to broaden the support required by a president and to protect minorities.

As I have already noted, the system worked well. But, of course, it has an obvious problem. It is possible that no candidate will meet the requirements of the formula. In 1979, this difficulty nearly came to pass. And so there must be a reliable backup formula to elect a president if no candidate secures the 25 percent distribution.^[52]

Two observations are in order. First, since these are competitive elections, candidates and their parties have every incentive to reach out

[52] In my discussions with members of the Nigerian Constituent Assembly in 1978, delegates—especially Ibo delegates—were much concerned with the prospect that, under a distribution formula, no candidate might meet the dual set of requirements. They feared the instability that might follow upon an uncertain outcome. Such concerns make it imperative to adopt an effective and definitive backup formula, so that some candidate is declared elected by a process that is rapid, authoritative, and clear-cut—in short, incontestable. For the 1978 calculations, see Donald L. Horowitz, "About-Face in Africa: The Return to Civilian Rule in Nigeria" Yale Review 68, no. 2 (Winter 1979): 192–206.

effectively to attain the 25 percent distribution threshold. They certainly cannot sit supine, because competing candidates must be presumed to be reaching out. As a result, there is less chance of failure to elect than might at first be thought. Still, there is some chance, and my second observation relates to the backup formula. That formula is likely to be based either on plurality alone or on a runoff, either by direct election or, as in the Nigerian case, by indirect election. In a federation, the indirect election could involve either central or state legislatures, or both. But note that, if distribution is the only incentive to interethnic cooperation, and the backup formula confers the power to elect on officials who themselves have been chosen by a system with different incentives, then these, too, will figure into the calculations of candidates and their parties, thereby likely weakening the accommodative effect of the distribution requirement.

In Nigeria, since federal and state legislatures, unlike the president, were elected without any incentives to interethnic accommodation, the backup formula did not reinforce the distribution requirement. Observers of the Nigerian preparations for the return to civilian rule in 1992 have, in fact, called for a system of legislative election different from first-past-the-post in largely homogeneous constituencies, a system that will reinforce the incentives for multiethnic distribution.^[53]

One such possibility might be to elect legislators by a formula requiring plurality plus distribution within legislative constituencies. The distribution principle of election need not be limited to a presidential system. But distribution requirements for legislative elections would have an accommodative effect only if constituencies were heterogeneous. Even then, the requirements would raise new problems, for there would have to be a backup formula for legislative elections, probably providing for a runoff or for indirect election by a lower tier of legislators. All of this is feasible, but it may seem unattractive or complex. If a president is to be elected on the basis of distribution, it is quite possible that legislatures will be elected by some other formula that provides a different set of accommodative incentives. The main objective is to build in accommodative incentives strongly for every office in each branch and each level of government.

The party system produced by the mixed Nigerian incentives was itself mixed. The same three ethnically based parties as prevailed in the

[53] Larry Diamond, "Nigeria: Pluralism, Statism, and the Struggle for Democracy," in Larry Diamond et al., eds., Democracy in Developing Countries , vol. 2, Africa (Boulder, Colo.: Lynne Rienner, 1988), p. 77.

First Republic reconstituted themselves, but they also broadened out to embrace minorities around the edges. Fairly clearly, this broadening was induced by the presidential electoral formula (and, in the north, by the effect of new federal states). No interethnic coalition emerged in the national legislature, except a coalition of convenience, based only on seat pooling. This is not surprising, since the presidency, with its unique mode of election, was only one piece of the system.

In South Africa, it is not clear whether a single principle of distribution could be made to work in legislative elections. The Nigerian technique uses geography as a rough proxy for ethnicity. It depends upon territorial concentration of groups—that is, a certain level of homogeneity within areas but heterogeneity between areas. The spatial distribution of all but the White group^[54] in South Africa, however, is uneven. Some areas, such as Natal and the Cape, have strong concentrations of large groups: Zulu and Indians in Natal, Xhosa and Coloureds in the Cape. Others, notably the Transvaal, have a good deal more ethnic heterogeneity. In the heterogeneous areas, it might be possible to impose distribution requirements in single-member parliamentary constituencies, using the standard of plurality plus distribution thresholds in some percentage of the total number of wards or electoral subdivisions. In more homogeneous areas, however, a party and its candidate in a single-member constituency could probably meet a low distribution threshold without making any commitment to intergroup moderation. Consequently, to make distribution work, complex electoral formulae in large multimember constituencies might be required. This might not be an altogether bad thing, because it might well produce ethnically mixed slates in those multimember constituencies. Since all voters would vote for all candidates, candidates could not win without relying on the votes of groups other than their own. Vote pooling among candidates would be essential to victory. The result would be to maximize intergroup conciliation.^[55]

Of course, if all that concerned us were distribution across racial (as opposed to ethnic) groups, the Group Areas Act has so segregated the groups for the time being (and will probably continue to do so for some time after its abolition) that geographic distribution requirements within

[54] Leaving aside, for the moment, distinctions between Afrikaans and English speakers.

[55] The rather clear analogy here would be the former Lebanese system of ethnically mixed lists in multimember constituencies, which produced exactly this result. See Horowitz, Ethnic Groups in Conflict, pp. 633-35.

multiracial constituencies would be easy to arrange. But the areas in which single racial groups predominate are so extensive that, again, large multimember constituencies would be required.

Although distribution requirements worked well in Nigeria, they are rather unusual. It bears mention, therefore, that analogous requirements have been proposed previously in Uganda and in South Africa itself.

The Uganda proposals were made shortly before Milton Obote was overthrown in 1971. They would have required each member of parliament to represent his own area plus three others. For victory, a candidate would need the highest vote in all four constituencies taken together.^[56] While this is not strictly a distribution requirement, it has the same aim of requiring panethnic support.

The South African proposals were made by the Molteno Commission that reported to the former Progressive Party in 1960. At the height of Verwoerdian apartheid, the commission recommended a bicameral legislature for a future South Africa seeking to alleviate its racial problems. It proposed that the upper house be elected by plurality vote in single-member constituencies, provided that the winning candidate obtain "at least one-fifth of the total votes cast by members of each [racial] community" in the constituency.^[57] The commission candidly acknowledged the important drawback that racial distribution would require "racial classification of voters for the purpose of Senate elections"^[58] —which, of course, would not be true for geographic distribution—but it urged distribution requirements because candidates "would know in advance that success depended on racial moderation and the support of some proportion, at all events, of all communities."^[59] The restricted franchise recommended by the commission drew several dissents, but the distribution provisions were recommended unanimously. Thereafter, the functions of distribution requirements were forgotten until the Nigerians discovered geographic distribution requirements independently in 1978.

If geographic distribution requirements are well designed, as in the Nigerian presidential election, they can produce an altogether benign

[56] Fred M. Hayward, "Introduction," in Fred M. Hayward, ed., Elections in Independent Africa (Boulder, Colo.: Westview Press, 1987), p. 9.

[57] Molteno et al., Final Report of the Commission Set Up by the Progressive Party to Make Recommendations on a Revised Constitution for South Africa , vol. 1, p. 19. Fallback provisions were also recommended. Ibid., p. 20.

[58] Ibid., p. 26.

[59] Ibid., p. 24.

impetus to ethnic inclusion. But it is easy to imagine quite different consequences in South Africa. Consider two hypothetical responses of electoral engineers to multiracial constituencies containing a heterogeneous array of Black groups.

To achieve maximum political inclusion of all the Black groups, the temptation might be to set the distribution thresholds so high as to produce recurrent uncertainty about whether any candidate could meet them. In the Nigerian presidential elections, distribution thresholds were only 25 percent and, at that, only in two-thirds of the states. Even so, a president was almost not elected in 1979. Even absent that uncertainty, if distribution requirements are set high, they might have the effect of consolidating the bonds among Black voters but exacerbating the cleavages between Black voters, on the one hand, and White voters, on the other. That is, parties and candidates might respond to the requirement of well-distributed support by securing their standing among all the Black groups. The way to do this is to emphasize racial issues.

If, however, distribution requirements are left very modest despite ethnic heterogeneity, those requirements might be satisfied even if parties broke cleanly along ethnic lines. If, for example, the requirements are plurality plus 20 percent support in two-thirds of the wards, that is something a party may well be able to achieve even if it appeals only to, say, Xhosa or Zulu. In short, distribution requirements can be so supereffective or so ineffective that they still leave many groups out in the cold.

Of course, this could be a problem with distribution requirements for a presidential election as well. But the likely effects of a distribution formula are easier to calculate for one whole country than they are for a multitude of individual constituencies. And so it is probable that distribution requirements for a presidential election can be designed to work as intended, just as they did in Nigeria.

Preferential voting may be more promising for the legislature. There are many varieties, but two are worth distinguishing. The single transferable vote we have seen at work—or, rather, not at work—in Northern Ireland. It is also utilized in composing the parliament of the Irish Republic and in composing several Australian bodies, including the federal Senate. Alternative voting is a system in which second and subsequent preferences of those voters whose first preference is not one of the top two candidates are reallocated until a candidate attains a majority.^[60] AV is used in composing other Australian bodies, including the

[60] Under another variant of AV, if no candidate wins an absolute majority of firstpreferences, the bottom candidate is dropped out, and that candidate's votes are redistributed in accordance with the second preferences of his supporters, and so on, using subsequent preferences, until one candidate attains a majority.

lower federal house, and in electing the president of Sri Lanka. It is generally regarded as a modest reform of first-past-the post, a sound way of diminishing the arbitrariness of the plurality system without yielding to the dominance of central party officials in candidate selection that tends to accompany list-system PR.^[61]

Quite obviously, both STV and AV give weight to second and third choices and open up the possibility that parties will bid across ethnic lines for the second preferences of voters whose first choice stands no real chance of election. This is especially true of alternative voting, since it is based on a majority threshold for election. The majority requirement necessarily disqualifies the first choices of a fair number of people whose candidate could meet a lesser standard. Indeed, an objection to alternative voting is precisely this: "If one party has, say, forty percent of the first preference votes, but those who do not rank the party first rank it last, then it may fail to win any seats."^[62] If it does win seats, it may be represented less than proportionately. This is labeled by critics "a major disadvantage of the AV system."^[63]

But notice that what is labeled a disadvantage can be turned to the service of moderation in a society in need of moderation. Why, after all, did the party with 40 percent of first preferences not succeed in negotiating across party lines to obtain the second and third preferences of voters who gave their first preferences to other parties? If the 40 percent party had done so, it would have come out better. But to do so, it would have had to compromise with supporters of those other parties.

Transfer this issue to a severely divided society with ethnically based parties, and the conclusion reads as follows: Under alternative voting, with a majority threshold for victory, many elections will turn on second and third preferences. Parties that succeed in negotiating for second and third preferences will be rewarded. The price of a successful negotiation is intergroup accommodation and compromise. The exchange of second and third preferences, based on reciprocal concessions on ethnic issues, is likely to lead to an accommodative interethnic coalition if no party can form a government alone. Under conditions of party proliferation, therefore, AV is likely to produce governments committed to

[61] Butler, "Electoral Systems," pp. 8, 20.

[62] R. J. Johnston, "Seats, Votes, Redistricting, and the Allocation of Power in Electoral Systems," in Lijphart and Grofman, eds., Choosing an Electoral System , pp. 63–64.

[63] Ibid., p. 63.

accommodative policies. To dislodge such governments at the polls, at least some incentives will be in place for the opposition parties to outbid the governing coalition at being moderate, although other incentives may point in the opposite direction—toward differentiating those parties and giving them reason to accuse the coalition partners of selling out group interests.

Before taking this line of reasoning any further, the realism of the key steps in the reasoning needs to be underlined. That realism depends on the willingness of parties and candidates actively to work to obtain second and third preferences from voters, and it depends on the willingness of voters to think in terms of second and third best. The willingness of both has been demonstrated. In the Republic of Ireland, where STV is used, the second and third strongest parties make deals to transfer votes; between 20 percent and 25 percent of all votes in Irish elections are actually transferred at least once.^[64] In Italy, where voters are permitted, even in a list system, to change the order in which they prefer candidates to be elected within party lists, it is very clear that candidates and factions set out to induce voters to do exactly that.^[65] Moreover, in 1972, some 56 percent of Italian voters exercised their option to change the candidate order; the tendency was not confined to sophisticated voters but was very common among traditionally oriented voters in the south as well.^[66] Experimental studies also find that voters are willing to deem acceptable a good many more candidates than they actually vote for; and, if preferential voting were in force, plurality winners in some currently first-past-the-post systems would become majority winners.^[67]

Apart from parties based on ethnic cleavages in severely divided societies, we know, then, that party organizations are easily geared up for various forms of preferential voting, and voters are receptive to it. And even if voters were not prepared easily to contemplate crossing ethnic lines, that is not an insurmountable problem, because second or third preferences could be made compulsory for a valid ballot.^[68]

[64] Gallagher, "The Political Consequences of the Electoral System in the Republic of Ireland," pp. 256–57.

[65] Richard S. Katz, "Preference Voting in Italy," Comparative Political Studies 18, no. 2 (July 1985): 229–49.

[66] Ibid. The same is true in Australia, where voters for federal Senate seats often do not follow party-recommended preference orders. See J. F. H. Wright, "An Electoral Basis for Responsible Government: The Australian Experience," in Lijphart and Grofman, eds., Choosing an Electoral System , p. 133.

[67] Gerald de Maio et al., "Approval Voting: Some Recent Empirical Evidence," American Politics Quarterly 11, no. 3 (July 1983): 365–74.

[68] Cf. Wright, "An Electoral Basis for Responsible Government," p. 133, to theeffect that in Tasmania it is required to cast a ballot for as many candidates as there are seats.

One point about proportionality and party proliferation. I said earlier that party proliferation is required to make this system work for accommodation, and I shall return to the proliferation theme in a moment. As Lijphart rightly points out, party proliferation depends in some measure on degree of proportionality of the electoral system, for, if the system gives the largest parties too big a seat bonus, smaller parties can be wiped out. Both AV and STV are both considerably better at insuring proportionality than is first-past-the-post, although not quite as good at this as list-system PR is.^[69] AV can provide quite enough proportionality for the requisite party proliferation. In any case, there are other ways to provide incentives for proliferation, which I shall describe in Chapter 6.

More than this, AV can provide the more rigorous incentive to vote pooling in a divided society that STV lacked in Northern Ireland. That incentive is the majority threshold. Under STV, a party could obtain representation without vote pooling. With a majority threshold, achieved by reallocated second and third preferences, it is harder to win election without vote pooling. Given party proliferation, if no one is elected without a majority and very few candidates can attain a majority without pooling second and third preferences across ethnic lines, that is about as strong a political incentive as can be devised. When supporters and leaders learn that the difference between power and no power lies in the mutual exchange of alternative votes, the sellout argument will still appeal to and motivate some voters, but it will not motivate all voters.

An analogy to the Malaysian coalition is not amiss. In the first Malaysian election, to be sure, there was only centripetal (multiethnic) competition, but in subsequent elections there was a flank (ethnic) competition exclusively. The coalition, which was rewarding because the logic of exchanging votes could win elections, stayed intact.

In 1978, the Sri Lankans adopted AV, with a majority rule, for their presidential elections. By now, therefore, we ought to know how vote pooling works under AV in actual elections in severely divided societies, but we do not. What we know, however, is still encouraging or, to put it more accurately, is not discouraging.

The two main Sinhalese parties were fairly well differentiated in their

[69] Part of the lesser proportionality of AV in the Australian House relates to the existence of single-member constituencies. Wright, "An Electoral Basis for Representative Government," pp. 129–31. On the proportionality of STV in Ireland, see Gallagher, "The Political Consequences of the Electoral System in the Republic of Ireland," p. 255.

willingness to be conciliatory toward the Tamils. The United National Party (UNP) of J. R. Jayewardene was, overall, more moderate than the Sri Lanka Freedom Party (SLFP) of Sirimavo Bandaranaike. In 1982, the first presidential election was held under AV. Unfortunately, no second-preference test was conducted. The SLEP was in factional disarray. Mrs. Bandaranaike had been deprived of her civil rights by the Jayewardene government and was ineligible to stand for election. The SLFP leadership gave limited support to its nominal candidate. More than that, because its demands had not been met, the Tamil United Liberation Front had expressed its indifference to the election in the Tamil areas and put forward no candidate. For the most part, legitimate Tamil party politics was no longer functioning by then.^[70] No arrangements for vote pooling were made. (Indeed, a defect of the system is that second preferences are not required for a valid ballot, and few voters cast any second preferences.) Under the atypical competitive circumstances of that election, second preferences were unnecessary. The UNP candidate won on a 52 percent majority of first preferences. In effect, since two main parties were far below their usual strength, party proliferation was insufficient to require vote transfers.^[71]

By the next presidential election, in 1988, conditions were worse. A civil war was being fought in the Tamil areas, and a reasonably effective boycott was conducted there. Again, the UNP candidate won on a small absolute majority of first preferences (50.4 percent).^[72] If and when the Sri Lankan Tamils (a 13 percent minority of the whole population) come back into the system, the Sinhalese parties will have every reason to bid for valuable Tamil second preferences. Since the Tamils' first choice, a Tamil presidential candidate, will have no prospect of winning, Tamil second preferences will be available for transfer. From the very narrow margins of first-preference victory under the most abnormal conditions, it seems perfectly clear that, under even modestly normal conditions,

[70] The rival Tamil Congress did put forward a candidate, who received 2.7 percent of the total vote.

[71] The election is described in full in C. R. de Silva, "Plebiscitary Democracy or Creeping Authoritarianism? The Presidential Election and Referendum of 1982," in James Manor, ed., Sri Lanka in Change and Crisis (London: Croom Helm, 1984), pp. 35–50.

[72] By now, however, the UNP quite clearly had majority support among all the minorities taking part in the poll—Indian Tamils (as opposed to the Sri Lankan Tamils in the war zone), Muslims, and Sinhalese Christians—as well as the support of a substantial minority of Sinhalese Buddhists. S. W. R. de A. Samarasinghe, "Sri Lanka's Presidential Elections," Economic and Political Weekly (Bombay), January 21, 1989, pp. 131–35; Sunil Bastian, "Two Elections," Thatched Patio (Colombo) 2, no. 2 (April 1989): 1–21.

no Sinhalese candidate will have a majority of first preferences, and elections will surely be decided by second preferences.

To illustrate just how AV would work and how it would produce conciliatory results, consider an imaginary conversation between party leaders seeking the presidency in Sri Lanka. Suppose two main Sinhalese candidates are contesting the election. The first estimates 40 percent first-preference support, and the second estimates 35 percent. There is also a Sri Lankan Tamil candidate, who can count on perhaps 10–12 percent of all first-preference votes. A meeting is convened between the first Sinhalese candidate and the Tamil candidate. Since no candidate will have a majority of first preferences, the discussion quickly turns to the subject of Tamil second preferences. The Tamil leader is asked whether he would be willing to urge Tamil voters to give their second preferences to the first Sinhalese candidate. He replies that his ability to do so depends on the Sinhalese candidate's willingness to be hospitable to Tamil aspirations. Otherwise, his appeal to Tamil voters to cast second-preference ballots for a Sinhalese candidate would be futile. Before long, concrete policy issues are being discussed. By the end of the negotiations, the first Sinhalese candidate has emerged as decidedly more accommodating on Tamil issues than the second Sinhalese candidate.

Now this conversation has reverberations. The second Sinhalese candidate will probably seek to differentiate himself from the first, by becoming less accommodating on Tamil issues, in order to attract hardline defectors who are displeased by the compromise commitments of the first Sinhalese candidate. For this reason, accommodation of Tamil interests will lose the first Sinhalese candidate some Sinhalese first preferences. He cannot go overboard in accommodating the Tamils, if he is to have a net gain from the transaction. But he can compensate, to some extent, for the loss of hard-line Sinhalese votes by appealing to the other minorities in Sri Lanka as well. They, too, can offer votes to conciliatory Sinhalese candidates. The 50 percent threshold means that, every time there are more than two candidates, parties need to search hard for pockets of votes to reach a majority. If there were more than two Sinhalese candidates, the majority threshold would be harder to reach, the extreme Sinhalese vote would be more divided, and the votes of minorities would be both more valuable and less risky to attract. In an AV system, then, intergroup compromise can become useful in getting elected, whereas in most other systems compromise makes it more difficult to get elected.

This transaction will not necessarily be limited to one election only. If reelection is to be sought later, the current commitments will be kept, and the partners will have found joint rewards in the arrangement. The defecting voters will likewise remember the concessions unfavorably, and the party of the second Sinhalese candidate will find rewards in continuing to attend to their concerns. In other words, a party spectrum, based on willingness or reluctance to compromise, will begin to differentiate itself.

Translating these parochial circumstances into general lessons about alternative voting, one point stands out. It was Sri Lanka's strong tendency toward party proliferation that led it to adopt a majority threshold and alternative voting to reach that threshold in the first place.^[73] Without party proliferation, AV is of less utility. If a party can win on first preferences, second preferences are irrelevant. The peculiar conditions that undid Sri Lanka's customary and long-standing party proliferation also undid, for the time being, some—but, as I shall note below, not all—of the conciliatory impact of AV.

If there is reason to think AV would generally encourage intergroup accommodation, is this likely to be true for South Africa? It will be recalled that a problem with applying Nigerian-style distribution requirements to South African parliamentary elections is that, for distribution to work and still provide certainty that someone will be elected, distribution thresholds must be kept at least moderately low. But modest distribution thresholds would afford no guarantee of inclusiveness. The same would not be true for alternative voting. With party proliferation and the need to get over the 50 percent threshold, there would be no real danger of not electing anyone. The reallocation of preferences takes care of that automatically. There would also be little danger of Black-White polarization. Just as all groups could make arrangements with each other, Whites could help elect a Black president, if AV were used in conjunction with a presidential system. And the same would be true for parliamentary seats. The whole notion of AV would be that voters of all groups could help elect parliamentarians of other groups, even if parties break along ethnic or racial lines. Not that any group could unilaterally dictate terms as the price of second preferences, for the simple reason that the party being dictated to is also in a competitive game for first preferences. The system fosters compromise but not surrender. Across the board, AV, with majority requirements, should en-

[73] Horowitz, Ethnic Groups in Conflict , p. 639.

courage intergroup conciliation and coalition in a major way. Indeed, as we shall see in Chapter 7, the percentages used in the Sri Lankan example above may not be so far removed from the South African electoral numbers when parties compete under universal suffrage.

There is, however, one qualification. Parliamentary constituencies would have to be heterogeneous. To achieve this, the constituencies may have to be large, and they may therefore need to be multimember constituencies. Alternative voting in multimember constituencies creates a complicated ballot. This is a major disadvantage, but the promise of alternative voting for intergroup compromise is so considerable that it is worth thinking long and hard about how to make it work.

There are certainly simpler systems, such as approval voting, which allows voters to cast ballots indiscriminately for all candidates they deem acceptable.^[74] In some circumstances, approval voting flushes out second choices. However, it will not produce conciliatory results in divided societies. Since approval votes are, in form, first preferences, voters are not likely to include candidates of other ethnic groups in their list. To induce the exchange of votes across ethnic lines, interparty agreements are necessary, and these are unlikely to be consummated without a preferential ballot.

This brings me to a third set of innovations, which can be dealt with quickly: ethnically mixed slates. The Malaysian example is the quintessential case in practice but not in prescription, for nothing in Malaysia requires mixed slates. The coalition partners simply found the single slate to be to their advantage, and they pursued that advantage. The Malaysian example shows how vote pooling can be accomplished with profit—it illuminates the mechanism—but, as indicated previously, because of the idiosyncratic formative conditions of the coalition, it does not show us how to get to that result.^[75]

The Lebanese had a way to achieve the same result through the electoral system, and it served Lebanon well from 1943 to 1976. The system consisted mainly of (1) seats reserved by ascriptive group, (2) in mixed, multimember constituencies, (3) on the basis of a common electoral roll. Because seats were reserved, competition for them was intraethnic, but because everybody voted for each seat, irrespective of the

[74] See Steven J. Brams and Peter C. Fishburn, Approval Voting (Bost Birkhäuser, 1983). I am indebted to Akhil Amar for suggesting the potential relevance of approval voting.

[75] However, in Chapter 7, I shall describe the emergence in South Afric of some patterns resembling those in Malaysia, and I shall point to some possibilities for coaxing them along.

voters' identity, there were inescapable incentives for candidates in the multimember constituencies to exchange with each other the votes of their supporters and thus to put up what were tantamount to mixed lists. Otherwise, each candidate might have been limited to the support of members of his own group and defeated by an opponent who had obtained support from other groups.

It was an ingenious system,^[76] but there is no need to consider it further, because, as I said in Chapter 1, it would be completely unacceptable in South Africa. A good many South Africans have made clear their determination not to have official group designations taint their electoral system. Reserved seats, balanced lists, communal rolls,^[77] and the like are all precluded. Even if an agreement were struck at the outset to create reserved seats or communal rolls, it would suffer from dubious legitimacy and might not prove durable. Were it not for the metaconflict, in short, a Lebanese-type system might have been considered to mitigate the conflict.

Apart from the Lebanese system, nothing I have discussed here rests on any official designation of group identity. The innovations I have considered should help mitigate group tensions, and yet all of the systems analyzed here are ethnically and racially neutral. As we have seen, several are used in more or less homogeneous (or at least not severely divided) countries.

Electoral Engineering:
Mechanisms and Consequences

The approach I have advocated is to adopt an electoral system that will make moderation rewarding by making politicians reciprocally dependent on the votes of members of groups other than their own. The dependence is only marginal, of course, but it will sometimes be the margin of victory. Since the parties must pool votes rather than pool merely seats, they must find ways before the election to communicate their ethnically and racially conciliatory intentions to the voters. After the election, they must deliver on those commitments or risk electoral retribution.

[76] I have described the Lebanese system and its conciliatory effects in Ethnic Groups in Conflict , pp. 633–35.

[77] For the Fijian electoral system, using communal and common rolls, see Ralph R. Premdas, "Fiji: The Anatomy of a Revolution," Pacifica 1, no. 1 (January 1989): 67–110, at 80–83.

This approach could hardly be further from that of consociational democracy. Two sharp differences should be underscored.

The first relates to the much-contested question of groups and alignments in South Africa. Consociational arrangements, it is true, do not require specification of group identities in advance of party formation and elections. But consociation assumes ascriptive groups will emerge; and, once they do, it requires a certain entrenchment of their rights to autonomy, mutual veto, and the like. (The fact that the formal holders of the rights in some schemes may be political parties, rather than ethnic groups, matters not at all, once there are ethnically based parties.) The electoral mechanism of vote pooling, on the other hand, makes no such provisions. It only assumes that, if distribution requirements are laid down, parties will seek to find well-distributed support; or, if alternative voting is adopted, parties will seek second preferences to assure a majority. The policies pursued by governments will reflect the exigencies of building support; they will flow from incentives, not constraints.

Quite obviously, it is my view that ethnic and racial divisions will be a prominent part of the South African political system. But if I am wrong and they are not, no harm will be done. These electoral systems will still encourage conciliation along whatever lines of difference emerge in the polity—including, but not limited to, differences based on policy, ideology, class, or region. Consequently, to adopt these innovations is not at all to bias the future political system in favor of ethnic and racial politics. It is only to take some precautions against the most severe polarization of any kind. And so these provisions will ameliorate the metaconflict as well as the conflict. One can hardly overemphasize the utility of incentives to conciliation in post-apartheid South Africa.

The second difference from consociation is that these mechanisms work at the voter level, not at the elite level. They contemplate no elite cartel, no monopoly control over the groups. Political leaders must act, of course, to take advantage of them; but, in acting, party leaders are responding to the political market in votes. They are not conspiring against the electorate. And, for this reason, we need make no assumptions about the politicians' own beliefs and whether they are more moderate than those of the voters. All we need assume is that politicians are rational electoral actors, that they like being elected and reelected—not exactly farfetched assumptions.

If these mechanisms work as they should, they will provide some modest boost to the moderate middle, on whatever issue dimensions emerge, at the expense of the extremes. But they do not, by any means,

rule out the extremes. On the contrary, they simultaneously provide a raison d'être for the extremes, for those extremes react to the compromises made by politicians to secure well-distributed support or second preferences. Those voters who are against vote pooling or conciliation will gravitate to those parties that do not advocate or practice it. It will be recalled that the emergence of the multiethnic Malaysian Alliance gave a fillip to parties on the Malay and Chinese flanks. Similarly, if a United National Party presidential candidate in Sri Lanka were conciliatory to the Tamils, in order to secure his margin of electoral victory on Tamil second preferences, the Sri Lanka Freedom Party would, no doubt, denounce the underlying interparty agreement that attracted Tamil votes. In short, the parties will sort themselves out by their willingness to compromise, and the center will be surrounded by less conciliatory flanks. But the rules of the game will generally favor accommodation—an enormous advantage in a divided society.

Sometimes, of course, if parties start out by being ethnically based, interethnic coalitions will emerge, not just to govern, but to contest elections. The Malaysian Alliance was such a coalition, which put forward a single slate. All of the electoral devices that might foster conciliation make pre-election agreements between parties electorally rewarding. That is how the devices will be brought to life.

It will be noted that, in advancing these electoral ideas, I have practically ignored the major conventional arguments for first-past-the-post and for list-system PR.^[78] The seat bonus typical of first-past-the-post systems is said to provide stable governments by converting pluralities of votes into majorities of seats and also to make those governments moderate. Ruling parties become risk averse, for if they offend marginal voters, small swings of votes can produce large swings of seats and thus alternation in office. In the case of ethnically or racially divided societies, however, these advantages are nullified by the common identification of party with ascriptive group and the concomitant tendency of majorities to shut out minorities.^[79] Floating voters are less important.

[78] For a good, concise statement, see Butler, "Electoral Systems," pp. 11–22.

[79] There is no evidence whatever that plurality systems minimize the the role of ethnic parties in divided societies. Compare Kader Asmal, "Electoral Systems: A Critical Survey" (paper presented at the In-House Seminar, African National Congress, Lusaka, Zambia, March 1–4, 1988), p. 5. Quite the contrary, such parties can flourish in plurality systems, as they did, for example, in the Nigerian First Republic or in Zimbabwe before the one-party state was declared. The most plurality systems will generally provide is an incentive to reduce the number of ethnic parties. By so doing, they decrease multipolar fluidity and increase bipolar rigidity and conflict. Cf. Horowitz, Ethnic Groups in Conflict , pp. 360–62. The Malaysian coalition, which grew up in a plurality system, is an unusual case.

Similarly, list-system PR is generally advocated on grounds of fairness, because it reduces the disparity between seats and votes; parties receive little or no seat bonus. But, again, in divided societies, list-system PR is as conducive to racial or ethnic exclusion as first-past-the-post is. The only difference is that, unlike first-past-the-post, list-system PR may compel formation of a coalition government before the ascriptive minority or minorities can be excluded.

And so these conventional arguments for plurality or proportionality are irrelevant to the problem of divided societies, which is the problem of exclusion and inclusion. As a matter of fact, close attention to the conventional arguments, mainly formulated in homogeneous societies, can produce harmful effects in heterogeneous societies. Two excellent illustrations, related to list-system PR, are at hand.

List-system PR, with the whole country as a single constituency, is, as we have seen, conducive to proportionality. However, there are two conventional objections to this form of national-list-system PR that have been noticed in South Africa,^[80] as they have elsewhere, and in some places have produced modifications of the system.

The first objection is that the relations between individual representatives and constituents are attenuated or fictitious when the whole country is a single constituency. This objection can be met by a change from national lists to separate lists for each multimember territorial constituency. Then, even if a constituency has several representatives, they are accountable to it.

The second objection is that voting by party-determined lists of candidates, with no voter ability to alter the order of candidates elected from the list, accords too much power to central party authorities and too little power to individual voters. This objection can be met by empowering voters to determine the order in which candidates on a party list will be elected. The mechanism is a simultaneous double ballot: the voter first votes for a party list and then votes for the order of candidates within his preferred list. In counting ballots, parties are first allocated percentages of seats, and candidates are then deemed elected or not according to voter preferences. As we have seen, candidates, party factions, and voters in Italy are actively mobilized in the process of altering the order of the party lists.

Both of these objections and both modifications are customarily de-

[80] Simkins, Reconstructing South African Liberalism , pp. 76–79; Slabbert and Welsh, South Africa's Options , pp. 152–53.

bated in ethnically neutral terms, but they are not ethnically neutral. The parliament of Sri Lanka is now elected by constituency-list PR, and voters have power, by means of an additional candidate-preference ballot, to determine the order in which candidates are elected within party lists. The first results under both modifications suggest that they have the ability to reduce the representation of minorities in both senses: office holding and accommodation.

Geographically dispersed minorities will fare less well under a constituency-list system than under a national-list system, all else being equal. In a constituency with three seats, it is perfectly possible that a 20 percent minority will neither gain a representative nor be in a position to affect the outcome, since party appeals will be directed to claims made by larger groups in the constituency. This, indeed, has been the result in some Sri Lankan constituencies.^[81] The same result follows from voter designation of candidate order within constituency lists. A party that generously places a minority candidate on its list will find that majority voters will change the winning order, to the detriment of that candidate. Any party tempted to go further and run a joint list and pool votes across ethnic lines will find that majority voters have undone the accommodative work of the parties by preferring majority candidates and defeating minority candidates. Again, Sri Lanka has experience on this score. Concludes C. R. de Silva, "The problem of marginalization through the system of preference voting in large electorates has to be viewed with concern by geographically dispersed minority groups."^[82]

Such considerations should give considerable pause before debating any electoral innovations for divided societies in terms of the conventional criteria used to evaluate electoral systems. The problem of ethnic and racial inclusion and exclusion should induce us to think differently about desirable electoral systems. In evaluating list-system PR, for example, its formidable ability to produce proportionality is far less important than its powerful impetus to exclusion of ethnic groups not represented in the majority coalition. In evaluating distribution requirements added to plurality systems, it is hardly a cogent objection that

[81] S. W. R. de A. Samarasinghe and C. R. de Silva, "The Development Council Election of 1981: Its Political and Electoral Implications," Sri Lanka Journal of Social Sciences 4, no. 1 (June 1981): 79–109, at 97. The elections analyzed in this article were the first conducted under the PR system also used in Sri Lankan parliamentary elections, but with seat apportionment that did not match the parliamentary apportionment.

[82] C. R. de Silva, "Preference Voting: Some Remarks on the Sri Lankan Experience, 1987–1989" (paper presented at the 18th Annual Conference on South Asia, Madison, Wisconsin, November 3–5, 1989), p. 7.

they are likely to reward, not just the largest party, but also the party most widely spread across a territory, for the most widely spread may also be the most ethnically inclusive. In evaluating alternative voting, it is not an important fault that it produces less perfect proportionality than does list-system PR, if, unlike list-system PR, it produces interethnic vote pooling. And it is a virtue that AV does not provide as much of a seat bonus as do plurality systems—though it would, in South Africa, as elsewhere, accord some modest seat advantage to the largest party.

Where AV produces majority rather than coalition governments, they are likely to result, not from artificial seat bonuses, but from the considerable incentives to broaden out support that are provided by the requirement of a majority.^[83] No racial or ethnic group will be permanently in office and no other group permanently out. Parties will have every reason to seek widely distributed support or to seek out the second and third preferences of many pockets of voters, enhancing the inclusiveness of the system. Since all this is the negation of apartheid—without being its mirror image, majority domination—it should be attractive to South Africans.

The broadening impact of alternative voting means it is more likely

[83] In fact, the incentives to broadened support can be very great in alternative voting systems. Some of the modest seat bonus provided by AV is a function of constituency size. See note 69, above. Evidence from a large, multicountry sample of elections suggests that majority-threshold electoral systems—unfortunately, the category merges AV and runoff systems—overall produce one-party majority governments (rather than coalition governments) more frequently than other PR systems do, although somewhat less frequently than plurality systems do. Moreover, a large number of these one-party majority governments are the product, not of an artificial seat bonus, but of the attainment of an actual majority. This suggests that majority requirements (AV or runoff) for each seat may provide such significant incentives to broaden support that the party that ends up with the majority of seats is likely to have the support of a majority of voters in the country as a whole. A. Blais and R. K. Carty, "The Impact of Electoral Formulae on the Creation of Majority Governments," Electoral Studies 6, no. 3 (December 1987): 209–18.

Two important qualifications derive from the sample. First, a disproportionate number of elections conducted under majority-threshold rules took place a long time ago. Second, and more important, as indicated, the category of majority systems embraces both runoff and AV systems. There are significant differences even within AV systems, but these have not been disaggregated; nor have AV and runoff systems been separated out.

Since the Blais and Carty study does not separate out elections in divided societies, it is not amiss to note that, after AV was introduced in Sri Lankan presidential elections, the UNP proceeded to broaden out its support among all the minorities taking part in the elections. Majorities among the Indian Tamils, Muslims, and Sinhalese Christians formed part of the UNP's overall majority. See note 72, above. For reasons I have described, these were unusual elections. Nevertheless, the majority threshold is likely to induce broadening out for first preferences as well as interparty bargaining for second preferences.

to produce a government with an outright majority of seats than list-system PR or STV is. AV does not stand in the way of majoritarianism, but makes majorities responsive to the interests of others as well. This is an important conciliatory feature—and one that builds legitimacy—in a divided society.

It should be emphasized that, if the vote-pooling mechanism works as planned, it will result in real participation in power by minorities and majorities alike . This participation will not occur by means of rigidly specified quotas, on the Cyprus model, and it will not consist of the token overrepresentation of minorities that Whites in Zimbabwe enjoyed until 1987. The numbers are not fixed but fluid. Some governments will have more representation of some groups than of others; other governments will be composed differently. Everything will depend on the vicissitudes of elections, which is as it should be. But the elections will be structured so that parties seeking to expand their seat totals will have special reason not to neglect the interests of voters, including voters across ethnic lines, who can help them to do exactly that.

The electoral systems that are most appropriate are somewhat complex, but that is to be expected, given the tendency of plurality elections and two-party systems to intensify conflict in a bifurcated society.^[84] Equally complex systems, with distribution requirements or preferential ballots, have been found workable and acceptable in other divided societies of the developing world that practice electoral democracy.^[85]

The benefits that come with the costs of complexity are great. Among other things, alternative voting (like distribution requirements) would be likely to limit centrifugal tendencies in the party system of a divided society. Because of the need for interparty agreements to exchange votes, there would be a pull toward the center of the system that would help counter polarization. As Juan J. Linz has pointed out, a divided society with a polarized multiparty system—which is what South Africa would have—needs something other than a "feeble electoral system, particu-

[84] Robert A. Dahl, "Some Explanations," in Robert A. Dahl, ed., Political Oppositions in Western Democracies (New Haven: Yale University Press, 1966), pp. 375–76; Lewis, Politics in West Africa , pp. 71–73.

[85] Nigeria, of course, has had the distribution requirements for presidential elections. Sri Lanka has alternative voting for its president. Malta has the single transferable vote with a procedure for "topping up" the seats of a party with majority support grafted onto the STV system. The complexity of that system has not prevented the acceptance of electoral results in Malta's divided society. See Stephen Howe, "The Maltese General Election of 1987," Electoral Studies 6, no. 3 (December 1987): 235–47.

larly one with pure proportional representation [that] exercises no restraint on the voters, and . . . furthers the persistence of fragmentation." Rather, it needs an electoral law that rewards cooperation by pro-system parties, preventing the creation by anti-system parties of "a negative majority" that impedes the work of parliament.^[86] The electoral systems proposed here thus avoid the problems associated with plurality systems, on the one hand, and PR systems, on the other, and build interparty cohesion. In Chapter 6, I shall argue that a presidential system would reinforce this centripetal effect in the party system.

The choice, as I have indicated, is not between racial and ethnic politics, on the one hand, and no racial and ethnic politics, on the other. The choice is between zero-sum, high-conflict contests along racial and ethnic lines—with a considerable potential for one or more actors to step in and end the democratic competition before it gets overheated or after it produces exclusionary results—and open-textured, fluid, low-conflict contests, mainly along racial and ethnic lines but with an admixture of intergroup cooperation. The challenge is to take an environment conducive to ethnic and racial allegiances in the party system and create incentives for parties to bid for floating voters who would otherwise vote their group identity. From what we know of the politics of severely divided societies, the choice may well be to see voters floating in the political system or floating in the river.

[86] Juan J. Linz, "Crisis, Breakdown, and Reequilibration," in Juan J. Linz and Alfred Stepan, eds., The Breakdown of Democratic Regimes (Baltimore: Johns Hopkins University Press, 1978), pt. 1, pp. 67–68.