Racial, Ethnic or Rational Voters? Splitting Tickets in South Africa

Abstract

Many scholars worried that political liberalization in South Africa would open the door to widespread racial and ethnic conflict or that elections would be little more than a ‘racial/ethnic census’. Today, after South Africa's third successful election, ideas about the extent to which South African voters cast their votes on the basis of ascribed racial or ethnic identities remain controversial. This paper offers a new way to assess competing hypotheses about the micro-foundations of South African vote choice: by studying the correlates of a strategic voting behavior, split-ticket voting. This is accomplished by analyzing data from a 1997 survey in which respondents participated in a ‘mock election’. The results show that ascribed identities do certainly play a role in predicting strategic voting in South Africa but that they are far from being the only important predictors. This research shows that strategic voting in South Africa is driven at least as much by political sophistication and political ideology as by loyalty to any racial, ethnic, or regional identity.

Notes

1. Although much of this violence was reported in the press and portrayed by elites on both sides as ‘ethnic’ in nature, closer examination suggests quite clearly that ethnic divisions themselves in no way explain ANC/IFP antagonism.

2. Because the ANC was known to be the party with the largest constituency by a wide margin, some speculated that negotiations about plurality vs. proportional representation (PR) would cause a major ‘hang-up’ between the ANC and its competitors. In all likelihood, the ANC was motivated to accept PR as the basis for South Africa's electoral system in order to secure broader international legitimacy and to maintain a foothold in the three provinces in which it would face the greatest challenge: KwaZulu Natal, the Northern Cape, and the Western Cape (see Reynolds, 1999).

3. Even so, Joel Barkan notes that the lack of geographical districts that often comes with PR voting schemes can eliminate the incentive for politicians to stay in touch with rural voters (1998). Some have also expressed concerns that proportional representation in divided societies can reinforce those divisions by eliminating incentives for pre-electoral compromise among competing groups; there is a sense in which systems of proportional representation can even offer elites the incentive to compete on the basis of race, ethnicity, or some other exclusive category (Lardeyret, 1996).

4. Independent election monitors reported considerable irregularities in KwaZulu Natal. Many ballot boxes were delivered to counters unsealed or with ballots inside neatly stacked—a relatively clear indicator of tampering (Reynolds, 1999). Despite these irregularities (the full extent and impact of which are still not known), the outcome of the elections in that province were accepted as ‘fair’ in a hastily negotiated compromise between the ANC and IFP. As a result, the IFP won the province by a slim margin.

5. The notion that Africa is a hotbed for ethnic conflict may be commonly held among laypeople and the non-African media, but many political scientists have in recent years sought to emphasize the enormous history of and potential for inter-ethnic cooperation on the continent (Fearon and Laitin, 1996).

6. Notable exceptions include studies of Australia (Rusk, 1970) and Israel (Arian and Weiss, 1969; Arieli-Horowitz, 2002).

7. Numerous studies treat divided government itself as the independent variable and examine its impact on policy and voters. Divided government in the US has been argued to influence, for example, legislative (policy) outcomes (Edwards et al., 1997), the vote choices of individuals (Nicholson and Segura, 1999) and voter turnout (Franklin and Hirczy, 1998).

8. Election results for all South African elections are accessible at www.elections.org.za, the website of the Independent Electoral Commission of South Africa.

9. These data were obtained at www.elections.org.za, where results from all South African elections are publicly available.

10. In this article, the survey data are analyzed in a manner that accounts for the sampling design. Probability weights, included in the dataset for each observation, are used to ‘weight the sample up’ to the population of South African voters of which it is meant to be representative.

11. Of course these frequencies cannot be directly compared due to differences in the standard errors that result from the fact that the frequency estimate for each population category is estimated using a different number of respondents.

12. A ‘dummy’ variable is simply a variable that is either scored as a 0 or a 1 (‘0’ for ‘off’ and ‘1’ for ‘on’).

13. Dummy variables are included for whether or not an individual chose to identify with a language and for whether or not an individual chose to identify with a race. The response ‘Afrikaner’ was coded as a language group. The response ‘Boer’ was not. The response ‘non-White’ was coded as a racial group despite its lack of specificity.

14. It is assumed here that ‘don't know’ responses signify an non-opinion or non-attitude regarding the relationship between the actual perceived status of one's home language and the ideal status of one's home language and thus are similar in meaning to ‘about right’ respondents. In other words, it is assumed that ‘don't know’ respondents have no reason to believe that their home language should be used any more or any less. They do not feel that their language is in any way privileged but they also have no grievances.

15. Use of the ‘income’ variable included in the IDASA survey results in a significant number of missing values. In South Africa, a country with significant income inequalities, controlling for income levels was thought to be highly important. The results reported in this article exclude the cases for which income was not reported. If income is omitted from the model (and the missing cases included), the results do not change, and the same key variables retain their statistical significance.

16. The threshold for ‘significance’ used in this study is 95 per cent statistical confidence. That is, we can be 95 per cent confident in the veracity of the relationships reported in this paper uncovered by the probit model.

Additional information

Notes on contributors

Eric S. Mclaughlin

* Department of Political Science, University of New Mexico, Albuquerque, New Mexico, USA. Email: [email protected]