Full terms and conditions of use:

Transcription

1 This article was downloaded by: [University of Cape Town Libraries] On: 15 August 2012, At: 10:26 Publisher: Routledge Informa Ltd Registered in England and Wales Registered Number: Registered office: Mortimer House, Mortimer Street, London W1T 3JH, UK Democratization Publication details, including instructions for authors and subscription information: Opposition Parties and the Voters in South Africa's General Election of 1999 R. Mattes & J. Piombo Version of record first published: 06 Sep 2010 To cite this article: R. Mattes & J. Piombo (2001): Opposition Parties and the Voters in South Africa's General Election of 1999, Democratization, 8:3, To link to this article: PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

2 83dem05.qxd 21/08/01 15:48 Page 101 Opposition Parties and the Voters in South Africa s General Election of 1999 ROBERT MATTES and JESSICA PIOMBO Variance in partisan choice among South African voters can be predicted on the basis of what is known about the way voters see economic trends, evaluate government performance, perceive political parties, and rate party leaders. However, in this analysis it is demonstrated that factors related to racial divisions shape and filter how voters perceive political performance, and to some extent lead different voters to emphasize different performance criteria. But race does not affect the way voters make decisions. Thus, South Africa s opposition parties are weak not because black voters, the overwhelming majority of the electorate, operate with a decisionmaking apparatus that emphasizes unity over performance or is hostile to pluralism and opposition. Rather, support for the African National Congress can be accounted for first, by positive ratings of its performance in government and second, by the fact that those black voters dissatisfied with the performance of the African National Congress (ANC) do not see a legitimate alternative among the existing opposition parties. South Africa s opposition parties are weak with very limited prospects of competing effectively for the support of black voters (the overwhelming majority of the electorate). 1 So much so that in late 1998 President Nelson Mandela abandoned his normally statesmanlike demeanour and called them Mickey Mouse parties. 2 What are the reasons for this problematic electoral imbalance and the spectre of a one-party dominant system? Most South African analysts have diagnosed the cause of this democratic malady to lie in the voters themselves. Due to their supposedly strong, even primordial, ties to political parties, few swing voters are seen to be open to persuasion, thus limiting chances for eventual electoral re-alignment. 3 The Dr Robert Mattes is Director of the Public Opinion Service at the Institute for Democracy in South Africa (IDASA), Cape Town. Jessica Piombo, of the Massachussets Institute for Technology (MIT), Boston, USA thanks the MIT GSC for enabling her to attend the conference on Opposition in South Africa s New Democracy in South Africa, June 2000, at which an earlier version of this article was presented. She is also grateful to the US National Science Foundation for the Award No , which supported her research at IDASA. The authors thank Irish Aid (South Africa) whose grant to the Opinion 99 research consortium supported the survey on which this article is based. Democratization, Vol.8, No.3, Autumn 2001, pp PUBLISHED BY FRANK CASS, LONDON

3 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION specific motivations for voter commitment to political parties have been seen to be: a desire to express national or communal liberation and independence by voting for the ruling liberation party (the ANC) 4 a desire to express communal solidarity with political parties that are seen to represent their racial or ethnic community 5 a fear of communal pressure and intolerance 6 cultural norms or predispositions hostile to the larger concept of political pluralism, multi-party politics, and opposition in general of hegemonically-oriented voters 7 a desire to maintain government patronage (on the part of current ANC supporters) or obtain it (on the part of current opposition supporters) which leads to a spiral of electoral domination on the part of the ruling party, 8 and disillusionment among minority party supporters at their minimal chance of selecting a government, leading them into either withdrawing from voting altogether or jumping on to the bandwagon of the majority party. 9 Thus, South African political analysts see voters to be motivated by a combination of primordial racial identities, fears and pressures, and narrow clientelist competition for patronage. Yet none of these hypothesized causes of electoral outcomes in South Africa considers the factors that have long occupied analysts of voter behaviour elsewhere, such as voter assessments of government and party performance, their evaluations of parties and candidates, and how they learn from election campaigns. But is the South African voter so different from voters in other areas of the world that the findings of such a voluminous literature should be ignored? This analysis argues that it should not. In a racially diverse and divided society emerging from two centuries of white minority domination and almost a half century of apartheid government policy, it would be fantastic to believe that race played no role in South African elections. Indeed, there is no denying the clear relationship between race, ethnicity, and voting patterns in South Africa s first two national elections. It remains by far the most salient line of political cleavage in the country. But what is important is how we interpret these relationships, for as is well known, correlation does not equal causation. As Christopher Achen has argued, correlations between demographic factors and voting do not explain voting; rather they themselves need to be explained. Social scientists need to specify what it is about a demographic variable or category that makes those voters vote differently than others. 10

4 83dem05.qxd 21/08/01 15:48 Page 103 SOUTH AFRICA S GENERAL ELECTION OF Thus, what exactly is the role of race in South African elections? How does it shape or affect voter choice? There are at least three broad possibilities. The South Africanist literature generally sees voters race as directly determining their vote choice as an expression of racial or communal identity, solidarity or loyalty. But race might instead have an indirect, moderating effect on how people vote because voters of different race groups use different criteria when deciding how to vote. Finally, race may shape the electoral context because voters of different races reach different judgements about the same set of criteria. The answer to this question holds fundamentally important implications for party strategies and the future of multi-party competition in South Africa. It also contributes directly to the understanding of voter behaviour in divided countries more generally. An earlier account devised a very simple three-step model to predict partisan behaviour and voter choice in South Africa. 11 Largely for heuristic reasons, demographic information, such as race, language, or class was excluded. Instead, knowledge of voters evaluations of economic trends, government, and political parties was found to predict the partisan choices of a large proportion of voters. The analysis here expands on that initial, simple model in several ways. First, the model is specified by the construction of conceptually and empirically grounded indices that help to reduce and simplify the large array of individual variables which were entered into the original model. This helps to identify more clearly the relative impact of various types of evaluations of the economy, government, and political parties. Moreover, the very exercise of constructing these indices also tells a great deal about how voters perceive the world around them. Secondly, the role of prior, or residual party identification is introduced to parry any assumption that each election is conducted de novo. Finally, race is introduced as a variable to explore the nature of its impact on voting. The Model Assumptions While a more detailed exposition of the conception of voter reasoning and choice can be found elsewhere, 12 the basic precept of the underlying model is that voters support parties and candidates who they think will govern best in the future. 13 People base this decision on available information about their own interests, how government policies may have affected their interests, and how alternative opposition parties might affect their interests if they were in government. In Achen s words: Voters do not ignore information

5 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION they have, do not fabricate information they do not have, and do not choose what they do not want. Thus, the voters need be neither geniuses nor saints. They are required only to do the best with the information they have. 14 Voters engage in what has been called low information reasoning. This can consist of drawing inferences from the information they do have; or of taking cues from others who they think have better information. The key point, however, is that what South African voters know (or think they know) about their interests, and how government and parties affect the latter, is heavily related to race. One source of available information is the immediate, everyday experience of going to the market, paying bills, bonds, mortgages and taxes, or keeping or looking for employment. In South Africa such experiences will vary greatly depending upon race. A second source of information is the past track record of government, parties, and candidates. Because almost all of the present parties or their key leaders have track records extending back before 1994, most voters can place most of them on one side or the other of the key issue of defending or reforming apartheid, or opposing it peacefully from within the system or fighting it violently from without. Since 1994, few of the public issues dividing the ANC and opposition parties have been framed in a way to transcend lines of racial cleavage. A third source of information is the style in which parties or candidates campaign, as well as simple demographic traits. Given no other information, where and how candidates campaign, as well as their skin colour, ethnic features, language, accent, or even dress can send powerful messages about where a candidate or party stands, what they value, and whether they could represent voters interests. 15 Yet important as it is to know what is inside voters heads, there is also need to focus on the context in which people learn information as well as the media through which they learn information. 16 People receive information through the lens of those they work or socialize with, or live next to, as well as the news media and political party debate. Moreover, they receive this information in light of their present values, and physiological and material circumstances. This is why, to varying degrees, voting patterns tend to differ in elections across the world along lines of class, ethnicity, religion, and race. This explains the peculiarly sharp voting cleavages along these lines in South Africa. The intricate matrix of apartheid era legislation ensured that people lived among, worked with, socialized and married within racially homogenous groups. People s entire worlds were circumscribed to the extent that they interacted primarily with those of the same race, language, culture, and basic material circumstances, and who received political information through the same news media as themselves. This has produced

6 83dem05.qxd 21/08/01 15:48 Page 105 SOUTH AFRICA S GENERAL ELECTION OF information networks that are largely culturally and racially bounded, internally consistent, and offer few cross-pressures. Thus, it is very likely that voters from diverse racial groups and ethnic communities see very different political and economic worlds when they judge political performance and reason about which parties and candidates will govern best in the future. Thus, where voters are situated in South Africa s racial matrix may shape not only what they learn about politics, but also which evaluations matter most in affecting their partisan choice. The Three Steps of Voter Choice While the actual process of reasoning about parties and candidates and forming identification with parties in an ongoing democratic political system is very likely to be fluid and seamless, it can be usefully deconstructed into three separate decisions. Decision One: Situating Yourself vis-à-vis the Governing Party A voter s initial task is to decide whether the government has done well enough to merit re-election. In a multi-party system, this step is best seen not as a simultaneous choice among all existing parties, but initially as a choice between the governing party and anything else. By making public policy, setting the policy agenda and dominating the news media, the prominence of the government between elections means that voters first need to decide where they stand vis-à-vis the governing party. Voters who supported the government in the previous election must ask themselves whether its record and achievements since then have been sufficient to keep them in their ranks. Voters who supported an opposition party must consider whether the government s performance has been sufficient to convert them to its column. Key factors in this decision are voters assessment of the economy (both personal and national conditions) and their sense of the overall direction of the country. Voters will have a general impression of how well the key actors of government (the president, parliament, provincial premiers and so on) have performed their jobs, as well as how government has performed in specific policy domains (like job creation, crime control). Voters will also look to their overall images of the governing party as a political party (specifically, its trustworthiness, credibility competence to govern, and representativeness). Voters pre-existing partisan dispositions are also an important part of this process. With the possible exception of founding elections in former communist systems where parties are totally new with few known leaders, 17 voters at any given time tend to have pre-existing predispositions leading

7 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION them to identify, or feel close to, particular parties, or to remain independent of parties. This stance, known as party identification, is best characterized as a standing decision that is constantly updated based on voters evolving evaluations of government, parties, and leaders. 18 New experience and information acquired about politics and governance in the post-1994 period must be processed in light of what voters already know about a governing party. Thus, existing partisan loyalties are a base that can be either eroded or reinforced by new information. 19 Decision Two: Situating Yourself vis-à-vis the Opposition Relative satisfaction with the government should lead people to remain committed to the ruling party or switch to identify with it. But while relative dissatisfaction should lead voters away from the ruling party, it will not tell them which opposition party they should support, if any. Thus, those voters dissatisfied with government performance face a second decision: Is there any opposition party that could do a good job in government? Or should they become, or remain independent? Since opposition parties, by definition, have not been in government, voters will look to their track records as political parties. Voters can look to the more visible actions of a party or their leading officials or spokespersons both inside and outside of the legislature to draw inferences about whether it could represent them, is credible, trustworthy, or competent to govern. And, as with the first decision, existing partisan identifications with any opposition parties will also constitute a base against which new information about politics and governance between 1994 and 1999 must compete. Decision Three: Situating Yourself vis-à-vis the Election Campaign The present model assumes that those who identify with a party will vote on election day, and vote for that party. 20 However, those voters whose experiences and evaluations with all parties since the previous election are so ambivalent or negative that they remain independent face a third task. They must then look to the rhetoric, messages, symbols, and personalities of the election campaign to decide whom to vote for on election day (as well as whether to vote at all). More specifically, people will consider what the parties say about a few salient issues that matter to them most. The Impact of Race: As mentioned above, the original model explicitly omitted any measures of demographic variables such as race, language or class, and found that a large proportion of variance in partisan choice could be explained based solely on a knowledge of what people thought of the economy, government performance and political parties. The causal process implied in this model is expressed in Figure 1.

8 83dem05.qxd 21/08/01 15:48 Page 107 SOUTH AFRICA S GENERAL ELECTION OF FIGURE 1 Evaluations of Government and Parties Partisan Choice An obvious criticism of this model is that the ability to predict successfully a large number of people based on their evaluations of government is not surprising since what is really being done is simply registering the impact of race without actually talking about it (Figure 2). On this logic, voters do not genuinely examine government and party performance or election campaigns. Instead, they decide which parties to support based on identity, networks of patronage, or intolerance (both intolerance of other groups, as well as intimidation or demands for conformity from their own community leaders). Thus, black voters know they support ANC (for reasons of racial solidarity) and thus offer more positive evaluations of ANC government than other racial groups, and more negative views of white opposition parties (the opposite would be true of white voters). Thus, the hidden variable of race simultaneously determines both evaluations and partisan identification, and either partially or totally explains the relationships which were observed in the earlier article and thus renders them either partially or totally spurious. FIGURE 2 Race Evaluations of Government and Parties Partisan Choice

9 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION Another possibility is that race and evaluations of government and party performance may interact. Thus, race has an indirect impact by modifying the relationship between evaluations and partisan choice because voters of different races either use different criteria on which to base their partisan allegiance, or weigh and prioritize the same criteria in very different ways (Figure 3). On this logic, race is actually a proxy for divergent cultural norms and values, different historical experiences of colonialism and apartheid, and different living conditions. All of these things create different imperatives and expectations that, in turn, lead members of one group to base their vote choices on different criteria than other groups. For example, voters of different races might place very different emphases on factors such as unity versus opposition, individuals versus parties, personality versus performance, or attach different weights to different aspects of government performance (such as economic growth versus redistribution). Evaluations of Government and Parties FIGURE 3 Race Partisan Choice But there is also a third possibility: race may shape voting patterns in South Africa because voters from different race groups form different evaluations of the same set of key factors. By this logic, race does not shape the way in which voters think about parties and government, but it does shape what they think. Rather than a fundamental difference in the criteria for political allegiances, people of all race groups use the same criteria though they might arrive at very different judgements (Figure 4). However, the causes of this type of impact have relatively little to do with different cultural traditions or norms, or even historical experiences. It has much more to do with the fact that radical differences in living conditions across racial lines leads to a homogeneity of economic and political interests within racial

10 83dem05.qxd 21/08/01 15:48 Page 109 SOUTH AFRICA S GENERAL ELECTION OF 1999 groups. Even where objective interests begin to diverge within a racial group, continued patterns of racially segregated communities may lead subjective perceptions of interest to converge. Moreover, internally consistent information networks may lead people of a specific race group to link their own interests and the performance of government and opposition in common ways. FIGURE Race Evaluations of Government and Parties Partisan Choice Testing the Model In order to test the model of voter choice and to specify the precise role of race, data is used from the first of a set of four surveys conducted by the Opinion 99 consortium during the campaign for the 1999 election. This specific survey was conducted in September 1998 with a nationally representative, random, stratified, proportionate cluster sample of 2,200 respondents. 21 First of all, the relevant question items were taken from the survey and were reduced from a large and unwieldy number of variables to a smaller and more coherent set of conceptually guided and empirically valid and reliable summary indices by using Factor Analysis and Reliability Analysis. Taken together, these procedures establish the validity and reliability of summary or constructed measures by testing for underlying patterns of responses to a series of question items as well as the consistency of those responses. This helps to identify and separate the relative impact of theoretically important factors more clearly. Moreover, the very exercise of constructing these indices also tells a great deal about how voters conceptually organize their economic and political world. Second, in order to test the differing causal linkages expressed in figures 1, 2, and 3, summary additive indices were constructed from the items identified by the Factor and Reliability Analysis and their relative impact assessed in explaining variance in partisan choice in each of the three models described above. To do this, a statistical test called Discriminant Analysis was used. This is useful to perform multivariate analysis when the dependent variable consists of unordered, discrete outcomes, such as which

11 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION party a voter identifies with or will vote for. Discriminant Analysis helps us understand whether and how a series of explanatory variables effectively discriminates cases (in this case, voters) into the various categories of the dependent variables (supporting or voting for various parties). It does this first, by calculating a set of functions that help discriminate cases between the categories of the dependent variable and determining the relative strength of the independent variables in the process. Second, it uses these discriminant functions to predict the group membership of each case (in this sense, the discriminant functions are analogous to the factors extracted by Factor Analysis). The rate at which the functions can correctly assign cases into categories (i.e. the party that the person actually says they support) informs us how powerful the independent variables are in discriminating among various categories. The relationships within race groups are then re-estimated to test whether race explains the observed relationship between voter evaluations of politics and governance and partisan choice, or whether it has an interactive effect that modifies the impact of evaluations of politics and governance across different categories of race. If the initial observed relationships reduce to zero, or depress considerably within race groups, then race totally or partially explains these relationships (Figure 2). If they change significantly in different ways for different categories of race, then race interacts with the relationship between evaluations and partisan choice and has a modifying effect (Figure 3). However, if the initial observed relationships remain intact, race has no direct impact on partisan choice. To then test for an indirect impact of race (Figure 4), the impact of race on voters evaluations of the political and economic environment was assessed by using Analysis of Variance, a more widely known statistical procedure that tests the association of a categorical independent variable with continuous dependent variables. Findings People s Views of the Political and Economic World The process of reducing the available items to a smaller set of summary indices yields several important, and perhaps surprising findings. First, South African voters do not automatically form opinions about the national economy from their personal economic conditions. While evaluations of personal economic circumstances, national economic conditions, and the overall direction of the economy are closely related to one another, Factor Analysis indicated that there were sufficient important differences to warrant creating separate factors for each concept. 22

12 83dem05.qxd 21/08/01 15:48 Page 111 SOUTH AFRICA S GENERAL ELECTION OF 1999 Second, voters make conceptual distinctions between overall government performance and performance within specific issue areas. They also make key distinctions in specific performance among different issue domains. Evaluations of the overall performance of (then) President Nelson Mandela, (then) Deputy President Thabo Mbeki, and of the national government reflect one common underlying dimension. 23 Evaluations of government performance across a range of 11 different issue areas reflected three broad underlying dimensions. First, people tended to evaluate government handling of the economy, job creation, and crime in a consistent fashion (a macro-management factor). 24 Government performance in a range of service delivery areas comprises a second factor; 25 and government performance with regard to appointing public officials, transparency, and fighting corruption formed a third factor ( good governance ). 26 A third important set of findings concerns how voters view parties. A wide range of items measuring evaluations of each of the seven largest political parties were factor analyzed. These comprised assessments of various characteristics of each party, as well as reactions to a range of specific actions taken by each party over the previous five years. It was found that voter evaluations consistently cluster by party, and then by whether the evaluation is more general or more specific. In other words, voters form a unique overall impression of each party that integrates opinions of how well a party could govern if elected, how trustworthy it is, how believable it is, how well it has performed basic party functions, and how inclusive it is. 27 Moreover, the relative importance of these items to each party-specific factor was consistent across all parties. For each party, the trust variable had the highest loading or correlation with the underlying dimension, followed either by evaluations of general party performance or competence to govern, then by credibility, and finally by inclusiveness. The second opinion of each party that people form is based on specific actions taken by the parties. 28 For each party, responses to these questions tended to cluster together, and could not be combined with responses to questions for any other parties. Nor could these specific evaluations be combined with the general evaluations. Thus, voters tend to see each party as a distinct entity and evaluate each one in a different light. They do not lump all opposition parties together, or make broad distinctions between the white and black opposition. They also form general impressions of a party s integrity and ability to govern that are distinct from whatever specific policy and leadership decisions or actions it had pursued. 111

13 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION Finally, it is important to note that voter ratings of party leaders do not cluster or factor together either with general party evaluations or evaluations of specific party actions, nor do they load together onto some common leader factor. Thus, they were used in this analysis as discrete, election campaign-specific items. Connecting Political and Economic Evaluations to Partisan Choice First, strong support was found for a three-step decisional process. The first two steps were compressed in an equation that models a choice of partisan identification based on a simultaneous judgement about the economy, government, and images of all parties. It performs worse (in terms of overall successful classification rates) than either the initial equation explaining ANC/non-ANC identification among all voters and the second equation predicting party identification among only non-anc supporters. Second, a combination of economic and political performance evaluations is able to place 79 per cent of all respondents into their actual ANC/non-ANC identification (Decision One). Notably, the model assigns 93 per cent of actual ANC identifiers to the ANC category based on what is known of their economic and political evaluations. However, it only has a 70 per cent success rate correctly placing non-anc identifiers into the non- ANC category; in other words, the model says that three out of every ten non-anc identifiers have an attitudinal profile that should lead them to identify with the ANC. Since the dependent variable in the first model has only two categories (ANC/non-ANC), Discriminant Analysis will only extract one discriminant function (it extracts a maximum of one less than all categories of the dependent variables). By examining the structure matrix loading of each item on that discriminant function (which represent correlations between variables and the underlying discriminant function), the most important predictor is the respondent s overall package assessment of the ANC as a political party (.93). The next most powerful predictors are pre-existing ANC partisanship (measured by recalling voting for the ANC in 1994, an admittedly imperfect proxy measure) (.68), overall evaluations of government performance (.67) and reactions to prominent ANC actions taken since 1994 (.62). However, a range of specific governance evaluations are also very important discriminators, including service delivery (.57), good governance (.50) and macro-management of the country and economy (.48). The least powerful predictors of ANC/non-ANC support are impressions of the overall direction of the country (.48) and national economic evaluations (.41). Importantly, personal economic evaluations are a very weak discriminator of ANC/non-ANC support (with a matrix loading of less than.30).

14 83dem05.qxd 21/08/01 15:48 Page 113 SOUTH AFRICA S GENERAL ELECTION OF TABLE 1 DECISION ONE Contribution of Independent Variables to Discriminant Function Structure Matrix Loading for the Discriminant Function Overall Image of the ANC.93 ANC Vote Overall Government Performance.67 ANC Specific Actions.62 Specific Government Performance on Service Delivery.57 Specific Government.Performance on Good Governance.50 Overall Direction of Country.48 Specific Government Performance on Macro-management.48 National Economic Evaluations.41 Canonical Correlation Between the Discriminant Function and ANC/Non-ANC Categories.63 Note: Only Structure loadings >.30 reported Source for this and all subsequent tables: Institute for Democracy in South Africa (IDASA), Opinion 99 Successful Classification Rate Total Sample All 79% ANC 93% Non-ANC 70% N = 2200 Note : Dependent Variable = ANC/non-ANC. Third, the analysis is able to place 79 per cent of all respondents who do not identify with the ANC (N=1617) into their actual partisan or independent categories based on their evaluations of opposition parties and prior partisan loyalties (dummy variables measuring a recalled vote for that party in 1994). Notably, the model is most successful classifying independents (84 per cent) and UDM identifiers (78 per cent) but only successfully categorises one out of every two DP (48 per cent) and NNP (43 per cent) identifiers. When non-anc identifiers decide which opposition party to support, or whether to remain independents, views of the IFP (having voted for it in 1994, its overall image, and evaluations of specific actions) were the most powerful discriminating function, followed by a function comprising views of the NNP (as well as previous PAC support) and then by functions comprising views and loyalties toward the DP, PAC, FF and UDM.

15 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION TABLE 2 DECISION TWO All Non-ANC Voters Structure Matrix Loading Contribution of Independent DF1 DF2 DF3 DF4 Variables to Discriminant Functions IFP Vote Overall Image of the IFP.70 IFP Specific Actions.38 Overall Image of the NNP.80 PAC Vote NNP Specific Actions.40 DP Vote Overall Image of the DP..62 Overall Image of the PAC.54 DP Specific Actions.44 PAC Specific Actions.42 FF Vote Overall Image of the FF.39 Canonical Correlation of DF and Groups (DV) Per Cent of Variance Explained 33% 19% 14% 12% by Function Note: Only structure loadings >.30 reported 4 DFs account for 78% of shared variance Successful Classification Rate Total Sample All 79% ANC X DP 48% FF 74% IFP 68% NNP 43% PAC 54% UDM 78% Independent 84% N = 1617 Note: Dependent Variable = Party Identification. Fourth, ability to predict correctly which party independent voters were most likely to vote for on election day or whether they were still undecided was significantly weaker (election day vote choice was measured by a secret mock ballot administered at the end of the interview). A model using ratings of party leaders and assessments of which party (if any) is best able to address the most important problem facing the country (as identified by the respondent) has an overall successful classification rate of 63 per cent. Of those independents who cast a secret mock ballot for the ANC, the model

16 83dem05.qxd 21/08/01 15:48 Page 115 SOUTH AFRICA S GENERAL ELECTION OF correctly placed 80 per cent into this category, the highest success rate. Rates were significantly lower for all other categories. The most powerful function discriminating independent voters into their correct vote choice categories comprises ratings of Nelson Mandela, Thabo Mbeki and an assessment of the ANC s ability to address the country s most important problem, as well as ratings of DP leader Tony Leon and the assessment of the DP s ability to address the country s most important problem. Importantly, the ANC and DP variables have opposite impacts on this function. The other three important discriminant functions combine assessments of whether the DP, IFP, NNP, UDM, or FF are able to solve the respondent s most important problems. All Independents TABLE 3 DECISION THREE Structure Matrix Loading Contribution of Independent DF1 DF2 DF3 DF4 Variables to Discriminant Functions Mandela Rating.56 Mbeki Rating.54 DP Able to Address MIP ANC Able to Address MIP.45 Leon Rating.36 IFP Able to Address MIP.48 NNP Able To Address MIP UDM Able to Address MIP FF Able to Address MIP Canonical Correlation of DF and Groups (DV) Per Cent of Variance Explained 35% 22% 13% 11% by Function Note: Only Structure loadings >.30 reported 4 Discriminant Functions account for 81% of the Common Variance Successful Classification Rate Total Sample All Voters 63% ANC Voters 80% DP Voters 57% FF Voters 45% IFP Voters 52% NNP Voters 51% PAC Voters 29% UDM Voters 46% Undecided 57% N = 1190 Note: Dependent Variable = Election Day Vote Choice.

17 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION The Impact of Pre-Existing Party Loyalties The data demonstrate the obvious: voters do not make up their minds anew at each election. Party identification (measured respondents recalled 1994 vote) works to ensure a degree of continuity in South African electoral outcomes. However, it is important to note that it does not play a determinant, or even a dominant, role. The equations modeling Decisions One and Two demonstrate that new information about political and economic developments since 1994 combines with pre-existing partisanship to account for voter dynamics in It is also clear that evaluations of government and parties are not simple reflections of pre-existing party loyalties. If they were, the multivariate analysis would indicate that previous partisanship rather than evaluations were the key driving force in shaping present party preferences, which it clearly does not. The Impact of Race: The clearest and most important conclusion from Tables 4, 5, and 6 is that controlling for race does not reduce the ability to explain partisan choice with economic and political evaluations. There are no consistent reductions in the overall power of the equations (as expressed by the successful classification rate). Thus the observed relationship between political and economic evaluations and partisan choice is by no means spurious, either totally or partially. In other words, there is no empirical support for the model represented in Figure Two. The second major point is that there are some significant modifications in the impacts of specific economic and political evaluations across different racial categories, especially when dissatisfied voters look for alternatives to the ruling ANC in Decisions Two and Three. The equations in Decision One show that voters of different races use the same set of economic, governance, and political criteria to decide whether or not to support the ANC (as reflected by the variables that load on the key discriminating function). At most, there are slight differences in the weighting, or emphases, given to these criteria by race (as reflected in the absolute size of the structure matrix loadings). When deciding whether to support the ANC or not, black voters place the heaviest emphasis on their overall image of the ANC as a party, followed by their opinion of overall government performance, pre-existing ANC partisanship, and government performance in service delivery. On the other hand, pre-existing ANC/non- ANC partisanship, and performance ratings on good governance and macromanagement of the country are the strongest discriminators for white voters. The relative strength of the discriminating variables also follows slightly different patterns for Coloured and Indian voters. The predictive power of this equation also differs along racial lines; it is much stronger among white (99 per cent), Coloured (98 per cent) and Indian

18 83dem05.qxd 21/08/01 15:48 Page 117 SOUTH AFRICA S GENERAL ELECTION OF TABLE 4 DECISION ONE BY RACE Contribution of Independent Variables to Discriminant Function Structure Matrix Loading Black White Coloured Indian (DF1) (DF2) (DF3) (DF4) Overall Image of the ANC ANC Vote Overall Government Performance ANC Specific Actions Specific Government Performance on Service Delivery Specific Govt. Performance on Good Governance Overall Direction of Country Specific Govt Performance on Macro-management Personal Economic Evaluations.32 Canonical Correlation of DF and Groups (DV) Successful Classification Rate Black White Coloured Indians All 75% 99% 95% 96% ANC 66% 41% 59% 95% Non-ANC 98% 99% 98% 96% N= Note: Dependent Variable = ANC/non-ANC. (96 per cent) voters than among black (75 per cent) voters. Finally, the ability to predict non-anc support is consistently stronger within each race group than it is for the overall equation. In the Decision Two equations, more racial differences in the variables that comprise the discriminant functions are seen. Views of the IFP and NP comprise the core of the first two discriminant functions that help non-anc supporting black voters discriminate among opposition parties or an independent status. However, views of the DP and the African Christian Democratic Party are the key discriminators amongst white voters. Again, different patterns of the key variables for Coloured and Indian voters emerge. Again, the overall predictive power of the equation increases once steps are taken to control for race, as well as the specific ability to predict people who decided not to support any opposition party but remain independent. This suggests that race was partially obscuring or deflating the true relationship between evaluations and partisan choice. 29

19 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION TABLE 5 DECISION TWO BY RACE Contribution of Black Non-ANC White Non-ANC Independent Variables Voters Voters to Discriminant Structure Structure Function Matrix Loadings Matrix Loadings DF1 DF1 DF2 DF1 DF2 DF3 Overall Image IFP Vote NNP Overall Image PAC Overall Image IFP Specific Actions.38 NNP Specific Actions.36 UDM Overall Image UDM Specific Actions.31 DP Specific Actions DP Overall Image.46 DP Vote ACDP Vote FF Overall Image.52 FF Vote FF Specific.39 PAC Vote Canonical Correlation of DF and Groups (DV) Per Cent of Variance 46% 46% 22% 27% 22% 17% 14% Explained by Function Note: Only structure loadings >.30 reported For Black Non-ANC Voters, 3 DFs account for 83% of shared variance For White Non-ANC Voters, 4 DFs account for 80% of shared variance Contribution of Coloured Non-ANC Indian Non-ANC Independent Variables Voters Voters to Discriminant Structure Structure Function Matrix Loadings Matrix Loadings DF1 DF2 DF3 DF1 SDF2 NNP Overall Image PAC Vote DP Vote DP Specific Actions.36 DP Overall Image.47 PAC Overall Image.33 IFP Specific Actions.62 UDM Specific Actions.58 ACDP Voter NNP Specific Actions.39 IFP Specific Actions.36 NP Vote UDM General Image.31 Canonical Correlation of DF and Groups (DV) Per Cent of Variance 41% 27% 17% 58% 42% Explained by Function Notes: Only structure loadings >.30 reported. For Coloured Non-ANC Voters, 3 DFs account for 85% of shared variance. For Indian Non-ANC Voters, 2 DFs account for 100% of shared variance.

20 83dem05.qxd 21/08/01 15:48 Page 119 SOUTH AFRICA S GENERAL ELECTION OF Successful Classification Rate of Partisan Black White Coloured Indian Stances Among Non-ANC Identifiers All 86% 85% 82% 87% ANC NA NA NA NA DP 100% 26% 34% FF 74% IFP 63% 100% NNP 89% 9.3% 43% 45% PAC 55% 100% UDM 56% 78% 100% Independent 90% 94% 94% 94% N = Note: Dependent Variable = Party Identification. In the Decision Three equations, independent black voters were most likely when deciding how to vote on election day, to take their cues from their attitudes toward Mandela, Mbeki and whether the they felt the IFP, NNP, DP, or UDM were able to solve their most important problems. However, election day preferences of white independents are most successfully discriminated by their views of DP leader Tony Leon, as well as whether either the DP or IFP were able to solve their most important problem. Further differences in criteria are seen among Coloured and Indian voters. TABLE 6 DECISION THREE BY RACE Contribution of Black White Independent Variables Independents Independents to Discriminant Structure Structure Function Matrix Loadings Matrix Loadings DF1 DF2 DF1 DF2 DF3 DF4 Mandela Rating IFP Able to Address MIP Buthelezi Rating NNP Able To Address MIP Mbeki Rating DP Able to Address MIP.65 UDM Able to Address MIP DP Able to Address MIP Leon Rating FF Able to Address MIP Viljoen Rating.30 Van Schalkwyk Rating.35 Other Party Able to Address MIP.30 Canonical Correlation Per Cent of Variance Explained 40% 29% 30% 23% 16% 13% by Function Notes: For Black Independents, 2 Discriminant Functions account for 69% of shared variance. For White Independents, 4 Discriminant Functions account for 82% of shared variance.

21 83dem05.qxd 21/08/01 15:48 Page DEMOCRATIZATION Successful Classification Rate Black White Coloured Indian All Voters 71% 62% 62% 69% ANC Voters 89% 62% 65% 60% DP Voters 64% 80% 29% 62% FF Voters 45% 0% IFP Voters 53% 80% 100% NNP Voters 54% 49% 70% 87% PAC Voters 29% 0% UDM Voters 38% 64% 61% 75% Undecided 37% 47% 58% 27% N = Note: Dependent Variable = Election Day Vote Choice. However, it is when we turn to evaluations of the actual decision criteria themselves that the greatest racial differences can be seen. Table 7 displays racially grouped mean scores for each of the key independent variables used in Decisions One, Two, and Three. It also displays for each variable the Eta coefficient (which expresses the magnitude of association between, in this case, racial groups and economic and political evaluation) and Eta2 (which expresses the total variance in economic and political evaluations accounted for by racial differences and is analogous to R2 in multiple regression analysis). What is important to note is that almost all of the crucial variables that discriminate people into ANC or non-anc partisan stances in Decision One are the very same variables that display the greatest racial differences in Table 7. In other words, it is on those economic, governance and political criteria that matter most in determining partisan choice that there is greatest disagreement across racial categories. Racial differences account for almost one-half the variance in the most important discriminating variable, how people view the ANC as a political party. Race accounts for one-quarter to one-fifth of the variance in reactions to specific ANC actions since 1994, as well as evaluations of overall government performance, and specific evaluations of performance on issues of good governance and service delivery. It accounts for approximately one-sixth of variance in evaluations of national economic trends and government performance on issues of macro-management. Race also plays a significant role in many of the key criteria in Decisions Two and Three. It accounts for one-quarter to one-fifth of variance in voter ratings of Mandela and Mbeki, and opinions about the ability of the soonto-be official opposition party, the DP, to address the country s most important problem. Race statistically explains one-sixth of variance in overall views of DP, reactions to specific DP actions, as well as overall views of the PAC and NNP.

22 83dem05.qxd 21/08/01 15:48 Page 121 SOUTH AFRICA S GENERAL ELECTION OF TABLE 7 RACIAL DIFFERENCE IN EVALUATIONS OF KEY DECISION CRITERIA Black White Coloured Indian Eta 2 Overall Image of the ANC (1 5) ANC Specific Actions (1 5) Overall Government Performance (1 5) Specific Government Performance In Good Governance Issues (1 5) Thabo Mbeki Rating (0 10) DP Able to Address MIP (0 1) Specific Government Performance In Service Provision Issues (1 5) Nelson Mandela Rating (0 10) Overall Image of the DP (1 5) National Economic Evaluations DP Specific Actions (1 5) Specific Government Performance On Macro-Management Issues (1 5) Overall Image of the PAC (1 5) Overall Image of the NNP (1 5) Overall Direction of the Country (1 3) NNP Able to Address MIP (0 1) Tony Leon Rating (0 10) ANC Able to Address MIP (0 1) PAC Specific Actions (1 5) Constand Viljoen Rating (0 10) Stanley Magoba Rating (0 10) Marthinus Van Schalkwyk Rating (0 10) Mangosuthu Buthelezi Rating (0 10) NP Specific Actions (1 5) FF Able to Address MIP (0 1) FF Specific Actions (1 5) Overall Image of the IFP (1 5) Overall Image of the FF (1 5) Bantu Holomisa Rating (0 10) Overall Image of the UDM (1 5) IFP Specific Actions (1 5) UDM Able to Address MIP (0 1) UDM Specific Actions (1 5) PAC Able to Address MIP (0 1) Personal Economic Evaluations (1 5) IFP Able to Address MIP (0 1) Implications for Voting Theory In contrast to the usual reasons cited to account for South African voter behaviour, it has been demonstrated that evaluations of government and party performance are fundamentally important determinants of people s partisan choices. The relationship of people s evaluations of the economy, governance and political parties on one hand, and their partisan choices on