PARTY VOTING in COMPARATIVE PERSPECTIVE: the UNITED STATES, TAIWAN, and JAPAN DISSERTATION Presented in Partial Fulfillment of T

Home , 1955 System, Elections in Japan, House Party (1996), Issue voting, Japan New Party, Party identification

PARTY VOTING IN COMPARATIVE PERSPECTIVE: THE UNITED STATES,

TAIWAN, AND JAPAN

DISSERTATION

Presented in Partial Fulfillment of the Requirements for the Degree Doctor of

Philosophy in the Graduate School of The Ohio State University

Chia-Hung Tsai, M.A.

*****

The Ohio State University

2003

Doctor’s Examination Committee:

Dr. Herbert F. Weisberg, Adviser Approved by

Dr. Paul A. Beck

Dr. Jakub Zielinski ______

Adviser

Department of Political Science

ABSTRACT

The purpose of this dissertation is to study the effects of political institutions on party voting. My hypotheses are that multi-member districts with presidential systems like Taiwan will generate the lowest level of party voting and that single- member districts with parliamentary systems like Japan after 1994 will generate the highest level of party voting. Multi-member districts with parliamentary systems like

Japan before 1994 and single-member districts with presidential systems like the U.S. will lie between the two extremes. I analyze survey data for these four cases to test these hypotheses.

Chapter 3 contains the analysis of congressional voting behavior in the United

States. Using 1990 and 1998 data, I generate multi-nomial logit estimates to compare the relative influence of partisanship and candidate evaluations. The result shows that the effects of candidate evaluations and partisanship are nearly even across the two elections, and that incumbency status is also critical to the voting act.

Chapter 4 is devoted to the analysis of voting behavior in Taiwan. The elections in 1992 and 1995 are examined to display the candidate-based voting behavior under multi-member districts with presidential system. It is found, however, that party voting still prevailed in Taiwan at that time, although the influence of candidate evaluations increased during the period.

Chapter 5 looks at the Japanese voting pattern in 1976 and 1996. According to the estimates generated by the multi-nomial logit model, the 1994 electoral reform indeed increased the relative influence of party labels. In other words, the small-sized district system contributes to party-based voting behavior.

Chapter 6 pools the 1998 U.S. data, 1995 Taiwan data, 1976 and 1996 Japan data and estimates the effect of candidate evaluations and partisanship controlling for the election system and government structure variables. By rescoring the value of the system-level variables, the multi-nomial logit model presents the main effect of the two explanatory variables. Partisanship remains critical to Taiwanese voting behavior, and candidate-centered pattern has been declining in Japan. The results are compatible with the individual case studies in the preceding chapters.

Thus, political institutions have an impact on individual voting behavior. The deviation of Taiwan calls attention to the factor of culture and other features of political systems.

iii

Dedicated to my parents

ACKNOWLEDGMENTS

Since the first day I stepped in the Department of Political Science, all of the faculty and staffs here have facilitated my study wholeheartedly. The department has also provided me with financial aid for four years. As an international student in

American politics, I am very thankful to this program’s demanding training. My greatest gratitude is to my adviser, Dr. Herbert Weisberg, for his intellectual inspiration and great patience in instructing this research and correcting my grammatical errors. Without his support, I would have not gone this far to complete this dissertation. I am also grateful to the other two dissertation committee members--

Dr. Paul A. Beck and Dr. Jakub Zielinski--for their suggestions to the early draft of this dissertation.

I also appreciate the Center for Survey Research at Ohio State for its outstanding research environment and two-year extensive training on survey research.

My classmates made my life in and out of the classroom very interesting. My friends at Columbus and other cities have enriched my life too, especially Chia-wei,

Gary, Yung-ming, Hiro, Kana, and Yu-chi. I would like to thank them all.

The Japan data and part of the American data is purchased with the research grant provided by “Program for the Enhancement of Graduate Studies”. The Taiwan data is provided by the Election Study Center at National Cheng-chi University. I am alone responsible for the interpretation of data herein.

VITA

August 3, 1971……………………………………….Born – Tainan, Taiwan

1993…………………..…B.A. Political Science, National Taiwan University

1996………………..…M.A. Political Science, National Cheng-chi University

1999-present………….…Graduate Research Associate, Ohio State University

PUBLICATIONS

Research Publications

1. Chia-hung Tsai, “The Categorical Analysis of the Social Context Position-A Case Study of the 1989 Legislative Election in the Second District of Taipei City” Journal of Electoral Studies 2,2: 171-203 (1995).

2. Chia-hung Tsai, “American Voter Responses to International Political Events and Economic Conditions: 1920-1996.” Euramerica, 30,3: 143-191 (2000).

3. Chia-hung Tsai, “House Party Leadership Selection: 1959-1994.” Soochow Journal of Political Science, 11:1-33 (2000).

4. Chia-hung Tsai, “Why Do Taiwanese Vote? ” Journal of Electoral Studies , 8,2:125-158 (2001).

FIELDS OF STUDY

Major Field: Political Science

TABLE OF CONTENTS

Abstract……………………………………………………………………………ii

Dedication……………………………………………………………………..…..iv

Acknowledgments…………………………………………………………………v

Vita………………………………………………………………………………..vi

LIST OF TABLES ...... x

LIST OF FIGURES...... xiv

Chapters:

1. Political system, partisanship, and vote choice...... 1 1.1 Introduction...... 1 1.2 The concept of partisanship and revisionist views...... 8 1.3 The assessment of party voting...... 12 1.4 Summary and remarks...... 14 2. A comparative perspective on party voting ...... 16 2.1 Origins and development of political parties...... 17 2.2 Review of comparative voting literature ...... 21 2.3 A comparative framework of party voting ...... 28 2.3.1 Election systems...... 32 2.3.2 Government structure ...... 35 2.3.3 Typology of election and political systems ...... 37 2.4 Context of four cases: the United States, Taiwan, and Japan...... 38 2.4.1 SMSP and presidential system: the United States...... 40 2.4.2 SNTV and presidential system: Taiwan ...... 41 2.4.3 SNTV and parliamentary system: Japan before 1994 ...... 43 2.4.4 SMSP and parliamentary system: Japan after 1994 ...... 45 2.5 Determinants of party voting ...... 46 2.6 Plan of the dissertation ...... 48 3. American congressional voting behavior in 1990 and 1998...... 50 3.1 Introduction...... 50 3.2 Incumbency and party competition...... 53

vii

3.3 Candidate evaluations and party identification...... 55 3.4 Data and variables ...... 58 3.4.1 Data...... 58 3.4.2 Operationalization of variables ...... 59 3.5 Party identification and candidate evaluations ...... 62 3.6 Party identification and issue positions ...... 70 3.7 Issue position and candidate evaluations...... 74 3.8 Evaluation on national economy in congressional elections ...... 81 3.9 Party images in congressional elections...... 82 3.10 Multi-nomial logit model of congressional voting behavior ...... 87 3.10.1 Model...... 87 3.10.2 Results...... 92 3.11 Conclusion ...... 102 4. Taiwanese legislative voting behavior in 1992 and 1995...... 105 4.1 Introduction...... 105 4.2 Party system and the unification issue ...... 107 4.3 Party system and the socio-economic equity issue ...... 115 4.4 Party identification in taiwan...... 117 4.5 Candidate evaluations and vote choice...... 127 4.6 Party images and vote choice...... 137 4.7 Multi-party voting model of legislative election...... 141 4.7.1 Variables and model...... 142 4.7.2 Multi-nomial logit estimation of voting choice in Taiwan...... 148 4.8 Conclusion ...... 156 5. Japanese voting behavior before and after the 1994 electoral reform...... 158 5.1 Introduction...... 158 5.2 Context of the Japanese party system...... 161 5.2.1 The party system of 1955...... 161 5.2.2 The party system of 1993...... 165 5.3 Long-term forces in Japan politics...... 167 5.3.1 Social groups and Japanese parties ...... 167 5.3.2 Partisanship in Japan ...... 170 5.3.3 Ideology and value cleavage ...... 176 5.4 Short-term forces...... 183 5.4.1 Candidate evaluations...... 183 5.4.2 Issue positions and voting...... 194 5.4.3 Party images in Japan ...... 203 5.4.4 Social networks and voting...... 208 5.5 Determinants of the 1976 and 1996 vote...... 215 5.5.1 Explanatory variables ...... 216 5.5.2 Results...... 220 5.6 Conclusion ...... 228 6. Comparison of four cases...... 230 6.1 Introduction...... 230

viii

6.2 Model of pooled-analysis ...... 233 6.2.1 Model...... 233 6.2.2 Data...... 243 6.3 Government structure and voting pattern ...... 244 6.4 Election systems and voting pattern...... 255 6.5 Pooled-data analysis of four political systems...... 264 6.6 Conclusion ...... 272 7 Conclusion ...... 274 7.1 Review of method ...... 274 7.2 Review of findings ...... 276 7.3 Implications ...... 280 7.5 Ideas for future research ...... 285 Bibliography ...... 289

LIST OF TABLES

Table Page

2.1 Party voting, election system and government structure...... 37

2.2 Party voting of four cases ...... 40

3.1 Voters’ mentions about things they liked and disliked about House candidates, 1990...... 64

3.2 Voters’ mentions about things they liked and disliked about House candidates, 1998...... 65

3.3 Positive feelings toward candidates and partisanship, 1990 ...... 67

3.4 Positive feelings toward candidates and partisanship, 1998 ...... 67

3.5 OLS estimates of comparative candidate evaluations, 1990 and 1998...... 78

3.6 Correlation coefficients of partisanship and party images, 1990...... 86

3.7 Correlation of partisanship and party images, 1998 ...... 86

3.8 Means and ranges of the independent variables, 1990 and 1998...... 92

3.9 Multi-nomial logit estimates for three-choice model, 1990 ...... 95

3.10 Multi-nomial logit estimates for three-choice model, 1998 ...... 97

3.11 Discrete change in the probability of Democratic and Republican vote, 1990...... 100

3.12 Discrete change in the probability of Democratic and Republican vote, 1998...... 101

4.1 Seat shares and vote shares of political parties in the legislative elections, 1989-1998 ...... 109

4.2 Perceptions of parties’ unification issue stands ...... 114

4.3 Perceptions of parties’ representation of social class...... 116

4.4 Social groups and partisanship, 1992...... 123

4.5 Social groups and partisanship, 1995...... 124

4.6 Voters’ contacts with candidates through social networks...... 132

4.7 OLS Estimates of candidate evaluations, 1992 ...... 134

4.8 OLS Estimates of candidate evaluations, 1995 ...... 135

4.9 Means and ranges of the independent variables, 1992 and 1995...... 146

4.10 Multi-nomial logit estimates for four-choice model, 1992 ...... 149

4.11 Multi-nomial logit estimates for four-choice model, 1995 ...... 152

4.12 Discrete change in the probability of party choice in 1992...... 154

4.13 Discrete change in the probability of party choice in 1995...... 155

5.1 Lower house seats and fate of Japanese parties, 1990-1996 ...... 166

5.2 Consistency level of voting choice, party attachment, party evaluation, and past vote by partisanship...... 173

5.3 Logistic estimates of consistent vote choice between 1972 and 1976...... 175

5.4 Discrete change of probability in consistent vote choice between the 1972 and 1976 election ...... 175

5.5 Ideology and party choice, 1976...... 178

5.6 Ideology and party choice, 1996...... 178

5.7 Value cleavage and party choices, 1976 ...... 181

5.8 Value cleavage and party choice, 1996...... 182

5.9 Average number of candidates of same party label by district magnitude ...... 187

5.10 The most important determinant of voting--party or candidate...... 188

5.11 Sources of candidate evaluations, 1976 ...... 191

5.12 Sources of candidate evaluations, 1996 ...... 192

5.13 Agreement between candidates and voting choice, 1976...... 193

5.14 Agreement between candidates and voting choice, 1996...... 194

5.15 Structure of political issues, 1976...... 200

5.16 Structure of political issues, 1996...... 201

5.17 Issue Position of Japanese Voters, 1976 ...... 202

5.18 Issue position of japanese voters, 1996...... 202

5.19 Mean and percentage of response of party images, 1976 and 1996...... 205

5.20 Correlation of party images and partisanship, 1976 ...... 207

5.21 Correlation of party images and partisanship, 1996 ...... 207

5.22 Social networks by political party, 1976 ...... 212

5.23 Social networks by political party, 1996 ...... 212

5.24 Logistic estimates of party voting, 1976 and 1996 ...... 214

5.25 Mean value and range of independent variables...... 219

5.26 Multi-nomial logit estimates for four-choice model, 1976 ...... 222

5.27 Multi-nomial logit estimates for four-choice model, 1996 ...... 224

5.28 Discrete change in the probability of the LDP and the JSP vote, 1976...... 226

5.29 Discrete change in the probability of LDP and NFP vote, 1996 ...... 227

6.1 Party voting of four cases ...... 232

6.2 Multi-nomial logit estimates of three-choice model, Taiwan and Japan...... 247

6.3 Multi-nomial logit estimates of three-choice model, the U.S. and Japan ...... 248

6.4 Discrete change of probability of vote choice with dummy variable of government structure ...... 250

6.5 Multi-nomial logit estimates of three-choice model, Taiwan and the United States ...... 257

6.6 Multi-nomial logit estimate of three-choice model, Japan before and after 1994 ...... 258

6.7 Discrete change of probability of vote choice with dummy variable of election system ...... 260

xii

6.8 Multi-nomial logit estimates of three-choice model, the United States, Taiwan, and Japan before and after 1994...... 266

6.9 Discrete change of probability of vote choice, pooled data ...... 268

6.10 Comparison of candidate evaluations and partisanship, pooled data...... 269

xiii

LIST OF FIGURES

Figure Page

1.1 The classical Michigan model ...... 3

3.1 Evaluation of candidates and partisanship, 1990...... 68

3.2 Evaluation of candidates and partisanship, 1998...... 69

3.3 Partisanship and two issue positions, 1990 ...... 72

3.4 Partisanship and two issue positions, 1998 ...... 73

3.5 Mean values of party images, 1990 and 1998 ...... 85

4.1 Attitudes toward unification and independence, 1991-1995...... 112

4.2 Distribution of Taiwanese partisanship, 1987-1996 ...... 119

4.3 Mean of party perceptions, 1992 and 1995 ...... 126

4.4 Partisanship and candidate evaluations toward each party, 1992 and 1995 ...... 131

4.5 Party images of the KMT and the DPP, 1992 ...... 138

4.6 Partisanship and positive feeling toward parties, 1992...... 140

4.7 Partisanship and positive feeling toward parties, 1995...... 140

5.1 Occupation categories of LDP supporter, 1955-1985...... 169

5.2 Party identification of Japanese voter, 1976-1996...... 171

5.3 Party popularity, vote share, and seat share, 1958 to 1986 ...... 186

xiv

CHAPTER 1

POLITICAL SYSTEM, PARTISANSHIP, AND VOTE CHOICE

1.1 Introduction

Contemporary representative democracies owe their existence to political parties. The two functions of political parties—representation and governance—are critical to the legitimacy and authority of representative government. Because the citizens cannot participate in every government’s decision and therefore rely on representatives, political parties integrate social interests and recruit candidates who can implement their policy goal. Political parties are linkages between government and citizens in terms of the integration and representation of social interests, which makes the support of parties indicative of the legitimacy of democracy (Epstein, 1967).

In the course of the development of political science, voting studies have been undoubtedly one of the major research topics. Voting studies have been so pervasive because of the behavioral approach that brought new methodologies to study of democracy starting in the 1920s (Dahl, 1961). The rapid development of the behavioral approach initially encouraged scholars to look at political behavior, and then enormous support from government and foundations stimulated more empirical inquiry into political behavior. Political scientists borrowed concepts from

psychology, economy, and sociology to measure political attitudes. In the meantime, they collected empirical data along the guidelines of the three approaches. With the predispositional approach prevailing, the concept of partisanship became the central concept of voting studies.

In the 1960s, the scholars at the University of Michigan characterized the voting decision process as a funnel of causality in which partisanship is the most important long-term force remote from the voting decision while short-term forces, like candidate perceptions and issues, are proximate to voting choice (Campbell,

Converse, Miller, and Stokes, 1960; Campbell, Converse, Miller, and Stokes, 1966).

People develop party identification in their family; partisanship acts like a group identification, such as religion and social class. People generally remain loyal to their party regardless of policy or leader change, thus party attachment is fairly stable across elections. Party identification also colors people’s perceptions of issues and candidates. Although candidate evaluations and issue positions are more proximate to the voting act than is partisanship, party identification has a direct and an indirect influence on vote choice (Niemi and Weisberg, 1993). Therefore, partisanship anchors an individual’s voting choices in the long term, which allows researchers to isolate short-term factors and short-term variables, and to identify the aggregate trend of voting choice (Converse, 1969; Beck, 1986). On the other hand, candidate evaluations and issue positions vary with time; voters may favor one party in one election and the other in another election (Stokes, Campbell and Miller, 1958; Stokes,

1966). Whiteley (1988) illustrates the classical model as in Figure 1.1.

Candidate evaluations

Partisanship Voting choice

Issue positions

Source: Whiteley (1988)

Figure 1.1: The classical Michigan model

The arrows in Figure 1.1 denote one-way influence. Issue positions and candidate evaluations are shaped by partisanship, but they have short-term influence on voting choice that may outweigh the influence of partisanship as the election approaches. However, partisanship can affect the voting act indirectly through candidate evaluations and issue positions. Stokes (1966: 23) contended that, “the stronger the voter’s party bias, the more likely he is to see the candidate of his own party as hero, the candidate of the other party as villain.” Considering its direct and indirect influence, partisanship is likely to dominate the process of electoral choice.

Since their early work, the analysis of single elections, multiple elections, and panel data has found that party identification has a consistent impact on the voting decision (Goldberg, 1966; Hartwig, Jenkins, Temchin, 1980; Bartels, 2000). The stability of the distribution of partisanship also suggests that partisanship is a more

prominent factor in voting choice than candidate evaluation and issue positions.1 As long as party organizations persist and no disruptive pressure changes the party system, people will remain attached to parties and party alignments will crystallize

(Wolinetz, 1990). With party attachment stable, moreover, political parties can convert ambitious candidates to partisans, who will implement parties’ policy goals

(Aldrich, 1995).

What makes the classical model developed by the Michigan team a theory instead of a bundle of propositions is that party attachment is a long-term force exogenous to candidate evaluations and issue positions. People can identify themselves as Party A’s supporters while voting for Party B because of Party B’s better candidate or issue positions. People can respond to short-term forces election by election, but they still identify with a given party and vote for it when they see no difference between candidates (Niemi and Weisberg, 1993). Normally people would vote for their favorite party unless there is an appealing candidate or a special issue that pushes people to deviate from their party loyalty. In other words, the classical model argues that partisanship drives voting behavior. In this regard, I call the classical model “party voting theory.”2

1 Although the relative importance of party, candidate, and issue is still under debate, partisanship is viewed as the most important causal factor. See Stokes, Campbell, and Miller (1958) on why partisanship cannot fully account for the voting act and Niemi and Weisberg’s (1993) discussion of the relative influence of political attitudes.. 2 The original Michigan model by no means denies the importance of either candidate or issue factor to the voting act. By setting partisanship behind candidate evaluations and issue positions in terms of the distance to the final voting decision, the classical model actually recognizes the possibility that partisanship may have smaller impact of the voting act in elections. That is the reason why I avoid the term “the Michigan model,” since the classical model indeed assesses both the long-term and short- term factors. Nevertheless, the classical model emphasizes partisanship for its voting function and argues that most Americans consider themselves either Republicans or Democrats. By highlighting the critical role of partisanship in the classical model and referring to the classical model as “party voting theory,” I can contrast it with candidate-based voting or candidate-centered pattern without mistakenly simplifying the original model.

The implicit assumption of party voting model is that partisan strength is determined by personal experience and socialization regardless of context. However, some scholars have questioned this assumption; they argued that in some political landscapes party identification and vote preference are the same attitude or “party identification is simply a reflection of the vote preference” (Thomassen, 1976: 71).

Party identification, in this view, is not necessarily antecedent to the voting act; it could be another short-term force or even the result of the voting act. And it has been found that the degree of independence between party identification and voting varies from country to country (Butler and Stokes, 1969; Thomassen, 1976; Kaase, 1976;

Bowler, Lanoue, and Savoie, 1994).

American scholars also have pointed out that party identification is less important now than before (Burnham, 1970; Pomper, 1975; Schulman and Pomper,

1975; Wattenberg, 1984). Burnham (1970:130) noted that: “The political parties are progressively losing their hold upon the electorate.” In some empirical studies, candidate evaluation instead of partisanship is put in the central position of the voting model (Jackson, 1975; Markus and Converse, 1979). Those findings imply that party- oriented politics has been moving toward, if not slipping into, candidate-centered or incumbent-oriented pattern, especially in party organizations and the government’s decisions (Cox and Katz, 1996). In congressional elections specifically, party identification has given its way to candidate assessment due to the enormous privilege of incumbency (Jacobson, 2001).

The weakness of the party voting theory lies in its lacking consideration of the system-level variables, such as society, institutions, and economic development, all of which set up a context for party attachment. For example, the stable American two-

party system contributes to the long-term personal identity with political parties as well as social groups, and the two-party system is actually molded by the single- member simple-plurality (SMSP) voting system. The presidential system, on the other hand, encourages the exchange of interests between the president and congressmen, which leads to the candidate-centered campaign in both presidential and congressional elections. The declining partisanship and party image starting in the 1960s should be considered as the consequence of the interaction between the presidential system and

SMSP system, but the party voting theory that ignores the constraint of context fails to account for it. On the contrary, comparative research on political parties has already shown that government structure and election systems condition party development (Schattscheider, 1942; Epstein, 1967). Samuels (2002), therefore, maintains that the presidential system imposes “vote-seeking” incentives on political parties, which leads to a candidate-centered campaign. While the variability of political behavior can be observed in panel data or multiple election data, party voting theory would not be a completely general theory without cross-national study.

Every nation is a context, and studying a country entails contrasting one context with another (Kohn, 1987). The theoretical significance of theory defined in terms of specific measurement can be tested through comparative descriptive statements or a sophisticated model (Przeworski and Teune, 1970). The purpose of this dissertation is to compare the determinants of vote choice across political institutions. If partisanship dominates voting behavior across boundaries, we will be confident of the generalizability of party voting theory. If the extent to which partisanship affects voting choice depends on the context, we can claim that party voting only exists in certain contexts and that, more importantly, political institutions

constrain political behavior (Asher and Richardson, 1977). Besides, previous single case studies can be revised by considering the influence of variables intervening between partisanship and voting choices.

It has been suggested that political institutions have considerable impact on individual’s voting behavior. For instance, the single-member single-ballot system that has only one seat and the one who wins the plurality takes it induces strategic voting, since voters choose the party that is most likely to get elected in order not to waste the vote (Duverger, 1954).3 This implies that the election rule dictates the way people choose among the candidates. A proportional representation (PR) system allocates seats based on the share of votes received by each party, thus voters can vote for their most preferred party. Due to strategic voting behavior, the level of party voting is supposed to be higher in the PR system than in the single-majority single- ballot system. In addition, the campaign pattern would vary from one institution to another institution (Myerson, 1993). The candidate-centric pattern, for example, prevails in multi-member districts, under the single-transferable voting (STV) systems and single-nontransferable voting (SNTV) systems, and open-list PR systems

(Samuels, 1999; Cox, 1997). In these systems, candidates cannot rely on party labels to get elected; party coordination will be a lot harder in multi-member districts than in single-member districts, and so is voting along party lines.

Party voting also varies with different types of executive-legislative relationships. It is argued that separation of powers as the American government induces split-ticket voting, meaning that voters support different parties for different

3 There two logics of strategic voting. The first one is the “wasted voting” as stated in the text, which refers to the maximization of utility by voting, and the second one is to prevent the least favorite party from getting elected, which was discussed by Downs (1957).

offices (Schattschneider, 1942; Fiorina, 1992). Under a presidential system, presidential candidates have the individualistic incentive and put their victory before everything (Samuels, 2002). Parties in America became campaign organizations of the presidential candidates because presidential elections are individualized affairs that fashion not party leadership but national leadership (Epstein, 1986). Political parties do not select the presidential candidate; instead, it is the mass public that selects him through primary elections held by state governments. On the other hand, fused power, like the British parliamentary system, favors strong party voting because parties undertake collective responsibility for policy-making. Party organizations are built with strong party discipline, particularly for the nomination of candidates. The strong tie between the cabinet and parliament puts every parliamentary member in the same trench, which leads to strong partisanship among the electorate.

The level of party voting refers to the propensity that partisanship determines election choices in the national election. The null hypothesis is that voters under different institutional arrangements would have same level of propensity to vote along with their partisanship. To test this proposition, I will examine individual-level data from the United States, Taiwan, and Japan and explain the variation in voting choices while controlling for political institutions. The implications of my findings will be directed toward the validity of party voting theory.

1.2 The Concept of Partisanship and Revisionist Views

The conceptualization of partisanship as a stable predisposition toward parties can be found in the study of American elections (Campbell, Converse, Miller, and

Stokes, 1960; Converse and Markus, 1979). Partisanship is acquired from socialization, and the partisan stability of voting behavior tends to accumulate over the adult years (Converse, 1969; Jennings and Markus, 1984; Niemi and Jenning,

1991). The party identification question measures how strongly people think of themselves as partisans or independents. The original question is as follows:

Generally speaking, do you think of yourself as a Republican, a Democrat, and

Independent, or what?

(IF REPUBLICAN OR DEMOCRAT) Would you call yourself a strong (Republican,

Democrat) or a not very strong (Republican, Democrat)?

(IF INDEPENDENT) Do you think of yourself as closer to the Republican or

Democratic Party?

Based on this set of questions, a seven-point classification (Strong Democrat,

Weak Democrat, Independent Democrat, Independent, Independent Republican,

Weak Republican, Strong Republican) was developed. The intensity of partisanship affects the level of political participation and stability of vote preference (Campbell,

Converse, Miller, and Stokes, 1960). Its direct and indirect influence on political participation, especially voting choices, is remarkable when considering other short- term factors (Markus and Converse, 1979).

Partisanship also affects the way people conceive of political issues and candidates (Belknap and Campbell, 1952; Campbell, Converse, Miller, and Stokes,

1960). The agreement between party preference and issue positions increases with the level of partisan strength. Key (1961) and RePass (1971) confirmed that people did

respond to issues, especially salient issues, and that party identification results in different issue perceptions. Candidate evaluation is also influenced by more durable party identifications. Rabinowitz, Prothro, and Jacoby (1982) accounted for the level of positive candidate evaluation with a battery of salient issues as well as party identification, and MacDonald, Prothro, Rabinowitz, and Brown (1988) stressed the role of ideology in making candidate evaluations.

Why is partisanship not only a voting guide but also a long-term, affective attachment toward politics in general? First of all, according to social psychology, people’s identification with religious, class, and racial groups gives rise to partisanship (Campbell, Converse, Miller, and Stokes, 1966; Weisberg, 1998). Parties are durable objects in terms of what they stand for and what they align with. Only if external dramatic change happens would political parties shift their policy positions and trigger a subsequent swing of party identification among the electorate.

According to this view, the long-term predisposition toward political parties is resistant to short-term forces and separate from vote choice. Secondly, voters use partisanship as a short-cut to make political inferences (Feldman and Zuckerman,

1974; Miller, 1976; Popkin, 1991). A party label can serve as a low-cost voting guide because of its long-term existence. Thus, partisanship defines the scope of one’s political world and structures political behavior.

Scholars argued that there is a reciprocal relationship between party identification and issues, which challenges the exogeneity of partisanship (Jackson,

1975; Franklin and Jackson, 1983; Page and Jones, 1979; Niemi and Jennings, 1991;

Whiteley, 1988). To some degree, party preference appears to accord with policy preference. Niemi and Jennings’ findings show that offspring may defect from

parents’ partisanship due to the decisive issues, such as Vietnam War or racial relationship. Considering the characteristics of presidential nominees, Rapoport (1991) suggested the long-term importance of candidate for partisan change. After reviewing the mounting revisions of the original Michigan model, Weisberg and Niemi (1993) asserted that it is difficult to isolate the relative influence among partisanship, issues, and candidate evaluation on American voting choices.

It has been claimed that individuals use their retrospective evaluations of parties to update their party preference (Fiorina, 1981; Howell, 1981). Despite that partisanship still strongly dominates the process of voting decision, it is not exogenous to the outside environment. Rather, it is responsive to previous political experience and information in the enviornment. Therefore, the stability of partisanship and its relationship to short-term stimuli are still open to debate (Green and Palmquist,

1990; Allsop and Weisberg, 1988). Observing the fluctuation of party identification, some scholars noticed that partisanship has declined since the 1970s due to the turbulent nature of the times (Converse, 1976; Wattenberg, 1984). The number of independents has been increasing slightly, and the number of strong partisans has decreased gradually. Walter Dean Burnham (1970) Nie, Verba, and Petrocik (1976) pointed out increases in split-ticket voting, alienation toward major parties, and negative evaluation of parties. Wattenberg (1984) and Klingemann and Wattenberg

(1992) used party images to demonstrate the source of decline; voters pay more attention to candidate’s issue positions and political philosophy than parties.

With controversy involving the concept of partisanship, more and more cross- sectional research on vote choice casts doubts on the utility of partisanship in explaining voting behavior. LeBlanc and Merrin (1978), using simple regression

analysis and data four elections, revealed the significance of candidate evaluations on

Canadian voting behavior. Hartwig, Jenkins, and Temchin (1980) looked at the six elections from 1956 to 1976 and concluded that the influence of partisanship is declining in the U.S. Whiteley (1988) confirmed the significance of partisanship, though he also found that partisanship is responsive to other short-term forces. Bartels

(2000) claimed that partisanship remains the most critical factor to vote choice between the 1960s and the 1990s in the U.S., yet he did not included short-term factors in his model.

From the standpoint of the comparative approach, the declining influence of partisanship seems inevitable because of the candidate-centered pattern that brings short-term forces to the elections. With the power of the presidency growing and party organizations losing their hold on the candidates, partisanship seems more responsive to issue positions and candidate evaluations rather than determining one’s preference over issues or candidates. It is expected that the influence of partisanship on vote choice through the short-term forces is also being downsized. The next section will examine aggregate-level party voting in the U.S.

1.3 The Assessment of Party Voting

Observing the trend of party voting provides us with insights into the strength of the predisposition approach across time (Beck, 1986). Normal vote theory assumes that party identification is a constant psychological attachment and that the electoral outcome results from the interaction of party identification and short-term forces

(Converse, 1966).4 Short-term forces generated by the current political situations may increase the turnout level of each group of identifiers, but in the long run the distribution of partisanship decides the electoral outcomes. The theory of normal vote is also employed to explain the consequence of electoral outcomes (Key, 1955;

Campbell, Converse, Miller, and Stokes, 1960; Flanigan and Zingale, 1974).

Similarly, the analysis of split-voting demonstrates the pattern of normal vote

(Burnham, 1965). Beck (1979) identified the “alignment-dealignment-realignment” election periods as a cycle that repeats itself every thirty years. During the dealignment period, party labels are much less useful as guides to electoral behavior than in other periods. Bartels’ (2000) analysis of partisan voting shows that the effect of partisanship on party choice sharply declined during the 1960s, but it has regained its momentum ever since.

The literature on party nationalization, which indicates voters respond to national interest instead of sectionalism, also helps account for the aggregate-level of variation in party voting. According to Key (1942) and Schattschneider’s (1960) findings, geographical interests used to dominate American politics until the extension of suffrage and government functions. The influence of constituency forces on turnout and partisan voting has declined over the last century due to the change of ballot form and the structure of mass communication. However, the national influence on partisan voting surged and declined sharply during the 1920s (Stokes, 1965; Stokes,

1967). Viewing nationalization as “the convergence of levels of partisan support,”

Claggett, Flanigan, and Zingale (1984) looked at the variation of election results in

4 Normal vote theory, albeit derived from the classical Michigan model, generalizes individual voting choices to national politics. The concern of this dissertation is not merely how voters make decisions, but how different political structures shape party system in terms of the ability to reach voters.

each county between the 1840s and the 1960s, and asserted that national party voting did not increase significantly over the course of one century. Kawato (1987) borrowed the concept of party swing from the British study, maintaining that nationalization has never happened.

Aggregate-level analysis gives us an overview of the regularity of party voting in terms of the electoral results, but it presents mixed evidence about the existence of party nationalization. Nevertheless, both individual-level and aggregate-level analyses suggest the decline of partisanship, especially its influence on voting. This result should be interpreted as the constraint of context that imposes candidate-centered incentives on American parties.

1.4 Summary and Remarks

Based on the research that analyzes American voting behavior, party voting theory has a strong theoretical ground. On the one hand, it conceptualizes the process of the voting decision. On the other hand, it accounts for the aggregate trend of electoral outcomes by isolating long-term and short-term forces. Despite the reciprocal relationship between partisanship and short-term factors and the resulting declining influence of partisanship on vote choice, partisanship remains essential to

American voting choices and its party system.

The broad influence of partisanship stems from the sense of group identity and low-cost information to voters. Partisanship, once acquired from socialization, affects perceptions of the political world and affects vote decisions. However, the decline of partisanship and of its effect on American voting choices calls attention to the importance of context. With the salience of parties diminishing and the link between

candidates and parties weakening, voters’ choices now seem to depend more on candidate evaluations instead of cues provided by parties. It is necessary, therefore, to look at political institutions and see how they affect political behavior.

A cross-national research design allows us to expand party voting theory with a comparative study of political behavior. In doing so, we can compare the extent to which partisanship determines voting decisions across countries where political arrangements are different than in the United States. A comparative framework for a systematic investigation of party voting will be presented in the next chapter.

CHAPTER 2

A COMPARATIVE PERSPECTIVE ON PARTY VOTING

Party voting theory states that the intergenerational stability of partisanship is evident due to the socializing influence of family and primary groups on partisanship. From the perspective of comparative political analysis, political parties are an agency between society and political institutions (Maor, 1997). As groups of people who seek power, political parties shape individual preferences in campaigns. Therefore, the extent to which individuals employ partisanship as the voting guide is conditional on the way election systems and government structures shape political parties. The origins and development of political parties will be discussed in Section 2.1. Section 2.2 reviews the literature dealing with political behavior outside the United States, especially European countries. After discussing previous cross-national research, a new comparative framework will be presented in Section 2.3. This study will use four cases for comparison, and the selection of cases will be explained in Section 2.4.

2.1 Origins and Development of Political Parties

Before the 1850s, the two parties of America were the only ones in the world. Not until 1867 had Great Britain granted a wide enough electorate franchise for the modern party to be formed. After the 1900s, universal suffrage was widespread in Europe. Responding to the political participation due to universal suffrage, political parties became fully developed within one century

(Duverger, 1954).

From the beginning of party formation, political parties have played a great role in governance (Key, 1942). In countries such as England and Japan, political parties that were built by notables or local organizations emerged and institutionalized constitutional government. In addition to governance, political parties take on interest integration. Eldersveld (1964:6) viewed parties as ‘a structural system seeking to translate or convert social and economic interests into political power directly.” He also argued that the party is a “social group” that seeks to control governmental positions through elections. Schattschneider (1960) also noticed the role of political parties in representing social cleavages, and stressed that political parties are different form social groups in interest integration.

The essence of parties can be summarized as three aspects: party-in- government, party-as-organization, and party-in-electorate (Key, 1964). The three parts of parties is aimed to aggregate diverse social interests and to provide policy outcomes. The main goal of political parties is to win political power and to implement policies (Downs, 1957). Sartori (1976) even claims that the electoral process distinguishes parties from interest groups. Through building coalitions in the parliament, political parties also organize government and exert pressure on

policy-making (Cox and McCubbins, 1993; Rohde, 1991). Aldrich (1995) argued that political parties structure the choice of the electorate and makes governance in an effective and integrated way.

Moreover, political parties have their own interests rather than being subject to the environment, including politicians or outside groups. The experience of European party organizations illustrates the tendency of hierarchy; party leaders of socialist parties tend to secure their power even if it is against the principle of social democracy (Michels, 1962). The Christian parties also pursued more power once churches formed them to protect their status (Kalyvas, 1991). In

America, on the other hand, political parties became public facilities when the state governments held primaries that decide party nominees (Epstein, 1967).

With the development of party systems and expansion of government functions, a political party is less a group of office-seeking politicians than a political institution. The protection of political parties can be seen in the primary elections, party organizations, and even the form of ballots.5

From the standpoint of rational choice theory, parties are merely concerned with office and vote-seeking rather than policy goals (Downs, 1957).

Ambitious politicians create parties to address the problems of collective action and social choice (Aldrich, 1995). With electoral politics developing, however, political parties need to cope with coordination problems in the legislative and electoral sides. Cox and McCubbins (1993) addressed the coordination problem in

Congress and argue that political parties are more important than standing

5 In some states of the U.S., only political parties can put the name of nominees on the ballot. For some nations, such as Great Britain, the party label is not on the ballot, but the influence of political parties exists because local people get involved in the nomination process.

committees in the policy-making process. Panebianco (1988) stressed that party organizations diffuse “collective incentives” among the electorate to build

“collective identity”, which is key to their survival. Aldrich (1995) maintained that American parties once effectively coordinate candidates because of their strong party machines but mass media and primaries have made party organizations less important since the 1960s.

Both perspectives of institutions and rational choice theory suggest competition between parties and their party members, especially candidates.

According to the “evolution theory of parties,’ on the institution side, political parties evolve to electoralist parties no matter whether they are organized as either mass-based or elite-based (Panebianco, 1988; Gunther and Diamond, 2001). They are focused on the conduct of the campaign, and substantially loosen their ties to specific clienteles in pursuit of votes and tend to emphasize the personal attributes of candidates. In this case, nominees are determined by the electoral resources of the candidates rather than by service to the party. The Democratic Party of the

United States, the Labor Party under Tony Blair, and Taiwan’s KMT, which completed its long-term transformation from a Leninist to a electoralist party, are among others mass-based electoralist parties (Gunther and Diamond, 2001).6

On the rational choice side, political parties are not always so organized that they can withstand the forces against institutionalization. Cox and

Rosenbluth’s (1993) study of the Japanese factions shows that the mainstream factions dominate the endorsement processes to win party nomination. Richardson

6 Here the notion of Lennist party refers to the type of political party that uses party organizations to control the government and every aspect of society, including labor unions, professional associations, religious groups, schools, and military.

and Flanagan (1984) and Richardson (1997) also suggested that Japanese parties are fragile coalitions of party factions, and not able to articulate social interests.

The problem of party organizations can be found in American parties as well. The ideal of “party government” is conditional on unified government at best (Rohde,

1991). The primary election opens the nomination process to the public, and the incumbency advantage further weakens the influence of partisanship in the electorate. It is also found that American party organizations have relied on incumbents in congressional elections since the dealignment period in the 1960s

(Cox and Katz, 1996).

Only if political parties diffuse interests outside party organizations and transcend party identification through generations, can political parties coordinate candidates by solving the collective action problem. 7 Party attachment, in this view, is considered as the instrument that parties rely on to tame candidates, and the voting act should be analyzed as the outcome of a battle between party identification and candidate evaluations. This position challenges the long-term, exogenous, stable status of party identification held by party voting theory.

Instead, candidate-based voting should be more prevailing in contexts that demand weak party organization. Reviewing the application of party voting theory to the countries outside the United States will help us understand why the level of party voting varies with context.

7 Aldrich (1995) argues that party affiliation helps voters reduce the cost of decision-making when they choose candidates. Candidates, as matter of fact, benefit from party identification because they are not able to provide partisan images. Since parties are assumed to be self-interested, it is safe to maintain that parties provide cues for their own interest, which is power and access to resources.

2.2 Review of Comparative Voting Literature

The cross-national study of political behavior and attitude has generated systematic and comparable evidence about how people have different attitudes due to context. Their findings concentrate on European countries, but extend to other political systems for their comparability. In addition to Britain that has a simple-majority-single-district election system, scholars studied partisan attitudes and association with voting choice in France, Italy, Germany, Japan and so forth.

The cross-national study of vote choice poses some challenges to the generalizability of party voting theory.

Almond and Verba (1963) collected data of political culture across five countries--Germany, Italy, the United States, Britain, and Mexico--and investigated the association between political participation and democracy. Their findings show that American voters have higher involvement in party activities than others, and citizens of Britain, Mexico, and Germany have more or less religion-oriented partisanship. Although their study is focused on political participation, it provides an insightful framework of comparative political behavior.

Bowler, Lanoue, and Savoie (1994) looked at party systems in three countries—Australia, Canada and Germany--and stressed that electoral systems not only affect voting behavior but also partisan strength. Personal factors, such as socio-demographical characteristics, are not related to the variation in partisan strength because they cannot polarize voters. Instead, party competition and the presence of extreme parties, which are shaped by electoral systems due to strategic voting, are expected to stimulate partisanship in that people may feel the

threat of extreme parties and acquire stronger identification with a given party.

Their research reveals the impact of political systems on partisanship through party competition.

Converse and Dupeux’s (1962) comparison of French and American voters is one of the earliest, and perhaps the finest, cross-national studies of political behavior. Their major finding is that only a small portion of French voters have party attachment and hold an issue concern. The lack of party attachment may be due to the multi-party system which is caused by the discontent of the mass public. The turmoil in politics inhibits the development of party attachment that is expected to take place in the socialization process.

Converse and Dupeux concluded that the political circumstances instead of any particular institution leads to different extents of party attachment in these two countries.

Inglehart and Hochstein (1972) drew on Converse and Dupeux’s findings and attempted to examine whether major political conflict would produce cross- pressure for citizens so people tend to abstain or that major conflict would facilitate party identification as a result of polarization of issues. They found that the 1968 crisis in France led to a single coherent attitude structure and, as a result, party alignment. However, Americans except blacks showed low rates of polarized attitudes and party preference after the Vietnam War, which started the period of party dealignment. These findings demonstrate that sudden dramatic events can either reinforce party alignment or cause party dealignment.

A comparison of American and English voters highlights the distinction between presidential and parliamentary systems. Political parties in these two

countries have long traditions of party competition; no abrupt change has struck their two-party systems. The similarity of these two countries includes their single-member district system. Butler and Stokes (1969) investigated party alignment, partisan attitude, and change of party support in Britain. The major concern in their study is to distinguish general partisan attitude, class, and vote choice. Butler and Stokes argued that people’s partisan and class attitudes come from family socialization, particularly from the father. With weakening socializing influence, the strong relation of class to party has declined among the young generation. Therefore, vote choice is in line with party identification. The decline in partisanship among the new cohort led to the partisan change that occurred in the 1960s; the conservative voters converted to or were replaced by the Labour Party. Butler and Stokes’ findings provide a different picture of partisanship from the American counterpart: the dealignment of partisanship happens in a shorter period of time, which is the case probably because the British voters did not develop partisanship separate from vote choice.8

Cain and Ferejohn (1981), comparing the stability of partisanship in the

United States and Britain, maintained that partisanship remains an independent and stable political attitude as long as the stability of minor parties and independents is also analyzed. By including them in the sample, the consistency between stable party preference and stable vote preference will decline, because they are likely to have less stable turnout and vote preference.9

8 Regarding the discussion on the relationship of age and partisanship, especially the test of Converse’s experience-based model, see Niemi, Powell, Stanley, and Evans (1985) and Cassel (1999). 9 Minor party supporters also tend to vote strategically. They may vote for the party that is most likely to win rather than their favorite party.

Canada is another Anglo-American country using plurality single-member districts. Unlike the United States and Britain, Canadian voters have shown low intensity of party identification since the 1960s, and declining party attachment leads to variability of vote preference (LeDuc, 1985). The effect of partisanship on vote choice varies with the level of political interest that is the function of responsiveness to short-term stimuli (LeDuc, Clarke, Jenson, Pammett, 1980).

Clarke and Stewart (1987) suggested that a loose federal system like Canada tends to create multiple party identifications at different levels of government due to separate national and local party systems. Moreover, the widespread regional conflict in the 1970s reinforced the inconsistent party attachment across the federal and provincial levels. Although their evidence is convincing, their findings may not extend to other countries using federalism. According to Eldersveld’s

(1973) study, for instance, Indian voters are very committed to their favorite parties at different levels.10

Germany is another country that attracts attention of scholars. Due to the collapse of the Weimar Republic and the Nazi era, Germany seemed to have no indication of the existence of party identification in the 1970s (Kaase, 1976). The volatility of party identification was so high that the concept of party identification in West Germany was not meaningful in the first place. Less than 30% of the respondents named a tie to a given party (Kaase, 1976). However, Norpoth (1978) pointed out a methodological problem when European analysts applied the notion of partisanship to the European countries, and he found that partisan attitude does influence the probability of turnout and the direction of voting decision. In 1972,

10 The level of decentralization in the federal government may contribute to the difference in the

for instance, Social Democrats and Christian Democrats voted along party lines.

As the level of partisan strength weakens, the congruence also diminishes.

Norpoth’s finding regarding the utility of partisanship is convincing, but it does not answer the question of whether German voters have separate vote preference and partisan attachment. Moreover, social cleavage is not considered in his research, which leaves us wondering if partisanship is independent from social class and regional consideration.

Echoing Almond and Verba’s (1963) concern over West Germany’s apathetic political culture, Klingemann and Wattenberg (1992) compared the strength of partisanship in West Germany and the United States from the 1960s to the 1980s. They found that West Germany’s parties link themselves to the issues of government management and political leaders but American party system features no connection between party image and issues. Both countries were moving toward party dealignment and candidate-centered politics, but German citizens appear to have more party images, in part because parties in Germany emphasize party leaders and issues than their American counterparts.

In their study of partisanship attitude in East Germany, Kaase and

Klingemann (1994) found that most East Germans adopted party ties through a long-term learning process, which the predisposition approach suggests. The correlation between party preference and voting intention is high, and the level of party identification during a two-year period is stable. They also found minimal similarity between the party images attached to political parties in West Germany and East Germany; East Germans have more positive views of political parties.

level of disagreement of partisanship, which is worth of further investigation.

Since social cleavages have played a great role in European party formation and individual case studies have found the significance of class and other social structures, social cleavages are expected to contribute to the stability of partisanship.11 Richardson (1991) found that parties founded along traditional social cleavages have voters with stronger ideology and closer issue distance to their preferred parties, which strengthens partisan ties in terms of vote preference.

Holmberg (1994) has compared partisanship of several European countries and the United States and found that voters in the European countries have more stable party identifications than in the United States. His analysis also lends support to the generation-replacement model regarding the decline of partisanship. All in all, his finding suggests that party voting theory applies to European party systems.12

In addition to partisanship, some scholars note the importance of party images measured by the open-ended like/dislike question. Feldman and

Zuckerman (1974) looked at partisan attitudes and vote choice in Britain, suggesting that partisan attitude consisting of evaluations of government and national leaders can have an independent effect on vote choice. Wattenberg (1982) found different patterns of party image in the United States, Britain, Australia, and

Canada. British voters have more polarized attitudes toward political parties, but

American voters seem to see their parties as irrelevant. Canada and Australia parties are moving toward multi-dimensional systems; parties are evaluated either positively or neutrally. The long-term decline of party images in the United States

11 Regarding the defense of class voting in Britain, see Wald (1982) and McAllister and Mughan (1987). 12 Considering the measurement error made by different panel data, Schickler and Green (1997) argued that the stability of partisanship in the United States is no less than that in Britain, Germany, and Canada.

results from declining socializing influence, but the party crisis in 1974 and long- term decline of international prestige contributes to polarization of parties in

Britain.

Two conclusions can be drawn from this review of previous research. First of all, the concept of partisanship is critical to the understanding of change of vote preference in the countries outside the United States. In spite of different political systems and party systems, the level of partisanship still accounts for variation in vote preference. In aggregate-level analysis, the outcome of elections and the performance of political parties are conditional on the distribution of party attachment. Although much research raises doubts about whether the notion of partisanship exists in Europe, party attachment is still vital to vote choice if we take account of the effect of national leaders and issues. Social class appears to affect vote preference through partisanship, which may be responsible for the polarization of the party system and the strong relationship of party attachment to vote preference (Richardson, 1991).

Secondly, it is found that more and more European voters show low party attachments, which introduces the influence of short-term factors to the voting decision process (Dalton, Flanagan, Beck, 1984). Although aggregate-level voting remains stable in Europe, European parties are vulnerable to short-term factors when rapid social change occurs in Europe.

The application of party voting theory to the European countries does not control for the effect of context; only a few studies mention the methodology of case selection (e.g. Almond and Verba, 1963). It remains unclear, in general, why cleavage-based voting transformed to party-oriented one in some countries but not

others. The roles of political leaders and party organizations are not given enough attention, considering the tradition of the mass public party organizations and social movements in Europe. Moreover, the development of political parties in each country is merely treated as a background or remote cause of the voting pattern. Only a few studies consider the configuration of the party system as an independent variable (e.g. Clarke and Stewart, 1987). To remedy the analysis deficiencies, we should not only address the influence of context, but also allow the context to explain the pattern of political behavior.

2.3 A Comparative Framework of Party Voting

Recall that a political party is defined as a rational unitary actor, and party voting is defined as the propensity that individual voters choose parties based on their partisanship. To explain why the level of party voting varies in different countries, I contrast the social structure model against the political structure model.

As for the political structure model, political arrangements set up the way that political parties act on voters. Cox’s (1987) analysis of party-based voting in

England suggests that party organizations are the mechanism between voters and parties. Strong party organizations were established in England because of the parliamentary system, in which individual candidates are only part of political parties. In the U.S., political parties were once strong in the 18th century, but presidential primaries as well as nationalized mass media since the 1960s shattered the importance of party organizations. Also, the presidential system by nature allows individual candidates to influence policy-making for the interest of their districts (Samuels, 2002).

The social structure model instead stresses the functions of political parties responding to social cleavages. Lipset and Rokkan (1967) contended that political parties “crystallize and make explicit the conflicting interests, …, and they force subjects and citizens to ally themselves across structural cleavage lines and to set up priorities among their commitments to established or prospective roles in the system.” The outcomes of struggles between state and church, center and periphery, landlord and labor, and owner and labor determine the structure of national politics.13 According to this perspective, the level of party voting depends on the level of group loyalties and how political parties transform the demand of social groups or politics (Wolinetz, 1990).

The major limitation of the social structure model is its explanation of the dynamic of party systems; it only provides an explanation of party formation based on social cleavages. The “freezing” effect of social cleavages on political parties explains why political parties originate from social groups, but it does not account for the transformation of political parties from closed communities to catch-all parties. Once they acquire legitimacy and institutionalize, political parties seek better access to resources through elections defined by political context (Mair, 1997). As to how political parties appeal to the electorate and transform to electoralist parties, the social structure model does not provide an adequate explanation.

With less measurement problems and an established classification of political systems, the political structure model can account for the strategy of

13 As Zielinski (2000) indicated, Lipset and Rokkan did not set aside the effect of election threshold on the entry of parties. Nevertheless, they mainly focused on how parties respond to social conflicts and integrate them. Therefore I view their theory as a social structure model.

parties and voting behavior in different party systems. Duverger (1954) noted the fundamental effect of the electoral system on the structure of party competition and summarized by the tendency: “the simple-majority single-ballot system favors the two-party system”(p.217), and that “the simple-majority system with the second ballot and proportional representation favors multi-partisan” (p.239).14

Scholars generalize Duverger’s law, maintaining that electoral rules for the most part dictate the number of effective parties (Rae, 1971; Lijphart, 1984; Taagepera and Shugart, 1989; Katz, 1997). It is also found that the sheer size of district magnitude, namely, the number of seats at stake, determines a voter’s strategy.

The voters can vote for the candidate who is most likely to get elected in small- sized districts, or they can equalize their votes in medium-sized districts (Cox,

1997). On the party side, the configuration of the political system provides various incentives for parties and candidates; electoral rules such as single-member districts encourage collective incentives and multi-member districts encourage individual incentives (Shvetsova, 1995; Katz, 1997; Samuels, 2002). The weakness of the political structure model is that the influence of political structure is not easy to observe in a short period of time, unless comparison of political systems is employed to show the pattern of party development. Moreover, social groups that parties align with can force parties to change their strategies in party coordination (Zielinski, 1999). In spite of these weaknesses, the political structure model is better than its counterpart in that the ultimate goal of political parties is

14 Riker (1976) argued that the existence of three-party system under simple-majority single-ballot system is possible but unstable; third parties can depart from the equilibrium of two-party system and the system returns to the equilibrium because third-party voting is disillusioned voting. In another article, Riker (1982) revised that formulation by considering if third parties are in the national competition and able to tip the balance between the two major parties.

political power and governance instead of representing social interests. The connection between state and society should go through political parties, and their relative positions between state and society depend on the characteristics of the state (Mair, 1997; Bartolini and Mair, 2001). Party voting, moreover, indicates the conformity of partisanship and voting choices. The way that political parties reach the electorate should therefore be examined under the constraints of political structures.

Electoral systems are singled out as the most critical determinant of the regularity of party systems (Duverger, 1954). The ballot structure, district magnitude, and electoral threshold are critical to the party system because these determine how the share of votes translates to the share of seats for each party in the contest. Government structure is also vital to the development of the party system because it dictates the access to resources and the degree to which political parties and candidates are tied together (Cox and Rosenbluth, 1995). Whether executive power is located in the legislative branch alone or diffused between the executive and legislative branches leads to several important distinctions among political systems. The most important one is whether the legislative branch and the cabinet take collective responsibility for policy making or whether the legislative and executive branches share the responsibility. For the former case, the discipline of political parties is stressed to ensure party cohesion. For the latter case, however, the president’s party cooperates with other parties to ensure the passage of bills.15

15 There are two assumptions for this argument. First of all, the presidential system tends to have a two-party system, which creates the incentive for individual candidates to deviate from their parties. Secondly, the term of the legislators in the presidential system is fixed so that the president

2.3.1 Election Systems

According to Lijphart (1984), the most important dimension of the election system is the electoral formula that classifies election systems as majority/plurality, semiproportional and proportional representation. The majority/plurality formulas apply to single-member single-plurality (SMSP) districts, where the candidate who receives the most votes gets elected.

Proportional representation formulas apply to PR systems, where seats are allocated according to political parties’ shares of votes. The semiproportional formula lies between majority/plurality and proportional representation in that it allows parties with a small share of votes to exist in a big district, while using majority/plurality formulas to determine the allocation of seats. The majority/plurality formulas favor parties with bigger shares of votes; the proportion of seat share to vote share increases convexly with the vote share. The semi-proportional system also favors major parties, even though it allows the same number of parties as the number of seats in the district to compete for seats

(Cox, 1996). Proportional representation formulas favor the small parties because vote share translates to seat share proportionately above a relatively low threshold, which leads to the multi-party system (Rae, 1971; Lijphart, 1984).

The single non-transferable vote (SNTV) system is one of the semiproportional formulas, using plurality rule and multimember districts. Each voter has only one vote, and each candidate’s votes are added up to determine the election result. In SNTV, multi-member districts encourage parties to support

cannot dissolve the legislature; the president has to work with parties other than his party in the

more than one candidate, but the first-past-the-post rule leads to intra-party competition (Cox and Rosenbluth, 1995). Individual candidates receive incentives from SNTV to conduct candidate-centric campaigns because of the single- plurality rule (Shvetsova, 1995). In the real political world, the SNTV system is arguably responsible for candidate-centered campaigns and factional politics in

Japan during the 1955 party system (Lin, 1996). To coordinate candidates and to reduce nomination error, the Liberal Democratic Party (LDP) divided particularistic resources among districts (McCubbins and Rosenbluth, 1995; Cox,

1997; Grofman, 1999). The consequence of the SNTV system is that Japanese voters refer to candidates more than political party in structuring their voting choices (Rochon, 1981).

For the PR system, the preferences over candidates and parties are non- separable, unless parties can change the ranking of candidates on the party list after the election. Essentially, voters will not mention candidates when they choose parties (Katz, 1997). For the SMSP system, however, the preferences over candidates and parties are not necessarily unified; voters can choose a candidate simply because of his or her personal character or because of his or her party label.

When voters have no information about candidates or too little information to pick one of them, they tend to vote in line with their party preference (Downs, 1957).

As for the SNTV system, the multi-member district produces the incentive of entry; political parties can secure seats with a small share of votes. The simple- majority rule, however, encourages candidates to pursue individualistic campaigns that involve intra-party competition. Each candidate can easily deviate from the

legislative branch in order to build a winning coalition.

equilibrium policy position to pursue vote gains, which makes parties into loose voting coalitions (Shvetsova, 1995). Therefore, SNTV lies in a intermediate position between the PR system and the SMSP system in terms of the individualistic incentive.

Both SMSP and SNTV provide individualistic incentives for candidates, but the fundamental difference between SMSP and SNTV lies in the district magnitude. Because of the multi-member district, political parties tend to nominate more than one candidate in a typical district (Reed and Bolland, 1999).

SNTV allows voters to choose only one candidate among their partisans, so that candidates are encouraged to pursue a personal vote outside party organizations to win a plurality. In this regard, particularistic interests outweigh party labels in structuring voting choices. In the SMSP system, each political party nominates only one candidate in each district. Despite candidates having their own personal agenda outside party labels, there is no intra-party competition that leads to candidate-centered campaigns (Cox and Rosenbluth, 1995; Grofman, 1999).16

With the same simple-majority rule, it is expected that the level of party voting is lower under SNTV than under SMSP. To highlight the importance of the district magnitude, the SMSP and the SNTV systems are chosen in this dissertation as the comparison cases.

16 In some SMSP systems, political parties use primary elections to decide their nominees, which may increase the intensity of intra-party competition. In some other countries, however, the

2.3.2 Government Structure

Presidential and parliamentary systems are two typical arrangements of governments in democratic countries (Lijphart, 1984). The essential difference between the two systems relevant to our concern is the relationship between the executive and legislative branches. In the presidential system, the separation of the president and the legislative body makes policy making entail cooperation between the two branches. The executive must cooperate with the legislative body to get his policies enacted, which lets individual lawmakers manipulate their influence on the policy making process. Instead of rallying behind political parties, lawmakers can opt for cooperation or for fighting for their best interests (Arnold,

1979, Fiorina, 1989). The president must build policy coalitions with his use of patronage to trade for the lawmaker’s cooperation (Mainwaring, 1997). Moreover, the fixed term of both bodies encourages individualistic incentives because confrontation between the two bodies would not lead to dissolution of the legislature (Cox and Rosenbluth, 1995). Without a collective responsibility to maintain the majority in the legislative body, law-makers are inclined to seek personal vote (Cain, Ferejohn, and Fiorina, 1987). The difference in the executive- legislative relationship is expected to influence the relative importance of partisanship and candidate evaluations. In presidential systems, individual incentives prevail and party discipline is hard to maintain, so that party-based voting is relatively smaller than in parliamentary systems.

By contrast, the parliament and the cabinet are not separable in parliamentary governments. The number of seats in the parliament determines

nomination is not entirely opened to the public. Therefore, the SMSP system should have lower

which party controls the executive power. The cabinet can be overthrown if the majority party or coalition loses a confidence resolution in the parliament.

Therefore, political parties stress party discipline to maintain party unity and to prepare for reelection in case they lose a confidence resolution. The concept of

“party government,” which indicates that parties take collective responsibility for the operation of government, is in some degree an ideal type of the parliamentary system (Rose, 1974).

From the point of view of the electoral side, parliamentary systems encourage parties to organize voters for a sudden election due to losing a confidence resolution. Political parties are motivated to pursue party-based voting in order to win the majority of seats in the parliament. In the presidential system, however, political parties are less interested in legislative elections than in the presidential systems. Candidate-oriented campaigns in presidential elections would not bring as much party discipline as do elections in parliamentary systems.

Rather, political parties become campaign organizations for the presidential candidates. Relatedly, the legislative candidates expect to win reelection by the personal vote, which depends on how well the incumbents work with the administration instead of party loyalty.

The working of the two ideal types of government also hinges on some additional factors. For instance, the federal system stresses local interests, which weaken the influence of parties and increase the importance of the legislators as representatives of local interests. The hegemonic party system and faction politics in Japan encourage candidate-centered campaigns, though parties maintain strong

level of intra-party competition than the SNTV system.

discipline in the Diet. In this dissertation, however, only electoral systems and government structure are considered because they contribute to party behavior

(Samuels, 2002).

2.3.3 Typology of Election and Political Systems

Table 2.1 shows a two-by-two matrix in which the four cells represent the predictions made by the two independent variables. The highest level of party voting should be derived from the interaction of SMSP and the parliamentary system, and the weakest one would be the product of SNTV and the presidential system. SMSP with a presidential system, and SNTV with a parliamentary system tend to have party voting levels that lie between the two extremes.

SMSP SNTV

Presidential System Intermediate Lowest

Parliamentary System Highest Intermediate

Table 2.1: Party voting, election system and government structure

A quasi-experimental research design allows us to control for the effect of political institutions. The null hypothesis is that there is no variation in the level of party voting in the four types of political systems. To test this hypothesis, I will construct four separate models to see if the influence of partisanship varies with

different cases. Then, I will pool all the individual data together in a cross- sectional model to compare the effects of partisanship and candidate evaluations on voting choices while controlling for the system-level variables.

I pick four regimes as my cases: the United States, Taiwan, Japan before 1994, and Japan since 1994. The United States is an important country using the presidential and the SMSP system, which is expected to have moderate party voting in congressional elections. As regard to SNTV, there are three countries that have used SNTV at various times: Japan, Taiwan, and Korea. Korea abandoned the multi-member district system after 1988,17 therefore it will not be analyzed in this project. Japan is another important country that used SNTV for its parliamentary election. Since 1994, however, the SMSP and party-list SNTV system have been used in general elections in Japan. Thus, by choosing Japan, we can observe the change in party voting before and after the transformation of election rules. Although it is based on the typology set up by the two institution variables, the selection of these four cases is non-random. The non-random selection inevitably introduces non-measurement errors to this study, thus conclusions regarding party voting theory should be viewed with care.

2.4 Context of Four Cases: the United States, Taiwan, and Japan

Considering the two dimensions built up in the last section, we fit the

United States, Japan, and Taiwan in the typology of party voting. The United

States is the typical country using a presidential system with SMSP rule for both

presidential and congressional elections. The winner-take-all rule reinforces the two-party system. Taiwan also has had a presidential system with SNTV system since the 1969 legislative election. Although Taiwan has a Prime Minister as the head of the Executive Yuan, he is not necessarily the majority leader in the

Legislative Yuan and his/her power is actually authorized by the president.18 The first popular presidential election in 1996 consolidated the power of the presidency.

Japan is a country with a parliamentary system using both SNTV and mixed SMSP-PR in legislative elections. SNTV was used in the lower house elections until 1993. In the election reform of 1994, Japan adopted mixed SMSP-

PR system. Every voter casts two ballots; one is for candidates in single-member districts and the other one for party lists. Candidates can run in single-member districts and put their names on party lists, but candidates who win in single- member districts leave party lists and let candidates behind them move up. Due to the fact that Japanese voters can vote in both districts and single-member districts are more emphasized, I view Japan after 1994 uses single-member districts.

Arranging the four regimes in the two-by-two matrix gives us more of a sense of the typology (see Table 2.2). Taiwan is expected to be the lowest level of party voting because of its presidential system and SNTV. On the other hand,

17 Korea used single-member-district for the National Assembly elections between 1948 and 1972, and two-member-district system between 1973 and 1988. In 1988, single-member-district plurality system was reintroduced and employed in the following elections (Lee, 1999). 18 The Constitution has been changed several times since 1947. According to the 1947 Constitution, the Legislative Yuan is the highest executive organ and the head of the Executive Yuan--prime minister—is nominated by the president and approved by the majority of the Legislative Yuan. The latest version of the Constitution that was announced in 2000, states that the president appoints the head of the Executive Yuan without the approval of the Legislative Yuan. In reality, the then ruling party—KMT--used party machine to control the administration; the president was

Japan since 1994 should have the highest level of party voting because of both its parliamentary and SMSP systems. The United States and Japan before 1994 are expected to be located between the two extremes. Given that variation across the four cases, it is necessarily to consider them in more detail.

SNTV SMSP

Presidential System Taiwan the United States

Parliamentary System Japan before 1994 Japan after1994

Table 2.2: Party voting of four cases

2.4.1 SMSP and Presidential System: the United States

American parties have been considered as federative, candidate-centered, and government-based. The national organizations are not as strong as their

European counterparts, although the link between grass-root and national organizations has been strengthened in the past few years. In the direct primary open to the public, political parties lose their control over candidates. Party platforms are even beyond the control of local and national party organizations, particularly in presidential nominations. The delegates to the national convention are selected through primary elections or open caucus meetings. Since the 1960s, party reform movements pressed party politicians, businessmen, and labor unions

the chairman of the party who moderated every policy in the meeting of the executive committee. Before the first popular presidential election in 1996, therefore, Taiwan had a presidential system.

not to intervene with bureaucrats to insure that local governments can receive federal grant to improve urban planning (Shefter, 1983). After independent agencies were established, parties lost a great deal of material incentives to mobilize voters. Political parties turned to interest groups for support, but broad appeals to interest groups weaken party unity (Lowi, 1979).

The caucus in Congress was the origin of American parties, but it was the patronage system that brought eligible voters to the polls (Epstein, 1967; Shafer,

1983). Party machines strictly controlled nominations in the 1920s and 1930s, but the New Deal system, civil service, and primary election laws considerably destroyed the way political parties secured votes, which allows politicians at the state level to deviate from the ideology of party organizations (Erikson, Wright, and McIver, 1989). The decline of party organizations due to the change of nominating process and government functions more or less accounts for the increase in the incumbency advantage (Cox and Katz, 1996). Nevertheless, scholars contend that partisanship remains an important influence in congressional elections (Jacobson, 1994; Abramowitz, 1993; Brown and Woods, 1991). The common conclusion is that the strength of party varies over time, which has been interpreted in election cycle theory (Beck, 1979). The change in the level of party voting in America is still open to debate, and we will compare it with other party systems.

2.4.2 SNTV and Presidential System: Taiwan

As mentioned above, Taiwan has used the SNTV system in legislative elections, while the presidency has been the ultimate authority through the

authoritarian and democracy periods. Since the governments do not alternate between political parties, policy-making and winning election can reinforce each other (Pempel, 1990). When the one-party system ended, the allocation of resources was not as lop-sided as before. Therefore, the strength of parties should be weak after the transition.

Under the SNTV system, political parties usually support more than one candidate in a district, which leads to intra-party competition during campaigns.

The once long-term ruling party, the Kuo-Ming-Tang (KMT) had used SNTV system efficiently in large districts (Cox and Niou, 1994), because the KMT developed party organizations and factions that were able to allocate resources and campaign geographically equally. In the meantime, however, the KMT was challenged by the emerging middle-class after the economic take-off and diplomatic crisis of the 1970s. A quasi-party organization had been operating under the label of “outside KMT” (tangwei), and it challenged the KMT’s legitimacy and authority. The alliance of the opposition politicians eventually became the Democratic Progressive Party (DPP) in 1987, and has taken about 30 percent of votes in every legislative election since then.

During the period of the KMT’s dominance, it controlled the nomination with strong party organizations and policy-making power. Its central headquarter selected party nominees who tend to cooperate party coordination during campaigns. As the one-party system transformed to multi-party system, candidates began to bargain with parties over endorsements. The level of party voting is expected to decline after the end of one-party monopoly. Besides, a presidential system that by nature encourages the weak link between political parties and

legislators is expected to weaken the influence of partisanship. Therefore, Taiwan politics in the 1990s should serve as strong evidence regarding the low level of party voting.

2.4.3 SNTV and Parliamentary System: Japan before 1994

Defeat in war and the American Occupation changed the Japanese political system. According to the Peace Constitution, the Diet was the supreme authority.

The conservatives were backed by the United States to create a party cartel in the

Diet. The conservative Liberal Democratic Party (LDP) founded in 1948 emphasized economic growth and conservative political reform. Economic development has won the voters’ mandate for the LDP since the 1960s. The leftist parties, such as Japan Socialist Party (JSP), the Japan Communist Party (JCP), and the Democratic Socialist Party (DSP), tried to reverse the dominant “GNP-ism”.

Although they could not stop the LDP from prevailing in Diet elections, the LDP gradually lost their support in urban areas because of a series of party scandals.

The SNTV system used before 1994 was of great importance to the

Japanese political parties. Because parties usually nominate more than one candidate, the politicians must coordinate themselves in the endorsement process, which leads to the fierce competition among party nominees from different factions (Tomita, Nakamura, and Hrebenar, 1986). Support by mainstream factions is critical to the fate of candidates looking for party endorsement (Cox and Rosenbluth, 1998). Moreover, party nominees must develop their own local campaign organizations (koenkai) to survive intra-party competition. Politicians, once they got elected, would support their faction leaders in the election of the

party president and even the prime minister in exchange for campaign finance.

They also rewarded the constituency by bringing projects or grants back to their districts, which boosted their electoral support.

The SNTV system combined with the parliamentary system contributed to a mixed style of policy-making. On the one hand, intra-party competition gave candidates an incentive to look out for particularistic interests. On the other hand, the parliamentary system requires political parties to undertake collective responsibility. Every policy proposal went to the LDP’s Policy Affairs Research

Council (PARC) for discussion and approval, and party members expressed their opinion there only. After the bill went to the Diet and its standing committees, policy debate was rarely seen. During the policy-making process, bureaucracy and factions aligning with social groups such as agriculture and transportation try to influence the final decision (Ricardson, 1984). The initiatives of policy-making and the final decision among alternatives are decentralized among legislative factions and bureaucracy, which opens the door to allocation of resource in local districts.

Due to the constraint of elections rules and policy-making process, therefore, partisanship is not the best predictor of voting choices in this system.

Social networks and candidate images are more salient factors in structuring voting choices. The level of party voting should be higher than in Taiwan but lower than in Japan after the electoral reform.

2.4.4 SMSP and Parliamentary System: Japan after 1994

In October 1996 Japan conducted the first House of Representatives (HR) election based on a new parallel system containing both single-member districts and proportional representation seats. Three hundred representatives were elected from SMSP districts, and two hundred PR seats were elected in eleven regional

“blocks”. Each regional “block” is composed of several small districts. Every voter has two ballots, one on which to write the name of a candidate in the single- member district and the other for the name of a party in the regional PR district.

Under the new system, candidates can run in both types of districts. Parties can place candidates among the four types of candidacies: pure SMSP, dual candidates with individual rankings, dual candidates ranked equally, and pure PR (McKean and Scheiner, 2000). The change in election system followed the LDP’s fall from power in 1993. The coalition government led by the Japan New Party (JNP) completed the electoral reform that was expected to terminate corrupt politics and unfair over-representation of rural districts.

Nevertheless, each party took advantage of the new system. They created

“dual candidacies” for politicians, nominating them in both SMSP and PR districts. In the 1996 general election, 59 out of 200 PR seats went to the losers in

SMSP districts (McKean and Scheiner, 2000).19

In spite of this manipulation, the new system encourages parties to stress party labels. The distinction between a candidate’s campaign organization and

19 McKean and Scheiner (2000) argue that only careful voters would understand the consequence of double candidacy. In this case, voters would split their votes for two districts. They found that most voters did not do split-voting, since most of the 59 double candidates who won the PR seats came in second or third in their SMSP districts. It indicates that institutional change results in party voting.

party branch becomes blurred, because each party can endorse only one candidate

(Gallagher, 1998; Curtis, 1999). The pattern of “personal voting” should decline with the replacement of multiple-member district. In the long run, both the parliamentary and the new mixed PR-SMSP districts should produce a more party-oriented Japanese politics.

2.5 Determinants of Party Voting

It is necessary to conduct individual case studies for better understanding of the voting patterns in the four systems that fit in the typology. Individual-level election data will be used to examine the effects of partisanship, candidate evaluations, and other context-specific factors. The effect of SNTV on Taiwan’s party system has been discussed at length (Cox and Niou, 1994; Grofman, 1999;

Lin, 1996). Candidate-based voting is often seen in large districts, and party voting merely exists in small districts. It has been found that there are partisan and candidate components in electoral choice, which coincides with the division between urban and rural areas (Chu, 1995). In urban areas, ideological considerations strongly influence voting choices. But in rural areas, local interest is more important than party labels. All in all, the Taiwanese voting pattern is determined by partisanship and candidate assessment together. The transition of party competition from a one-party to a multi-party system preserves the influence of party labels, but more and more voters prefer candidate-based voting (Hwang,

1999).

As to the Japanese voter, social networks along with candidate evaluations were singled out as the determinants of voting choice (Miyake, 1999). Flanagan,

Kohei, Miyake, Richardson, and Watanuki (1991) stressed the role of partisanship and party image in making voting choice. Richardson (1997) found that the electoral connection between parties and voters is not as prominent as that between social groups or candidates and voters. Social groups, such as family members, friends, and social organizations, seemingly play a bigger role during campaigns than do parties.20

In the case of the United States, the determinants of the congressional elections are mixed. Voting in congressional elections is strongly associated with assessments of the candidates (Mann and Wolfinger, 1980). But partisanship is also no less important than candidate evaluations (Brown and Woods, 1991;

Abramowitz, 1984). Besides, the national economy and presidential popularity are also important factors (Born, 1986).

On the other hand, the decline of competition in the congressional elections is remarkable since the 1960s. The ninety-percent House reelection rate is attributed to high incumbency benefits (Jacobson, 1997) and low quality of challengers (Abramowitz, 1991). American voters rate incumbents higher than challengers (Mann and Wolfinger, 1980; Hinckley, 1980). Compared to Japan and

Taiwan, American voters seemingly tend to vote for the incumbents.21 Overall,

20 The underestimation of the influence of parties is partly due to the inadequate questions in the survey. Nevertheless, this recent survey has included some questions that allow us to estimate the effect of parties. For instance, in CNEP survey the respondents were asked “What is the reason for your voting for the party? Is that for securing the seats of the party in the Diet, or for your favoring toward the candidate because of this career, political ability, personal character?” and “For each of the following organizations, which do you belong to?” Both questions, albeit not very relevant to party organizations, would be useful in the estimation of party effect on voting choice. 21 Although there is no research for this point, the rough statistics may support it. In the latest legislative election in Taiwan, only half of the incumbents kept their office. That does not mean that, of course, there is no incumbency advantage in Taiwan. For example, the current Speaker of the Legislative Yuan came in second in his district, and a young DPP nominee received more votes than him. As far as I know, the margin between the incumbents and challengers is not as large as in the United States.

individual-level analysis of congressional elections confirms the influence of both the relative difference of candidate evaluations and partisanship (Hinckely, 1981;

Abramowitz, 1980; Abramowitz, 1991; Brown and Woods, 1991).

To understand the effect of political attitudes on party voting in each country, a comparable-case model is constructed in this dissertation. For a country with a long history of elections and considerable data like the U.S., comparable- case study allows us to see the fluctuation in the effect of independent variables over time. For countries with less data, like Taiwan and Japan, this type of study serves as a common ground for comparison.

A cross-national study requires comparable measurements. Due to the limitations of secondary analysis, however, this requirement may not be met for every single variable. Furthermore, it is not easy to find same concepts across different context. Even though the face validity of each variable used in this dissertation has been examined, the comparability problem should be recognized and taken as a caution regarding the findings of this dissertation.

2.6 Plan of the Dissertation

Chapter 3 will be devoted to voting behavior in United States House elections.

Chapter 4 will estimate the effects of determinants of long-term and short- term variables on Taiwanese voting behavior.

Chapter 5 will explore the change in Japanese party voting before and after

1994.

Chapter 6 will pool the election data and examine the influence of the system-level variables—government structure and election systems.

Chapter 7 will summarize the findings of both the comparable-case study and the cross-sectional study, particularly the extent to which institutional differences affect differences in the amount of party voting in these countries.

CHAPTER 3

AMERICAN CONGRESSIONAL VOTING BEHAVIOR IN 1990 AND 1998

3.1 Introduction

As discussed in Chapter 2, four hundred and thirty-five seats in the House of Representatives are subject to single-member district election every two years.

In addition to the size of district, there are several differences between presidential and congressional elections. The first difference lies in the incumbent factor.

Incumbents run for reelection to the House in almost 80 percent of districts, and the successful rate has been seldom below 85 percent (Jacobson, 1997). Secondly, the margin of vote shares between the two candidates is larger in House elections than in the presidential race. Last, the turnout rate in House elections is lower than in the presidential one, especially in mid-term elections. The turnout rate in presidential elections is always higher than that of the next congressional elections.

While scholars have pointed out the distinctive features of House elections, the efforts to explore choices among congressional candidates are not as extensive as for presidential elections. Political scientists collect survey data in the off-year elections, but before 1978 the study of congressional elections received less

attention than it deserves (Hinckley, 1980). Most literature is focused on the advantages enjoyed by incumbents and its effect in congressional elections.

Thanks to the perquisites granted to members, such as free mailing, generous travel allowances, and large staff, incumbents are able to keep contact with their constituents and to make themselves widely recognizable (Fiorina, 1977;

Yiannakis, 1981). A portion of literature is aimed to find out why voters favor incumbents. Scholars drew the conclusion that congressmen exploit their advantage of incumbency over potential challengers (Cox and Katz, 1996).

Incumbents are also better financed than challengers, because incumbents receive more political action committees (PACs) contributions than challengers (Herrnson,

1998).22 Political parties raise a huge amount of money to support candidates, particularly marginal incumbents. Casework is also singled out as one of the factors that contribute to the high successful rate of incumbents (Cain, Ferejohn, and Fiorina, 1987). By delivering services to constituents, incumbents build up positive images in the districts. (Cover, 1977; Fiorina, 1989). In short, voters choose incumbents rather than challengers because incumbents have higher visibility, better office support, and higher reputation than challengers (Mann and

Wolfinger, 1980; Jacobson, 2001; Cox and Katz, 1996). Scholars also stress the strategic entry of candidates; either national events or local motivation can influence the decision to enter the race (Jacobson and Kernell, 1983; Bond,

Covington, and Fleisher, 1985).

However, there are several political realities that are not in accordance with the great influence of incumbency. First of all, national economic conditions

22 Hernson broke down the allocation of PAC contributions and found that most of this money was

considerably affect voting behavior in House elections (Tufte, 1978; Kinder and

Kiewiet, 1981; Kramer, 1983). Had voters only considered incumbents versus challengers, the national economy should have little effect on voting decisions.

Secondly, divided partisan control occurs more often than united government since the World War II. Fiorina (1992) has suggested that a significant portion of the American electorate purposively split their votes between two parties in the presidential and House elections. This implies that incumbents get reelected not because of their personal traits or services. Rather voters support incumbents who are not the president’s party’s nominees. Also, scholars characterize congressional elections as low-stimulus ones, compared to the presidential race. Partisanship should be more salient than other factors since there is no short-term force like high-profile candidates or dramatic issues (Campbell, 1960; Beck, 1986). Up to

1994, the national House vote approximated the division of partisanship, which is about 54 percent Democratic (Erikson and Wright, 1993).

In addition to the high victory rate of incumbents in House elections, there is strong evidence that incumbency plays a major role in House elections. First of all, the dominance of the Democratic Party in House elections for four decades is in part because voters consistently vote for incumbents (Epstein, 1986). Even though the presidency alternates between the two parties, the Democrats always defied the Republicans’ bid for the majority party in the House of Representatives.

The one-party southern politics to some degree contributed to the Democrats’ majority in Congress. In 1994, the Republican Party successfully turned around the south region and ended the Republicans’ forty-year drought for the majority in

spent on the competitive incumbents, but not the competitive challengers.

the House of Representatives. Moreover, the margin between incumbents and challengers, which was sharply depicted by Cover (1977), also underscores the existence of an incumbency advantage.23

The mixed evidence indicates that congressional elections are characterized by both national and district-level factors. In comparison with the presidential election, the choice among congressional candidates entails more local interest that reflects candidate evaluations. However, evaluations of the national economy and issues that are well beyond local interest and associated with political parties also have been taken into account. In addition, partisanship, which is developed in the process of socialization, with or without consciousness, also influences voting choices. Analyzing voting behavior in congressional elections entails considering both national and local factors, including candidate assessments and incumbency.

3.2 Incumbency and Party Competition

Rather than party choice, most congressional election studies attempt to account for incumbency voting. More formally, voting choices between incumbents and challengers are regressed on the explanatory variables, such as party identification (with the incumbent’s or the challenger’s party), challenger quality, and ideology (Hinckely, 1981; Abramowitz, 1984; Cain, Ferejohn, and

Fiorina, 1987). The major findings are that voters tend to vote for incumbents when partisanship is congruent with incumbent’s party and that incumbents have

23 Regarding the different measurements and estimates of incumbency advantage, see Gelman and

better evaluations than challengers do. It is meaningful to view congressional elections as a way to assess the performance of incumbents, but that fails to capture another aspect of electoral politics: party competition. Only when we model choices between candidates nominated by the two political parties, can we link congressional election results with individual’s attitudes toward political parties.

From the perspective of political institutions, Congress is a national legislature and the outcome of a congressional election is critical to law-making.

Although Congress shares power with the president and the courts to some degree, it is powerful in initiating, modifying, approving, and rejecting laws. Sometimes it has different ideas on the government budget and public policies from the president, which can lead to a gridlock. We can view congressional elections as not only an evaluation of incumbent representatives, but also an assessment of the performance of Congress as a whole.

Every individual congressman can influence law-making in standing committees (Epstein, 1986). As matter of fact, congressmen build policy coalition with the bureaucracy (Arnold, 1979). Nevertheless, only political parties are able to push forth legislation. The party caucus, particularly after the 1970s, chooses the speaker of the House, majority party leaders, and sets the schedule of legislation (Rohde, 1991; Cox and McCubbins, 1993). Political parties also fund party members running for election or reelection through the congressional campaign committees. It would be fair to assume that voters have certain perceptions toward political parties behind their candidates and make their voting

King (1990).

choices more or less based on their evaluations of Congress. More specifically, voting in congressional elections is partly grounded in voters’ evaluations of congressional parties.

In U.S. congressional elections, voters also consider the national economy and other salient issues (Kramer, 1971; Tufte, 1975; Jacobson, 1997; Brown and

Woods, 1991). As the theory of economic voting implies, people care about their own personal and the national economy. The president is held accountable for the national economy, thus voters who are not satisfied with the national economy would send a signal on the president’s performance. In some senses, congressional elections are characterized as a referendum on the president, instead of Congress

(Kernell, 1977). That makes congressional elections national races between political parties, instead of local events.

It remains an open question as to why voters consistently pay attention to the economic issue in congressional elections. However, we should accept the assumption, based on previous literature, that voters would consider the national situation as the evaluative standard and vote accordingly. 24 In this case, congressional elections should be studied from the perspective of party competition, while putting some emphasis on candidate evaluations.

3.3 Candidate Evaluations and Party Identification

The comparison between party identification and the incumbency advantage in congressional elections is of great interest. As Congress has become

more and more ideological (Jacobson, 1997), the performance of parties should bear on the fate of the incumbents. Moreover, members of Congress are less familiar to most voters than is the president, thus the short-term force resulting from candidate evaluations is likely to be less influential. Therefore, we expect that party identification will dominate voting behavior. However, the incumbency advantage is still a prominent factor. Incumbency serves as an electoral force that highlights the importance of candidate evaluations (Hinckley, 1981).

The roles of political parties and candidates in voting choices have become more and more complicated in congressional elections. On the one hand, national party organizations support candidates financially (Herrnson, 1998; Jacobson,

2001). On the other hand, however, we see a steady increase in the incumbency advantage (King and Gelman, 1991; Cox and Katz, 1996). Party-line voting in

House elections also declined after the 1970s, partly because of diminishing one- party politics in the South (Mann and Wolfinger, 1980). Despite that political parties remain influential in congressional campaigns and Congress, they turn out to be less widely perceived among voters. Why is it the case? How do we reconsider our voting behavior model?

The presidential candidates are well known because they are national political figures. They receive attention when they compete in the primary elections, and they become better and better known as the election day approaches.

On the other hand, voters are less exposed to media coverage of congressional candidates. Congressional candidates must use personal contact and community meetings to elicit their visibility. In Mann and Wolfinger’s analysis of the 1978

24 Regarding the discussion and analysis of retrospective and prospective voting in the presidential

survey data, incumbents have more voter contact with House candidates than challengers, particularly in “met personally”, “saw at meeting”, and “received mail”. 25 Obviously, incumbents benefit from contacts with voters more than challengers. They use personal contacts and other services to distinguish themselves from challengers. Incumbents are engaged in personal contact with voters, and they force the challengers to compete with them on this basis

(Abramowitz, 1980; Krasno and Green, 1988).

However, party and ideology are still mentioned by congressional voters as the reasons they like or dislike candidates (Jacobson, 2001). Cain, Ferejohn, and Fiorina (1987) used David Mayhew’s (1974) analysis about the three activities that congressmen conduct—advertising, credit claiming, and position taking—to explain why congressmen’s voting record is important to congressional voting behavior. While the fact that incumbency does not necessarily reflect every aspect of candidate evaluations, we can propose that how the electorate evaluates the candidates determines voting choices (Brown and Woods, 1991). I expect to find equal effects of party identification and candidate evaluations concerning the voting act.

The contrast between candidate evaluations and partisanship is very evident in American congressional elections. One of the main objectives of this project is to explore the relative importance of each factor in voting behavior. In addition to partisanship, candidate evaluations, incumbency, and retrospective economic evaluation, party images are considered as an important variable. Party

election, see Miller and Wattenberg (1983)

images—whether the voter likes or dislikes parties—are also viewed as an intermediate attitude between long-term party attachment and vote decisions

(Hinckley, 1981). They tap into the campaign-related aspect of party loyalty, separate from party identification because party identification is acquired from socialization and party images from specific information.

3.4 Data and Variables

3.4.1 Data

To explore how voters choose among political parties, I will use the data from the national survey of the American National Election Studies that are carried out by the Survey Research Center (SRC) and by the Center for Political

Studies (CPS) of the Institute for Social Research at the University of Michigan. I chose the surveys conducted for the 1990 and 1998 Congressional elections. Post- election interviews were conducted with 1980 eligible voters in 1990, and 1281 eligible voters in 1998. Each of the two datasets includes questions on voting choice, party identification, candidate evaluations, issue positions, and voters’ personal economic situations. Besides, both datasets contain a code indicating whether or not there is an incumbent running in the district. Two surveys are analyzed because the comparison of two years can show if the short-term forces vary with time and if the influence of long-term forces responds accordingly.

One reason to choose the 1990 and the 1998 elections as the cases is that party control of Congress was different in these two elections. The Democratic

25 Actually, voters also report that they read much more about incumbents than challengers in a newspaper and magazine. I suspect that the difference results from incumbents’ constituency

Party was the majority party in 1990, and the Republican Party took over after

1994. Having both eras of Congress can let us observe if voters had different reaction to the performance of Congress. In addition, both governments were divided ones, which means the majority party in Congress had to compete against the president’s party in each off-year election. Divided control of Congress and presidency makes responsibility for policy-making less obvious. In this circumstance, voters would be expected to rely on candidate evaluations for voting guidance.

3.4.2 Operationalization of Variables

Both 1990 and 1998 survey data requested similar information and political attitudes, which facilitates comparison between datasets. The most important two independent variables are partisanship and candidate evaluations.

Both surveys include a same set of questions asking people their partisanship. The question reads: “Generally speaking, do you usually think of yourself as a

Republican, a Democrat, an Independent, or what?” If the respondent chooses either party, she will be asked the follow-up question: “Would you call yourself a strong Democrat (Republican) or a not very strong Democrat (Republican)?” If the respondent chooses “Independent” for the first question, she will be asked another follow-up question: “Do you think of yourself as closer to the Republican

Party or to the Democratic Party?” The series of three questions can be organized

service.

as a seven-point scale that indicates the strength of single-dimensional partisanship from strong Democrat through independent to strong Republican.26

The candidate assessment variable is derived from a series of questions.

The first question reads: “Was there anything in particular that you liked about the Democratic (Republican) candidate for the U.S. House of Representatives?” If the respondent says “yes,” there is a follow-up open-ended question asking the particular thing that people like about the Democratic or Republican candidates. If the respondent says “no,” he will be asked: “Was there anything in particular that you liked about the Democratic (Republican) candidate for the U.S. House of

Representatives?” And the probe question follows up. The more times that people mention about what they like about candidates, the greater degree to which they are expected to like them. Only few of those people who answer ‘yes’ cannot think of any particular thing that makes them like the candidate. For those people who cannot give any specific reason, I assigned 0 to them. For those of people who can give one or more specific reasons for their support, I assigned 1 to 5 to them, depending on how many responses they give. Here I merely deal with the positive perceptions toward the two candidates, because of the need for cross- country comparability.27

The issue variables use individual responses to two 7-point-scale questions in both surveys. The first issue is whether government should provide fewer or more services. I assume that more services means greater involvement of

26 Here I assume that partisanship is a one-dimensional attitude, which is based on Fiorina’s perspective. Weisberg (1987) and others has argued for the multi-dimensionality of party attachment, but most studies still set it up as one-dimensional continuous variable. 27 In Taiwan and Japan, the surveys merely asked the respondents which candidate meets certain requirements--there is no negativity questions. I will count the number that people mention about the candidates, who are also coded as parties.

government in society, so I label this issue “big government”. People who support more government services would be coded as 7, and people who prefer fewer services coded as 1. The second one is whether government should let every person go ahead on their own or find everyone an equal job. I call this issue

“individualism.” The response for getting equal job is coded as 1 and letting everyone go her own way is coded as 7.

Ideology comes from the question: “Do you consider yourself as liberal or conservative?” For the purpose of comparable measurement, the responses are recoded as a three-point scale of three categories--liberal, neutral, and conservative. Party images are derived from the responses to the feeling thermometer questions in order to use comparable measure for each country. The respondents are asked to give a 0 to 100 rating for each of the two parties.

There is a battery of questions on people’s assessment of their personal and the national economy. For theoretical reasons, I only chose the retrospective evaluation of national economy. The question reads: “Now thinking about the economy in the country as a whole, would you say that over the past year the nation’s economy has gotten better, stayed about the same, or gotten worse?” I assigned 1 to people who chose “gotten better”, 0 “stayed about the same”, and –1

“gotten worse”.

In addition to political attitude questions, the surveys include many demographic questions. I will use the information about the respondent’s age, income, education, occupation, geographic region of residence, and race. Those variables would be treated as exogenous to both political attitudes and voting choices.

3.5 Party Identification and Candidate Evaluations

Two theories suggest that party identification affects candidate evaluations.

The first theory states that partisanship colors people’s perceptions of candidates

(Campbell, Converse, Miller, and Stokes, 1960; Stokes, 1966). If candidates are not at all different, or the difference among them is not salient, then voters may evaluate the candidates based on their party attachment. Party identification is a long-term force, thus those voters who have predispositions toward one party or the other would prefer their party nominees. The second theory states that voters have images regarding the candidates, but their images come from political parties.

Loyal voters in their districts would support candidates who push their parties’ policy in Congress (Erikson and Wright, 1993). Both theories suggest that partisanship is the antecedent factor to the perceptions of candidates (Weisberg and Rusk, 1970). As the incumbency factor becomes more and more prominent in congressional elections, however, candidate evaluations could be more important than partisanship. Mann and Wolfinger (1980) pointed out that a significant portion of people defect from their political parties and support the incumbents who they can recognize.

Before examining how political parties shape evaluations of candidates, it is necessarily to break down how voters evaluate candidates. Jacobson (1997) analyzes voters’ mentions of things from 1978 to 1994 found that voters mentioned less and less about personal characteristics of candidates. Instead, voters noticed more and more about candidates’ ideological positions and policy positions. Political parties rather than candidates became the focus in

congressional elections.28 To examine if the electorate have different perceptions toward candidates of the two parties, I acquired the original responses to the like/dislike open-ended questions and coded them by party. There are seven categories for positive and negative comments about candidates. “Personal” refers to comments regarding candidate characteristics, such as “I just believe him,” or

“He was new,” or “He is a crook.” “Performance” means general assessment of the candidate, such as “He has done a good job,” or “He tried hard to help people.” “District service” refers to specific mention that the candidate has done something for the area, industry, or projects. “Party” category would cover the comments regarding candidates’ party label. “Ideology/policy” refers to comments on candidates’ issue position or ideology—conservative or liberal.

Some respondents cannot give a reason as to why they like/dislike candidates and they are coded “don’t know.” In addition to these six categories, there is a

“campaign” category for the 1998 data, which means that the respondents like/dislike candidates for their campaign advertising. Table 3.1 shows what voters like and dislike about the candidates in the 1990 election, and Table 3.2 shows what voters like or dislike about the candidates in the 1998 election.

28 Of course, voters may rationalize their feelings when they answer those questions. Since those surveys were conducted after elections, it is possible that voters gave rationalized answers instead

Democratic Party Republican Party Like Dislike Like Dislike Personal 34.5% 57.1% 31.0% 27.3% Performance 13.8 0 10.3 0 District Service 6.9 0 6.9 0 Party 13.8 14.3 10.3 9.1 Ideology/Policy 10.3 28.6 6.9 18.2 Group Associations 0 0 13.8 18.2 Don’t Know 20.7 0 20.7 27.3 Number of Mentions 29 7 29 11

Data: 1990 NES data

Table 3.1: Voters’ mentions about things they liked and disliked about House candidates, 1990

Because the 1990 data was not recorded well, there are only 71 cases to be analyzed.29 Table 3.1 makes it clear that most people comment on candidates’ personal traits as their liking/disliking reasons. There are 34.5 percent of people who like the Democratic candidates mentioning candidates’ personality, and there are 57.1 percent of people who dislike the Democratic candidates citing candidate characteristics. Around 10 percent of people consider party label and ideology respectively as their liking reason for the Democratic and Republican candidates.

District service and performance for both parties’ candidates are as important liking reasons as party and ideology. As for the disliking reason, the percentage of party and ideology mentioned by the respondents is second only to personal characteristics.

of the first thing on their mind. We should not underestimate the importance of affective evaluations. 29 The 1990 data is not included in Jacobson’s (2001) analysis.

Democratic Party Republican Party Like Dislike Like Dislike Personal 43.2% 36.2% 38.7% 36.6% Performance 17.8 5.5 16.9 5.7 District Service 8.7 4.5 9.9 2.4 Party 3.6 4.0 4.2 6.5 Ideology/Policy 14.1 35.2 21.5 32.5 Campaign 0.5 9.0 0.7 8.1 Group Association 9.8 0.5 5.6 3.3 Don’t Know 2.2 5.0 2.5 4.9 Number of Mentions 583 199 284 123

Data: 1998 NES data

Table 3.2: Voters’ mentions about things they liked and disliked about House candidates, 1998

In the 1998 survey, the respondents mentioned more about personal traits of candidates as their liking/disliking reason than other reasons, which is similar to the 1990 survey. The percentage of response regarding ideology increases for both parties, but that of party decreases. Overall, people still pay less attention to candidates’ issue position, ideology, and party label; instead, they receive very general perceptions toward candidates. Moreover, this pattern is the same across both parties’ candidates; they deliver general personal traits rather than specific ideology or issue positions to the electorate.

Although the open-ended questions reveal that the major source of candidate evaluations is candidates’ personal traits, party identification may implicitly influence whether people perceive candidates positively or negatively.

Using simple cross tabulation analysis can show us the effect of party identification on candidate evaluations measured by whether people like anything about candidates of two parties.30 I ran the cross tabulation analysis between the strength of partisanship and the two one-dimensional variables for both 1990 and

1998 survey data. Table 3.3 and Table 3.4 reveal the strong association between partisanship and candidate evaluations in 1990 and 1998 respectively, which confirms my hypothesis. The percentage of people who like Republican candidates is declining with increasing attachment with the Democrats. On the other hand, the percentage of people who have positive feelings toward the

Democratic candidates increases with the increasing strength of Democratic partisanship. At first glance, people are more responsive to the Democratic Party than the Republican Party in 1990 and 1998. In 1990, 50 percent of very strong

Democrat partisans indicated that they like the Democratic candidates, but only 6 percent of them mentioned they supported the Republican candidates. However,

33 percent of very strong Republican partisans liked the Republican candidates, and 17 percent of this type of voters liked the Democratic candidates.

This pattern remains in 1998. The percentage of the Republican partisans who supported the Republican candidates actually increased, and the percentage of the Democratic partisans who likes the Democratic candidates decreased.

Overall, the percentage of people who have positive feelings toward their own party increases significantly with the level of partisanship. This implies the association between partisanship and candidate evaluations.

30 The open-ended responses for both surveys are assigned random case identification number, so

Very Strong Rep. Independent Dem. Strong Very Strong Rep. Dem. Strong Rep. Dem. Like Rep. 33.86 22.53 21.89 10.59 8.98 9.84 6.87 No 66.14 77.47 78.11 89.41 91.02 90.16 93.13 Opinion Column 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Total Like Dem. 17.46 15.70 17.60 14.83 28.57 31.38 50.64 No 82.54 84.30 82.40 85.17 71.43 68.62 49.36 Opinion Column 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Total

Note: Table entries are the column percentages. N=1965; p<=.001

Table 3.3: Positive feelings toward candidates and partisanship, 1990

Very Strong Rep. Independent Dem. Strong Very Strong Rep. Dem. Strong Rep. Dem. Like Rep. 54.74 32.32 34.59 13.10 17.68 18.64 21.34 No 45.26 67.68 65.41 86.90 82.32 81.36 78.66 Opinion Column 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Total Like Dem. 14.60 12.1 18.80 14.48 32.60 25.00 43.10 No 85.40 87.88 81.20 85.52 67.40 75.00 56.90 Opinion Column 100.00 100.00 100.00 100.00 100.00 100.00 100.00 Total

Note: Table entries are the column percentages. N=1269; p<=.001

Table 3.4: Positive feelings toward candidates and partisanship, 1998

the tabulation analysis here uses data that codes which party that the respondents like/dislike.

Furthermore, I counted the number that voters mention the candidates as the level of positive feelings toward candidates. Figure 3.1 shows that the level of positive responses toward Democratic candidates is strongly associated with strength of partisanship, but that is less the case for Republican candidates. The difference in evaluation of Democratic candidates between very strong

Democratic supporters and very strong Republican supporters is more impressive than that for Republican candidates between the two groups of people on both ends. Those people who identify with Democrats and Republicans actually have the same judgment about the opposite party’s candidates, no matter how strong their party attachment is. Nevertheless, they have many responses to the questions about the candidates nominated by the parties they support, and the level of responses is a function of their strength of identification. The linear relationship is more evident among Democratic partisans than among Republicans.

0.8

0.6 Democratic Candidates 0.4 Republican 0.2 Candidates Mean Value Mean

0 -3 -2 -1 0 1 2 3 Partisanship (Democratic Party)

Figure 3.1: Evaluation of candidates and partisanship, 1990

1.4 1.2 1 Democratic 0.8 Candidates 0.6 Republican 0.4 Candidates Mean Value Mean 0.2 0 -3 -2 -1 0 1 2 3 Partisanship (Democratic Party)

Figure 3.2: Evaluation of candidates and partisanship, 1998

In 1998, both strong Democratic partisans and strong Republican partisans gave far more credit to their party’s candidates than to the other party’s. It seems that the supporters of both political parties are more polarized than in 1990.

Noticeably, the Republican partisans have about the same level of responses to the evaluation questions on the Republican candidates as the Democratic partisans do in 1998. I suspect that the majority in the House of Representatives helps the

Republicans obtain great attention in the electorate.

I ran an ANOVA analysis to check if we can accept the null hypothesis that the difference of candidate evaluations for each party across categories of partisan strength happens by chance. The F value for the 1990 Democratic candidate evaluations is 26.76 and for Republican candidate evaluations 18.02.

Because both are significant at the .05 level, we can reject the null hypothesis that every category of partisan strength has the same level of positive feelings toward the candidates of concern. Regarding the 1998 Democratic and Republican candidate evaluation, the F values are 16.902 and 14.302 respectively. With 6 degree of freedom, the null hypothesis that there is no difference in candidate

evaluations among the categories of the partisan strength can be again rejected. In sum, the association between partisanship and candidate evaluations is statistically significant.

3.6 Party Identification and Issue Positions

Traditionally, both American parties have distinctive social bases that are based on domestic and foreign issues. The level of strength of party identification affects the intensity of issue opinions. The stronger is party identification, the more relevant issue position would be. Party identification is one of the antecedent factors that have a great impact on political attitudes, which are specific to particular elections (Campbell, Converse, Miller, and Stokes, 1960; Stokes, 1966).

Unlike partisanship or ideology, the voter’s attitudes toward issues are sensitive to context. It is even less reliable than people’s perceptions toward political leaders

(Converse, 1964; Converse and Markus, 1979). Besides, the authors of The

American Voter discovered that only few American voters employ their issue perceptions in voting.31 The predisposition model states that issue positions are proximate to voting choices, but voters would act in accord with their partisanship if they cannot perceive issue difference between parties. The notion of “standing decision” exemplifies the influence of partisanship on the direction of policy preferences as well as on voting behavior.

Scholars have ascertained the cause-and-effect relationship between issues and party identification (Brody and Page, 1972; Jackson, 1975; Page and Jones,

31 Regarding the discussion of the changing role of issue position in voting behavior since the 1970s, see Nie, Verba, Patrocik (1976).

1979; Wright and Niemi, 1983). On the one hand, voters may project their preferred party’s issue position on their policy positions. On the other hand, they may adjust policy positions to their perceived party’s position.32 With panel data and new statistical models being used, some scholars now conclude that partisanship is not exogenous to issue positions. Instead, partisanship is also sensitive to the policy preferences (Jackson, 1975; Franklin and Jackson, 1983).

More and more evidence indicates that issue positions as short-term forces could affect the strength and even the direction of partisanship over time.

Fiorina (1981) makes the most notable argument that current political issues have an effect on how people identify with political parties. He suggests that partisanship is a running tally of past political affairs, thus the continuity or change of partisanship is based on evaluations of political happenings. Panel data shows that people update their partisanship to some degree, therefore it is very possible that issue positions can cause change in partisanship.

Acknowledging the revisionist view of partisanship, I propose that in congressional elections partisanship has a strong one-way association with issue positions. In theory, party identification is learned prior to policies as a part of primary socialization in the family. Moreover, political issues in congressional elections are not as important a factor as in presidential elections; national issues for the most part are remote from local districts (Hinckley, 1981). Accordingly, it is safe to assume that party identification is the cause of issue positions. Before using a more complicated model to weigh the relative influence of partisanship

32 Page and Jones (1979) using the 1972 and 1976 survey data discussed the two-way causation between policy positions, candidate evaluations, and party attachment. They found that the

and issue positions on vote choice, I examine the association between them in the

1990 and 1998 survey data. I use two issues: whether the government should provide fewer or more services (big government), and whether the government should let every person go ahead on their own or see to equal opportunity

(individualism). In theory, the measurements of issue positions return interval- level data. Figure 3.3 shows the association between the two issues and the partisanship in 1990. It is apparent that there is a fairly linear relationship between partisanship and each of the two issue positions. With change in the level of party attachment, issue positions also change accordingly. The Republicans support more individualism, but fewer government services. On the contrary, the

Democrats favor more government services, but less individualism.

6 5 Individualism 4 3 2 Big Government Mean Value Mean 1 0 -3-2-10123 Partisanship (Democratic Party)

Figure 3.3: Partisanship and two issue positions, 1990

comparative policy distances between voters and parties strongly affects current party attachment at least in 1976.

6 5 4 Individualism 3 Big Government 2

Mean Value Mean 1 0 -3-2-10123 Partisanship (Democratic Party)

Figure 3.4: Partisanship and two issue positions, 1998

Figure 3.4 shows the effect of party loyalty on issue stands in 1998. The

Democratic partisans favor more government services but less individualism. On the contrary, Republicans support individualism and like to see smaller government.

The effect of partisanship on issue positions has been confirmed by the

ANOVA analysis, which tests the null hypothesis that there is no difference in the issue positions among the seven groups of partisans. The probability that the difference in issue positions across the seven groups of partisans occurs by chance is so small that we can reject this possibility. The F values are 16.70 and 20.81 for the individualism and big government issues respectively and both are significant at the .05 level, therefore we can reject the null hypothesis that every category of partisan strength has the same issue positions. Regarding the 1998 issue positions, the F values are 23.95 and 18.15 for the individualism and big government issues respectively. With 6 degree of freedom, the null hypothesis that there is no difference in issue positions among the categories of the partisan strength can be rejected. This result implies that American voters use their party preference to

conceive of their policy positions. The conformity between these two political attitudes, however, constrains one of them in terms of the explained variance of the dependent variable. The relative influence of partisanship and candidate evaluations will be explored in the multi-variate model.

3.7 Issue Position and Candidate Evaluations

In the study of House races, the issue component of candidate evaluations has seldom been examined. Unlike presidential elections, there is no single national candidate who can represent the whole party. Besides, candidates tend not to engage in any substantial policy debate, avoiding taking responsibility for government performance. Therefore, it is difficult for voters to classify either the incumbent or the challenger as supporters of any policies. Nevertheless, with

Congress becoming more and more ideologically-oriented, the way that voters evaluate their representatives can become more issue-oriented. As seen in

Jacobson’s (2001) analysis, the frequency that voters choose ideology and policy as the reason for liking or disliking candidates has been increasing. The other evidence is the rising correlation between district-level House and presidential voting since the late 1980s. In the 1990s, the tension between the administration and Congress reached an unprecedented level. Congress can get involved in an issue that sharply divides the electorate in districts, which makes candidate more vulnerable to challengers.

The reciprocal relationship between candidate evaluations and issue position is therefore of concern for scholars who apply the predisposition model to election studies. Markus and Converse (1979) found a persuasion effect, which

means that individual’s issue positions are partly determined by the perceived candidates issue stands. If voters have very limited information on issues, they may pick up their favorite candidate’s issue positions as theirs. Page and Jones

(1979) studied the reciprocal relationship between policy distance and candidate evaluations in two presidential elections, concluding that the assessment of policy distances is greatly contingent on candidate evaluations. By the same token, voters may perceive candidates’ issue positions on the basis of their issue positions

(Brody and Page, 1972). Brown and Woods (1991) confirmed the reciprocal influence between issue distance and candidate evaluations. Certainly, the influence of candidate evaluations depends on the setting of the election, and it is an open question concerning how important they are in House races.

Because I use voters’ actual issue stands, instead of the issue distance between voters and the candidates, I posit that the relationship between candidate assessment and issue stands is one-way instead of two-way. In other words, voters’ judgment of the candidates would not shape their issue positions on national issues. The first explanation is that congressional candidates are viewed as more local representatives than national delegates. Voters cast their ballots for someone who can bring something back to the district, but not a national leader.

The second reason is that issue positions tend to conform with partisanship when there is no dramatic event. In the House races, which lack charismatic leaders and nationwide issues, voters’ actual issue stands should be based on more long-term forces than short-term ones.

To investigate how candidate evaluations vary with issue positions, partisanship, incumbency of candidates and demographic variables, I estimate the following model for both 1990 and 1998 survey data.

E(Y│X)=β´X------(3.1)

where

Y: Candidate evaluations of Democratic candidates minus candidate evaluations of Republican candidates

X: vector of the explanatory variables

β´: inverse vectors of the coefficients

In Equation 3.1, I include partisanship, issue positions, race, residence area, occupation, and whether their incumbent is the Democrat or not as the explanatory variables. On the left-handed side of the equation, I computed the difference between the positive evaluation of the Democratic candidates and that of the

Republican candidates. The conversion of candidate evaluations is based on the utility theory, which states that people tend to support the candidate who has a better evaluation. The reduction of the number of dimensions in candidate evaluations will also save us from redundant analysis.33

Since I subtract the positive evaluation of the Republican candidates from that of the Democratic candidates, I expect that the higher the level of individualism is, the lower the value of the dependent variable. The level of

partisanship, which also places the Democrats on the positive side, should have a positive impact on the dependent variable. People whose representative is a

Democrat should tend to have better evaluation of the Democratic candidate, thus the sign of the coefficient of incumbent’s party should be positive. Among the demographic variables, race is expected to be the most prominent predictor.

Blacks are expected to evaluate Democratic candidates better than Republican ones. Thanks to the racial issue, the Democratic Party has been dominant in the south. Therefore, I assume that southern voters are more likely to support the

Democratic candidates. The test of simultaneity is a test of whether the regressor is correlated with the error term. I use the Hausman specification test that is suggested by Gujurati (1995: 670-673) to confirm that there is no simultaneity problem between issue position and candidate evaluations.34

33 I also ran the OLS equations for the evaluation of the Democratic and Republican candidates for both years separately, but the influence of issue positions is unstable and vague. 34 According to the method, the difference in candidate evaluation is regressed on both issue positions, partisanship, incumbency, a set of demographic variables and the predicted values of the two issue positions that are acquired from regressing the two issue position variables by ideology

Comparative Candidate evaluations 1990 1998 Intercept -1.270(.230)*** -.824(.274)** Big Government .057(.025)* .069(.030)* Individualism -.033(.022) -.034(.027) Partisanship .141(.020)*** .165(.025) *** Incumbent’s Party 1.416(.084)*** 1.343(.098)*** (Democrat=1) Age .158(.051)** -.081(.065) Education .188(.054)** .023(.062) Race (Black=1) -.306(.122)* .222(.150) Income .002(.007) .000(.007) Occupation

Professional -.116(.106) .052(.113) Farm -.139(.252) 1.301(.695)* Labor .002(.123) .130(.155) Residence Area

East -.282(.110)* .152(.125) South .292(.101)** .595(.116)***

Number of 1222 1125 Observations Adjusted R-square .271 .247

Data: 1990 and 1998 NES data Note: Standard errors are in parentheses. *: p<.05, **: p<.01, ***: p<.001.

Table 3.5: OLS estimates of comparative candidate evaluations, 1990 and 1998

The OLS estimates presented in Table 3.5 show the strong impact of partisanship, the issue of big government, and the status of the incumbent in 1990 and 1998. The coefficients of the major explanatory variables in both years are similiar. First of all, only the issue of big government is influential in terms of candidate evaluations. When the level of supporting big government increases by

respectively. The null hypothesis that both purged issue position variables are insignificant is not rejected, thus both variables are exogenous.

2, the difference between the Democratic evaluation and Republican evaluation will increase by about .1 on average. Secondly, partisanship is quite important across two elections. When the individual moves toward the Democratic Party by

1 category, the difference in the Democratic evaluation and the Republican evaluation will increase by .1 on average. Thirdly, the influence of the incumbent on the candidate evaluations is rather significant. The individual who has a

Democrat as his representative will have more positive evaluation of the

Democratic candidate than someone who has no Democratic incumbent by 1 category. Last, southern voters tend to give higher evaluation to the Democratic candidates than the Republican ones. In 1990, being a southern voter will increase the level of the preference of the Democratic candidates over the Republican ones by .292 unit on average, and in 1998 the increase is .595.

Other than the four major explanatory variables, the effects of the other control variables are not impressive. The influence of race is not in the right direction, partly because of the intervention of the residence area. Age, education, and living in East are only significant in 1990 survey, and farming has great impact on the dependent variable in 1998.

Essentially, this is a preliminary analysis to explore how issue positions shape the assessment of candidates, controlling for the relevant variables. I found that partisanship is the most consistent source of candidate evaluations. The issue of big government is second to partisanship in terms of effect. Whether or not people have a Democrat as their representative strongly influences their preference over the two parties’ candidates in 1998. Living in the south will significantly increase the level of supporting the Democratic candidates, but being

black does not necessarily lead to higher evaluation of the Democratic candidates—the race variable has a significant effect only on the relative assessment in the 1998 election. I speculate that the residence area variable weakens the effect of race.

Instead of exploring the two-way relationship between the two short-term forces, I used OLS equations to find out the determinants of the difference in candidate evaluations. Overall, people carry their perceptions toward issues when they make their judgment on the candidates. When they support “big government,” they tend to agree with the Democratic candidates. When they oppose “big government,” they would praise the Republican candidates. The strongest determinant is whether or not the incumbent is from the Democratic

Party. If an individual is exposed to the Democratic incumbent, she will mention one more good things about the Democratic candidate than the Republican candidate. The one-way relations between candidate evaluations and issue positions suggest that candidate evaluations are responsive to issue positions, controlling for partisanship, incumbent’s party, and the other demographic variables.

Before assessing the effect of each independent variable, there is one more step to take. National conditions, such as presidential popularity, Congress approval and economic conditions are often used to predict the aggregate-level outcome of Congressional elections. I assume that those considerations would influence the actual relationship between voting choices and the independent variables mentioned above. Among the national-level factors, economic concern attracts much attention regarding both presidential and congressional elections.

3.8 Evaluation on National Economy in Congressional Elections

The economic model of voting states that the voter would opt for the party that can improve social well-being (Downs, 1957). The level of satisfaction with the national economy is related to congressional voting behavior, because economic downturns would induce negative attitudes toward the government

(Campbell, Converse, Miller, and Stokes, 1960; Bloom and Price, 1975). The underlying assumption is that voters hold the president accountable for economic situations. The relationship between evaluations of personal and national economy, and political behavior has been explored with cross-section data (Kramer, 1971;

Tufte, 1975; Kinder and Kiewiet, 1981; Feldman, 1982; Markus, 1986). It is apparent that voters take their evaluations of economic situations into account, particularly when the economy is not in a good shape (Bloom and Price, 1975).

Although congressional elections are characterized as choices between the two candidates in a given district, the performance of the administration, particularly the state of economy, has both a direct and an indirect influence on voting behavior.

First of all, congressional voting is responsive to how people feel about the national economy (Tufte, 1975; Kinder and Kiewiet, 1981). Dissatisfaction with income or unemployment of the society can contribute to change in preference for political parties. The president’s party would be blamed in House elections for economic downturns. Secondly, quality challengers are not likely to appear when the administration has favorable political and economic conditions (Jacobson and

Kernell, 1983). The economic conditions and, as a result, sentiments toward the government sometimes are so overwhelming that the electorate wants to send a

signal to the president’s party by casting their ballots against it. It is almost a trend that the president’s party will lose the mid-term election, in part because of the national economy.

There are four types of evaluation on economic conditions: prospective evaluation of the national economy, prospective evaluation of personal financial situation, retrospective evaluation of the national economy, and retrospective evaluation of personal financial situation (MacKuen, Erikson, Stimson, 1992).

Among them, retrospective judgment of the national economy is widely recognized as the most important one (Fiorina, 1981; Kinder and Kiewiet, 1981;

Feldman, 1982). Here we employ the “sociotropic” voting theory and assume that people focus on the current economy in making their judgment about the performance of government (Downs, 1957). It is also expected that personal grievance, albeit proximate to individual perceptions of politics, is less influential on voting decisions than judgments of the national economy. Therefore, retrospective evaluations of the national economy will be included in the voting choice model as an exogenous variable. The question is to what degree do the voters reflect their concern about the national economy on their voting choices, holding partisanship and candidate evaluations constant.

3.9 Party Images in Congressional Elections

Party images refer to people’s perceptions of political parties, which can be very general or specific. Using open-ended questions asking people what they like or dislike parties, party images are conceptualized as a cognitive sense toward parties. It is a concept more of substantial evaluation than of psychological

influence (Trilling, 1976). It is a useful notion, first of all, because it considers the negative sense of affect toward a party as well as positive affect (Wattenberg,

1982). In a two-party system, the negative feelings toward one party may result in strong support for the other party (Trilling, 1976).

The second point that makes the consideration of party image desirable is the short-term aspect of partisanship. Party identification is a stable predisposition toward parties but remote from vote choice, while party image is characterized as party evaluations responsive to campaign-related events; people’s sentiment toward parties can be more significant than candidates or issues. For people who have a weak party attachment, party image adds a broad picture of parties to their views toward parties. Party images are different from party identification, and less ephemeral than candidate evaluations or issue positions (Matthews and Prothro,

1966).

In the previous research, the like/dislike questions were employed to measure party image and to summarize it by subtracting positive images from negative images toward a given party (Matthews and Prothro, 1966; Trilling, 1976;

Wattenberg, 1982; Sanders, 1988). The more positive the party image is, the higher is the likelihood that people vote for the party. Although like/dislike questions are useful for a single case study, they are not standardized in the comparative study. For instance, Wattenberg’s (1982) comparison of party image among four countries has different number of responses about party images.

Moreover, the question wordings are not the same. In the NES study, party images are coded from open-ended questions: “Is there anything you like/dislike

Republicans/Democrats?” In some research on other countries, however, the

responses to a set of structured questions are coded as a scale of certain points to measure party images (Richardson, 1988). In short, they are not as standardized a measurement as party identification. Finally, like/dislike open-ended questions contain more than one issue underlying the responses, and each issue has a different weight on general feelings (Sanders, 1988).

Wattenberg (1984) suggested that the feeling thermometer of parties is a useful rating of evaluations of parties. The feeling thermometer is a score that ranges from 0 to 100 and 50 is the neutral point. Weisberg and Rusk (1970) also used the feeling thermometer to assess how warm or cool people feel about candidates. They found that the thermometer question measures favorable/unfavorable evaluation as well as saliency of candidates. Weisberg and

Rusk (1970) argued that both sources of evaluations are distinguishable and that the like/dislike rating can be obtained from the thermometer question. Since the feeling thermometer taps into broad party evaluations and they are comparable across countries, I operationalize party images as the score of 0-100 rating for each party, even though thermometer scores reflect affect rather than cognition.

Figure 3.5 shows party images of the two parties in the 1990 and the 1998 election.

70 60.9 59.6 54.6 60 49.9 50 Democratic Mean 40 Party Value 30 Republican 20 Party 10 0 1990 1998 Year

Figure 3.5: Mean values of party images, 1990 and 1998

According to Wattenberg’s (1984) research, the Democratic Party enjoys better party images than does the Republican Party and the difference has been fairly consistent through the 1960s to the 1980s. What Figure 3.5 shows conforms with Wattenberg’s finding. Moreover, the average ratings of both parties range from 50 to 60, which also is the extension of the trend from the 1980s.

Because the feeling thermometer is simply a summary of positive and negative party images, it merely indicates how people feel about two parties on average. To operationalize party images as a predictor to vote choice, we need to know the association of party images between the two parties. If most people see the two parties in different way, then the feelings toward the two parties should be handled separately. Furthermore, it is necessary to look at whether party attachment correlates with party images.

Republican Images Democratic Images Partisanship

Republican Images 1.000 ------

Democratic Images -.1373 1.000 ---

Partisanship -.4922 .5636 1.000

Data: 1990 NES data Note: Table entries are Pearson correlation coefficients.

Table 3.6: Correlation coefficients of partisanship and party images, 1990

Republican Images Democratic Images Partisanship

Republican Images 1.000 ------

Democratic Images -.3277 1.000 ---

Partisanship -.5894 .6212 1.000

Data: 1990 NES data Note: Table entries are Pearson correlation coefficients.

Table 3.7: Correlation of partisanship and party images, 1998

According to Table 3.6 and Table 3.7, the Pearson correlation coefficients are -.1373 and -.3277 for the 1990 and the 1998 election respectively. That suggests that in both congressional elections people’s attitude toward one party is only modestly related to the other one. Regarding the association between

partisanship and party images, Table 3.6 shows that the Pearson correlation coefficients are -.4922 and .5636. In other words, partisanship can explain one quarter of variance in party images. Table 3.7 also shows a high degree of association between partisanship and party images; the correlation coefficients for the three indicators are -.5894 and .6212. The results suggest some multi- collinearilty among partisanship and party image variables, which will be addressed later.

3.10 Multi-Nomial Logit Model of Congressional Voting Behavior

3.10.1 Model

The voting model used in this dissertation considers abstention as one alternative. By including abstention, the estimators with regard to the influence on the categorical dependent variables would be more consistent (Lacy and Burden,

1999). Moreover, there will be more cases included in the model to avoid selection bias in the dependent variable. Besides, the turnout rate is relatively lower in the U.S. congressional elections than in Taiwan and Japan. Treating abstention as a choice can facilitate the comparison of these three countries. The theoretical basis of this treatment of abstention is considered more in Chapter 6.

Here I assume that the status quo is one option facing the electorate.

Voters opt for either the Democratic candidates or the Republican candidates not only because they want to express their preferences, but also because they are not satisfied with the status quo. This assumption makes sense particularly in congressional elections because there usually are no national issues or no national leaders calling for voters’ attention. Abstention is an alternative when individuals

do not want to change the status quo, which usually favors the incumbents.

Therefore, a unified model is needed to explain voting choices as well as abstention.

McKelvey and Ordeshook (1972) modeled n-candidate contests as the following form:

Ek-E0=Pk1Bk1 + Pk2Bk2 +…+ PkiBkd ---(3.2)

where Ek is the expected utility for candidate k if she gets elected; E0 is the expected utility associated with abstention; Bki is the utility difference between candidate k and candidate d (d,k ∈n; d≠k); Pki is the probability that the vote is decisive between candidate k and i.

According to Equation 3.2, we can deduce that the two-candidate contest is determined by the utility differential between two candidates, with abstention as the comparison group. In other words, we can rewrite Equation 3.2 as follows.

E2-E1=f(Ui2-Ui1)---(3.3)

where i denotes voter i (proof is given in Appendix).

If E2-E1>0, individuals will vote for party 2. Likewise, individuals will vote for party 1 when E2-E1<0. I assume that there is no tie in the expected utility of two parties.

Therefore, Equation 3.3 can be rewritten as follows.

V21= f(Ui2-Ui1)---(3.4)

where V21 denotes vote choice between party 1 and party 2

Substantially, voting choices among two candidates and abstention is the function of utility comparison between the two candidates, controlling for the other variables. In other words, three-choice model can be estimated with

Equation 3.4.

Basically, we view the categorical dependent variable as the observed outcome of a latent attitude. On the surface, the observed outcome is discrete and limited, thus we have to transform the limited value to continuous in order to obtain the estimators of the independent variables. In other words, we want to estimate the coefficients given that we can observe certain values of the dependent variable. In this study, there are three categories in the dependent variable: voting for the Democrats, voting for the Republicans, and abstention. I assign abstention as 0, voting for the Democrats 1, and voting for the Republicans 2. The multi- nomial model is set up as follows (Long, 1997).

Pr(Y=m X)= Xβ' m+µi---(3.5)

β' m is a vector of m parameters, m is one of the alternative in the set of M alternatives, X is a vector of observable exogenous variables, and Y is a nominal

2 variable with J outcomes. µi is the disturbance term with distribution N(0, σ ).

One of the well-known underlying assumptions of multi-nomial model is that the probabilities of choosing alternative m to alternative m' is independent of

the probability of choosing the other alternatives.35 In this project, the ratio of the probability of choosing the Democratic Party to the probability of choosing the

Republican Party is unchanged by the availability of abstention. If this assumption holds, maximum likelihood estimation in multi-nomial logit model will generate consistent estimates of the parameters (Alvarez and Nagler, 1995; Alvarez and

Nagler, 1998). Based on our prior information, this behavioral assumption holds and multi-nomial model can be employed to estimate the effect of explanatory variables.

According to Equation 3.4 and Equation 3.5, I model the probability of voting choices as a linear equation of the utility differential and other controlled variables. I use the measurement of comparative candidate evaluations to represent utility differential. In some senses, comparative candidate evaluations are an instrument variable of political attitude variables and demographic variables. However, there is no two-stage multi-nomial logit model for this purpose. To avoid biased estimation by using two-stage ordinal logit model, I treat comparative candidate evaluations along with the other factors as exogenous variables and use the multi-nominal logit model.

In addition to the utility variable, I include partisanship, issue positions of big government and individualism, party images, incumbent’s party, retrospective economic evaluation, and the other demographic variables in the model.

Partisanship and incumbent’s party should be strong predictors. Issue positions are also expected to influence voting decisions, whereas the effects are not as strong

35 To keep the probability nonnegative, we transform the right-handed side of the equation into exponential form: exp (Xβ' m). The probabilities of choosing the m alternative from a set of M alternatives are given by Pm=exp(Xβ' m)/Σexp(Xβ' M). Therefore, Pm / Pj = exp(Xβ' m)/ exp(Xβ' j)

as partisanship and incumbent’s party are. Party images are included because they are the short-term aspect of party identification that reflect how people feel about parties in the short-run. However, they are also correlated with party identification, thus the interpretation of the coefficients should be taken with caution 36 .

Retrospective economic evaluation is usually a good predictor of voting choices at the aggregate and individual level, and it can be influential when we model voting choice as utility comparison. To avoid the specification problem, I controlled for some demographic variables, such as residence area, age, race, occupation, and education. Those demographic variables may have occasional impact on the dependent variable, but the effects may not be across the board. That is because of the intervention of political attitude variables, which are proved to have major influence on political behavior.

Before moving on to the estimation of the multi-nomial logit model, I ran the descriptive statistic analysis for the explanatory variables. Table 3.8 displays the demographic characteristics of the sample, how they rate the two issues, and their partisanship.

36 Section 3.9 has already mentioned the Pearson correlation between partisanship and party images. I also computed the partial correlation coefficients between partisanship and party images. For the 1990 election, the partial correlations between partisanship and the Democratic party images and the Republican party images are .508 and -.485 respectively. For the 1998 election, the partial correlations between partisanship and the Democratic party images and the Republican party images were .466 and -.444 respectively.

Variables 1990 1998 Party Identification (-3 to 3) .398 .335 Candidate Evaluations (-5 to 5) .370 .035 Big Government Issue (1 to 7) 4.393 3.965 Individualism Issue (1 to 7) 4.142 3.607 Ideology (1 to 7) 3.240 .148 Republican Party Images (0 to 100) 54.63 49.93 Democratic Party Images (0 to 100) 60.98 59.64 Incumbency (0 to 1) .673 .354 Economic Retrospective (-1 to 1) -.693 .324 White (0 to 1) .840 .851 Male (0 to 1) .452 .448 South (0 to 1) .349 .366 East (0 to 1) .195 .165 West (0 to 1) .192 .206 Elementary Education (0 to 1) .587 .433 Secondary Education (0 to 1) .213 .285 18-35 years old (0 to 1) .379 .300 36-50 years old (0 to 1) .264 .330 51-65 years old (0 to 1) .178 .202 Manager (0 to 1) .079 .088 Professional (0 to 1) .267 .294 Labor (0 to 1) .178 .129

Table 3.8: Means and ranges of the independent variables, 1990 and 1998

3.10.2 Results

The main proposition of this chapter is that candidate factors have relatively higher influence than partisanship on American’s voting decisions.

Comparative candidate evaluations and incumbent’s party are employed to measure the influence of candidates in the House races, which are more localized and more candidate-oriented than presidential elections. Under the context of the presidential system and single-member district, the incumbency factor and the candidate factor should prevail in the Congressional elections. The influence of

the American political parties remains steady and goes through candidate evaluations. Voters’ issue stands, however, may not have impressive impact on voting choices, since there are few national issues in these two congressional elections because partisan confrontation is limited. National survey data that measures individual’s political attitude and codes the status of the incumbent will reveal the voting pattern of the electorate.

Note that the dependent variable is the nominal choice among voting for the Democratic candidates, the Republican candidates and abstention, with abstention the comparison group. Therefore, the logit coefficients in the model indicate the extent to which the independent variable can change the probability that voters turn out to opt for the Democratic or the Republican candidates against abstention. Table 3.9 and Table 3.10 show the estimated coefficients for the two elections, and Table 3.11 and Table 3.12 the corresponding change of the probability caused by the unit change of the selected variables.

Upon first glance at Table 3.9, partisanship, candidate evaluations, and incumbency are the three most important variables for the 1990 vote choice.

Comparative candidate evaluations and incumbent’s party exert significant influence on the probability. Partisanship also has the right sign and is significant as to whether people vote for the Democratic Party or the Republican Party.

People’s issue positions on both issues are not as important a factor as candidate characteristics and partisanship. Up to this point, the findings accord with

Jacobson (2001), which stresses that vote choice depends on the level of partisanship, candidate evaluations, and especially incumbency.

Retrospective evaluation of the national economy is not significant, though it has the right sign. Most of the demographic variables do not exhibit any significant effect on voting choices, except education and age. Elderly and educated people are more likely to turn out than younger and less educated people.

Surprisingly, only the influence of race has the right sign regarding the

Republican vote in 1990. Residence area and occupation are not good predictors, either; their influences are only significant occasionally.37

37 To double-check the estimators, I ran binary logit analyses for each year. The results resemble what I got from the multi-nomial model.

Democratic Republican Constant -.865(.770) -2.580(1.089)* Democratic Partisanship .244(.063)*** -.174(.084)* Democratic Candidate Evaluations .453(.066)*** -.683(.112)*** Big Government Issue -.052(.060) .000(.080) Individualism Issue .016(.053) .184(.074)* Ideology .054(.409) .001(.068) Republican Party Images -.006(.004) .012(.006)* Democratic Party Images .013(.005)* .000(.006) Democratic Incumbency .859(.240)*** -1.115(.251)*** Evaluation of Economy -.206(.171) .049(.214) White .219(.244) 1.262(.539)* Male -.289(.188) -.020(.240) South -.713(.228)** -.429(.285) East -.995(.276)*** -.572(.345) West -.275(.245) -.347(.301) Elementary Education -.821(.245)*** -.694(.293)* Secondary Education -.141(.248) -.241(.304) 18-35 years old -1.474(.299)*** -1.531(.383)*** 36-50 years old -1.054(.310)*** -.516(.387) 51-65 years old -.344(.321) -.034(.392) Manager .315(.340) .077(.381) Professional .280(.235) .326(.291) Labor .100(.286) .403(.383) Number of Observations 1021 Log Likelihood -690.187 Chi-Square 670.11*** Pseudo-R-square .3268

Data: 1990 NES data Note: Coefficients are maximum likelihood estimates with standard errors in parentheses. Abstention is the comparison group. *: p<=.05, **: p<=.01, ***: p<=.001.

Table 3.9: Multi-nomial logit estimates for three-choice model, 1990

Table 3.10 displays the corresponding result for the 1998 election. Partisan strength positively affects the probability of voting for the Democratic Party, but not the Republican Party. Nevertheless, the difference in candidate evaluations simultaneously influences the likelihood of the Democratic and the Republican vote. So does incumbency. Party images have more significant effect on the

Republican vote than the Democratic vote, which may account for the weak impact of partisanship on the Republican vote. Economic evaluation is not an important predictor; the electorate cannot blame either party for the booming economy. In terms of demographic variables, southerners, low educated and young voters are less likely to vote for either party in 1998.

Democratic Vote Republican Vote Constant .196(.817) .792(.788) Democratic Partisanship .329(.091)*** -.026(.076) Democratic Candidate Evaluations .432(.086)*** -.543(.089)*** Big Government Issue .013(.066) -.131(.062)* Individualism Issue -.039(.056) .066(.052) Ideology .266(.161) -.144(.156) Republican Party Images .003(.005) .020(.005)*** Democratic Party Images .000(.006) -.015(.005)** Democratic Incumbency .801(.243)*** -.762(.257)** Evaluation of Economy -.233(.158) -.028(.145) White .152(.332) -.198(.378) Male .059(.228) -.416(.212)* South -1.552(.317)*** -.576(.274)* East -.384(.330) -.120(.325) West -.184(.311) -.082(.312) Elementary Education -1.331(.299)*** -1.064(.268)*** Secondary Education -.548(.279)* -.473(.255) 18-35 years old -2.555(.392)*** -1.862(.347)*** 36-50 years old -1.394(.340)*** -1.098(.329)*** 51-65 years old -.637(.352) -.049(.336) Manager 1.613(.398)*** 1.067(.372)** Professional .237(.280) .447(.248) Labor .103(.390) .406(.353) Number of Observations 896 Log Likelihood -583.981 Chi-Square 649.68*** Pseudo-R-square .3572

Data: 1998 NES data Note: Coefficients are maximum likelihood estimates with standard errors in parentheses. Abstention is the comparison group. *: p<=.05, **: p<=.01, ***: p<=.001.

Table 3.10: Multi-nomial logit estimates for three-choice model, 1998

To demonstrate the relative weights of the major explanatory variables on the voting choices, I calculated the discrete change in the dependent variable caused by changing values of the independent variable. Table 3.11 displays the great role played by the candidate assessment variable in 1990. A standard deviation change in the candidate evaluations difference centered around the mean will increase the probability of voting for the Democratic candidates by .04, holding all other variables constant, which is greater than the effect of the other selected variables. Likewise, comparative candidate evaluations will drop the probability of voting for the Republican contenders by .007. For the effect of partisanship, the probability of voting for the Democratic Party and the

Republican Party will increase by .029 and decrease by .002 respectively due to a one standard-deviation change in partisanship. Republican party images and

Democratic party images influence the likelihood that people vote for Democratic

Party; the discrete changes in are -.008 and .016 respectively. However, they have nearly zero impact on the Republican vote. Whether or not the voter has a

Democratic incumbent in his district also affects his voting for either party to a great degree. When a Democratic incumbent runs for re-election, the probability of Democratic voting will increase by .035 but that of Republican voting drops by .001. Over all, the most two important determinants of vote choice for the

Democratic Party are incumbency and candidate evaluations. Partisanship runs in third because candidate evaluations combined with incumbency largely shape people’s perception toward candidates. As to the Republican Party, relative candidate evaluations have an impact.

The results of Table 3.12 also indicate the importance of incumbent’s party as well as comparative candidate evaluations in 1998, particularly for the

Republican candidates. An increase in candidate evaluations by one standard deviation will increase the probability of voting for the Democratic Party versus abstention by .002 and will decrease that of voting for the Republican Party versus abstention by .012. Party attachment has a similar magnitude of effect on the likelihood of voting for Democratic candidates, but only a minor impact on that of voting for Republican ones. Party attachment is not an important factor across the board as in the 1990 election, which indicates the decline of party loyalty in the congressional elections (Jacobson, 2001). Incumbency remains critical to voters’ judgment on the candidates. When there is a Democratic incumbent running for office, the probability of voting for the Democratic Party nearly remains the same but the probability of Republican voting drops by .016. The same situation applies to the effect of party images. Regarding party images, Republican candidates enjoy the advantage brought by better party images, but better party images for the

Democratic Party barely influence the likelihood of voting for the Democratic

Party. The negative impact of Democratic party images on Republican voting is -

.005. In summary, the most important determinant of vote choice for the

Democratic Party and the Republican Party is candidate evaluations.

Democratic Vote Republican Vote Variable Centered Standard Centered Standard Deviation Change Deviation Change Democratic Partisanship .029 -.002 Democratic Candidate .040 -.007 Evaluations Big Government Issue -.004 .000 Individualism Issue .001 .002 Ideology .006 .000 Republican Party Images -.008 .001 Democratic Party Images .016 .000 Democratic Incumbency* .035 -.001 Evaluation of Economy -.006 .000

Note: Centered standard deviation change is calculated as follows. (Long, 1997) Because [p(y=m)/1-p(y=m)]=exp(Xβ'), thus p(y=m)=[exp(Xβ')/1+ exp(Xβ')]. Let th p1(y=m|x, xi+½si ) the predicted probability that i variable at its mean plus one- half of its standard deviation, holding the dummy variable 1 and all other th variables at mean, and p2(y=m|x, xi-½si ) the predicted probability that i variable at its mean minus one-half of its standard deviation, holding the dummy variable 1 and all other variables at mean. The difference between pi1 and pi2 is the centered standard deviation change. *: Because the incumbency variable is a dummy variable, whose value ranges from 0 to 1, the discrete change is calculated as p1(y=m|x, 1 )- p2(y=m|x, 0).

Table 3.11: Discrete change in the probability of Democratic and Republican vote, 1990

100

Democratic Vote Republican Vote Variable Centered Standard Centered Standard Deviation Change Deviation Change Democratic Partisanship .003 .000 Democratic Candidate .003 -.012 Evaluations Big Government Issue .000 -.004 Individualism Issue .000 .002 Ideology .001 -.002 Republican Party Images .000 .007 Democratic Party Images .000 -.005 Democratic Incumbency .002 -.016 Evaluation of Economy .000 .000

Note: the calculation of the centered standard deviation is same as Table 3.11

Table 3.12:Discrete change in the probability of Democratic and Republican vote, 1998

Overall, multi-nomial logit estimates provide a clear picture: candidate evaluations are critical to voting choice in the House elections and the importance of partisanship varies with time. Partisanship bears upon candidate evaluations; it facilitates the perceptions of candidates. Nevertheless, its influence is smaller than candidate evaluations for both parties in the 1990 election and for the Republican

Party in the 1998 election. Individuals vote for either Republican or Democratic candidates primarily because they recognize the candidates, which incumbency contributes to in some degree. Party images also have a relatively strong impact on vote choice, particularly for the majority party in the Congress. The

Democratic Party in 1990 and the Republican Party in 1998 enjoyed the power of

101

agenda setting, thus they were exposed to mass media more than their counterparts. Issue stands, ideology, and national economic concern have only marginal effects at best.

3.11 Conclusion

One of the premises in American congressional voting research is that incumbents are preferred because of their visibility and the lack of national factors in congressional elections. The context of congressional elections—campaign money, campaign organizations, incumbent privileges—is favorable for incumbents (Jacobson, 1997). Cain, Ferejohn, and Fiorina (1987) maintained that the separation of powers also contributes to the constituent factor in congressional voting, though they also found the personal vote in Britain. Instead of looking at incumbent voting, I investigate the determinants of party choice in two congressional elections. In doing so, national factors such as partisanship, ideology, and political issues are weighted with local factors, including candidate evaluations and incumbency.

Presidential systems embrace candidate-centered pattern because the separation of powers institutionally emphasizes the cooperation of the executive and legislative branches, which allows members of Congress to manipulate policies for their constituency interest. The source of candidate evaluations is more on the basis of personal traits and district services than ideology or party label. In other words, the presidential system encourages voters to use candidate characteristics as the evaluative standard.

102

The SMSP system used in the U.S. encourages party competition because each party nominates only one candidate. Despite that primaries open the nomination to the citizens and weaken party organizations’ control over candidates, the SMSP system maintains the importance of party label in campaigns. Partisanship, a long-term psychological attachment in the U.S. context, still colors people’s candidate evaluations and issue positions.

Hinckley (1981) claimed that partisanship and incumbency can nearly explain all of the variation in vote choice and that candidate evaluations have limited impact on vote choice when partisanship and incumbency are identified.

Brown and Woods (1991) also found the impressive effects of partisanship and candidate evaluations on the choices of voters. A model that includes all of the three voting decision—voting for the Republicans, for the Democrats, and abstention—confirms the importance of party identification and incumbency.

More importantly, it considers the effect of candidate evaluations while controlling for the incumbency status.

The result of the analysis shows that partisanship is significantly associated with voting choice. It also confirms the hypothesis that candidate evaluations and incumbency are critical to congressional elections. Party images are salient to the electorate for the party in power and national issue positions are remote, if not being put aside, from the final voting decision. After all, the increasing relative influence of candidate evaluations in these two elections demonstrates the prevailing candidate-centered pattern in U.S. congressional elections.

103

Appendix

The purpose of proof is to show that the two-candidate contest is the function of the utility differential of two candidates.

According to Equation 3.2, E2-E0=P12(U2-U1)

and E1-E0=P21(U1-U2)

Supposed that U=f(U)

c.d.f. of f(·)=1

E2-E1=P12 ·f(U2)- P12 ·f(U1)- P21·f(U1)+ P21·f(U2)

=( P12+ P21)( f(U2)- f(U1))

Because P12=1- P21, therefore

E2-E1=( f(U2)- f(U1))

Besides, f(a)- f(b)= f(a-b)

E2-E1= f(U2-U1) □

104

CHAPTER 4

TAIWANESE LEGISLATIVE VOTING BEHAVIOR IN 1992 AND 1995

4.1 Introduction

Taiwan has a very different tradition of party competition than the United States.

Because of the Temporary Provisions of the Constitution, only part of the seats in the two legislative bodies--the National Assembly and the Legislative Yuan—are subject to reelection. Those members elected in Mainland China by no means represented the residents in Taiwan, but they were used to justify the legitimacy of the KMT government. Though not unanimously supporting every public policy, they served as the symbol of constitutional democracy in contrast to the communism politics of

China. With the constitutional arrangement, the KMT government survived the setback of international relations and uplifted economic development with success in the 1960s.

Two phenomena beneath the surface of rapid economic growth and stable administration are noteworthy. First, the institutional arrangement granted the one- party politics legitimacy, but changing society and international environment resulted in a crisis of legitimacy. The one-party regime went to its limit in the 1970s, when a larger middle class emerged with the takeoff of the economy and the U.S.

105

attempted to court China against the former Soviet Union. Secondly, the

Temporary Provisions of the Constitution froze popular elections for the legislators, but the opposition movement gained strength from local elections and supplementary elections for the legislative bodies. As more and more elites questioned the anti-democracy nature of the Temporary Provisions, the ruling party could no longer resist the demand of sweeping political reform. Responding to the pressure of the opposition movement, the KMT opened up more seats but could not stop the downward trend of its popularity.

In this chapter, I will investigate Taiwanese voting behavior in the 1992 and 1995 legislative election. These two elections will allow us not only to examine the determinants of voting behavior, but also to look at change in the relative influence of the determinants.

The reasons why I chose these two elections are twofold. First of all, the

1992 election was the first one in which all of the seats in the Legislative Yuan were open; none of them was reserved for the representatives elected in China before 1949. 38 The 1992 election, therefore, opened a new era of party competition and legislative politics (Liu, 1996). Second, the 1995 legislative election was held a few months before the first popular presidential election in

1996. Both the presidential candidates of the ruling party and the opposition parties traveled around to endorse legislative candidates. Being mindful of the coming presidential election, the opposition parties geared up to vie for every seat.

38 In 1969, 11 new members who served life-time term were elected in Taiwan and overseas Chinese communities. In 1971, an additional 51 seats were added to the Legislative Yuan. 15 out of the 51 seats were from overseas Chinese communities. In 1980, the number of new legislators was increased to 97, of whom 27 were assigned to overseas Chinese. In 1986, 74 legislators were elected in Taiwan. Since 1992, PR system has been adopted. In 1992, 36-30 seats are elected reserved for the nationwide constituency and 6 for overseas residents.

106

Besides, they warned their nominees not to accept any offer made by the KMT, including cabinet-level positions. Although the 1998 election was the first one after the 1996 presidential election, unfortunately, the survey data cannot provide us with compatible candidate evaluation measurement. Therefore, 1992 and 1995 election data will be used in this project.39

Section 4.2 will delineate the evolution of the party system and the most important issue: national identity. Section 4.3 will discuss the social issue. Section

4.4 analyzes party identification in Taiwan. Section 4.5 continues to explore candidate evaluation and its effect on vote choice. Section 4.6 focuses on the short-term aspect of party identification—party images. Section 4.7 gives a summary of data and variables and presents the voting choice model and results.

Section 4.8 concludes this chapter.

4.2 Party System and the Unification Issue

Along with the series of constitutional reforms starting in 1986, Taiwan has undergone a series of changes in its party system. In 1989, the then opposition party, the Democratic Progressive Party (DPP), made a breakthrough by winning

18 out of 78 popular seats. Three years later, the DPP improved to 39 out of 116 district-level seats when all of the seats were open to both district-level and group- level elections. In 1994, the China New Party (CNP) competed with the DPP and

39 Data analyzed in this chapter were collected by the research project of “A Study of Voting Behavior and Political Democratization in Taiwan: the 1992 Election for the Members of Legislative Yuan (Post Election)” and “A Study of Voting Behavior and Political Democratization in Taiwan : the 1995 Election for the Members of Legislative Yuan”, and directed by Chen Yih- yan. The Election Study Center of National Chengchi University is responsible for the data distribution. The author appreciates the assistance in providing data by the institute and individuals aforementioned. The views expressed herein are the author's own.

107

the KMT in the Taipei mayor election, and began an era of multi-party system. In the 1995 legislative election, the DPP gained 41 out of 115 seats, but the CNP shocked the country by securing 15 seats. The fledgling three-party system consolidated in the 1996 presidential election, in which four sets of presidential and vice-presidential candidates vied for the office and the KMT nominees won it with 54 percent of votes. The fierce three-party competition continued in 1998, as the size of the Legislative Yuan increased to 166 seats. The CNP lost 8 seats but the DPP gained 11 more seats. Although the KMT retained its majority in the

Legislative Yuan throughout the 1980s and 1990s, its decline in vote shares is evident. Table 4.1 shows the change of vote shares and seat shares from 1989 to

1998. In 1989, the KMT had 54 out of 78 district-level seats, which is a 69% seat share (not shown in Table 4.1). In 1998, however, the seat share was 53%; the

KMT dropped 16% of seats. The change in the percentage of vote is also against the KMT. The DPP has kept 30% of vote share for four elections, while the KMT lost nearly 14% of vote shares. The CNP gained strength in the 1995 election, but it suddenly lost 8 seats three years later. The competition between the DPP and the

KMT was the main axis of the party system during the 1990s.

108

1989 1992 1995 1998 KMT 54 (59.33%) 66 (52.68%) 59 (45.89%) 88 (45.87%) DPP 18 (29.91%) 39 (32.23%) 41 (33.58%) 52 (30.05%) CNP 0 0 15 (1.72%) 7 (7.18%) Independent 6 (10.76%) 11 (15.09%) 4 (18.81%) 19 (16.90%)

Data: Compiled by the author from the Databank of Election Study Center Note: Parentheses are the percentage of vote. Only district-level seats are counted.

Table 4.1:Seat shares and vote shares of political parties in the legislative elections, 1989-1998

The three-party system in 1992 was founded on two decisive issues: democratization and national identity. The KMT, which fled to Taiwan in 1949, claimed that Taiwan is part of China and that Taiwan will reunify with it when it abandons communism. The two KMT presidents, Chiang Kai-shek and Chiang

Ching-kuo, established an authoritarian regime, using the Temporary Provisions to constrain freedom of assembly, speech, and political organization. Although the

KMT held elections to show off the democratic form to the Western world, it actually prohibited any organized opposition force from emerging. In the elections, however, more and more local elites, idealistic or power-oriented, appeared and trained themselves. They attacked the privilege enjoyed by the KMT and the political arrangement as a whole, demanding sweeping political reform and the alternation of power. Also, the sweeping democratization wave in the 1980s pressed the KMT to change its iron-fisted policy. When the U.S. resumed diplomatic relations with the People’s Republic of China in the 1970s and cut off its official diplomatic relationship with Taiwan in the early 1980s, moreover, the legitimacy of the KMT administration diminished (Wang, 1989). Facing with

109

demands for democracy from society and the global democratization tide, the

KMT accepted the organization of the DPP in the fall of 1986, which led to the end of the four-decade one-party politics.

The DPP was founded by a group of lawyers and representatives in 1987.

Since the 1970s, the opposition elites, mindful of martial law, maintained strong ties in the campaigns without formal party organizations. They called themselves

“tangwei”, which means “outside KMT”, and shared beliefs in speech freedom and democratization. Thanks to the national and local elections, the opposition movement learned how to mobilize people with loose organizations and platforms.

Political liberalization was the common belief shared by the opposition activists in the first place, but the DPP members realized that the issue of independence could distinguish them from the KMT (Lin, 1989). First of all, the legitimacy of the

KMT government came from the claim that Taiwan is part of China and the KMT is the only legal political party that took over Taiwan in 1945. Second, 85 percent of the residents and their families immigrated to Taiwan back to the 18th century.

They speak Taiwanese and have a similar but distinct culture from China. When

Japanese colonial government ruled Taiwanese between 1891 and 1945, they held local elections and even sent Taiwanese to the upper parliament of Japan. The separation of Taiwan and China for five decades resulted in different cultural, economic, and political experiences. The 228 incident--the KMT government in

China sent troops to arrest and to execute native elites who criticized the agency that took control of Taiwan before the KMT’s fall from power in 1949--sharpened the conflict between Taiwanese and mainlanders. After the DPP was established, the pursuit of the sovereignty and national identity for Taiwan became its primary

110

agenda. In its party platform, the DPP not only emphasizes democratic reforms, such as reelecting the representatives of the three national legislative bodies and abolishing martial law, but also calls for the self-determination for Taiwanese residents. The DPP challenged the KMT’s legitimacy, citing the principle of self- determination (Lu, 1992; Wang, 1994). As a result, the lingering democracy issue gradually transformed into a national identity issue.

The ethnic cleavage not only distinguished the DPP and the KMT but also cut across the KMT factions. The KMT argued that every opposition party should respect the 1947 Constitution and the Temporary Provision, which claim that

Taiwan is part of China. But the DPP held the KMT accountable for the international isolation due to KMT’s claim to sovereignty over China. As the debate continued, more and more KMT Taiwanese members lent their support to democratization and Taiwan nationalism, but mainlander members tried to preserve the goal of unification and authoritarianism. Lee Teng-hui, who succeeded Chiang Ching-kuo in 1988 as the first president born in Taiwan, received support from the DPP in removing the hard-line KMT members. Some young KMT legislators, who found themselves being marginalized by Lee Teng- hui’s cooperation with the DPP and KMT’s local factions, founded the Chinese

New Party (CNP). The CNP argued that the KMT under Lee’s leadership had became corrupted, and it vowed to promote unification with China.

The surveys that were administered by the Mainland Affairs Council make it clear that Taiwan is divided on the unification-independence issue. Figure 4.1 shows that most people supported unification in 1991 and 1992, and that people who prefer the status-quo rather than any radical change are about one-third of the

111

population. The proportion of people who support independence has been increasing, though the number of no-opinion people has also gone. In 1995, the number of people who support independence and unification was nearly the same, and no-opinion and status quo advocates account for another one-third of population. This indicates that Taiwanese are still engaged in this complicated cross-straits issue.

50 45.3 44 Independence 40 33.2 Unification Status Quo 30 25.3 30.4 23 23 No Opinion 20 16.5 17 Percentage 14.6 13.4 10 12.5 0 1991 1993 1995 Year

Data: Compiled by the author from You (1996)

Figure 4.1: Attitudes toward unification and independence, 1991-1995

Another indicator of the cleavage is whether people identify themselves as

Taiwanese or Chinese.40 The separation of Taiwan and China in 1949 and the

40 There are three very similar conceptions. The first one is sub-ethnic division, which means whether people are descendants of the mainlanders or native Taiwanese, which consists of Hakka and Min-nan groups. Politically, mainlanders strongly favor the KMT or the CNP, Hakka modestly the KMT, and Min-nan strongly the DPP. Usually we measure it by asking the respondent if his father is mainlander or native Taiwanese (either Hakka or Min-nan). For clarity, I will call it “ethnicity”. The second conception is “Taiwanese” and “Chinese”. In some research, they are treated as two ethnic groups and measured by the question reads like: “Would you call yourself (or people in Taiwan) Taiwanese or Chinese or both?”. See You (1996). But I view them as two

112

following confrontation between the two countries sharpened the line between the two ethnic identities.41 The two terms-- “Chinese” and “Taiwanese”-- stand for two ethnicity camps involved in political conflict. According to the results of polls, each of the three categories of ethnic identification receives support evenly (You,

1996). It is also claimed that people who choose “Taiwanese” are likely to support

Taiwan independence, people who prefer “Chinese” tend to support unification, and people who are indifferent to both terms are in favor of the status quo rather than the two extreme options (Shyu, 1996).

With trade and transportation between Taiwan and China increasing, both the KMT and the DPP agreed to negotiate with China on cross-straits economic affairs, but the DPP still insists on the priority of nationalism. In addition, the DPP strongly promotes Taiwanese identity. Hence, the DPP is widely regarded as a pro-independence party (You, 1996). Because of the background of the CNP members, people tend to view the CNP as a strong pro-unification party.

Compared to the CNP, the KMT is perceived as moderate on the unification issue.

Table 4.2 displays people’s perception of those three parties on this issue.

Obviously, the KMT and the DPP are on the two extremes of the issue continuum in 1992 and 1993. After the CNP joined them, however, the CNP took the extreme position and the KMT moved toward the median point of the issue. In the 1996

political groups, since “Taiwanese” or “Chinese” is self-identification. See Hsu and Fan (2001). Finally, some scholars view attitude toward unification-independence issue as “ethnic identity” or “national identity”. See Wu (1996) and Shyu (1996). Here I treat the decision between “Taiwanese” and “Chinese” as “ethnic self-identity” and “independence” and “unification” as “national identity”. The former is related to how people categorize themselves rather than their choices about where their country should go to, and the latter is involved explicitly in whether Taiwan should be nation or not. Both are long-term issues that are responsive to events, elite’s interpretations, and international factors.

113

survey, 33 percent of respondents viewed the KMT as the pro-unification party, which is much less than in 1992 and 1993. However, 36 percent of respondents regarded the CNP as the pro-unification party.

KMT DPP CNP Indepen- Unifi- Indepen- Unifi- Indepen- Unifi- dence cation dence cation dence cation 1992 1% 68% 66% 2% ------1993 2% 64% 66% 3% ------1996 7% 33% 66% 1% 4% 36%

Data: Lin (2000) Note: Entries are the percentage of people who locate political parties as pro- independence or pro-unification.

Table 4.2: Perceptions of parties’ unification issue stands

The unification issue stemmed from the tension between the Mainlanders and native Taiwanese in the 1950s when the KMT wiped out many Taiwanese intellectuals who refused to cooperate with the KMT. The “tanwei” dissidents extended the social cleavage to party competition, attacking the KMT’s legitimacy that is based on its claim on the sovereignty of China. Thanks to the unification issue, the DPP aligns with native Taiwanese and people who support democratization (Hu and You, 1984). The DPP also favors the ethnic mobilization strategy, which reinforces its pro-Taiwanese perception shown in Table 4.2.

41 Here I prefer “ethnic identity” than “ethnic groups” because “Chinese” and “Taiwanese” are not actually groups in terms of anthropology. Most Taiwanese are Chinese descendants; only the aboriginal people who came to Taiwan far earlier than Taiwanese (or Chinese in the 17th century).

114

4.3 Party System and the Socio-economic Equity Issue

In addition to the unification issue, socio-economic equity is a potential social cleavage (Chu and Lin, 1996). The social movement emerged during the

1980s, when economic growth caused grievances for the lower social class. A wide variety of issues and social groups, such as consumers, workers, farmers, veterans, indigenous habitants and people who suffered from industrial pollution, confronted the KMT government (Hsiao, 1992). Responding to the social movements, the KMT softened its control over society, which encouraged the opposition party and mass public to challenge the KMT’s authority. Many intellectuals joined the social movements and the opposition parties.

In spite of their conflict over the unification issue, the DPP and the CNP attacked the KMT on the government corruption and its close tie with the big corporations. The opposition parties asserted that the localized KMT traded its resources with local factions to save its power from collapse. The KMT’s close relationship with big business also led to unequal wealth distribution and corporate fraud. Both opposition parties tried to capitalize on cross-ethnic social issues, but the DPP had better connections with disadvantaged groups, including farmers, labor unions, and indigenous habitants. The CNP, whose party members mainly came from urban areas and had mainlander background, appealed to the middle class and veterans.

Previous studies show how people perceive the social base of the three parties, confirming the existence of the socio-economic equity issue. In Table 4.3, nearly half of the people in 1995 think that the DPP represents poor people, and only around 5 percent of people agreed that the DPP represents rich people. The

115

KMT used to have a double party image on this issue. In 1992, 25 percent of people thought that the KMT represents poor people and another 25 percent think the other way around. After Lee Ten-hui regrouped the KMT, more people viewed it as the wealthy people’s party. In 1998, 46 percent of people think that the KMT stands for the upper class; only 10 percent of respondents still view it as the poor people’s party. The CNP has less perceivable position on the socioeconomic equity issue; 24 percent of respondents think that the CNP represents poor people, and 8 percent think it the other way.

It is evident that the three parties occupy different issue positions on the independence and socioeconomic issues. The DPP is perceived to be the advocate of nationalism and the representative of the poor people, and the KMT is perceived to be close to the wealthy people and unification. The CNP, which vowed to distinguish itself from the DPP and the KMT, is viewed as a pro- unification party, but voters are uncertain about whether it is close to poor or wealthy people.

KMT DPP CNP Wealthy Poor Wealthy Poor Wealthy Poor 1992 25% 23% 5% 36% ------1993 30% 21% 4% 45% ------1996 46% 10% 6% 40% 8% 24%

Data: Lin (2000)

Table 4.3: Perceptions of parties’ representation of social class

According to Chu and Lin (1996), people’s attitude toward ethnic identity and social equity issues, and their partisanship are independent from each other.

116

They ran a logistic model using partisanship, independence issue, social welfare, and money politics to predict voting choice and found that social welfare is far from a good predictor. Partisanship and the unification issue are more important than the emerging socioeconomic equity issue, but we cannot ignore the potential of the new issue in dividing the parties internally (Lin, 2000). 42

Drawing on the analysis above, we can conclude that the unification- independence issue is critical to the party system. After the seats of the Legislative

Yuan were open for reelection and the Temporary Provisions were abolished, the democratization issue was resolved and replaced by the unification issue. The socio-economic equity issue is a potential social cleavage, but it is not as salient as unification. The original ethnic cleavage and the emerging cross-straits affair will keep the unification issue critical to Taiwan politics.

4.4 Party Identification in Taiwan

The conception of party identification developed in the U.S. is new to

Taiwan, where people had no formal opposition party until 1987. Before then, martial law prohibited press freedom and permitted no opposition party to be founded. Under the one-party regime, few independent candidates ran in the local elections, and they labeled themselves as “no party”. In a sense, “no party” is equivalent of “outside KMT” (tangwei), which means that the candidates, if elected, would make their decisions independent from the KMT or even against it.

After the DPP was established, some independent candidates joined the DPP and

42 Both Chu and Lin (1996) and Lin (2000) noticed the coalition in the Legislative Yuan on the two front lines--independence and social welfare issue. They argued that the power struggle among

117

some still ran for office themselves. Until now, independent candidates appealed to the disillusioned voters, who were not satisfied with either the KMT or the DPP.

Figure 4.2 makes it clear that independent voters were the largest bloc, although the number of party identifiers has been increasing since 1990.43 The percentage of Independents has been in decline. It was 53.7 percent in 1987 and it dropped to 30.8 percent in 1996, which is very different from the American case.

In 1995, the number of independent Taiwanese voters is about the same as that of the KMT and the DPP partisans together. The proportion of people who identify with the DPP has been increasing from 7.9 percent to 15 percent, but the proportion of the KMT identifiers decreased from 37.5 to 27.1. Notice that the number of party identifiers is always underestimated because the concept of party identification is not widespread yet. For example, the percentage of the DPP party identifiers was under 20 percent from 1987 to 1996, but it received 30 percent and

35 percent of the popular vote in the 1992 and 1995 legislative elections respectively.

elites along the two issues extends to the electorate. 43 In Taiwan, the question used to construct party identification scale reads like: “In our society, some people say they support the KMT, and some say they support the DPP. Do you support the KMT, the DPP, or other parties?” In the context of Taiwanese society, “support” means “help” or “preference”, which is somewhat different from “identification”. If the respondent picks one party, the next question asks whether the degree of support is “strongly” or “so-so”.

118

60 KMT 53.7 50 45 DPP 39.5 37.8 Independent 40 42.4 37.5 37.5 30.8 CNP 30 32.8 27.1 20 14.1 Percentage 7.9 7.8 7.8 15 10 12.4 7.8 0 1987 1987 1990 1993 1996 Year

Data: Compiled by the author from Lin (2000). Note: The percentages do not add up to 100%. The remaining percentage is “don’t know” or “no opinion”.

Figure 4.2: Distribution of Taiwanese partisanship, 1987-1996

Nevertheless, the numbers shown in Figure 4.2 by no means suggest that

Taiwanese voters lack party identification. To look for evidence of the existence of partisanship, we resort to the conformity between partisanship and the social groups from individual-level survey data. The underlying assumption is that a long-term stable social base is one of the necessary conditions of partisanship. The authors of The American Voter (1960) asserted that individuals would strengthen their partisanship if their social groups conform to the party. Since political parties are the bridge between social groups and the political world, the way that social groups match political parties should correspond to what we understand through the development of political parties. If this is the case in Taiwan, the existence of party attachments can be confirmed.

119

Table 4.4 and Table 4.5 demonstrate the distinct social bases for the KMT and the DPP; 84 percent of mainlanders support the KMT, and only 35.7 percent of Min-nan people—the native Taiwanese whose ancestors immigrated to Taiwan in the 17th century--support the KMT. On the other hand, 84.5 percent of the DPP supporters are Min-nan people, yet 54.6 percent of the KMT partisans are from

Min-nan group. Hakka people prefer the KMT and the independents than the DPP.

This suggests that the DPP only attracts Min-nan people, and mainlanders as a minority group (13.6 percent of the population) one-sidedly support the KMT. As a catch-all party, the KMT appeals to the three ethnicity groups, but it disproportionately emphasizes the mainlanders. Likewise, the DPP largely counts on the support of Min-nan people.

As to age, elderly people like the KMT; 48.5 percent of people between 51 and 65 years old and 51.5% above 66 identify themselves as the KMT. Nearly a half of the DPP partisans are in the young generation (20 to 35 years old), born in the late 1950s at the earliest. Before they had the right to vote, they may have witnessed the repression of the opposition movement in the 1970s and the emerging social movement in the 1980s. On the contrary, the KMT receives support from different generations evenly. The independent candidates also receive slightly more support from the young generation than the old one.

The association between education and partisanship is very interesting. On the one hand, the number of the DPP supporters from each level of education is in proportion to the size of each level. On the other hand, the KMT receives a great deal of support from college-educated people, and the supporters of independents are mainly from the low-educated group.

120

Before I ran the cross tabulation, I expected that the DPP would draw many supporters from the southern Taiwan, since the “tangwei” politicians initially tried to build their organizations there. The KMT is not as popular in the southern Taiwan as in the northern Taiwan, yet its far-reaching party organization maintains considerable partisans in each area.

Last, the KMT received support overwhelmingly from the public sector, which includes teachers, police, and military servicemen. A large portion of the

DPP supporters come from the unemployed and the private sector, which are not well penetrated by the KMT party organizations. Overall, the KMT’s social base is mainlander, elderly, highly educated, northern residents, and public sector. The

DPP found its supporters among Min-nan people, young generation, average educated, southern residents, and private sector.

In 1994, the KMT split and the CNP became another competitive party after the 1995 legislative election. It highly focused on mainlander, young generation, high-educated, public sector, and people in the northern Taiwan, in part because its party members came from the northern Taiwan. The DPP maintained its social base, but the KMT lost many supporters belonging to those five categories. For instance, 26 percent of the KMT partisans were mainlander in

1992, but the proportion drops to 16 percent in 1995.

It is evident that each party has its own social base, and the correspondence nearly fits what we have seen throughout the development of party system (Lin, 1989). The change and continuity of social bases of parties indicate that each party has its distinctive bloc of supporters. Among the five categories, ethnicity, area, and occupation are salient geographical and

121

sociological distinctions. The firm association between partisanship and social groups supports the stability of partisanship, and it may continue to exist for generations.

122

Party Identification KMT DPP Independent Row Percentage Ethnicity Hakka 42.7(19.1) 9.4(13.8) 47.9(21.8) 19.6 Min-nan 35.7(54.6) 16.8(84.5) 47.5(73.8) 66.8 Mainlander 84.4(26.3) 1.6(1.7) 14.0(4.4) 13.6 Age 20-35 43.9(39.4) 16.4(49.2) 39.7(36.6) 39.5 36-50 39.1(28.4) 13.8(33.5) 47.1(35.1) 32.0 51-65 48.5(20.9) 8.3(11.9) 43.2(19.1) 19.0 65 and above 51.5(11.2) 7.5(5.4) 41.0(9.2) 9.6 Education Primary 37.1(32.5) 12.1(35.1) 50.8(46.1) 38.6 Secondary 45.2(43.4) 13.7(43.8) 41.1(40.8) 42.4 College 55.8(24.1) 14.7(21.1) 29.4(13.1) 19.0 Area Taipei City 49.7(23.2) 16.3(25.4) 34.0(16.3) 20.5 Taipei Area 47.7(9.9) 8.6(5.9) 43.9(9.3) 9.1 Hsinchu 42.7(17.0) 6.9(9.2) 50.4(20.6) 17.5 Taichung 45.3(18.2) 13.0(17.3) 41.7(17.1) 17.6 Tainan 37.3(13.6) 15.6(18.9) 47.1(17.6) 16.0 Kaohsiung 38.1(13.0) 17.6(20.0) 44.3(15.5) 15.0 East 53.4(5.0) 10.3(3.2) 36.2(3.5) 4.1 Occupation Public 66.5(20.0) 9.7(9.8) 23.8(7.4) 13.3 Business 38.5(12.9) 14.1(15.8) 47.3(16.2) 14.7 Farming 31.1(6.7) 13.6(9.8) 55.3(12.2) 9.5 Labor 43.1(43.2) 12.5(41.8) 44.4(45.7) 44.1 Unemployed 41.1(17.3) 16.3(22.8) 42.6(18.4) 18.5

Data: 1992 ESC data Note: Table entries are row percentages. Figures in parentheses are column percentages.

Table 4.4: Social groups and partisanship, 1992

123

Party Identification KMT DPP CNP Independent Row Percentage Ethnicity Hakka 50.3(18.4) 9.7(7.6) 14.4(16.5) 25.6(15.3) 15.2 Min-nan 39.2(65.4) 24.6(88.0) 8.5(44.7) 27.6(75.2) 69.5 Mainlander 44.6(16.3) 5.6(4.4) 33.8(38.8) 15.9(9.5) 15.2 Age 20-35 37.2(39.4) 21.2(48.8) 19.5(65.3) 22.1(37.7) 44.2 36-50 43.1(32.5) 20.9(34.3) 9.6(22.9) 26.4(32.0) 31.5 51-65 43.5(16.7) 15.9(13.3) 7.2(8.8) 33.3(20.7) 16.0 65 and above 57.0(11.3) 8.4(3.6) 4.7(2.9) 29.9(9.6) 8.3 Education Primary 47.0(30.7) 15.4(21.8) 2.3(4.7) 35.3(37.3) 27.3 Secondary 42.4(49.1) 20.4(51.2) 12.4(45.3) 24.8(46.4) 48.3 College 34.6(20.3) 21.3(27.0) 27.0(50.0) 17.1(16.3) 24.5 Area Taipei City 34.9(18.0) 21.6(24.3) 19.4(31.8) 24.1(20.1) 21.6 Taipei Area 45.9(9.5) 14.4(6.5) 14.4(9.4) 25.2(8.4) 8.6 Hsinchu 54.4(27.5) 11.4(12.6) 13.2(21.2) 21.0(17.1) 21.1 Taichung 38.8(16.3) 19.4(17.8) 9.3(12.4) 32.6(22.2) 17.6 Tainan 42.9(15.2) 23.0(17.8) 3.7(4.1) 30.4(17.4) 14.8 Kaohsiung 34.6(10.4) 25.3(16.6) 16.0(15.3) 24.1(11.7) 12.6 East 35.4(3.2) 22.9(4.5) 20.8(5.9) 20.8(3.0) 3.7 Occupation Public 49.7(14.5) 9.6(6.0) 22.3(21.2) 18.5(8.8) 12.2 Business 39.8(19.7) 19.2(20.5) 15.4(24.8) 25.6(20.5) 20.7 Farming 42.9(7.3) 19.8(7.2) --- 37.4(10.3) 7.1 Labor 39.9(34.3) 20.6(38.2) 13.4(37.6) 26.0(36.3) 36.0 Unemployed 42.3(24.2) 22.8(28.1) 8.8(16.4) 26.1(24.2) 23.9

Data: 1995 ESC data Note: Table entries are row percentages. Figures in parentheses are column percentages.

Table 4.5: Social groups and partisanship, 1995

124

From the standpoint of the classical Michigan model, partisanship determines people’s perceptions of parties (Campbell, Converse, Miller, and

Stokes, 1960). If people lack party attachment, then they merely draw on random information to construct their perceptions of the political world. In this case, the direction and level of party perceptions will not be stable. If we do not observe fluctuation in the overall perceptions of parties, we can infer that there is an underlying psychological attachment that shape party perceptions. Party perception refers to the extent to which people evaluate parties by several criteria, such as leadership, policy, and candidates. Figure 4.3 shows us that, between 1992 and 1995, people mentioned the KMT more frequently than any other party, and the gap enlarged in 1995, when the number of times the KMT was mentioned by respondents is nearly 4 times than that of the DPP. The reason why people mention the DPP and the CNP far less frequently than the KMT is primarily because the KMT was the ruling party. The increase of the KMT perception does not conform to the downward trend of the KMT partisanship shown in Figure 4.2, but the distribution of party perceptions reflects the relative strength of partisanship and vote share. The sheer magnitude of party perceptions varies with time, in part because party perceptions are campaign-specific. The relative advantage of party perceptions enjoyed by the KMT remains unchanged through the three-year span, which partly confirms the proposition that partisanship is a stable political attitude.

125

0.3804 0.4 0.35 0.3096 KMT 0.3 DPP 0.25 0.2 CNP 0.1449 Mean 0.15 0.0858 0.1 0.0771 0.05 0 1992 1995 Year

Data: 1992 and 1995 ESC data Note: In the 1992 survey, there are four questions about the perceptions of the parties: nominees, platforms, leaders, and contribution to social well-being. In the 1995 survey, there are six questions: leaders, contribution to social well-being, handling economic problem, maintaining social order, handling national security, and improving cross-straight relationship. To compare the two indicators with different scales, I divide the scores by 4 and 6 respectively to conform them to 0~1 scale.

Figure 4.3: Mean of party perceptions, 1992 and 1995

Several researchers also suggested that partisanship is an excellent predictor of voting choice (Hu and Chu, 1992; Liu, 1996). The level of partisanship among the electorate and its impact on voting behavior has been growing substantially since party competition was widely conceived. In Liu’s

(1996) study of the 1992 election, for example, the linearity between one- dimensional partisanship, which ranges from strong KMT to strong DPP, and voting choice is nearly perfect. Whereas there is no panel data to show the stability of party attachment, the correlation between partisanship, party perceptions and voting choices demonstrates the potential influence of party

126

identification. More research on the stability, dimensionality, and source of party identification is needed; however, it is legitimate to apply it to the study of

Taiwanese voting behavior.

4.5 Candidate Evaluations and Vote Choice

Candidates play an important role in Taiwan elections in that the SNTV system with multi-seat districts encourages candidate-centered elections. First of all, each party nominates more than one candidate to win the majority, which leads to intra-party competition. Whereas political parties promise coordination, candidates count on the personal vote instead of party coordination to avoid being deserted. Cox and Rosenbluth’s (1995) study on electoral cohesiveness in England,

Japan, and the United States has asserted that multimember districts “makes it impossible for them (co-partisans) to run simply on the party platform.” (p.28)

Grofman’s (1999) comparison of Japan, Taiwan, and Korea also proves the significance of personal vote in multi-member districts. Hsieh (1996: 208) also maintained that candidates “have little incentive to remain loyal to parties” under

SNTV.

Second, party competition lags behind elections. The local-level elections had been held under the rule of Japanese, and many of those elected continued to participate in the elections after KMT opened the local-level and limited national- level offices to elections. Without the party endorsement, independent candidates broke through the KMT’s control of public opinion and campaign resources, which gradually transformed the one-party politics (Chu, 2001). Many independent candidates joined political parties afterwards, but still keep their

127

personal supporters for intra-party competition. Despite that major political parties attempt to emphasize party discipline and coordination, they do not convert campaign-oriented politicians to party mandates.

For instance, Chen and Hwang (1992) pointed out that in legislative elections the respondents mention candidate considerations more frequently than other reasons. Their pre-election study also showed that the image of candidates strongly influences their final rankings in the district. Lin (1998) also found that people consider candidate characteristics as the major determinant of voting choices.44 Descriptive statistics show the same pattern of voting as for Japanese voters (Rochon, 1981).

Nevertheless, the conflict between the KMT and the opposition parties was intense since the 1980s. The policy debate often intensified the level of confrontation between the two parties in and out of the Legislative Yuan. Both the

KMT and the DPP also emphasized the support of social groups and even the social movements. Therefore, it is necessary to estimate the relative influence of candidate assessments and partisanship on the voting act.

To estimate the influence of candidate evaluation, questions similar to those in the U.S. congressional election study are employed here. There is a battery of questions regarding the requirements of candidates, such as the most considerate and the least corrupted. The respondents indicate which candidate fits each of these requirements respectively, and then we recode the candidates into the

44 Part of the two findings above is derived from the multiple-choice question that reads like: “Generally speaking, people take many thing into account when they make their voting decisions. Among the following things, select ones that apply to your voting decisions.” The answer to this question simply looks into what consideration is in people’s mind when they make their decisions, though it does not support any causality relationship.

128

categories of political parties. In doing so, we obtain a scale of the positive feelings toward political parties related to their nominees. Because more than one candidate is nominated by one party in a given district, we must assume that each party nominee has the same characteristics. In addition, we must assume that people know the party labels of the candidates.45 As Chu (1993) asserted sharply in his study, in the Taiwan context it is inevitable to oversimplify voting behavior in order to construct a parsimonious model. In this case, we must bear in mind that party preference may contaminate candidate evaluations. The scale that we created actually measures people’s perceptions toward the parties in terms of their nominees; the more the respondents mention the candidates, the more positive to the party behind the candidate the respondents are. With the two assumptions-- people know the party labels behind the candidates and each candidate has the same characteristics--it is legitimate to use this variable to estimate the impact of candidate evaluation on voting choices. We can expect that the impact of candidate evaluation on voting choices would be lower than that of parties because we know the two assumptions do not always hold for every respondent.

In spite of the imperfect measurement of the variable, its validity holds.

First of all, people mention candidates from different parties through the set of questions, therefore the scale that codes responses into parties is based on people’s perceptions toward parties and candidates. Moreover, the cross tabulation analysis indicates that people who consider “candidate characteristic” as their top reason to choose their candidates are more likely to respond to the questions of candidate evaluations; on the contrary, the “party-oriented” voters are less likely to select

45 For 1992 data, there are follow-up questions as to which party the candidates belong to. Despite

129

the best candidate that meets the conditions. 46 In other words, the candidate evaluation measurement should not be contaminated by the influence of partisanship.

In recognition of the potential flaw of the measurement, we move on to explore how people evaluate the candidates based on the five criteria. Figure 4.4 presents the mean values of candidate evaluation of the two parties in 1992 and of the three parties in 1995 respectively. The KMT’s candidates are more popular than the opposition parties’ nominees are, which is partly because the KMT had more incumbents in these two elections.47 Whereas the KMT candidates enjoy more favorable images, the gap between the KMT and the DPP decreased somewhat in the interval of three years; the DPP’s steady gain of seats and the

CNP’s breaking up with the KMT explain the change in candidate evaluations.

that, the two assumptions are still required for the following analysis. 46 Both 1992 and 1995 surveys contain the following question: “Generally speaking, people take many thing into account when they make their voting decisions. Among the following things, select the options that apply to your voting decisions.” Not surprisingly, “candidate” is the most popular option among political party, issue, social network, and so forth. 47 According to the author’s study (2002), the KMT had 39 incumbents running for re-election and the DPP had 12 in 1992. In 1995, the KMT endorsed 48 incumbents and the DPP 27. The CNP had not been established until 1994, therefore the CNP had zero incumbents in 1992 and 1995.

130

1.4 1.1786 1.2 1.0847 1 0.9309 0.8513 KMT 0.8 DPP

Mean 0.6 CNP 0.4 0.1577 0.2 0 1992 1995 Year

Data: 1992 and 1995 ESC data Note: Y-axis reflects the average number that the respondents mention the party’s candidate is the best one who fits the requirements.

Figure 4.4: Partisanship and candidate evaluations toward each party, 1992 and 1995

According to the literature on American congressional elections, favorable candidate images result from personal contact as well as partisanship. Political issues are not important because voters would not link the national issues with district interest. Personal contacts are critical to candidate evaluations because people who receive particularized interest would support the candidate (Fiorina,

1977; Yiannakis, 1981). Partisanship also enhances candidate evaluations when they are consistent (Mann and Wolfinger, 1980). Richardson’s (1996) study of the Japanese voter also stresses the importance of mobilization. Here I count the number of contacts initiated by the respondents’ relatives and friends, and political

131

parties during the period of campaign. Presumably, the social network that carries out the contacts is grounded in the casework of candidates. Therefore, the effect of contacts can be interpreted as the influence of candidates through district service.48 Table 4.6 demonstrates the pattern of contacts. Contacts from friends are mentioned more often than contacts from either relatives or parties. Although the partisan influence of contacts is not shown in the data, it is safe to assume that the

KMT has better access to social networks because of its far-reaching organizations.49

1992 1995 Friends 41.2% 27.5% Relatives 24.3 15.5 Party Organizations 27.1 9.7

Data: 1992 and 1995 ESC data Note: Table entries are percentages of contacts initiated by each source.

Table 4.6: Voters’ contacts with candidates through social networks

Candidate evaluations are not only contingent on how much information that people receive, but also how people receive information. According to Cox

(1997), SNTV is a super-proportional system in that it favors the major party in coordination; the governing party is the only one that is able to create particularized benefits to coordinate candidates.50 As a result, the KMT candidates should have more positive feelings since voters can recognize more KMT

48 The questions read like: “During the period of this campaign, who has solicit your vote? Please check the options that apply.” 49 Regarding the disadvantage of the DPP’s party organization, see Chu (2001). 50 Because of different definitions and measurements of coordination success, Cox’s argument is not entirely confirmed by later research. See Tsai (2002).

132

candidates in large districts than in small districts. To test the effect of district size,

I include the number of seats in the district that the respondents reside in. Table

4.7 and Table 4.8 make it clear that partisanship largely determines how people evaluate the candidates in 1992 and 1995. Personal contacts also bear on candidate evaluations, yet party identification is more influential. As to the KMT candidates, a one-point of change in the 7-point scale of partisanship will lead to an increase in evaluation by .3, holding the other factors constant. As to the DPP candidates, a one-point of change in the 7-point scale of partisanship will lead to a decrease in evaluation by .2, holding the other factors constant. Change in personal contacts causes consistent but less impressive change in the KMT candidate evaluation. For the KMT and the DPP candidates, more contacts bring in higher evaluations, which implies that both camps reach the electorate and present their candidates. The unification issue, albeit a political agenda, has an insignificant effect on candidate evaluations. Finally, the district size is another predictor of candidate evaluations. Holding the other variables constant, a one-seat increase in the district will lead to a .05-point increase in KMT’s candidate evaluation. On the other hand, the DPP candidates have better evaluations in smaller districts.

In 1995, partisanship still played an important role in the making of candidate evaluations. For instance, people who identify with the KMT have a better image of the KMT candidates than people who do not have KMT attachment by .1 point. People with different partisanship have distinct feelings toward their candidates; they only pay attention to the candidates of their own party. Contacts also benefit the KMT and the CNP candidates, but do not affect

133

the way that people evaluate the DPP candidates. Like the case of the United

States, issue position does not have an impact on candidate evaluations. The district size indeed affects candidate evaluations, but the direction is somewhat confusing; people in the small districts favor both the KMT and the DPP candidates, which is against the super-proportional hypothesis. The emergence of the CNP may cause this result because it successfully won seats in the large districts, such as Taipei City, Taipei County, and Tao-yuan County.51 Voters in those districts embraced the CNP not because of any district interest, but because of the strong negative feelings toward the KMT and the DPP on the unification issue.52

KMT candidate DPP candidate evaluations evaluations Constant 2.177(.129)*** .218(.096)* Party Identification1 -.386(.028)*** .238(.021)*** Contacts2 .132(.041)*** .137(.030)*** Unification3 .194(.084)* -.247(.062)*** District Magnitude4 .050(.009)*** -.017(.007)**

Observation Number 1129 1129 Adjusted R-square .197 .123

Data: 1992 and 1995 ESC data Note: 1 The partisanship variable is a 7-point scale from strong DPP to strong KMT. 2 The contact variable is the number of contacts that respondents receive during campaign. 3 The unification variable is a 3-point scale: pro-independence, neutral, and pro-unification. 4 The district magnitude variable is the number of seats in each district. Parentheses are standard errors. *: p<.05, **: p<.01, ***: p<.001.

Table 4.7: OLS Estimates of candidate evaluations, 1992

51 Among its 15 seat gains in 1995, the CNP won 5 seats at Taipei City, 3 at Taipei County, and 2 at Tao-yuan County. 52 According to Lin (1996), these three districts are among the top five list of the number of the mainlanders.

134

KMT DPP CNP candidate candidate candidate evaluations evaluations evaluations Constant 1.167(.166) 1.168(.170) .172(.062) Partisanship1 KMT .120(.051)* .069(.053) .004(.019) DPP .120(.051) .160(.063)** -.028(.023) CNP .068(.062) .198(.072)** .081(.026)** Contacts .108(.029)* .020(.051) .078(.018)*** Unification .003(.036) -.011(.036) -.024(.013) District Magnitude -.050(.009)*** -.046(.009)** .005(.003)

Observation Number 1255 1255 1255 Adjusted R-square .029 .024 .031

Data: 1992 and 1995 ESC data Note: 1 The partisanship variable is 3-point scale from each party from strong to lean partisan. 2 The contact variable is the number of contacts that respondents receive during campaign. 3 The unification variable is a 7-point scale from pro- independence to pro-unification. 4 The district magnitude variable is the number of seats in each district. Parentheses are standard errors. Parentheses are standard errors. *: p<.05, **: p<.01, ***: p<.001.

Table 4.8: OLS Estimates of candidate evaluations, 1995

The analysis of the determinants of candidate evaluations uncovers the impressive impact of partisanship on candidate evaluations across different years and parties. Contacts and the size of district only affect people’s perceptions occasionally. In 1992, for example, a one-point increase in the KMT partisanship will cause a .386-point increase in the KMT candidate evaluations, and a one- point increase in the number of contacts will increase the positive feelings toward the KMT by .132. In 1995, when a voter has a one-point increase in his attachment with the KMT, his positive responses to the KMT candidate will increase by .120. For the DPP, partisanship and ideology are equally important in

135

the formation of candidate evaluation in 1992, but both the attachment with the

DPP and the CNP positively influence the DPP candidate evaluation in 1995.53

The CNP candidate evaluations come from both partisanship and contacts, which is partly because of the CNP’s penetration into the Mainlanders’ social network held by the KMT before. This finding suggests that candidate evaluations are mostly party-driven and partly crafted by candidates. Moreover, voters gave more weight to partisanship in 1995 than in 1992. For the supporters of both parties, the importance of partisanship increased between the two elections, but contacts and social network became less significant. Personal contacts and district magnitude merely strengthen candidate images occasionally.

Despite party identification strongly affecting candidate evaluations, it does not entirely account for the variation in candidate evaluations. We also find that personal contacts as well as the type of districts have influence on candidate assessments. The changing political environment and the permissible SNTV allow individual candidates to seek their own support. If there was more information as to the way that candidates perform their district services, we could force a more nuanced understanding of evaluations of candidates. In spite of this speculation, the finding suggests that political parties had strong ties with candidates when all of the seats in the legislative body were available for the first time. The strong tie still held three years later when the new party challenged the existing parties and new elites tried to replace the old ones. While the number of parties increased

53 The puzzling influence of the CNP party identification, I speculate, comes from the effect of the unification issue, which holds people on the same ground of the unification issue. The “Great Consolation” rose by the DPP and the CNP in an attempt to weaken the majority of the KMT in the Legislative Yuan after the 1995 election explains why people can accept both parties in some aspects. It deserves more attention as to how people perceive of political parties when the issue stand is set aside.

136

from two to three and ideology tension became higher due to the unification issue, it is not surprising that candidate assessment is more party-oriented than candidate-fashioned.

Under the presidential system, political parties tend to give their way to individual candidates and play the role of campaign organizations. Due to the fact that political parties just proceeded to a meaningful party competition, however, political parties in Taiwan still take the lead in recruiting candidates and setting the campaign tone. On the other hand, SNTV encourages intra-party competition, which is confirmed by the finding that district size positively affects the KMT’s candidate evaluations. Hence, the effect of political institutions is not compromised.

4.6 Party Images and Vote Choice

Since the conflict between the KMT and the DPP was intense, the sentiment toward either party was very strong. Both the KMT and the opposition parties aligned with certain social groups, which made their images salient to the electorate. Furthermore, both parties endeavored to frame the issue agenda in every election (Chu, 1998). In the eve of the 1992 election, the DPP promoted the socialeconomic equity issue and forced the KMT Administration to implement its social welfare platform. In 1995, the cross-strait war loomed in the election; China maneuvered its offshore missiles after the president’s visit to Cornell University.

That pushed the three parties to debate the issue of independence or unification.

You (1996) and Lin (2000) provided a picture of party images in which the

KMT is praised for its images of governance, peaceful reform, and democracy but

137

loses credits on mass interest and anti-corruption. The DPP is well known for its fight for mass interests, anti-corruption, fairness, and civil rights, but it has poor images on public policy, peaceful reform, and governance. In short, the KMT and the DPP have sharply opposing party images. Figure 4.5 illustrates the sharp contrast between the KMT and the DPP during the 1992 campaign. When the impact of the issue position is controlled for, party images have considerable effect on party identification (You, 1994; You, 1996).

-12.4 Public policy 35.2 -53.7 7.5 -15.2 Peaceful reform 48.4 -60 5.3 -0.9 Goverance 39 KMT -44.1 14.1 -24.8 KMT Mass interest 23 -5.3 35.6 -21.1 DPP Fairness 14.6 29 Images -11.4 -17.8 DPP Anti-corruptions 17.5 -3.3 36.2 -7.5 Civil rights 43.4 -12 28.9 -5.6 Democracy 51.2 -13.7 27.2 Percentages

Data: Compiled by the author from You (1996). Notes: The right-hand side of the X-axis stands for the positive images of parties and the left-hand side the negative images of parties.

Figure 4.5: Party images of the KMT and the DPP, 1992

Because of the KMT’s long-term dominance in education and the mass media, the opposition parties can hardly compete with it. Wu (1995:79) pointed out that “The opposition in Taiwan had a serious image problem, and this was directly and indirectly caused by the KMT’s control of the mass media.” On the

138

one hand, the KMT enjoyed positive images for the rapid economic growth but negative images for the corruption. On the other hand, the opposition parties won acclaim for their calls for sweeping political reform but lost support for their radical measures. Figure 4.5 shows the significant advantage of party images enjoyed by the KMT.

Figure 4.6 and Figure 4.7 demonstrate that partisanship and positive feelings toward parties are consistent in 1992 and 1995 respectively.54 In Figure

4.6, the lengths of the bars increase with partisanship moving from the one end to the other end. It confirms the conflict between the DPP and the KMT. In Figure

4.7, the strong KMT partisan has the strongest perceptions of the KMT; the darkest banner stretches beyond 4. The strong DPP partisans and CNP partisans have significant better feeling toward their parties as well. It seems that some of the DPP and the CNP supporters have good evaluations of the KMT, but the KMT supporters only praise their own party. The vague distinction between the three political parties implies multiple dimensions in the party system. Some of the DPP supporters like the KMT because of its candidates, and some of the CNP supporters like the KMT for its unification policy. This phenomenon prevents us from treating partisanship as one-dimensional.55

54 For the 1992 data, the positive feelings toward parties are acquired from a battery of questions asking the respondents which party between the KMT and the DPP has a better party leader, cares about people like us the better, brings more benefits to the district, and is closer to people’s idea. For the 1995 data, the set of the positive-feeling questions asks the respondents which party among the KMT, the DPP, and the CNP has the best party leader, cares about people the most, is the most capable one to solve the national problems, is the most capable one to maintain law and order, is the most capable one to protect the national security, and is the most capable one to enhance the cross-straits peace. 55 I ran the correlation analysis to see if the perceptions of parties are independent from each other, particularly in 1995. The Pearson correlation coefficients between the KMT and the DPP evaluation is -.138, the KMT and the CNP -.197, and the DPP and the CNP -.072. The result indicates that people conceive of each party as different objects.

139

3.5 KMT 3 DPP 2.5 2 1.5 Mean 1 0.5 0

T nt T nd KMT KMT KM e g DPP k ep an ean KM Le Ind Weak DPPL Strong Wea Stron Partisanship

Data: 1992 ESC data Note: The horizontal axis is the level of partisanship and the vertical axis is the level of positive feeling toward the two parties.

Figure 4.6: Partisanship and positive feeling toward parties, 1992

6 KMT DPP 5 CNP 4 3 2 1 0

P P T P PP T D D CNP EN G G K D N N INGCN ONGKMT O EA N EN TR WEAKDPP W TR WEAKKMT S STRONGCNP EP S LEANI LEA D LEANINGKM N I

Data: 1995 ESC data Note: The X-axis is the level of partisanship and the Y-axis is the level of positive feeling toward the three parties.

Figure 4.7: Partisanship and positive feeling toward parties, 1995

140

4.7 Multi-Party Voting Model of Legislative Election

The 1992 legislative election was the first in which all of the seats in the

Legislative Yuan were elected popularly. Because the KMT made popular election possible, the KMT took the credit for democratization. While the DPP looked for a new political reform issue, its pro-Taiwan independence image became more and more striking against the KMT’s “one China” policy. The KMT argued that Taiwan independence would devastate Taiwan, but the DPP contended that people should have the right to opt for self-determination to avoid

China’s takeover. Some KMT party members were sympathetic with the DPP’s nationalism, which led to a split in the KMT. In 1994, the CNP emerged, promoting the pro-unification policy and attacking the KMT for its move to a moderate position on the unification-independence issue. While the KMT incorporated local factions and became moderate on the national identity issue, the

DPP tried to distinguish itself from the KMT by aligning with pro-independence supporters.

Knowing that party identification strongly directs people’s evaluation toward the KMT and the DPP candidates, we expect that voting choice is also a function of party identification. Candidate evaluations are supposed to be a significant predictor only second to party identification, because candidates are better known to voters than is the unification issue. Finally, ethnic identification with Taiwanese or Chinese or both influences voting choices occasionally, controlling for party identification as well as candidate evaluations. Political parties are the leading force in the process of voting decision-making due to the newly implemented popular election as well as unsettled political issues.

141

Moreover, new political parties tend to strengthen party discipline as they enter into politics (Hsu and Tsai, 2002). The transformation of the Legislative Yuan contributed to strong influence of parties as well. Considering both the constraint of measurement and politics, any impact of candidate evaluations should be taken as evidence of the influence of political institutions.

Before proceeding to detailed analysis, I describe the variables and data in the next sub-section.

4.7.1 Variables and Model

Party identification is measured by the set of questions that read like: “Do you support any political party? If so, how strong is your support?” In the 1992 election, there were only two parties. As a result, we can view partisanship as a one-dimensional political attitude that ranges from strong KMT to strong DPP with independent in the middle. In the 1995 election, however, there were three parties. According to our analysis above, the three parties were treated as more like three distinct political organizations than two camps.56 In this case, I will treat partisanship in 1995 as multi-dimensional. I recoded it as three variables, each of which represents the strength of party attachment with a given party.

The candidate assessment variable has been constructed in the last section.

Basically, it is a 5-point scale for candidate evaluation of each party. Due to the fact that there is more than one candidate from a given party, the notion of candidate evaluation is not as same as that used in the United States. Here I tallied

56 As matter of fact, the three-party system continued to exist until 2001, when two new parties replaced the CNP. From the aspect of history and the permeability of SNTV, small parties can find their spaces in districts.

142

people’s positive responses for each party’s candidates. In the 1992 election, there are two variables standing for assessment of the KMT and the DPP’s candidates.

In the 1995 election, there are three candidate evaluation variables for the KMT,

DPP, and CNP candidates respectively. I assume that the variation in candidate evaluations can explain people’s party choices.

The independence issue represents different prospects for Taiwan. It also is associated with the division between the KMT, the DPP, and the CNP. In the

1992 survey, the respondents were asked if they support Taiwan independence, status quo, or unification. In the 1995 survey, the scale had more categories to indicate the intensity of attitude. The respondents can opt for “unify with China as soon as possible”, “maintain the status quo and then unify with China,” “unify with China when it is democratic,” “maintain the status quo forever,” “claim independence if China attacks Taiwan,” “maintain status quo and then claim independence”, and “claim independence”. According to the categories, I treat this as a 7-point scale.

The last political attitude variable—party images—stands for how warm people feel about parties. In the 1992 survey, people are asked how well they like the KMT and the DPP respectively. In the 1995 survey, a ten-point scale is employed to measure the extent to which people like the three parties. The 1992 question is constructed as a five-point scale. Like candidate evaluations, the party images variable is treated as utility comparison by subtracting the KMT images from the DPP images.

People’s conversations with friends, relatives, and parties may enhance their strength of party attachment and candidate preferences. I include this

143

variable in order to exploit its influence independent from candidate evaluations.

The hypothesis is that the KMT candidates benefit from the number of contacts because they have better party organizations than the DPP and the CNP.

In addition to those long-term and short-term forces, the institutional factor should be considered. In multi-member districts, political parties must coordinate candidates to avoid seat loss (Cox and Niou, 1994). However, the multi-member district system also encourages personal campaigns, which makes the task of candidate coordination extremely challenging. One view is that factional competition can reduce the number of non-incumbents being endorsed (Cox and

Rosenbluth, 1994). The other view is that major parties have access to particularistic benefits and they can allocate them among candidates (Cox, 1996).

Tsai (2002) also found the KMT could gain more seats if the chairmanship of the standing committees in the Legislative Yuan was distributed evenly among the

KMT incumbents. After all, the district size should positively affect the fate of the

KMT candidates because they can divide the votes more evenly and thereby increase the probability of being elected.

Given the importance of social groups that align with political parties, it is necessary to control for demographic variables, such as education, age, occupation, and ethnic identity. While demographic variables merely have remote influence on the voting decisions, including those variables can avoid the specification problem.

For the 1992 data, as shown in Section 4.4, partisanship is a standard 7- point scale from pro-DPP through independent to pro-KMT. For the 1995 data, however, each of the three parties has its own supporters and some partisans

144

identify with more than one party. The descriptive statistics show that the KMT enjoys stronger partisan strength than other parties in both elections. As for candidate evaluations, the KMT candidates enjoyed high popularity in 1992, but the DPP candidates made a lot of progress in 1995. The mean values of the KMT and the DPP candidate evaluation are .810 and .045 in 1992 respectively, and .886 and .824 in 1995 respectively. People’s stand on the independence issue is located around the neutral point: .182 in 1992, in which the scale ranges from –1 to 1, and

4.265 in 1995, in which the score is between 1 to 7. The number of contacts is up to 3, and people on average have .9 and .5 contacts with their friends, relatives, and parties. District magnitude ranges from 1 to 16 and 17, and the average number is around 6 or 7. It is evident that parties must nominate more than one candidate in a typical district, which stresses the importance of political parties.57

57 The CNP did not nominate at least one candidate in every district. For the convenience of analysis, however, I include all of the cases no matter whether they can choose a CNP candidate or not. Unless the research design uses oversamplings in some districts where the minor parties indeed nominates candidates, we have to recognize this drawback of analysis by assuming that every individual chooses from three parties regardless of the districts.

145

1992 1995 KMT Partisanship (0 to 3) .641 DPP Partisanshipa 3.298 .307 CNP Partisanship (0 to 3) .183 KMT Candidate Evaluations (1 to 5) .886 DPP Candidate Evaluationsb -.765 .824 CNP Candidate Evaluations (1 to 5) .158 Independence Issuec .182 4.265 KMT Party Images (0 to 5) 3.196 DPP Party Imagesd -.213 2.758 CNP Party Images (0 to 5) 2.781 Social Networks (0 to 3) .925 .511 District Magnitude 6.717 7.060 Taiwanese (0 and 1) .197 .312 Hakka (0 and 1) .290 .184 Male (0 and 1) .501 .488 20-35 years old (0 and 1) .383 .331 36-50 years old (0 and 1) .311 .341 51-65 years old (0 and 1) .200 .195 Elementary Education (0 and 1) .411 .366 Secondary Education (0 and 1) .404 .432 Public Sector (0 and 1) .126 .124 Private Sector (0 and 1) .304 .188 Farmer (0 and 1) .191 .090 Labor (0 and 1) .434 .334

Note: a: it is 7-point scale (-3 to 3) for the 1992 data. b: For the 1992 data, it is the deference between the DPP candidate evaluations and the KMT candidate evaluations. c: For the 1992 data, the 3-point-scale issue position ranges from –1 to 1. d: For the 1992 data, it is the difference between the DPP party images and the KMT ones.

Table 4.9: Means and ranges of the independent variables, 1992 and 1995

Since the legacy of one-party politics still lingered in 1992, the effect of partisanship on vote choice should be greater than candidate evaluations. After the first popular legislative election, however, candidates had more resources to mobilize voters and craft candidate-based voting. Moreover, the expected

146

presidential election and the new presidential system after democratization would create more individual incentives for candidates. Therefore, I expect that the influence of candidate evaluations will increase. The influence of issue positions will be remote from vote choice, since the multi-member district system encourages intra-party competition and downplays the role of issues behind party labels.

Two multi-nomial logit models as follows will be estimated, and the results will be presented in the next section.

Model 1: Pr(Y=Voting for the KMT, DPP |x)={exp(xβ)/[1+ Σexp(xβ)]} where x= [DPP Partisanship, DPP Candidate Evaluations, Independence Issue

Position, DPP Party Images, Social Networks, District Magnitude, Taiwanese

Ethnic Identification, Hakka Ethnic Identification, Male, 20-35 years old, 36-50 years old, 51-65 years old, Elementary Education, Secondary Education, Public

Sector, Private Sector, Farmer, Labor]

β=[β1,…,β19]′

Model 2: Pr(Y=Voting for the KMT, DPP, or CNP |x)={exp(xβ)/[1+ Σexp(xβ)]} where x= [KMT Partisanship, DPP Partisanship, CNP Partisanship, KMT

Candidate Evaluations, DPP Candidate Evaluations, CNP Candidate Evaluations,

Independence Issue Position, Social Networks, District Magnitude, Taiwanese

Ethnic Identification, Hakka Ethnic Identification, Male, 20-35 years old, 36-50 years old, 51-65 years old, Elementary Education, Secondary Education, Public

Sector, Private Sector, Farmer, Labor]

β=[β1,…,β21]′

147

4.7.2 Multi-Nomial Logit Estimation of Voting Choice in Taiwan

Based on the multi-nomial logit estimates, we can tell which variables are significant and whether their signs are in the right direction. At first glance, party identification exerts greater influence on voting choices than any other variables in the 1992 election. Table 4.10 shows that both the KMT vote and the DPP vote are associated positively with partisan strength. Under the far-reaching influence of partisanship, candidate evaluations and the unification issue are lackluster but in the right direction. The relative candidate evaluations only work for the KMT vote; the KMT candidates benefit largely from positive candidate evaluations but the DPP candidates solely rely on partisan strength. Although both parties sharply disputed the national identity issue, people’s position on this issue is by no means a good predictor for KMT or DPP voting. Neither do party images play an important role, which is probably because the KMT has overwhelming party images and the variation is small. The number of contacts that individuals receive only has significant impact on voting for the KMT. The number of seats in the district has a positive influence on the probability of voting for the KMT, which confirms the super-proportional theory of SNTV (Cox, 1996).

Regarding the demographic variables, none of them has noteworthy influence across the board, which reflects the fact that those demographic variables have only distal impact on voting behavior. Interestingly, the direction of ethnic identification is opposite to what we anticipate, probably because of the control for the issue position.58

58 I put different independent variables in the model and found that it is the issue position changes the direction of the ethnic identification.

148

All in all, partisanship and candidate assessment account for variation in voting for the KMT, voting for the DPP, and abstention. People’s perceptions toward candidates are less influential than their party attachment, particularly for the opposition party. The explanation is that people have more exposure to the candidates of the governing party, thus they vote for the party because of their preference for candidates as well as the party.

KMT vote DPP vote Independents Constant 2.441(.661) -3.143(1.063) -.465(.687) DPP Partisanship -.584(.083)*** .646(.106)*** -.002(.083) DPP Candidate Evaluations -.549(.089)*** .228(.143) -.155(.093) Independence Issue -.042(.218) -.265(.268) -.135(.224) DPP Party Images -.075(.088) .338(.116)** -.227(.091)* Social Networks .203(.103)* .155(.127) .261(.104)* District Magnitude .085(.022)*** -.035(.029) .069(.022)** Taiwanese .328(.308) -.003(.344) .708(.290)* Hakka -.108(.228) .284(.296) .007(.237) Male -.059(.213) .080(.265) .283(.218) 20-35 years old -1.513(.506)** .149(.902) -.994(.519) 36-50 years old -.407(.493) 1.010(.882) -.080(.505) 51-65 years old -.060(.511) .770(.915) -.040(.529) Elementary Education -.431(.349) -.287(.446) .362(.365) Secondary Education -.176(.259) .179(.341) .530(.280) Public Sector -.133(.366) -1.034(.570) .221(.375) Private Sector .254(.240) -.009(.287) -.037(.244) Farmer .668(.325) -.207(.412) .142(.332) Labor -.244(.220) -.081(.264) -.168(.224) Number of Cases 1076 Log Likelihood -1076.8575 Chi-Square 605.09*** Pseudo R-Square .2193

Note: Standard errors are in parentheses. Abstention is the comparison category. *: p<=.05; **:p<=.01; ***:p<=.001

Table 4.10: Multi-nomial logit estimates for four-choice model, 1992

149

Table 4.11 shows a different voting pattern for the 1995 election than in the previous one. As for the KMT vote, the attachment to the KMT, candidate evaluations, and its party images are significant predictors. Regarding the DPP vote, the level of partisanship, candidate evaluations, party images, and Taiwanese ethnic identification account for the variation. As for the CNP, however, only partisanship and party images have impact on it. For the DPP in 1992 and the

CNP in 1995, candidate evaluations have considerably less impact than partisanship. This implies that in Taiwan party labels are important to new parties, which confirms Aldrich’s (1995) theory that parties are created to solve the collective choice problem. As candidates can manage their constituencies, parties will gradually give their way to individual candidates, which reflects on the influence of candidate assessments on the KMT vote choice.

Independence issue position is not influential in either election, which is partly because of the strong impact of partisanship and candidate evaluations.

Because of multi-member districts, in addition, intra-party competition inevitably weakens the importance of political issues. Furthermore, candidates vie for local interest and personal vote within the district, which also contributes to the low level of attention to issues (Chu, 1995).

Neither the level of contacts nor district size is a significant predictor across the board. Is it because a new party joins the party system? Or is it because electoral competition at the local level realigns? Or is it because candidate evaluations capture all of the influence of particularized benefits? I suspect that district magnitude has little impact on people’s perception toward candidates when holding candidate evaluations constant. The effect of district size in the

150

1992 election is probably due to a brand new level of electoral competition that is favorable to the governing party.59 In the 1995 election, the minimal effect of district magnitude suggests that people still can identify candidates as the number of candidates increases to 17 at most. More precisely, the number of parties would not increase with district magnitude, thus the effect of district magnitude merely applies to the beginning of a new party system.

Few demographic variables contribute to party choices. People under 35 years old who just enter politics are less likely to vote for the KMT than other party or abstention. People in the public sector are inclined not to vote for the DPP.

Moreover, people who merely receive elementary or secondary education tend not to support the DPP or the CNP. Compared to the 1992 election, the KMT remains unpopular among the young generation and it loses its attraction to farmers. A connection between education and party support appeared in the 1995 election, but the small influence of occupations in the 1992 election nearly went away.

59 Cox’s (1996) seat bonus analysis is derived from the 1992 election. Hsu and Chen (2002) extended the analysis to the 2001 election, which shows the learning process of the opposition party and the decline of the seat bonuses of the governing party.

151

KMT vote DPP vote CNP vote Constant -1.555(.665)* -.988(.805) -1.635(1.255) KMT Partisanship .620(.096)*** .214(.124) .052(.236) DPP Partisanship .071(.150) .601(.124)*** -.218(.385) CNP Partisanship .046(.166) -.224(.232) .809(.202)*** KMT Candidate .272(.057)*** .176(.068)** .091(.119) Evaluations DPP Candidate .205(.057)*** .338(.063)*** .171(.101) Evaluations CNP Candidate .058(.148) -.104(.222) .245(.181) Evaluations Independence Issue -.085(.054) -.061(.060) -.206(.113) KMT Party Images .510(.102)*** -.263(.116)* -.242(.163) DPP Party Images -.184(.091)* .469(.109)*** -.444(.166)** CNP Party Images .170(.090) -.279(.104)** .877(.181)*** Social Networks .092(.097) .022(.118) .173(.170) District Magnitude .016(.017) .029(.020) .036(.030) Taiwanese -.005(.193) .629(.210)** -.351(.510) Hakka -.315(.210) -.306(.282) .279(.339) Male -.303(.182) -.098(.206) .183(.332) 20-35 years old -1.120(.339)** -.236(.498) -.890(.654) 36-50 years old -.542(.308) .128(.471) -.283(.625) 51-65 years old -.254(.324) .330(.488) -.833(.714) Elementary Education .044(.272) -.405(.324)* -1.031(.563) Secondary Education .120(.208) .032(.234) -.649(.324)* Public Sector .097(.294) -1.014(.402) -.879(.532) Private Sector -.085(.255) -.114(.283) -.331(.474) Farmer -.041(.372) -.530(.448) -.545(1.162) Labor -.094(.222) .040(.250) -.180(.430) Number of Cases 1201 Log Likelihood -116.3275 Chi-Square 812.31*** Pseudo R-Square .2668

Note: Standard errors are in parentheses. Abstention is the comparison category. *: p<=.05; **:p<=.01; ***:p<=.001

Table 4.11: Multi-nomial logit estimates for four-choice model, 1995

152

To demonstrate the change in the probability of voting for a given party against abstention due to change in the value of independent variables, I transformed the log odds to probabilities and calculated the discrete change caused by change in the value of each independent variable. Table 4.12 and Table

4.13 make it evident that voting choice largely responds to partisanship, but candidate evaluations affected vote choice more in 1995. Although candidate assessment has smaller effect on voting choices than expected, its influence is only second to partisanship. In 1992, for instance, a standard deviation change in partisanship centered on the mean will decrease the probability of voting for the

KMT candidates by .192, holding all other variables at their means. A one standard-deviation change in candidate evaluations just increases the probability of voting for the KMT by .162. For the DPP candidates, a one standard-deviation change in partisanship increases the likelihood of voting for them by .196, but the same magnitude of change in the DPP candidate evaluations will raise the probability of voting for the DPP by .062. As regards to party images, its influence on the KMT vote against abstention is not as noticeable as other variables except the independence issue. For the DPP and third party vote, however, party images have tremendous impact. District magnitude and social networks are positively associated with the KMT vote, which sheds some light on the success of the KMT and the candidate-centered pattern of the SNTV.

153

KMT DPP Indendents DPP Partisanship -.192 .196 .000 DPP Candidate Evaluations -.162 .062 -.050 Independence Issue -.004 -.025 -.015 Party Images Difference -.020 .087 -.068 Social Networks .044 .031 .062 District Magnitude .084 -.032 .074

Note: Centered standard deviation change is calculated as follows (Long, 1997). Because [p(y=m)/1-p(y=m)]=exp(Xβ'), thus p(y=m)=[ exp(Xβ')/1+ exp(Xβ')]. Let th p1(y=m|x, xi+½si ) the predicted probability that i variable at its mean plus one- half of its standard deviation, holding all other variables at mean, and p2(y=m|x, th xi-½si ) the predicted probability that i variable at its mean minus one-half of its standard deviation, holding the dummy variable at 1 and other variables at their mean. The difference between pi1 and pi2 is the centered standard deviation change.

Table 4.12: Discrete change in the probability of party choice in 1992

Table 4.13 shows the trend of the candidate-dominated voting pattern.

When the KMT partisanship increases by one deviation, the probability of vote support will increase .138. The same magnitude of change in KMT candidate evaluation only leads to .096 increase in the KMT support. The probability of voting for the DPP candidates would increase by .093 for a one standard-deviation change in people’s attachment with the DPP, but .104 for the same magnitude of change in candidate evaluations. The CNP partisanship and unification issue are more crucial than the CNP candidate assessment: one-deviation changes in CNP partisanship and unification support increase the probability by .028 and .020 respectively, but a one-deviation change in CNP partisanship increases the probability by only .008.

154

Variables KMT DPP CNP KMT Partisanship .138 .041 .002 DPP Partisanship .012 .093 -.009 CNP Partisanship .006 -.027 .028 KMT Candidate Evaluations .096 .053 .008 DPP Candidate Evaluations .074 .104 .015 CNP Candidate Evaluations .007 -.012 .008 Independence Issue -.034 -.021 -.020 KMT Party Images .117 .025 -.013 DPP Party Images -.045 .100 -.027 CNP Party Images .041 -.058 .054 Social Networks .018 .003 .008 District Magnitude .018 .028 .010

Note: Centered standard deviation change is calculated by the same method as Table 4.12

Table 4.13: Discrete change in the probability of party choice in 1995

Overall, partisanship is the center of decision-making for voting choices, but candidate evaluations get their influence as the presidential election approaches. Notice that KMT candidate evaluations and DPP candidate assessment have impressive weights on party choice, which implies that candidate evaluations gain their strength as parties undergo the transition of democratization and politicians recognize the importance of seats in the Legislative Yuan as the presidential system consolidates. The CNP still had strong control over individual candidates; candidate assessment is not as important a consideration as partisanship. Should Taiwan have a longer tradition of the presidential system and new generations of voters enter the election, the influence of political parties would decline and the characteristics of candidates would be a rising factor.

155

4.8 Conclusion

Since the 1970s, Taiwan has used multi-member districts with the SNTV system, in which one party nominates more than one candidate and encourages intra-party competition. In practice, candidates in the Taiwan context tend to have strong personal campaign organizations and downplay the influence of party labels, which should reinforce the tendency of candidate-oriented campaigns.

Moreover, the presidential system, which was the legacy of one-party politics, is expected to enhance the candidate-centered pattern because of the separate power of the executive and legislative bodies. On the other hand, however, political parties in Taiwan have undergone dramatic change since 1987. While Taiwan transitioned from one-party politics to a multi-party system, the tension among parties in terms of ideology and power struggle highlights the role of party labels in elections.

The analysis of this chapter shows that candidate evaluation is not as critical to voting choices as was predicted, in part because of the short history of party competition. Besides, the measurement of candidate evaluation may be contaminated by party identification; candidates mentioned by the respondents are coded according to their party labels. In addition, partisanship, personal contacts, and district magnitudes are found to have great influence on candidate evaluations.

The relatively new political system has merely pushed forth the role of political parties; individual candidates are not strong enough to direct voting choices. However, the increasing importance of candidate evaluations, though overshadowed by partisanship, proves the influence of political institutions—the presidential system with SNTV. On the one hand, the presidential system pushes

156

individual candidates to seek positions in the legislative body as access to policy- making; the KMT and the DPP candidates are more influential in 1995 than in

1992. On the other hand, the size of district has marginal influence on candidate evaluations, which suggests that voters have considered the candidates, instead of merely following their partisanship. In terms of voting choices, however, candidate evaluations and party identification wash out the influence of district magnitude; the electorate simply relies on the cues offered by parties and candidates regardless of district magnitude. The national identity issue—whether

Taiwan should reunify with China or should claim independence—has insignificant influence on voting choice. Party images are important for new parties, the DPP in 1992 and the CNP in 1995.

This interpretation of the results should be taken with caution because of the imperfect measurement of candidate evaluations. Moreover, the dependent variable is party choice rather than “candidate” choice, which actually simplifies the voting act in the SNTV context. Therefore, the conclusion is tentative that political parties and candidates are equally important to Taiwanese voting behavior. This comprehensive analysis nevertheless assessed the determinants of

Taiwanese voting behavior and the results are comparable with the U.S. election and the following Japanese case.

157

CHAPTER 5

JAPANESE VOTING BEHAVIOR BEFORE AND AFTER THE 1994

ELECTORAL REFORM

5.1 Introduction

The study of Japanese political behavior is a very interesting topic because of their century-old party politics as well as their parliament government. There has been some research on the development of the party system, which has evolved from a one-and-a-half party system to a multi-party system.

Scholars have studied Japanese voting behavior from the sociological and social psychological approaches (Richardson and Flanagan, 1977; Miyake, 1989;

Flanagan, 1980; Richardson, 1988; Flanagan, Kohei, Miyake, and Richardson,

1991; Richardson, 1997). Using national survey data on both the upper and lower houses elections, they revealed complicated patterns of long-term reinforcement of political attitudes, multi-dimensional partisanship, overlapping and changing issues, and media influence on party evaluations. In their research, new concepts, such as habitual voting and cultural politics, have been developed and compared with elements of the two existing models. They also referred to political

158

institutions as the cause of political behavior, and public policies as the outcome.

The findings not only enrich our understanding of Japanese politics but also extend the horizons of the study of voting behavior.

Recognizing the unique context of Japanese politics, scholars very carefully sorted out the factors of voting choices with the two existing models.

They identified short-term and long-term influences and the mechanism that triggers the two clusters of factors. After considering those factors in the process of voting decisions, scholars argued that party identification, which is endogenous to voting at some level, largely shapes voting behavior (Flanagan, Kohei, Miyake, and Richardson, 1991). The long-term forces, such as value cleavage, ideology, and social groups, also anchor individual’s prepositions toward politics and, as a result, political parties. The short-term forces, which are responsive to political leaders and international environment and reflect on party image, issue positions, and images about politics, have minor, if not marginal, influence on the partisan vote. In the end, the scholars conclude: “partisanship drives the vote in Japan exactly as it has often been asserted to do in other industrialized countries.”

(Flanagan, Kohei, Miyake, and Richardson, 1991: 427) In a comparative study, however, Japan was found to have the smallest percentage of strong party identifiers (Verba and Pye, 1978). Japanese voters mention candidate characteristics as the main reason to select a candidate more than political parties, which is different from the five industrialized countries (Rochon, 1981). In other

159

words, whether or not partisanship is the most important determinant to the voting act in Japan remains a puzzle, in part because of different research designs.60

In this chapter, the 1976 and 1996 HR elections will be used for the analysis of Japanese voting behavior. The analysis presented below will explore the determinants of vote choice before and after the change of the electoral system, which has not been examined by other researchers to this date. The primary reason for choosing these two elections is the comparable measurements of candidate evaluations and party preference. The 1976 election was held two years after the

1974 House of Councillors election, when the LDP obtained 44.3% of the popular vote in the National Constituency. It is representative of other elections because it is half way through the economic development era and public criticism against the

LDP’s money politics and domestic policies has grown over the 1970s (Baerwald and Nobuo, 1977). The 1996 election is chosen because it was the first one after the 1994 electoral reform. The mixed electoral system—300 members elected in single-member districts and 200 members elected in the proportional representation districts according to their rankings on the party list—allows political parties to put the same candidates in both single-member districts and party lists. Most candidates took dual candidacy—they ran for the SMD seats while being listed on the party list (McKean and Scheiner, 2000).61 The mixed system encourages the electorate to split their votes; the LDP won more votes in single-member districts than in PR districts (Kohno, 1997; Mair and Sakano,

60 In some research, investigators merely looked at the way that Japanese voters reach the final voting decision instead of assessing the weight of different factors systematically (Richardson, 1977; Rochon, 1981). 61 Each candidate can run in a single-member district or multi-member district or both. When he wins the single-member-district seat, he takes it and leaves the party list. All candidates ranked

160

1998). Because of the dual candidacy and split-voting, the mixed electoral system may preserve candidate-oriented voting in single-member districts. Together, these two elections allow us to observe voting behavior before and after the electoral reform. The 1976 data was collected in the 1976 general election and is called JABISS data. The 1996 data--the Japan Election and Democracy Survey

(JEDS)--was collected in the 1996 general election. Both datasets are distributed by the Bukutakusha company.

Before we explore the determinants of voting behavior, it is necessary to understand the party system in Japan. Essentially, the evolution of their party system coincides with the emergence of long-term social cleavages and short-term political issues. Therefore, a brief review of the party system back to 1955 will be presented.

5.2 Context of the Japanese Party System

5.2.1 The Party System of 1955

Before the Meiji Restoration in the 1860s, Japan was an agrarian, feudal society. The arrival of American Commodore Matthew C. Perry with a naval fleet forced Japan to open its foreign trade to the world and to adopt economic and political reform, including strengthening the power of the emperor. During the process of social mobilization, some of the old social classes lost their original privileges granted by local nobles. Under the imperial authority, the Meiji elites crafted a constitution that preserved the emperor but invested power in the cabinet

beneath him on the party’s list can move up. Some dual candidates who are defeated in single- member districts could still win PR seats.

161

and the Diet. Wide political participation not only bolstered the legitimacy of the new government, but also pushed the old social groups to become organized. In the 1880s, the small local groups running at the prefectural level became larger units of organizations, and the conference of local groups turned into the Liberal

Party in 1881. The Constitutional Progressive Party, which consisted of a higher percentage of public servants and intellectuals than the Liberal Party, was founded in 1882. Like the party formation in England, small organizations emerged and expressed their interests in the new parliament. However, geographically-based factionalism, and, as a result, intra-party conflict were intense in the Liberal Party and the Constitutional Liberal Party, which was formed in 1890 (Fukui, 1970).62

After World War II, the first HR election under the Peace Constitution was held in April 1947. The lower house consisted of 466 members chosen from 117 multi-member districts. The 1947 general election under the new constitution resulted in the first Socialist government. Due to a government scandal and a series of intra-party struggles over the San Francisco Peace Treaty and the U.S.-

Japan Security Treaty, the right and left wings within the Socialist party temporarily split. In 1955, however, the Socialist party reunified because they expected to win the election. The Liberal and Democratic parties, which took over the socialist government after its fall in 1948, combined as the Liberal Democratic

Party (LDP) to combat the unified Socialist party. The first postwar party system began with the formation of the LDP and the JSP in 1955. After winning the 1958 election, the LDP maintained its power for the next thirty-eight years. In the

62 The Progressive Party succeeded to the Constitutional Progressive Party in 1896, and merged with the Constitutional Liberal Party in 1898 to form the Constitutional Government Party (Kenseito).

162

twelve elections held during that LDP-dominant period, the LDP was short of a majority in four elections (1976, 1979, 1983, 1990). Hrebenar (1986) therefore contends that Japanese party system is multi-party system in nature. Thanks to the fragmentation of the opposition parties, however, the LDP controlled the position of the prime minister until 1993, when non-communist parties .

The JSP inherited the legacy of the protest in the 1920s; it was considered to be the party of “opposition to anything.” (Stockwin, 2000). It opposed the alliance of the U.S. and Japan, and promoted the labor union movement, women rights, and the abolition of the emperor system. The JSP consisted of three main factions: right, left, and moderate.63 The leaders of the three factions had different views of the problems of Japan’s foreign policy but shared a belief that the emperor system must be abolished because it was a legacy of the pre-war period and was against democratization. Moreover, they believed that Japan must avoid of further involvement in war. Because of the fear of war and occupation, the

Japanese welcomed the JSP’s “permanent neutrality” policy adopted in 1949. The outbreak of the Korean War and the rising communist power in China increased the tension between conservatives and progressives, and the Right and Left

Socialist party reunited in 1955 by reconciling their foreign policies. In 1960, however, the increasing intra-party struggle over ideology and foreign policy led

63 The right wing faction supported the security principle behind Japan’s alliance with the U.S., but the left wing faction promoted neutralism, which includes the abolition of indigenous armed forces, and opposed the Japan’s integration into the U.S.’s defense system. The center faction leaned towards either the right wing faction or the left wing faction sometimes (Stokwin, 1968). Moreover, the right wing faction wanted to shake off the Marxist ideology that pursues the rule of the proletarian class; instead, it supported social democracy and parliamentarism. To achieve this goal, the right wing faction proposed that the Socialist party extends its base from labor unions to the mass public. However, the left wing faction, which was led by the intellectuals, demanded more radical social reform than the right wing faction called for. The left socialists contended that social problems can be resolved only through socialism or even revolution.

163

to a split of the JSP; the right wing faction founded the Democratic Socialist Party

(DSP). The emergence of the DSP as well as the Clean Government Party

(Komeito) weakened the electoral success of the JSP through the 1970s. The

JSP’s turning point came in the 1989 upper house election, when the LDP was toppled by the Recruit Cosmos scandal and its proposal of a consumption tax that irritated Japanese voters.64 After the 1993 election, the non-communist parties organized a coalition government.65 The JSP finally shook off their reputation as a

“permanent opposition party,” but new parties that reshuffled the existing parties gained more seats and the JSP only came in fifth in the 1996 election.

The LDP singlehandedly ruled Japan between 1955 and 1993, in part because of a stable succession of party leaders, the president of the LDP.

Richardson (1997) argued that the institutionalized factions had ensured the smooth transfer of power since the 1950s. The president of the LDP is selected in the periodic elections and will form the government once selected. Factions are assigned cabinet positions in proportion to their number of Dietmembers.

Factional leaders also decide which candidate can receive the LDP endorsement, and each faction has its own money resources (Scalapino and Masumi, 1964;

Tomita, Nakamura, and Hrebenar, 1986). The “parties within party” politics eventually led to a split in the LDP in the 1990s. A new-generation of

Dietmembers who saw the demise of one-party politics convinced other members

64 Recruit Cosmos was a property holding company donated 1.3 billion yen (about $10 million) to the LDP politicians through pre-issue stocks and purchase of tickets to the fundraising parties. Prime Minister Takeshita and four LDP cabinet ministers were forced to resign, and all of the major LDP faction leaders except the Komoto were found involved in the bribery scandal. Additionally, both the DSP leader and the CGP chairman resigned their posts for receiving the money.

164

within the same faction to organize new parties, whose emergence began the era of the Second Party System. In short, it was the power struggle rather than policy or ideology rift among factions led to the collapse of the LDP in the 1990s.

5.2.2 The Party System of 1993

In May 1992 the former Dietmember Hosokawa defected from the LDP and organized a new conservative political party, the Japanese New Party (JNP), which called for clean politics and the end of the LDP monopoly. With

Hosokawa’s departure, the remaining members of the Takeshita faction, which lost many party leaders during the Recruit Cosmos scandal, formed the Shinseito.

The Shinseito later merged with the JNP to form the New Frontier Party. Another

LDP splinter is the Harbinger Party (Sakigake), which was founded by several former LDP Dietmembers in 1993. After the Harbinger Party failed to cooperate with the LDP to form a cabinet in 1997, its key members left and organized another new party: the Democratic Party (Minshuto). Party fragmentation involved policy struggle more than policy debate (Mair and Sakano, 1998). Table

5.1 shows the distribution of seats in the lower house between 1990 and 1996.

65 Before then, there had existed possibilities of inter-party coalition government after the 1976 election. However, the government coalition did not occur until 1993 because of ideological cleavages and complicated factional politics. See Richardson (1997: 86-88).

165

1990 1993 1996 Party Seats/Fate Seats/Fate Seats/Fate LDP 275 223 239 JSP 136 70 15 CGP 45 51 joined the LDP DSP 14 15 faltered JCP 16 15 26 SDF 4 4 faltered Minor 1 0 1 Independent 21 30 9 JNP not founded 35 merged Shinseito not founded 55 merged Harbinger not founded 13 2 NFP not founded not founded 156 DP not founded not founded 52 Total 511 511 500

Data: Hrebenar (2000)

Table 5.1: Lower house seats and fate of Japanese parties, 1990-1996

In 1993, the LDP fell from power for the first time since 1955. The non- communist coalition government, however, did not last more than one term of three years. The LDP called for a general election in 1996 and won 15-seats short of a majority in the Diet. The NFP and the DP came in second and third. The NFP and the DP turned down the LDP’s offer to join a coalition government, so the

LDP formed a minority government by itself.

The political landscape changed a lot during the 1990s with the emergence of new conservative parties, the non-communist coalition government, and the resurgence of the LDP after 1996. This dissertation assumes that relative influence of party attachment is positively responsive to the change of the electoral system from SNTV to mixed SMSP and SNTV system. Under the SNTV system, intra-

166

party competition weakens the importance of party labels and encourages particularized interests (McCubbins and Rosenbluth, 1995). The SMSP system, however, strengthen the role of parties for each party nominates only one candidate. Scholars have claimed that party organizations have responded to it by emphasizing party discipline in the nomination for the single-member districts

(Gallagher, 1998; Curtis, 1999). To confirm the hypothesis that the change in the electoral system will result in different voting patterns, I will investigate the long- term forces in voting choices first, and then examine short-term forces.

5.3 Long-Term Forces in Japan Politics

As mentioned earlier, prefectual groups rather than individuals joined the party organizations in the 1880s, so political parties were confederations of local groups. This geographically based membership led to group loyalty and exchange of voting support for representation of group interests. Therefore, we start from the sociological perspective of voting behavior and will move on to a review of partisanship and ideology.

5.3.1 Social Groups and Japanese Parties

In general, Japanese society has less differentiated social cleavages than in

European countries, but economic development created diverse social groups. On the one hand, Japan society is very homogenous in terms of ethnicity, religion, language, and culture; no major immigration has taken place and no religional or regional conflict has occurred since the 19th century. On the other hand, Japan has been also a highly industrialized country and a variety of social groups were

167

organized to campaign for their interests, such as labor unions, business, and farmers. With modernization of society and change in the international environment, social groups demand and attempt to preserve their interests, which led to the emergence of social issues and, as a result, policy-making. For example, suburban area residents protested against pollution from factories, which cost the

LDP voting support from those areas. When the government was faced with pressure to open markets, farmer groups asked for higher taxes on imported agricultural products or more subsidies for domestic products. Tax reform aroused disputes among business groups; retail associations and industry associations have different views on collecting and reporting of the income tax. Business groups among others lent strong support to the LDP by donations and consulting services.

Although they play an important role in Japanese society and have relevance for voting, Japanese occupational groups are not viewed in the social class terms of other industrialized countries. While some social groups with a stake in public policies traditionally support a specific party, there is no clear-cut relationship between voting choices and occupations (Richardson, 1997).

According to Watanuki’s (1991) study on the 1958 election, farmers and self- employed salaried workers prefer the LDP, but salaried workers and employees also are close to the JSP. Richardson (1997) reported that a significant portion of farmers and small-business support the LDP, and a third of salaried employees and manual workers favor the LDP. Because of the large number of small-firm employees, the LDP enjoyed stable support over the course of the 1955 system.

Figure 5.1 shows that farmers were an important base of the LDP before the 1960s. Beginning in 1965, the percentage of farmers among the LDP

168

supporters shrank. Nevertheless, the LDP received support consistently from people who belong to the categories of labor, salesperson, manager, and employee.

The emerging middle class lent their support to the LDP and made up for the shrinking farmer population. Miyake (1989) also pointed out that intra-class mobilization weakened the structure of social class and strengthened the association between occupations and parties.

3 3 3 5 5 7 15 13 Other 29 19 ory 34 43 Farmer 23 19 25 Business 28 27 14 16 Labor 14 25 12 12 14 Salesperson 11 10 ccupation Categ ccupation 18 8 7 8 Manager O 15 4 23 14 18 16 19 21 Employee

1955 1962 1965 1975 1980 1985 Year

Data: Miyake (1989:88) Note: Entries are the percentage of each occupational category among the LDP supporters

Figure 5.1: Occupation categories of LDP supporter, 1955-1985

Although occupation appears to anchor party choice, a multi-variate regression model shows it has very little impact on voting choice (Richardson,

1997). Besides, parties penetrated into the occupational groups that used to support other parties. For instance, the socialist parties win some seats in agricultural districts because they support the individual farmer movements, which does not fit the sociological prediction. In spite of that, social groups exert

169

influence on people through social networks. Different occupational groups have involvement in different types of social networks: union membership, neighborhood contact, and trade associations (Watanuki, 1991). The members of the social groups share the same concern over policies and vote along with group leaders. Therefore, the effect of social groups shall be estimated with social networks.

5.3.2 Partisanship in Japan

Meaningful party competition in Japan dates back to the late 19th century.

Originally, the lower house consisted of 300 members elected from single- member districts. In 1925 a bill was passed to change small districts to medium sized districts. Under 1947 constitution, 466 Diet members were elected from medium sized districts, each with 3 to 5 seats. In the 1976 general election, there were six parties competing for 511 seats. The New Liberal Club (NLC), led by a group of defecting LDP members, won 17 seats and 4.1 percent of vote share. Its success encouraged party fragmentation in the later elections (Hrebenar, 2001).

Therefore, the NLC is included in the analysis. The 1955 one-and-a-half party system became a multi-party one after the 1993 election. The LDP regained a majority in the lower House, and the NFP and the DP came in second and third.

The JSP and the JCP retained the socialist forces, and those five parties nearly put the Harbinger party (Sakigake) to rout (see Table 5.1). All in all, five parties were regarded as effective parties in the 1996 election. The partisanship variables in both surveys are constructed from a series of questions. The first question is

“Which of Japan’s contemporary political parties do you support?” When the

170

respondent supports anyone of them, we probe his intensity of support. When the respondent does not like any party or has no opinion, they are asked, “Well, is there some political party that you ordinary like?” As for the 1996 survey, the question reads like, “Aside from elections, what party do you normally support?”

If the respondent picks one, he will be asked, “Do you support the party enthusiastically? Or, don't you support it enthusiastically?” If the respondent cannot pick any of them, he will be asked, “Is there a party which you favor normally, even though you do not support it?” An ordinal-level variable is created for each party, indicating the strength of partisanship from 0 to 3.

40 LDP 35 JSP

30 CGP JCP 25 DSP 20 Independent Percentage Don't know 15 NFP 10 SDP DP 5

0 1976 1983 1986 1989 1990 1992 1996 Year

Data: 1976 JABISS data, Miyake (1985), 1996 JEDS data.

Figure 5.2: Party identification of Japanese voter, 1976-1996

171

Figure 5.2 displayed the volatility of party identification across time.

Japanese voters turned their backs on the LDP in 1976 and 1989 because of the

Lockhead scandal and Recruit bribery scandal respectively, but their support returned to its prior level in 1990. When the LDP was caught in the scandal, the support for the JSP and the independents increased. The JSP went up to 22% and the independent 24%. When the LDP courted voters by moderate political reform, however, the JSP also lost its momentum and its party support slipped to 18.9% in

1990 and 12.1% in 1992. Despite the short-term resurgence of the LDP in the

1990s, some people have lost their confidence in the ruling party. Therefore, the percentage of independents soared in 1992 again. Overall, Figure 5.2 shows that the instability of the LDP and the JSP party attachment. The percentage of LDP party identification ranges from 25% to 40%, and that of JSP party identification from 10% to 22%.

The unstable party attachment and increasing number of independents may result from weak party strength. Compared to the U.S., the percentage of Japanese who hold strong party attachment is relatively small. A series of surveys conducted between 1966 and 1986 show that less than 30 percent of people have strong party attachment (Miyake, 1989). However, up to 40 percent of respondents in the SRC/NES surveys conducted between 1952 and 1982 reported strong party identification (Niemi and Weisberg, 1993).

Another possibility is that partisanship in Japan is more cognitive than affective. Previous research confirms that Japanese partisanship consists of both affective and cognitive components (Richardson, 1986; Richardson, 1997). People not only have predisposition toward a given political party, but also pay attention

172

to how party leaders respond to the current issues, particularly scandals. For instance, the electorate votes for socialist parties as a protest against the conservative parties, which can be seen in the 1989 upper house election. The

Recruit scandal destroyed the LDP’s credibility, and it lost its majority in the upper house.

Moreover, partisanship is not exogenous to voting choice; instead, it reflects the previous voting choice. In Table 5.2, the consistency between party support and party choice is higher than that between party support and party attachment and party evaluation. For the LDP partisans, party attachment and party evaluation are equally important as to their party support. For partisans of other parties, however, party evaluation is less important than party attachment.

Richardson (1997) argued that political alienation accounts for the weakness of partisanship, which is due to Japanese culture. The level of political trust and efficacy is lower than in other countries, so people have a weak affective tie with political parties.

Party Attachment Party Evaluation Past Vote LDP 87.8 87.8 89.6 JSP 82.1 70.0 90.8 JCP 75.2 56.1 92.9 Komeito 91.6 74.5 92.5 DSP 88.5 70.5 82.9 Independent 71.5 42.9 36.0

Data: Manabe (1983: 154) Note: Entries are the percentages that vote choice matches party attachment, party evaluation, and past vote, 1973.

Table 5.2: Consistency level of voting choice, party attachment, party evaluation, and past vote by partisanship

173

Richardson (1986) argued that Japanese voters consistently support certain parties in the absence of strong party attachment, which he called “habitual voting”. Past voting habits provide cues for current voting choices as people lack stable party identification, so Japanese voters consistently vote for the same party with weak partisanship. Richardson pointed out that habitual voting could result from voting for the same candidate, which leads to voting for the same party.

Therefore, habitual voters are more likely to vote on the basis of candidates than are stable identifiers. Miyake (1985) also found that weak partisans increase their level of partisanship after an election. The conception of habitual voting challenges the theory of exogenous partisanship, but there is actually no panel data to examine its existence. More significantly, the consistent past voters chose the LDP overwhelmingly, which may be due to a lack of competition in districts.

To explore why people vote for the same party across two elections, I regress the consistency of two vote choices by the strength of partisanship and candidate evaluations. 66 If people vote for the same party because of their candidate preference instead of strong partisanship, the effect of partisan strength on the consistent vote choice should be no greater than or equal to candidate evaluations. In other words, the stability of the party system results more from the extent to which people like the candidates than the level of psychological attachment. Table 5.3 and Table 5.4 confirm this hypothesis.

66 The dependent variable is whether the vote choices in 1972 and 1976 are consistent. Partisan strength is measured by the 0-3 ordinal scale for each party. I measure candidate evaluations as the number of requirements that respondents think the candidate can fit, and obtain the strongest candidate evaluations among the six parties. Detailed operationalization of candidate evaluations is given in Section 5.4.1

174

Dependent Variable: Whether or not the 1972 vote and the 1976 vote are consistent Coefficients Standard Deviation P Constant -.747 .0935 .000 Partisan Strength .538 .056 .000 Candidate Evaluations .589 .077 .000 Number of Observations 1332 Pseudo R-square .1026

Data: 1976 JABISS data

Table 5.3: Logistic estimates of consistent vote choice between 1972 and 1976

Variable Centered change of one standard deviation around the mean Partisan Strength .130 Candidate Evaluations .135

Table 5.4: Discrete change of probability in consistent vote choice between the 1972 and 1976 election

According to Table 5.4, the effect of candidate evaluations on consistent voting is nearly equal to that of partisanship. This suggests that Japanese voters take candidates and parties into account as they vote for the same party across the two elections. Whether or not candidate evaluations anchor Japanese voting behavior across time, though, remains an empirical question.

Third, partisanship is acquired through informal groups and formal organizations (Miyake, 1985). The homogenous society that lacks deep social cleavages handicaps the socialization of partisanship (Kabashima and Ishio, 1998).

Group-oriented living and working styles shape the collectivism, so that

175

individual preference over political parties is not as strong an inherited attitude as in other countries. The intermediate effect of social networks on partisanship is reinforced by candidate contacts; candidates reach voters through organizations.

Miyake (1991) analyzed the influence of formal and informal organizations and concluded that they are more important than the family in instilling partisanship.

Though American scholars also confirmed the role that secondary groups play in socialization, Japanese voters appear more likely to acquire partisanship from their peer groups and working places. This suggests that we must consider the way that social networks interact with partisanship while constructing a voting model.

The analysis and literature review above provide several points on the

Japanese party system. First, Japanese partisanship is unstable, in part because people acquire partisanship more through social networks than from a socialization process. Second, Japanese voters are responsive to both long-term and short-term forces. Furthermore, Japanese voters tend to vote for the same party even in the absence of partisanship, which highlights the importance of candidate-centered campaigns in multi-member-district elections.

5.3.3 Ideology and Value Cleavage

The 1955 system was based on the post-war ideological division between socialism and conservatism. The socialist parties argued that Japan should be independent from the United States in both politics and economy, but the conservative parties pursued a military alliance with the U.S. to ensure economic growth. The socialist parties also demanded clean politics and less political funds from business, which the LDP relied on heavily. The ritual power of the emperor

176

and the territory issue are also related to the theme of party competition. All of these issues boil down to a fundamental ideology: socialism and conservative. The two ideological camps share the same nationalism, but they disagreed with each other on the issues of political reform and labor unions.

In addition to left-right ideology, value cleavage is another likely factor that influences the Japanese voters. Watanuki (1967) argued that the underlying value system is critical to political behavior. A paternalistic pattern of interpersonal relations and collectivism still persists in the older cohort of

Japanese, but economic growth and the Occupation period brought in modern values that challenged the role of conformity and authority. Miyake (1985) developed “conservative-progressive” and “traditional-modern” dimensions. The two dimensions both carry the general perceptions toward the six parties; the conservatives view the JCP and the JSP as traditional parties, but the progressives see the DSP and the New Liberal Club as traditional parties. Flanagan (1991) found several dimensions of value cleavages and suggested that value cleavage is associated with ideology and partisanship. He asserted that there are two dimensions in the Japanese value system: authority/autonomy and conformity/openness. Both value system and ideology, therefore, determine voting choice to the extent that partisanship and mobilization cannot account for.

Under the 1955 party system, left-wing and right-wing supporters had obvious parties to vote for. When Japan entered into a multi-party system in 1993, however, the confrontation between the two ideology poles dissolved and the JSP lost the protest votes. A certain portion of the progressives stayed with the JCP, and some left-wing voters left for the NFP. Table 5.5 shows that in 1976 the

177

conservatives supported the LDP and the progressives tended to vote for the JSP and the JCP, and Table 5.6 reveals the replacement of the JSP with the NFP in

1996.

LDP JSP JCP CGP DSP NLC Total Conservative 85.5% 3.2 0.2 1.9 3.8 5.5 100% Neutral 33.5 29.1 6.4 13.3 9.4 8.4 100% Progressive 4.1 62.2 18.9 6.2 7.4 1.2 100%

Data: 1976 JABISS data Note: Table entries are the percentages within the ideology categories that support each party. Pearson Chi-square value: 679.396, d.f.: 10; p<=.001

Table 5.5: Ideology and party choice, 1976

LDP NFP DP JSP JCP Total Conservative 67.1% 24.4 5.1 .9 2.5 100% Neutral 43.0 31.6 9.7 2.8 12.8 100% Progressive 21.0 34.1 17.4 4.3 23.2 100%

Data: 1996 JEDS data Note: Table entries are the percentages within the ideology categories that support each party. Pearson Chi-square value: 113.547, d.f.: 8; p<=.001

Table 5.6: Ideology and party choice, 1996

As regards to value cleavage, four variables are created to represent four dimensions of existing values: conformity, tradition, individualism, and localism. 67 Conformity, showing respect for group decision and avoiding

67The value system question wordings are as follows. Conformity question in the 1976 survey reads, "Do you agree or disagree that it meetings in your neighborhood, place of work, PTA or the like, when opinions are sharply divided, it is better to cut the discussion short in order to avoid bad feelings afterward." In the 1996 survey, it reads, "Do you agree that there are many political views

178

confrontation, is viewed as a requisite for social harmony. The second aspect of value cleavage is tradition, which refers to broad-based social custom or even obligation to some people. Individualism has been viewed as a revolt against collectivism in Japan, thus it is also included in value cleavage. The fourth aspect of value cleavage is the choice between local and national interests. The distributive politics in Japan makes this issue salient to voters. These four aspects of value system should affect the distribution of voting support to some degree; people with high levels of conformity, traditionalistic attitudes, individualism, and localism will support the conservative parties because of their long-term association with those values. However, the importance of those issues will differ between elections. For the generation who has just experienced war and still struggles to pursue economic growth, the distinction between tradition and modern is not an issue yet. In 1996, however, the new generation of Japanese voters may see a difference between traditional and modern values.

Simple cross tabulation analysis in Table 5.7 shows that the perceptions about conformity and localism indeed affect voting behavior in 1976. The

in society, so it is natural for politicians to oppose each other or that politics should be carried out in an orderly fashion, even if disputes among members of parliament can't be avoided." Tradition question in the 1976 survey reads, "Do you agree or disagree that these time-honored Japanese morals must be strongly supported." As for the 1996 survey, it reads, "Do you agree that one should preserve the traditional way of doing things so as to maintain social harmony even though new ways may be better, or that it is better to follow one-s own inclinations without sticking to the traditional way of doing things than to put too much importance on social harmony?" The individualism question in the 1976 survey reads, "Do you agree or disagree that respecting individual rights and freedom is more important than filial piety and the repayment of favors?" In the 1996 survey, respondents are asked, "Do you agree that it is better to have a society where people respect each other-s individual life styles and do not interfere with what others do or that it is better to have a society where neighbors and acquaintances help each other when they are in trouble?" Last, the localism question in the 1976 survey reads, "Do you agree or disagree that events in the area one lives in are more important than national incidents such as the Lockheed scandal?" As for the 1996 question, it is worded as: "Do you agree that I prefer a candidate who makes serious efforts on behalf of our local area such as carrying a petition to the central government on our behalf, or that I prefer a candidate who is mainly active in solving national

179

proportion of the LDP supporters increases with the level of conformity, which implies that the LDP appeals to people who praised the value of mutual help and interpersonal dependencies. Flanagan (1991) pointed out that the use of modern machines and urbanization in the 1960s has shattered the importance of mutual dependencies since then. Therefore, the value of conformity was salient to people of different ages and areas. In 1976, localism had an impact on voting choice, but the effect of other attitudes is not impressive.

As for the 1996 election, localism remains influential, but traditionalistic attitude becomes the significant factor. In Table 5.8, the percentage of voting for the LDP, for instance, linearly increases with the level of localism. People with a high level of traditionalistic attitude are also more supportive of the LDP because the LDP places emphasis on stability. In both elections, localism differentiates

LDP supporters and opposition party supporters, which is due to the fact that the

LDP cultivates districts by seeking credit from distributing public projects and subsidies. As for the opposition parties, they can only appeal to the people who care about national interests.

problems such as those relating to foreign affairs even though she may be indifferent to our local

180

Party Choices Level of Values LDP JSP JCP CGP DSP NLC Total Conformity1 Low 35.3% 35.8% 14.5% 4.6% 4.6% 5.2% 100% 36.4 35.6 9.3 2.5 6.8 9.3 100 50.0 29.9 5.9 6.9 5.2 2.1 100 52.2 22.7 4.3 9.2 5.8 5.8 100 High 45.7 25.2 8.6 4.6 11.3 4.6 100

Tradition2 Low 46.7 40.0 6.7 6.7 0 0 100 32.4 37.8 13.5 5.4 2.7 8.1 100 39.9 30.3 10.7 3.9 7.9 7.3 100 47.8 27.8 7.2 7.8 4.7 4.7 100 High 52.0 26.2 5.7 5.7 6.8 3.6 100

Individualism3 Low 45.9 27.6 5.4 5.4 9.7 5.9 100 46.7 24.4 9.6 6.7 5.2 7.4 100 44.7 31.1 7.2 6.4 5.7 4.9 100 57.4 23.3 6.2 3.9 5.4 3.9 100 High 41.3 33.7 9.8 8.7 6.5 0 100

Localism4 Low 26.5 43.4 10.6 8.8 8.0 2.7 100 34.1 42.9 6.6 1.1 8.8 6.6 100 45.7 28.8 8.0 5.8 5.1 6.7 100 53.7 30.3 7.3 8.1 6.1 4.5 100 High 59.3 20.8 6.8 5.9 4.5 2.7 100

Data: 1976 JABISS data Note: Entries are the percentages that people in different level of value cleavages vote for a given party. 1: Pearson Chi-Square Value: 56.208, d.f.: 20, p: <=.001. 2: Pearson Chi-Square Value: 27.498, d.f.: 20, p <=.5. 3: Pearson Chi-Square Value: 24.184, d.f.: 20, p <=.5. 4: Pearson Chi-Square Value: 66.320, d.f.: 20, p <=.000

Table 5.7: Value cleavage and party choices, 1976

affairs?"

181

Party Choices Level of Values LDP NFP DP JSP JCP Total Conformity1 Low 49.0% 30.4% 5.5% 3.2% 11.9% 100% 45.6 28.4 14.2 1.8 10.1 100 50.4 28.7 8.5 3.1 9.3 100 High 52.0 25.3 9.8 3.1 9.8 100

Tradition2 Low 51.3 24.4 8.6 3.0 12.7 100 41.4 32.7 13.6 3.2 9.1 100 53.7 28.4 5.3 3.2 9.5 100 High 59.6 27.9 7.7 0 4.8 100

Individualism3 Low 51.4 29.5 7.3 2.5 9.3 100 46.8 26.9 13.5 2.6 10.3 100 35.5 29.0 16.1 9.7 9.7 100 High 32.3 32.3 12.9 3.2 19.4 100

Localism4 Low 37.0 32.0 15.0 5.0 11.0 100 38.5 30.2 12.5 1.0 17.7 100 48.8 33.0 7.9 3.7 6.5 100 High 56.3 27.7 6.8 1.5 7.7 100

Data: 1996 JEDS data Note: Table entries are the percentages that people in different level of value cleavages vote for a given party. 1: Pearson Chi-Square Value: 12.065, d.f.: 12, p <=.5 2: Pearson Chi-Square Value: 24.418, d.f.: 12, p <=.05 3: Pearson Chi-Square Value: 20.383, d.f.: 12, p <=.1 4: Pearson Chi-Square Value: 33.455, d.f.: 12, p <=.001

Table 5.8: Value cleavage and party choice, 1996

Social groups, partisanship and ideology account for vote choice to different degrees. Due to the homogenous society in Japan, only occupation groups associate with party choice. The social base of Japanese parties is fluid; one-party politics encourages the catch-all party instead of class-based or issue- based parties. Ideology and culture issues anchor vote choice, but their influence may be artificial since they are also related to party identification. In terms of the voting function, the level of agreement between vote choice and party

182

identification decreased and the stability of voting behavior depends on candidate evaluations and partisanship simultaneously. Further consideration of short-term forces with multi-variate analysis will make the assessment of their importance in voting more accurate.

5.4 Short-Term Forces

5.4.1 Candidate Evaluations

One of the distinctive features of Japanese politics is the candidate- centered campaign because of its SNTV system. Not only do candidates have their own campaign organizations, but also they receive funds through their factions or affiliated interest groups.

The distant cause of the candidate-centered voting pattern is the norm of patron-client and community solidarity in Japan. Richardson (1975) pointed out that localism and mobilization in exchange for parochial interests let individual candidates carve out personalized service. Besides, political parties were composed of elites, who pursued civil rights and political participation (Scalapino and Masumi, 1964). Japanese culture encourages personal loyalty to the leader of the group, rather than the group as a whole, according to Richardson and Flanagan

(1984). The voter is more concerned about what a candidate can do for him personally or for his local area. Their findings also suggested that voters pay more attention to local politics than national politics, which supports the theory that most Japanese voters are concerned about what candidates can do for them rather the major policy issues being discussed by Dietmembers.

183

With urbanization and the decline of the traditional community, the diversity of new communities increases and personal connections become less important. Therefore, formal and institutional organizations substitute for traditional personal followings on the basis of particularized interests and community solidarity. The emergence of small parties, which stress their party labels instead of individual candidates, highlights the fact that an increasing portion of the electorate is aware of the role of political parties.

The proximate cause of the candidate-oriented tendency is the multi- member district with the SNTV system. Tomita, Nakamura, and Hrebenar (1986) pointed out that LDP candidates run their own support organizations because the

LDP nominates more than one candidate in multi-member districts. Hrebenar

(2000) also suggested that SNTV resulted in intra-party competition in that each candidate builds his own campaign organization and wins the seat at the expense of his party mate.

Rochon (1981) presented strong evidence that the consideration of candidate evaluations in voting choice increase with district magnitude. As the number of seats in a district increases, the percentage of people who mention candidate as the basis of voting also increases, which implies that individuals would rely on candidate evaluation when the number of candidates increase. Also of interest is that voters stress the candidate factor even controlling for political party and the residence area. Rochon’s convincing findings not only emphasize the role of the election system in shaping voting patterns, but also reject the cultural explanation provided by Richardson and Flanagan (1984). In spite of his institutional theory and innovative test, however, he did not really explore the

184

determinants of voting choice. The respondent’s report of the basis of voting can be misleading because voters actually choose the candidate rather than the party.

Moreover, voters can pick party first and then candidate, or vice versa; the answer to this question is not complete without asking the respondent if he decides the candidate or the party first.

The success of candidate-centered campaigns depends on party coordination in the endorsement process. Cox and Rosenbluth (1993) showed that, other things being equal, factional-backed candidates are more likely to receive party endorsement than non-factional-backed candidates. New candidates will shop around factions and districts to seek support, and factions may support new candidates in districts where they have no incumbent (Curtis, 1988). It takes several elections for political parties to learn how to coordinate candidates with success (Reed, 1991). One of the reasons that the LDP could maintain its one- party dominance is because it could maximize its seat share by coordinating candidates. Figure 5.3 shows the party popularity of the LDP and the JSP and their vote shares, which makes clear that, through coordination, a candidate can win more seats than the party deserves. Before the 1970s, the LDP won more votes than its popularity predicted because of the strength of its candidates. After the

1970s, however, the JSP candidates went beyond the ceiling of party popularity.

Although the JSP continued to decline through the 1970s and the 1980s, its success in nomination reduced the seat bonus of the LDP.

185

LDP--Vote Share 40 JSP--Vote Share LDP--Popularity 30 Percentage JSP--Popularity

0 1958 1960 1963 1967 1969 1972 1976 1979 1980 1983 1986 Year

Source: Compiled by the author from Curtis (1988: 198-199)

Figure 5.3: Party popularity, vote share, and seat share, 1958 to 1986

Individual-level data also confirms the importance of candidates in elections. First, social networks constructed by interpersonal relationship provide candidates with access to voters. Flanagan (1991) made a point that casework activities include playing the role of mediator in settling conflicts and assisting constituents. Voters can deliver their requests to politicians through the leaders of neighborhood associations, and politicians are active in soliciting their services.

As a result, voters are exposed to the influence of candidates.

Second, there is more than one candidate sharing the same party label in a district. Political parties, especially the LDP, nominate more than one candidate in the district to win a majority in the Diet. In this case, the personal characteristics of candidates rather than party labels will affect the way that voters perceive candidates. Table 5.9 shows that the districts with more than one seat at stake

186

typically have more than one candidate of the same party label. In 1958, for example, there were 2.16 candidates of the same party label in districts with three seats, and 3.23 candidates from the same party in districts with five seats. The average number of candidates of the same party label declined over the twelve elections between 1958 and 1990, yet most districts still have more than one candidate endorsed by the same party. Obviously, candidates must appeal to the electorate with resources in addition to party labels.

District Magnitude

1 2 3 4 56 1958 1.00 2.16 2.50 3.23 2.67 1960 1.50 1.85 2.25 2.84 2.35 1963 1.00 1.79 2.17 2.59 2.22 1967 1.25 1.67 1.96 2.37 2.03 1969 1.50 1.58 1.80 2.12 1.86 1972 1.00 1.69 1.91 2.13 1.93 1976 1.67 1.48 1.85 1.96 1.78 1979 1.00 1.56 1.81 1.89 1.77 1980 1.00 1.46 1.78 1.87 1.71 1983 1.00 1.57 1.83 2.11 1.85 1986 1.00 1.53 1.57 1.87 1.86 1.86 1.78 1990 1.00 1.42 1.61 1.86 1.94 2.00 1.81 Average 1.18 1.47 1.67 1.97 2.25 1.93 1.99

Data: Compiled by the author from the Lijphart Archive.

Table 5.9: Average number of candidates of same party label by district magnitude

In some opinion polls, voters indicate their attention to both political parties and candidates. In the wake of the 1990 general election, 46 percent of respondents in a poll conducted by Asahi News (Naikasusouridaijin, 1990)

187

indicated that party is more important than candidates, and 45 percent considered candidates as a more important reason to vote than party. Miyake’s (1989) study of the 1983 election suggests that the level of candidate recognition positively affects the tendency to vote on the basis of candidates. Several similar routine surveys also show a tendency for people to put more weight on candidates than political parties. As shown in Table 5.10, more voters based their decision on candidate personalities than on party attachments before 1967, but the pattern reversed beginning in 1972. Still, candidate evaluations affected the voting act.

Year Party Candidate Both Don’t Know N 1958 32% 45 10 13 2157 1960 33 43 19 5 2228 1963 31 51 14 3 2107 1967 37 47 11 4 2163 1972 48 38 12 2 2151 1976 46 40 12 2 2048 1978 41 46 12 1 2003 1980 49 38 11 1 2112

Data: The Kommei Senkyo Renmei, cited from Richardson and Flanagan (1984). Note: Table entries are the percentage of responses.

Table 5.10: The most important determinant of voting--party or candidate

Despite the compelling evidence of a candidate-centered voting pattern, scholars have pondered the influence of partisanship on voting choice. Richardson

(1988), analyzing the 1976 JABISS data, argued that the electorate’s regular exposure to information about parties shapes their partisan attitude and voting choices. Candidates and local politics are not as significant as often seen because

188

the mass media are highly centralized in Japan and its coverage of local politics or candidates is very limited. Therefore, people possess more information about parties than candidates, which puts more emphasis on partisan attitude than candidate image or evaluation. His test shows conclusive evidence that party identification, past vote, and party image have higher impact on voting decisions than candidate image and campaign activity do. Watanuki’s (1986) analysis of the

1983 JES data revealed a similar voting pattern; party preference prevails over candidate evaluation and mobilization. Finally, a path model that summarizes the determinants of the 1976 voting decision confirms the strong impact of party image and partisanship; constituency candidate images and requests are not as important as their counterparts (Flanagan, Kohei, Miyake, Richardson, Watanuki,

1991).

Before proceeding to the complete voting model, we shall look at the source and the level of candidate images and compare them with party image and partisanship. According to Watanuki’s (1986) report on the 1983 JES data, there are significant differences in the level of candidate recognition between different groups of genders, age group, and level of education. In this dissertation, I extend the demographic variables to social networks, which will be addressed later. The operationalization of the candidate evaluation variable is based on the number of times that each party’s candidates fit the requirements such as “has done something for the people of this area” or “did some special favor for you or some member of your family.” The more the respondent mentions about the candidate, the more credits that the respondent gives to the candidate. According to Hinckley

(1981), candidate qualifications represent how voters rate the candidates.

189

Therefore, the constructed candidate evaluation variable should capture the effect of the candidate factor for voters. Moreover, this measurement is comparable with

Taiwan and the U.S.

People derive information from social networks and campaign activities as well, since Japanese candidates emphasize personal contacts and services instead of party labels. Additionally, candidate judgment is expected to come from both mass communication and long-term party attachment, which in theory filters political information for individuals. For the conservative parties, the recommendation or request from neighbors, coworkers, family members, friends, and other organizations is more critical to the assessment of candidates because of the strong linkage between supporters and candidates. The opposition parties, though relying on partisanship, also emphasize the mobilization of labor unions, supportive organizations, and social networks. It is believed that personal contact and service are what drive the recognition and support of candidates that occur in group endorsements and personal requests.

Table 5.11 and Table 5.12 show the strong tie between candidate evaluations and social networks in both elections. For the 1976 election, participation in support organizations and partisanship have cross-board impact on candidate evaluations. Partisanship also was strongly associated with people’s perception of candidates, particularly in 1996. For the LDP candidates in the 1976 election, high evaluation is associated with campaign mobilization, such as meeting, support organization, and postcards or flyers. But for the JSP candidates, union endorsement and family requests are important. Regarding the JCP candidates, requests from people in the same working place have enormous

190

impact on their images. Candidate support organizations and meetings are important to nearly every party candidate; candidates undoubtedly resort to their personal organizations regardless of their party labels.

For the 1996 election, campaign mobilization is still influential, yet the effect of partisanship becomes stronger. Hinckley (1981) noted that both incumbency and party identification provide voters with low-cost information. As the district size decreased, partisanship accounts for more variations in candidate evaluation. The transition of party systems contributes to the rising importance of parties in candidate assessment.

LDP JSP JCP CGP DSP NLC Constant .078*** .009 .022** .012 .011* .015*** Association1 .126*** .112** .042 .169*** .146*** -.007 Union2 -.058 .218*** -.083 -.211 .012 .182* Request3 Family .187* .391*** .094 -.028 .476*** .092 Coworker .283*** .229*** .680*** .144** .021 -.015 Neighbor .044 -.117 .050 -.014 -.188** -.015 Friend .092 .172 .091 .027 .269*** -.015 Campaign4 Support .453*** .762*** .906*** .488*** 1.027*** .536*** Organization Postcard .168*** .150*** .050 -.079* .021 .015 Meeting .644*** .434*** .637*** .534*** .523*** .111 Partisanship5 .067*** .161*** -.005 .491*** .070*** .041**

Adj.R-square .257 .311 .259 .429 .314 .040

Data: 1976 JABISS data Note: 1: the number of associations that the respondent participate; 2: dummy variable indicates if the respondent is union member; 3: the source of personal requests; 4: participation in or getting information from campaign organizations; 5: 3-point scale partisan strength.

Table 5.11: Sources of candidate evaluations, 1976

191

LDP NFP DP Constant 1.265*** .082*** .178*** Association1 .704*** .512* .724*** Union2 1.001 .530 2.532*** Requests3 .092 .177 .406* Campaign4 Support -.081 2.235*** 2.532*** Organization Postcard .431*** .536*** .794*** Meeting 1.992 2.195*** -.924 Partisanship5 .713*** .669*** .391*** Number of Cases 1243 1243 1243 Adj. R-square .204 .187 .244

Data: 1996 JEDS data Note: 1: the number of associations that the respondent participate; 2: dummy variable indicates if the respondent is union member; 3: the number of personal requests; 4: participation in or getting information from campaign organizations; 5: 3-point scale partisan strength. Due to the small number of respondents that participate in campaign support organizations, receive postcard, and attend meeting for JSP and JCP, only the candidate evaluation of the three major parties is estimated.

Table 5.12: Sources of candidate evaluations, 1996

Since social networks that deliver messages for political parties shape candidate evaluation, the choice of candidate should agree with party choice. To some degree, the preference over candidates should agree with the preference order of parties. Due to the large number of parties, only the top ranked candidate is used to predict party choice. Table 5.13 and Table 5.14 show evidence of the agreement between the preference order of candidates and parties. For example,

59 percent of people who pick the LDP representatives as their favorite candidate

192

among others vote for the LDP in the 1976 election. By contrast, only about one- third of the people who favor JSP candidates voted for the JSP. The level of agreement between candidates and party choice is higher in the 1996 election.

More than 60 percent of people who give the most credits to LDP candidates vote for the LDP eventually, and more than half of voters who like NFP candidates vote for the NFP. What is surprising is how few people who like JSP candidates vote for the JSP. Only ten percent people who prefer the JSP candidate also choose the JSP.

Party Choice Candidate Preferred LDP JSP JCP CGP DSP NLC Row Total LDP 59.0 21.7 5.6 4.1 5.5 4.0 100 JSP 43.3 35.7 6.1 4.8 5.5 4.5 100 JCP 48.6 24.3 10.7 5.1 6.2 5.1 100 CGP 46.7 23.0 10.3 9.2 6.0 4.7 100 DSP 58.5 20.5 5.7 3.9 7.4 4.0 100 NLC 47.5 24.3 10.2 5.0 6.0 6.9 100

Data: 1976 JABISS data Note: Entries are the percentages that the more preferred candidate agrees with voting choice.

Table 5.13: Agreement between candidates and voting choice, 1976

193

Party Choice Candidate Preferred LDP NFP DP JSP JCP Row Total LDP 65.3 20.2 5.4 0.7 8.4 100 NFP 31.6 54.0 6.3 0.9 7.2 100 DP 33.5 26.2 28.2 2.9 9.2 100 JSP 41.9 30.6 8.8 10.0 8.8 100 JCP 36.2 30.0 7.5 1.4 24.9 100

Data: 1996 JEDS data Note: Entries are percentages that the most preferred candidate agrees with voting choice.

Table 5.14: Agreement between candidates and voting choice, 1996

The findings above confirm the critical role played by candidates in

Japan’s elections. Perceived by voters through social networks, candidate evaluations indeed agree with party choice. Because of the candidate-oriented historical context and the electoral rules, individual candidates have more influence on people’s vote choices than does partisanship. When the election system was switched from SNTV to simple-majority single-member district, however, the linkage between candidates and parties strengthened and partisanship now shaped candidate evaluation to a higher degree. In this case, the agreement between party choice and candidate evaluation should be interpreted as evidence of the rising role of partisanship.

5.4.2 Issue Positions and Voting

As discussed in section 5.1, Japanese parties are not only concerned about mobilization; they also compete with each other on critical policy issues. The

194

confrontation between the conservative parties and opposition parties ranged from the constitutional issue to political reform. At the beginning of the 1955 system, the conservative and progressive camps took the revision of the 1947 constitution as their primary political agenda. The conservative party favored development of military forces, which requires the revision of Article 9 that forbade the use of force as an instrument of Japanese foreign policy and forbade the possession of armed forces. In other words, the conservatives intended to restore the military power prior to World War II. The opposition parties were concerned about the basic values like labor union rights and the principle of peaceful polity. Moreover, the conservative party emphasized the Japan-U.S. alliance, which became very important when the U.S. sent troops to Korea. The opposition party, on the other hand, opposed any involvement in conflict. The Japan-US alliance was still a sensitive issue, particularly because the U.S. navy uses several Japanese harbors as its military bases to monitor China and North Korea. Parties also clashed on economic issues like inflation and social welfare programs. In the 1970s, ideological conflict was replaced by environmental issues that were caused by the rapid economic growth with little environmental regulation. Lee (1971) found that the LDP and JSP cannot satisfy the electorate’s growing concern over diversified social conditions, notably in urban areas. The public dismay with the two major parties gave rise to minor parties, like the DSP and CGP.

In addition to political and economic issues, the LDP and opposition parties debated foreign affairs. Diplomatic relations with China, North Korea, and

Soviet Union were related to the issues of territory and power conflict in northeastern Asia. On most issues, the opposition parties echoed the position of

195

the LDP; they merely disagreed with the LDP on the negotiation agenda and the way to set up a new relationship. For example, the JSP and JCP were sympathetic with the Soviet Union on the territory issue, though they supported the return of the small northern islands to Japan.

The electorate was also concerned about corruption. Before the 1976 general election, a former Prime Minister and three other LDP Diet members were arrested for taking bribes from the Lockheed Aircraft Corporation. The opposition parties asked for “purification of politics”, though some opposition party members were involved in the scandal. The Lockheed scandal indeed cost the LDP around

20 seats. The LDP proposed reform bills to ease the discontent of electorate, but

LDP members were divided on the scope of reform.

In 1988-1989, the Recruit Cosmos scandal shocked the Japanese society again. More LDP and opposition party members were indicted for taking political funds in the form of pre-issued stock shares in the company. Both scandals not only made the Japanese voters lose their faith in the governing party, but also created a split in the LDP. Following the Recruit scandal, Prime Minister Uno

Sosuke was found to have an affair with a mistress. Uno’s sex scandal became a crisis of the LDP in the aftermath of the Recruit scandal, and the LDP lost its majority in the upper house in the 1989 House of Councilor election. Prime

Minister Hosokawa, who organized the non-communist coalition government after the 1993 general election, was indicted for receiving gifts from a delivery company, and quickly resigned his post. Overall, the battle among the LDP factions and between it and opposition parties centered on the scandal events, which can explain why the Japanese possess low level of political trust and that

196

issue voting model may not apply to Japanese voters.68 In this regard, power corruption should be a major issue in the two elections that we are investigating.

These four sets of issues--ideology, environment, foreign policy, and power corruption--mirror the evolution of the Japanese party system and the concerns of Japanese voters. Kabashima (1986) reported that tax reduction and political ethnicity are the two most important issues. In addition to these two issues, respondents are also familiar with national defense, the Japan-US Security

Pact, and administration reform and are aware of the parties’ positions on those issues. Miyake (1989) found that people’s perception of political issues can be arrayed on a dimension called “conservative-reformism” image. The most important issue on the conservative side is the return of the northern islands and reinforcement of national defense. On the reformism side, the most important issues are reform of money politics and the regulation of big business. As mentioned above, the disagreement over the territory issue among political parties is slim to none; only the reform issue remains a concern of Japanese voters.

Although Japanese voters recognize political issues, issue voting model does not necessarily apply to Japanese voters. Kabashima (1986) employed issue preference weighted by the level of salience to voters, and found that issue preference and party image predict voting choice well. However, his model does not include the direct influence of party identification and candidates. Miyake

(1989) assessed two voting models in which people’s issue preference on the

Japan-US alliance and money politics as well as party support at the prior time point are used to predict their perceptions toward party policy images and party

68 Regarding the comparison in political efficacy and trust between Japan and the other industrial

197

support, and he found that only previous party support has a significant impact on current party support. Moreover, his analysis suggested that strong partisans have weaker preference over policies; they solely make judgments based on their party attachment. Flanagan (1980) suggested that both economic and value cleavages have important impact on the voting behavior of Japanese, but he did not control for party identification and candidate evaluation when he assessed the relationship between issue preference and vote choice. Flanagan (1991) also found that political corruption, social welfare, and public interest against big corporations are the three most salient issues, but only those people who have high levels of political interest and partisanship would recognize political issues. For a substantial proportion of the Japanese electorate, the link between the choice of the candidate or party and the respondent’s issue position is not very strong.

In a parliamentary system, every single election should entail at least one decisive issue that the cabinet can dissolve the parliament or resign after the passing of a no-confidence resolution. Nevertheless, a low level of political trust and efficacy can weaken the linkage between issue preference and voting decision; people cannot attach great importance to any election. Japanese voters paid attention to only the power corruption issue. For instance, the “Lockheed incident” cost the LDP 22 seats in the 1974 election, and the LDP severely limited the number of its endorsed candidates (Baerwald and Tomita, 1977). The one- party politics indeed made ideological issues (or cultural politics issues) the sharpest difference between the conservative and socialist parties in the mid-1970s

(Kohei, Miyake, and Watanuki, 1991). As a new generation of voters joined the

countries, see Richardson (1997:24).

198

electorate, however, they were no longer engaged in the ideological conflicts

(Richardson, 1997). The new parties emerged in the early 1990s also showed no policy ideas but general promises of political reform (Blaker, 1995). Little issue voting is expected among the Japanese electorate.

Due to the early confrontation between the two ideological camps that alternated power once, political issues indeed played a role in Japanese elections at the early stage of the 1955 system. With the decline in the socialist parties, however, “issue-less” elections prevailed over the course of one-party politics and only political reform remains pivotal to politics.

The 1976 and 1996 datasets have different batteries of issue questions, which makes the comparison of voting pattern between two election systems difficult. To explore the underlying structures of issues, discriminant analysis is used to differentiate groups of voters in the two elections. For the 1976 election data, there are five functions derived from twelve issues. According to the Wilks’ lambda significance test, a single dimension can represent all of the observed difference between groups (Klecka, 1980). I name the function “conservatism” because of the large coefficients for defense issues and the right to strike for public employees. In the 1996 survey data, two functions are extracted from four issue questions. The first function represents national defense, which is one aspect of conservatism. The second function is correlated with the issue position on the protection of agricultural products. At the time, the Japanese government decontrolled the domestic rice market, as part of efforts to liberalize Japan’s economy (Curtis, 1999).

199

Issues Function 1—Conservatism

Japan-U.S. security system should be .706* strengthened The right of public employees to strike -.685* should be recognized Japan’s defense forces should be .636* strengthened A treaty of peace should be concluded with .171 China as soon as possible Stronger measures should be devised to -.054 stimulate business activity The oppression of big business should be .194 checked Contributions by business leaders should be .374 prohibited The ‘money power’ politics should be .178 reformed Social welfare program should be .186 strengthened even if taxes go up The return of the territories from the USSR .057 should be demanded More effort should be put into friendly .322 relations with North Korea The emperor should have a stronger political -.406 voice

Eigenvalue .454 % of Variance 84.3

Note: Table entries are structure coefficients of discrimant analysis

Table 5.15: Structure of political issues, 1976

200

Issues Function 1— Function 2— National Defense Agriculture Protection The Japan-US Security .922* .279 Treaty should be maintained Liberalizing imports -.360 .819* for agricultural products should not be allowed Increasing tax in order -.154 -.311 to improve social welfare and public services For politicians to .206 .375 receive money and other valuables from supporters is all right

Eigenvalue .081 .039 % of Variance 63.5 30.7

Note: Table entries are structure coefficients of discrimant analysis

Table 5.16: Structure of political issues, 1996

Discriminant analysis not only provides us with the linear functions of variables based on the difference between groups of objects, but also shows the centroid of each group, which helps us identify the issue position of each group.

Based on Table 5.17, the JSP and the JCP were on one extreme of the conservatism continuum and the LDP on the other extreme in 1976. The conservative issue continues to exert influence on voting behavior until the 1990s.

The security issue still lingers in the 1996 election, and the LDP and the two socialist parties remain in opposite positions, which is shown in Table 5.18. The

201

agricultural product issue is the other dimension of Japanese politics in 1996. The

LDP and the JCP voters opposed the liberalization of the agricultural market, but the other conservative parties and even the JSP advocated this policy. The unusual alliance of the JCP and the LDP is reflected in their supporters in the election.

However, there are not enough question on diverse issues in the 1996 survey to assess the issue divide that year fully.

Function Vote Choice Conservatism LDP .574 JSP -.683 JCP -1.208 CGP .229 DSP .499 NLC .336

Note: Table entries are the group centroids of observations.

Table 5.17: Issue Position of Japanese Voters, 1976

Function Vote Choice Security Agriculture LDP .139 .174 NFP .084 -.241 DP .022 -.287 JSP -.468 -.104 JCP -.785 .090

Note: Table entries are the group centroids of observations.

Table 5.18: Issue position of japanese voters, 1996

202

Discriminant analysis presents the issue cleavages among different parties and shows similar issue dimensions across two decades. The results confirm the expectation that political issues and ideology are correlated, so it is plausible to use ideology to predict voting behavior. However, the small number of issues in

1996 makes it difficult to compare the issue structure of the two elections. Overall, political issues can explain the difference between groups of voters, but partisanship and candidate evaluations should be more proximate to the voting decision.

5.4.3 Party Images in Japan

When comparing the attitudes and behavior of partisans across countries, scholars have stressed the theoretical role of party images. Party images are a useful concept for Japanese political system. In multi-party system, partisans may or may not support a given party and dislike others. They can like or dislike more than one party. Or they can be neutral toward every party (Wattenberg, 1982).

Furthermore, party image reflects the short-term aspect of partisanship.

Party image is characterized as party evaluations responsive to campaign-related events. In a circumstance that the general election is called for to resolve the national debate, people’s sentiment toward parties can be more significant than candidates or issues. Party images are more campaign-specific than party identification, but are less ephemeral than candidate evaluations or issue positions

(Matthews and Prothro, 1966).

As pointed out in Chapter 3, feeling thermometer questions instead of the like/dislike questions were employed to measure party image. The feeling

203

thermometer is a score that ranges from 0 to 100, and 50 is the neutral point. The warmer people feel toward the party, the higher is the likelihood that people vote for the party.

Table 5.19 summarizes the party images in the 1976 and the 1996 elections.69 Voters feel warm toward the LDP, but cool toward the other parties to varying degrees. The lowest score is for the JCP, while the other parties enjoyed moderate negative party images. The level of party images is not necessarily related to the number of responses on the thermometer question. The LDP enjoyed warmer party images than the JSP in 1976, for example, though it got only three more percent of responses than that party. The parties in the 1993 coalition government did not have higher saliency; instead, fewer people responded to this set of thermometer questions than in 1976. The LDP seemed more popular for the electorate in 1996, but only around sixty percent of people have any image regarding the other parties. The volatility of party images may come from the improvement of the LDP images after the 1994 reform. The overall party images for the 1996 parties are a little bit better than for the 1976 parties. Considering the relatively high level of party images of the NLC, NFP, and DP and their relatively short history, the defecting politicians created positive images for their new parties in a short time.

69 The feeling thermometer variable is derived from the one to one hundred scale on people’s feeling toward each party. The question reads like: “People have all kinds of feelings toward the people and groups which are influential in politics. In answering the following questions, if you don't like or dislike a person or group would you indicate this by giving them a rating of 50 degrees? If you like the person or group would you give them a rating between 50 and 100 degrees,

204

Mean Percentage Difference Score of between Mean Responses Score and 50 1976 LDP 50.23 85.36 +0.23 JSP 47.68 82.28 -2.32 DSP 31.18 79.57 -8.82 CGP 34.49 77.70 -5.51 JCP 39.98 76.35 -10.02 NLC 47.85 63.28 -2.15 Average 41.90 77.42 -8.10

1996 LDP 54.10 80.16 +4.10 NFP 47.40 66.44 -2.60 DP 47.76 57.90 -2.24 SDP 40.92 65.31 -9.18 JCP 31.89 65.18 -18.11 NPH 35.88 53.77 -4.12 Average 42.99 64.79 -17.01

Data: 1976 JABISS and 1996 JEDS data

Table 5.19: Mean and percentage of response of party images, 1976 and 1996

Before using party images to predict the voting act, it is necessary to ascertain the association between partisanship and party images. If they are highly correlated, then they may measure the same political attitude and one of them must be set aside. Table 5.20 and Table 5.21, respectively, present the correlation between partisanship and party images, which is measured by the 0-100 feeling thermometer toward each party in 1976 and 1996. For the electorate in 1976, the constructed partisanship variable appears to have little in the way of party image,

and if you dislike them a rating between 0 and 50 degrees?” Detailed discussion on this variable is

205

which is displayed in Table 5.20. Except the LDP partisans, the magnitude of correlation coefficients for the partisans is below .5, which means party identification can explain less than 25 percent of variation in party image for the partisans. This is probably because voters were rarely exposed to news coverage of the opposition parties. The small association makes clear that party images and party identification are different political attitudes.

With party system reshuffling, the 1996 election is expected to somewhat bolster the role of party images. For the existing parties like the LDP, JSP, and

JCP, the association between party image and partisanship should remain the same.

Regarding the new parties, however, party image should be ahead of partisanship, which requires longer time to develop. Thus the association between partisanship and party image for the NFP and the DP should be small. Table 5.21 displays the strong correlation between party image and partisanship for the LDP voters. The correlation between these two variables for the JCP comes in second, which implies the success of the JCP in the single-member districts. The NFP and the DP are too new to develop strong party images that are associated with partisanship, but the party leaders attract attention of voters.

given in Section 4.4.

206

Partisan Strength Party Images LDP JSP JCP CGP DSP LDP .565*** -.218*** -.277*** -.212*** -.037 JSP -.250*** .438*** .083** .026 -.061* JCP -.095** -.053 .117*** -.041 -.068* CGP -.126*** -.054 -.062* .440*** .037 DSP -.025 -.065* -.107** -.006 .306***

Data: 1976 JABISS data Note: Table entries are Pearson correlation coefficients. *: p<.05, **:p<.01, **:p<.001.

Table 5.20: Correlation of party images and partisanship, 1976

Partisan Strength Party Images LDP NFP DP JSP JCP LDP .549*** -.085** -.058 -.067* -.223*** NFP .020*** .277*** .003 -.055 -.015*** DP -.094** -.102** .320*** .058 .050 JSP -.078** -.056 .027 .275*** .031 JCP -.210*** -.204*** -.153*** -.044 .400***

Data: 1996 JEDS data Note: Table entries are Pearson correlation coefficients. *: p<.05, **:p<.01, **:p<.001.

Table 5.21: Correlation of party images and partisanship, 1996

The analysis above demonstrates the existence of party images among the

Japanese voters and its weak association with the strength of party affiliation.

Party images are useful to predict vote choice for new parties in 1996. For people who still identify with the LDP, party images may be less important than their partisan strength in determining whether they vote for the LDP.

207

5.4.4 Social Networks and Voting

The influence of social networks on individual voting behavior cannot be overstated; many scholars have investigated the role of social networks

(Richardson, 1967; Flanagan, 1980; Miyake, 1989). Essentially, their analyses confirmed the function of social networks; social networks can reinforce, convert, or activate political attitudes (Richardson, 1991). Moreover, it is found that people receive different kinds of information from different networks, such as neighbors, friends, co-workers, campaign organizations. The way they process that conforming or nonconforming information constitutes another topic, but it requires panel data to ascertain the direction of influence. Here we simply treat social networks as mechanisms that reinforce political attitudes, and whose influence varies with the amount and source of information. Moreover, we assume that social networks, unlike mass media, convey biased messages to the electorate because political parties affiliate with groups of similar ideology or related interest.

By contrast, mass media only increases the level of political interests of the electorate.

To begin the analysis of social networks in Japan, it is necessary to understand the operation of the political system. According to Richardson’s (1997) study on Japanese politics, political power is fragmented and bounded by particularistic interest, which is semi-institutionalized and horizontal-patterned.

Political parties and their supporters exchange voting support for particularistic interest, and political parties cooperate with the bureaucracy to build policy coalitions. Factions within the LDP have their own access to interest groups and

208

the bureaucracy, and they run candidates in districts to represent the interests of voters. The distribution of resources favors candidates of the governing party, which basically accords with the geographical districts they represent (Lin, 1996).

To meet constituency demands, the LDP spent government money on farmers, rural dwellers, and public work programs in rural areas (Curtis, 1988). The era of the LDP hegemony was built upon the coalitional relationship between parties, bureaucracy, and voters.70 Although political parties of Japan do not have strong grass-root organizations, the exchange of voting support and representation of interest between voters and candidates ensures the influence of social networks.

Given that social networks entail exchange of interest and support, people who are engaged in campaign mobilization must provide incentives for likely supporters. Constituency service is always mentioned as the foremost incentive that voters cannot resist. This includes public projects, subsidies to farmers, help for obtaining employment or education, and appearances at weddings, funerals, or festivals. Of those services, public projects obviously cover more voters, but subsidies to farmers are also influential. For Japanese voters, showing up at a festival, funeral, or wedding is seen as a sign of respect to the host; the appearance of politicians grants the host honor before his friends and relatives. In a poll conducted by Yomiuri News (Naikasusouridaijin, 1990) in 1989, 35 percent of respondents agree with the statement that politicians should at least consider appearing at local events. Among those people who agree with the presence of politicians, 63 percent consider it as a social custom, 20 percent view it as payback for voting support, and 16 percent regard it as a display of relationship.

70 Regarding the policy coalition between factions, interest groups, and bureaucracy, which is

209

Through personal or public services, candidates can build up direct contacts with voters, which is the first step of campaign mobilization through social networks.

Social networks refer to family, friends, relatives, neighbors, party organizations, candidate support organizations (koenkai), occupational associations, religious groups, community meetings, and labor unions. Individuals receive information from those organizations or their friends, who may or may not receive services or benefits from candidates or parties. The occupational associations and labor unions can influence their members by endorsing candidates and community members can help candidates as they solicit voting support door to door. Besides those existing organizations, candidates organize koenkai to enlarge their local bases. Candidates even pay the party membership for their supporters. A candidate support organization consists of local assemblymen, activists, party members, and its size varies positively with the size of the district.

Party organizations, whose role is often ignored because of the conventional wisdom that Japanese parties have heads but no feet, constitute another potential source of mobilization. According to Foster’s (1982) case study of party organizations in Hyogo prefecture, parties except the JCP reach a variety of social groups. Thus the national party, which is the caucus of assemblymen in the Diet, can reach voters through local organizations that incorporate fragmented interest groups.

With only one cross-section reading of the number of party members and social groups, we simply assume that local organizations can account for the

called zoku (tribe), see Richardson (1997), chapter 3.

210

electoral performance. However, the relationship can be reversed; the more

Dietmembers a political party has, for instance, the more party members it can recruit and more social groups it can connect with. Regardless of the causal direction, political parties can utilize social networks for voting support, which is not different from any party in a democracy.

The cross-pressure theory developed by Lazarsfeld, Berelson, and Gaudet

(1944) suggests that voters who receive cross-pressure from social networks would postpone their voting decisions until Election Day. On the contrary, voters who simply decide their choices before the beginning of a campaign should have no cross pressure. More precisely, people who make up their mind well before the campaign begins will either receive no further information or discard any nonconforming information. In other words, the influence of social networks may not be seen on those people who already decide whom they will vote for.

To assess how social networks associate with political parties, I count the number of requests from different sources and show the results in Table 5.22 and

Table 5.23. In 1976, the LDP and the JSP edge out other parties in bringing people in support organizations and meetings, sending postcards, and finding potential supporters in associations. The JSP is good at reaching union members. So is the

DSP. The CGP shows outstanding ability to solicit voters through their neighbors and friends.

In 1996, voters receive personal requests from the LDP and the new conservative party, the NFP. Surprisingly, the JCP also emphasize personal request and campaign mobilization. Overall, voters receive more requests from the two conservative parties, the LDP and the NFP, which is the same as in 1976.

211

Party Affiliation LDP JSP JCP CGP DSP NLC Requests Family 55 15 10 27 4 4 Coworker 48 46 14 24 14 1 Neighbor 46 22 10 58 8 1 Friend 70 23 26 69 10 1 Campaign Koenkai 93 31 13 15 12 3 Postcard 310 132 49 49 32 9 Meeting 89 44 20 22 6 3 Union 7 135 6 3 22 4 Association 193 72 20 43 13 1

Data: 1976 JABISS data Note: Table entries are the number of respondents who receive requests regarding voting, participate campaign activities, or hold membership for associations related to political parties. The operationalization of the variables is given in Table 5.17.

Table 5.22: Social networks by political party, 1976

Sources LDP NFP DP JSP JCP Requests 195 248 31 8 41 Campaign Koenkai 58 21 4 0 3 Postcard 536 401 82 29 206 Meeting 62 28 3 1 6 Union 8 23 8 6 5 Association 145 69 8 1 13

Data: 1996 JEDS data Note: Table entries are the number of respondents who receive requests regarding voting, participate campaign activities, or hold membership for associations related to political parties. The operationalization of the variables is given in Table 5.17.

Table 5.23: Social networks by political party, 1996

212

Granted that the information that voters receive has influence on how they make their voting choices, social networks should agree with party choices.

Miyake (1989) comparing respondent’s voting intention and final decision, demonstrated three effects of social networks during the campaign: crystallization, reinforcement, and change. The effect of social networks mainly occurs when people already decided their voting choices in the campaign period, which is called reinforcement. When people have no voting intention before the election, personal requests provided cues for choice, which refers to the crystallization effect. Since most people decide whom they will vote for right before the election, there is little room for social networks to change party choice. Therefore, the effect of reinforcement should prevail in elections; social networks will explain very little variance in voting choice.

To re-examine the reinforcement effect of social networks, I set up party voting as a dichotomous variable as to whether voters vote along with their partisanship or not. The assumption is that partisanship can be treated as pre- existing voting guide. If people use their long-term attachment toward the party to decide their vote choice, then the strength of partisanship should account for the consistency of voting. If people rely on information from their preferred parties and use it as a guide for voting, then the level of engagement in social networks will explain why people do not deviate from their party identification. In the former case, social networks can merely reinforce political attitudes. In the latter case, however, social networks crystallize people’s predispositions of political behavior. Table 5.24 confirms the reinforcement hypothesis, showing that only partisan strength taps into the source of party voting. The coefficients for the

213

number of requests from social networks are insignificant across the two elections, but partisan strength has a great impact on party voting. The implication is that voters receive information from social networks but they still use their predisposition to make their voting choices. After controlling for partisanship and candidate evaluations, the effect of social networks on party choice should be minimal.

Party Voting in Party Voting in 1976 1996 Constant -3.395(.211)*** .964(.100)*** Partisan Strength 1.489(.099)*** .181(.058)** Social Networks1 -.075(.064) .006(.040) Number of Cases 1332 1535 Pseudo R-square .2596 .005

Data: 1976 JABISS and 1996 JEDS data Note: 1: the total number of requests from various social networks. Standard errors are in parentheses. **: p <=.01; ***: p <=.001

Table 5.24: Logistic estimates of party voting, 1976 and 1996

Section 5.4 presents strong evidence that candidate evaluations and party images are potential factors in vote choice, but that issue positions as well as social networks are less important to people’s choice. Among the four short-term forces, candidate evaluations are more dominant than others because it is associated with social groups and candidates’ campaign activities. Moreover, the electoral system creates individual incentives that push candidates to pursue a personal vote. While the candidate factor prevails in campaigns, issue positions in both elections are wildly diversified. Discriminant analysis finds one salient issue for the 1976 election and two issues for the 1996 election, but neither issue

214

dimension accounts for large variance among voters. Regarding party images, the correlation analysis shows its independence from partisanship. It is also found that the old parties like the LDP and the JCP enjoyed stable party images, while the new parties built up their moderate images in a short period of time. This implies that the partisans of the old parties can use partisanship to decide their voting intention and that the supporters of new parties rely on party images to make decisions. Finally, social networks have reinforcement impact on voting behavior.

This suggests that social networks may not be able to sway people and, as a result, they have minimal impact on actual vote choice.

5.5 Determinants of the 1976 and 1996 Vote

Due to Japan’s multi-member district system, candidate-centered campaigning has prevailed in Japan politics since the merging of the conservative parties (Fukui and Fukai, 2002). In medium-size districts, one party nominates more than one candidate with the hope that each candidate can win a seat with the minimal share of votes, which encourages candidates to rely on geographically local machines and personalized interests rather than their party label. The making of political parties from local councils sets the bases for local factions, and the group-oriented culture reinforces the candidate-centered pattern. When the electoral reform was implemented, political parties gained more media coverage and nominated only one candidate in each single-member district, which was expected to increase the importance of partisanship. The change of political behavior, though, should be more evident after more elections (Curtis,1999).

215

5.5.1 Explanatory Variables

Having presented the association between long-term and short-term variables and voting choice individually, here I briefly reiterate the operationalization of each variable before moving on to the multivariate analysis.

The measure of partisanship, as noted in Section 2.2, is based on the respondents’ report about which party they support and how enthusiastic they are. Because of the multi-dimensional party system, I coded party support as several ordinal-level variables, each of which stands for partisan attachment toward a party. To test the validity of the measurement, I ran correlation analysis for party image, party feeling on a scale thermometer from 0 to 100, and party identification. The results not only confirm the validity of the measurement, but also reveal the affective and cognitive aspects of party identification in Japan.

Ideology is the other long-term psychological variable that taps into the general direction of policy. Under the 1955 system, the socialists and conservatives aligned with the JSP and the LDP respectively. Although the socialist parties are “permanent opposition parties,” the ideological conflict reflects several issues, such constitution revision, labor unions, social welfare, and even political reform. The way to measure ideology is straightforward; respondents are asked if they are close to the conservative or the progressive camp.

In addition to ideology, it is believed that there exist some salient value cleavages among voters. Four variables are created to represent four dimensions of existing values: conformity, tradition, individualism, and localism. All of them are constructed directly from agree/disagree questions and coded as 5-point scales.

216

Turning to the short-term forces, the candidate evaluation variable is the most critical one in terms of our research interest and its measurement. Because the voting choice of different candidates is coded as the same party choice, the dependent variable is actually party choice rather than candidate choice. The candidate evaluation variable counts voters’ responses to candidates, then transferring the number of responses to parties. In light of the dependent variable, the operationalization of the candidate evaluation variable assumes that every candidate can represent the party that nominates him/her. Moreover, party choice is expected to vary with change of the level of candidate evaluation, which is different from the relative evaluation between two candidates in the United States.

Due to the nature of data and research interest, candidate evaluation will be measured by the number of responses to the candidate’s party. The response is coded from respondents’ answers to the question that reads: which candidate is the best to fit the requirement. Only when voters know about the candidate, can they answer the question. As Miyake (2002) noted, this type of question considers both parties and candidates as they turn out to vote.

Congressional scholars are skeptical about the predicting power of issue positions in congressional elections (Hinckley, 1981). In Japan, however, money politics and personal scandal are more appealing issues for the mass public and may lead to a cabinet crisis. As discussed in Section 5.1, the conflict between the conservative parties and opposition parties ranged from the constitutional issue to political reform. Through discriminant analysis, one issue dimension made of three issues is identified for the 1976 election and two issue dimensions for the

1996 election. I also analyzed culture issues, such as localism, conformity,

217

individualism, and traditionalistic attitude. Because of concern for missing data, I merely consider two culture issues and assume that ideology can stand for those political issues.

Social networks comprise of personal requests, campaign organizations, and associations membership. I simply counted the numbers of social networks from which the respondents receive partisan information. Although it is expected that social networks have a minor role in voting decisions, they may still anchor some uninformed or non-partisan voters.

Finally, some demographic variables, such as occupations, age, gender, and education level, will be controlled for. It is expected people who characterize themselves as farmers would tend to support the LDP. It is also reported that elder people prefer the LDP but young people are distant from politics (Miyake, 1989).

Due to the strong influence of partisanship and candidate evaluation, however, demographic variables should be relevant to voting decisions only to a limited extent. Table 5.25 presents a summary of those explanatory variables.

218

1976 1996 LDP Partisanship (0 to 3) 0.68 .985 JSP Partisanship (0 to 3) 0.363 .616 Progressivism (-1 to 1) -0.117 .633 Localism (1 to 5) 3.359 1.354 Traditionalism (1 to 5) 4.101 1.326 Candidate Evaluation (-5 to 6) 0.15 2.768 LDP Party Images (0 to 100) 50.20 52.89 JSP Party Images (0 to 100) 48.09 NFP Party Images (0 to 100) 48.27 LDP Social Networks (0 to 6) 0.543 1.426 JSP Social Networks (0 to 6) 0.379 1.140 Gender (0 or 1) 0.416 .483 Education (1 to 3) 1.664 .571 Age (1 to 4) 2.033 .979 Farmer (0 or 1) 0.094 .189 Labor (0 or 1) 0.23 .372 Executive (0 or 1) 0.232 .385

Table 5.25: Mean value and range of independent variables

Two multi-nomial logit models as follows will be estimated, and the results will be presented in the next section.

Model 1: Pr(Y=Voting for the LDP, JSP, or third parties |x)=exp(xβ)/[1+

Σexp(xβ)] where x= [LDP Partisanship, JSP Partisanship, Progressivism, Localism, LDP

Candidate Evaluations, JSP Candidate Evaluations, LDP Party Images, JSP Party

Images, LDP Social Networks, JSP Social Networks, Male, Elementary Education,

Secondary Education, 20-35 Years Old, 36-50 Years Old, 51-65 Years Old,

Farmer, Labor, Manager]

β=[β1,...,β19]-1

Model 2: Pr(Y=Voting for the LDP, NFP, or third parties |x)=exp(xβ)/[1+

Σexp(xβ)]

219

where x= [LDP Partisanship, NFP Partisanship, Progressivism, Localism, LDP

Candidate Evaluations, NFP Candidate Evaluations, LDP Party Images, NFP

Party Images, LDP Social Networks, NFP Social Networks, Male, Elementary

Education, Secondary Education, 20-35 Years Old, 36-50 Years Old, 51-65 Years

Old, Farmer, Labor, Manager]

β=[β1,…,β19]-1

5.5.2 Results

It is difficult to disentangle the influence of partisanship and candidate evaluation, particularly in a multi-party system where there are only few cases for each party that are available for analysis. To avoid the inflation of parameters due to a small number of observations, I merely look at the respondents who voted for the first and the second largest parties and coded other cases as “third party vote”, which includes voting for either political parties or independent candidates. The comparison group is the group of people who did not turn out to vote71. Moreover, the candidate evaluation variable is constructed as two dimensions for both elections. The reason that I keep the candidate evaluation variable as two- dimensional is because the first and second largest parties have some overlap in candidate evaluations. Furthermore, the intra-party competition may override the inter-party competition, and keeping candidate evaluations separate would tell us if candidate evaluations anchor vote choice. Party images are also separate because the catch-all parties are not necessarily against each other, especially the parties organized by politicians defecting from the LDP.

220

The results of the multi-nomial logit analysis shown in Table 5.26 provide us with coefficients representing the likelihood of voting for the LDP and the JSP among other two choices. The specification of the model is aimed at comparing the effects of candidate evaluations and partisanship on voting choices.

Partisanship plays a big role in voting choice against abstention; both parties’ supporters tend to vote for their parties rather than staying at home. Ideological position also has an impact on voter turnout and voting choice. People who call themselves liberal are more likely to vote for the JSP than to stay at home, but people who are in the conservative camp tend to vote for the LDP. The localism issue position does not associate with voting choice, which is what the theory expects. The candidate evaluations variables have a significant impact on voting choices, and their directions are as predicted. The LDP and the JSP party images associate with the LDP and the JSP voting choice positively and significantly.

Social networks for both parties’ candidates are not closely related to voting choice, which implies that social networks are more remote to voting decision than expected.

Due to the homogenous society and “middle-class awareness,” social cleavages make very little difference on political behavior between different groups of people. The demographic variables, therefore, have minimal influence on voting choice.

71 Abstention is chosen as the comparison group because the results can show why Japanese voters

221

LDP Vote JSP Vote Third Party Vote Constant -1.454(.971) -1.308(1.034) 1.477(.880) LDP Partisanship .620(.169)*** .098(.216) .000(.179) JSP Partisanship .293(.245) .667(.201)*** .075(.205) Progressivism -1.306(.195)*** 1.122(.201)*** .373(.166)* Localism .101(.110) -.080(.114) .019(.101) LDP Candidate 1.512(.381)*** .329(.428) .465(.397) JSP Candidate .457(.793) 2.424(.709)*** .828(.729) LDP Party image .017(.008)* .011(.008) .007(.007) JSP Party image -.004(.009) .025(.010)** -.006(.008) LDP Contacts .051(.134) .069(.148) .143(.124) JSP Contacts .001(.160) .260(.159) .025(.144) Male .223(.307) .686(.325)* .403(.285) Elementary .457(.424) .02(.445) .108(.375) School High School .279(.389) .217(.399) -.109(.340) 20-35 years old -.652(.693) -1.473(.788) -1.249(.669) 36-50 years old -.386(.696) -.919(.789) -.927(.675) 51-65 years old -.136(.708) -.73(.808) -.685(.693) Farmer .517(.559) .025(.675) -1.113(.664) Labor .307(.356) -.244(.375) -.031(.322) Manager .799(.407)* .298(.438) .173(.384) Number of 1100 Observations Log-Likelihood -887.787 Chi-Square(57) 1069.95*** Pseudo R-square .3760

Data: 1976 JABISS data Note: Coefficients are maximum likelihood estimates with standard errors in parentheses. Abstention is the comparison group. *: p<.05, **: p<.01, ***: p<.001.

Table 5.26: Multi-nomial logit estimates for four-choice model, 1976

are willing turn out to vote instead of opting for abstention.

222

As for the 1996 election, Table 5.27 shows that partisanship plays a significant role. Both LDP partisanship and NFP partisanship positively contribute to the LDP voting choice, partly because political leaders share the similar principles of politics. However, only NFP partisanship has a positive impact on voting for NFP rather than abstention. Party attachment with the LDP would make voters less likely to vote for NFP candidates than staying at home.

For the LDP, conservatism is still their top ideology. Progressivism would make people less likely to vote for the LDP than not turning out, but the NFP voters appear not to be ideology-oriented. Regarding the two issues of cultural politics, only traditionalistic attitude affects the likelihood of voting for the NFP.

It is unexpected that the LDP candidate evaluation is influential on voting for the

NFP compared to abstention, which implies the intriguing context of post-reform politics. Party images indeed determine vote choice; both coefficients of the LDP and the NFP party images are in the right direction and significant. Like in the

1976 election, social networks still have very limited influence. Among the social- demographic variables, age is significant only for the NFP and the third party vote versus abstention; elder people are less likely to vote for new parties.

223

LDP Vote NFP Vote Third Party Vote Constant -1.057(.701) -.965(.718) 1.611(.628)** LDP Partisanship .449(.153)** -.441(.174)* -.612(.164)*** NFP Partisanship .875(.271)*** 1.269(.269)*** .939(.257)*** Progressivism -.417(.203)* -.395(.206) -.092(.186) Localism -.087(.096) -.166(.097) -.185(.088)* LDP Candidate .104(.045)* .112(.044)* .020(.043) NFP Candidate .132(.075) .314(.070)*** .133(.070) LDP Party image .038(.008)*** .000(.007) .013(.007) NFP Party image -.023(.008)** .032(.008)*** -.015(.007)* LDP Contacts -.123(.096) -.010(.095) .027(.085) NFP Contacts .155(.117) .135(.115) .060(.108) Male -.040(.260) -.219(.263) -.034(.239) Elementary School .477(.629) -.603(.675) -.431(.626) High School -.291(.307) -.447(.303) -.255(.276) 20-35 years old -.896(.535) -1.602(.551)** -1.688(.504)** 36-50 years old .156(.518) -.508(.539) -.377(.495) 51-65 years old -.010(.483) -.473(.510) -.350(.412) Farmer 1.585(1.054) .746(1.127) .850(1.100) Labor -.130(.323) .007(.323) -.265(.298) Manager -.198(.368) -.526(.369) .022(.317) Number of 965 Observations Log-Likelihood -971.491 Chi-Square(57) 649.01*** Pseudo R-square .2504

Data: 1996 JEDS data Note: Coefficients are maximum likelihood estimates with standard errors in parentheses. Abstention is the comparison group. *: p<.05, **: p<.01, ***: p<.001.

Table 5.27: Multi-nomial logit estimates for four-choice model, 1996

Table 5.28 and Table 5.29 display the discrete change in the probability of voting for the two parties against abstention. In 1976, candidate evaluations were more important than partisanship and party images. A change in LDP partisanship by one standard deviation around the mean will cause an increase of probability for voting for the LDP by .059, but the same change in the LDP candidate evaluations will lead to increase in the probability by .106. As for the

224

JSP, the effect of candidate evaluations is .240, but partisanship is only .099.

Ideology is critical to the voting act in the 1976 election. A one standard deviation change in progressivism positively contributes to the probability of voting for the

JSP versus abstention by .177, and drops the probability of voting for the LDP versus abstention by .105. Party images are the fourth most important factor to both party choices. Its impact is .031 for LDP voting versus abstention, and .040 for JSP voting.

The relative difference between candidate evaluations and partisanship is not as impressive in 1996 as in 1976. The likelihood of voting for the LDP versus abstention in 1996 increases by .063 when people gain one more standard deviation of LDP partisanship, but it only increases by .041 when people have one more standard deviation of positive LDP candidate evaluations. A one standard deviation change in partisanship positively contributes to the probability of voting for the NFP versus abstention by .083, and the same amount of change in candidate evaluations increases the probability of voting for the NFP versus abstention by .078.72 Party images became more important in the 1996 election. A one standard deviation change around the mean of the LDP party images increases the probability of voting for the LDP by .103, and the probability of voting for the

72 The results can be interpreted from other perspectives. One perspective is to compare discrete change based on the multi-nomial logit coefficients across the two equations. From this view, the effect of the LDP partisanship increases from .059 to .063, which suggests the rising impact of partisanship. My interpretation is solely based on the comparison of the discrete changes in one equation. I consider the comparison of the effect of the explanatory variables across the equations inappropriate because the estimates are from two different samples. The only way to compare them is to pool the samples together and use the dummy variable to pick up the variance due to different samples, which I shall do in Chapter 6. Moreover, discrete change in the probability varies with different means and standard deviations of the independent variables. The sheer size of the discrete change due to one standard deviation change around the mean cannot be compared across two different samples. Furthermore, there is no way to tell if the difference in the effect is significant or not; we cannot reject the null hypothesis that the discrete change from different samples is the same.

225

NFP would increase by .059 for a one standard deviation change around the mean in the NFP party images.73 Ideology has a great impact on vote choice in 1976, but the influence of ideology is behind partisanship and candidate evaluations in 1996.

This suggests that the 1993 party system was less ideological than the 1955 system. The localism issue is far from a sharp distinction to both parties, since they all pursue a personal vote.

LDP Vote JSP Vote LDP Partisanship .059 .018 JSP Partisanship .021 .099 Progressivism -.105 .177 Localism .011 -.018 LDP Candidate evaluations .106 .045 JSP Candidate evaluations .022 .240 LDP Party image .031 .040 JSP Party image -.006 .078 LDP Contacts .004 .012 JSP Contacts .000 .038

Note: Centered standard deviation change is calculated as follows (Long, 1997). Because [p(y=m)/1-p(y=m)]=exp(Xβ'), thus p(y=m)=[ exp(Xβ')/1+ exp(Xβ')]. Let th p1(y=m|x, xi+½si ) the predicted probability that i variable at its mean plus one- half of its standard deviation, holding dummy variables at 1 and other variables at th their mean, and p2(y=m|x, xi-½si ) the predicted probability that i variable at its mean minus one-half of its standard deviation, holding dummy variables at 1 and all other variables at their mean. The difference between pi1 and pi2 is the centered standard deviation change.

Table 5.28: Discrete change in the probability of the LDP and the JSP vote, 1976

73 As to why the LDP party images are more influential than the LDP party attachment but the NFP partisanship is more important than the NFP party images, I suspect that the LDP partisans have consistent party images toward the LDP but the NFP supporters justify their vote choices by referring to their partisanship. In other words, they identify with the NFP because they voted for it. More empirical data is required to answer this question.

226

LDP Vote NFP Vote LDP Partisanship .063 -.045 NFP Partisanship .077 .083 Progressivism -.037 -.026 Localism -.015 -.021 LDP Candidate evaluations .041 .032 NFP Candidate evaluations .044 .078 LDP Party image .103 .000 NFP Party image -.057 .059 LDP Contacts -.025 -.001 NFP Contacts .025 .016

Note: the calculation of the centered standard deviation is same as Table 5.28.

Table 5.29: Discrete change in the probability of LDP and NFP vote, 1996

Both the multi-nomial logit estimates and transformed probabilities due to the change in the explanatory variables confirm the increasing importance of partisanship. The change in the electoral system from multi-member district to single-member district indeed increases the importance of political parties. Even though only two conservative choices are analyzed, the confrontation between the two parties is still evident seen from the discrete change in the probability. The effect of candidate evaluations is relatively weaker than partisanship in the 1996 election, and party images have larger weight on vote choice in 1996 than in 1976.

Candidate evaluation did not entirely diminish in 1996, partly because of the continued candidate-centered pattern in multi-member district system. Overall, the result of this analysis shows the effect of the changing electoral system; it is changing because the 1996 election is the first one after the reform. Further research is needed to confirm this finding.

227

5.6 Conclusion

In Japan, partisanship is an unstable psychological attitude. The homogenous society prevents political attitudes from polarizing, but it also limits the socialization process. Voters tend to respond to short-term events, such as party scandals. Voters may also consistently vote for the same party because of candidates instead of parties. Examining the differences between party popularity and their seat shares across elections and the context of multi-member districts, it is believed that candidate evaluations rather than partisanship account for voting behavior in Japan. The SNTV system allows as many candidates as the number of seats plus one to campaign for multiple seats, which creates individualistic incentives for candidates to cultivate and rely on a personal vote.

Due to the legacy of one-party politics and the electoral system, the LDP dominated the government for four decades. However, the scandals undermined people’s faith in politics, and intensified the internal rift over political reform.

After the 1970s, meanwhile, the JSP lost its appeal to progressive voters.

Responding to people’s call for clean politics, new parties emerged in 1993 and ended the LDP’s thirty-eight year’s of dominance. The SNTV system provided incentives for political reform to both opposition parties and the LDP politicians

(Christensen, 2000). The change of the electoral rule is expected to increase the importance of party labels, because small districts can only allow one candidate from each party.

The multi-variate analysis of voting behavior confirms the hypothesis that partisanship plays a more significant role in the process of vote choice after the

228

reform of 1994. The effect of candidate evaluations is less significant in the 1996 election than in the 1976 election, and party attachment becomes more influential, particularly for new parties. The heavy weight of the ideological variable in 1976 results from the confrontation between two ideological camps. In 1996, however, the reform issue did not distinguish the conservatives from opposition parties any more. Instead, party images became the critical variable for the new party system because of the new parties and a new generation of voters.

The new electoral system indeed enhances the effect of partisanship. The single-member and simple-plurality district system leads to a unification of parties and candidates, and the Japanese voters respond to the change by giving partisanship more weight in voting decisions. The questions for the future would be whether the change of voting pattern will continue and, as a result, if Japanese parties will become modern mass parties. If that indeed happens, the change of institutions not only changes political behavior but also transforms party organizations.

229

CHAPTER 6

COMPARISON OF FOUR CASES

6.1 Introduction

With comparable measurements and concepts, we are able to conduct a cross- national study concerning the relationship between voting and partisanship. The purpose of comparison is to find out if the relationship between the dependent variable and the independent variables holds across different political systems.

The four political systems included in this research have very different cultures, histories, economic activities, and party systems, but they share partly similar election systems and government structures. The dimensions of culture, history, and other social characteristics may account for the aggregate phenomena, such as nation-building, revolution, and mobilization, but it is very difficult to determine which aspect of society influences individual political attitudes and diffuses the same attitude among people within a society and makes them behave the same way. Even when party organizations represent certain social groups (Lipset and

Rokkan, 1967), we have to make a strong argument regarding why parties still represent their social bases after the formation of party organizations (see Section

230

2.3). Other than socio-cultural factors, however, dimensions of policy-making reflect conflicts among political elites and the population they want to mobilize

(Rokkan, 1996). This would strongly suggest three-level analysis: macro political institutions, intermediate political parties, and micro behavior.

Institutional-setting factors are widely utilized as determinants of political behavior (Przeworski and Teune, 1970). The consequence of electoral laws, centralization of political power, and societal systems “influence properties of the system that in turn affect individual behavior.” (Przeworski and Teune, 1970:54)

In this dissertation, institutional factors refer to election systems and the legislative-executive relationship in these countries. What makes the setting critical to individual political behavior is the political party. The political party positively contributes to partisan strength through its large coverage through the media and far-reaching organizations, though individual candidates try to take advantage of party labels for their best interest. Aldrich (1995) pointed out that politicians built political parties to implement their goals in the first place, but parties gain their own functions and become institutionalized. With nominating candidates, holding primary elections, and steering public policies in legislatures, parties became enduring political objects. The extent to which parties are dominant in politics depends on political arrangements. In the case of party government, political parties are cohesive, responsive to national forces, and dominant in making governmental decisions. Consequentially, voters are party- oriented in this circumstance. However, individual candidates are more likely to be independent from national tides and less responsive to party’s decisions when decision-making of parties is decentralized and politics is localized. In the case of

231

decentralized politics, however, voting tends to be candidate-oriented. In other words, the level of party-based voting would be lower in decentralized politics.

The two-by-two typology in Chapter 2, which is replicated here as Table 6.1, is designed to capture the effect of institutions on political behavior through the mechanism of political party. Government structures and election systems are used to construct the typology, which reflects the contrast between fused powers and separated powers, and single-member and multi-member districts. The difference between fused and separated powers leads to different party disciplines; parliamentary parties emphasize party loyalty of candidates rather than local interests, while presidential parties are inclined to allow lawmakers to cooperate with bureaucrats in pursuit of particularistic interest. As regarding the difference in election systems, the number of seats at stake is the key. In the single-member district system with non-transferable vote, each party will nominate one candidate and voters can attain information from parties to make their voting decision, which strengthens the role of party attachment. In multi-member districts, however, not only parties but also candidates send messages to voters, thus the voting decision involves considerations about candidates and parties. Political parties, as an intervening variable, set up the context of the propensity of party voting.

SNTV SMSP Presidential System Taiwan the United States Parliamentary System Japan before 1994 Japan after1994

Table 6.1: Party voting of four cases

232

Due to the level of data, the correlation between individual political attitude and vote choice is the comparable object in this project. Published comparisons of party strength using indicators such as campaign resources, national party ties, and legislative election outcomes have addressed the question of which type of political institutions has greater party cohesion (Bawn, Cox and

Rosenbluth, 1999; McCubbins and Rosenbluth, 1995). The features of the party systems have been presented during the discussion of party voting in the previous chapters. Here the focus rests on comparison of the effect of candidate evaluations and partisanship on vote choice across the four cases. The propensity of deviation from the party line is expected to vary with the type of political arrangement, which is assumed to reflect party strength that constrains individual candidates.

6.2 Model of Pooled-Analysis

6.2.1 Model

Party Voting theory states that most voters consistently identify with the same party and party loyalty increases steadily with age. Because parties are the most enduring political objects, the affective party attachment is the carrier of public response to issues and perception of candidates is at most campaign- specific. Party identification is therefore the principal determinant of vote choice across time (Campbell, Converse, Miller, and Stokes, 1960; Shaffer, 1972;

Converse, Miller, Rusk, and Wolfe, 1969; Converse and Markus, 1979). The alternative theory suggests that candidate evaluations largely shape vote choice because candidates are the main objects that people vote for. Candidate

233

evaluations summarize people’s feelings toward parties and their issue positions

(Kelly and Mirer, 1974; Page and Jones, 1979; Markus and Converse, 1979;

Miller and Shanks, 1996). The assumption is that inter-system differences will influence the relative weight of candidate evaluations and party identification on vote choice. If that is the case, there should be an interaction effect between system-level variables and attitudinal variables; the effects of attitudinal variables, such as party identification and candidate evaluations, depend on whether they are consistent with certain system-level characteristics. Furthermore, we intend to test whether the influence of party attachment goes up in certain contexts and recedes in others. Equation 6.1 is the model that considers the main effects of the two attitudinal variables and control variables. It assumes that the value of the dependent variable changes with the value of each of the independent variables, controlling for the value of control variables.

P(Y│X)= β1X1+β2X2+AZ ---(6.1) where Y= vote choice;

X1=partisanship;

X2=candidate evaluations;

A=the vector of coefficients of a set of control variables;

Z=the set of control variables.

The control variables refer to demographic variables, such as education, age, race, and occupation, and political attitudes, such as issue positions and party images. According to the previous individual studies, they are not significant

234

across the board. Unlike partisanship and candidate evaluations, therefore, they are considered less important in accounting for the voting act across different political institutions.

In order to verify the generalizability of the party voting model, it is necessary to take system-level differences into account. The system-level variables are assumed to interact with the quantitative variables already in the model; the effect of the independent variable will change with the level of the system-level variables holding the other variables constant. Therefore, the regression of vote choice on two system-level and two quantitative variables is formulated as follows.

P(Y│D,X,Z)=β0+β1X1+β2X2+β3D1+β4D2+β5D1X1+β6D1X2+β7D2X1+β8D2X2+β9

D1D2X1+β10 D1D2X2 +AZ ---(6.2) where D1=1 when election system is multiple-member district and 0 otherwise;

D2=1 when government system is presidential system and 0 otherwise.

Equation 6.2 allows us to estimate the main effects of the attitudinal variables, controlling for the interaction effect of system-level variables and attitudinal variables. To obtain the main effects, the value of the dummy variables will be changed, in this case, between 0 and 1 in turn. By redefining the value of the dummy variables, we can estimate the main effects in different combinations of the system-level variables (Jaccard, 2001). If an inter-system difference exists, the main effects of X1 and X2 will vary with the rescoring of D1 and D2. More formally, the ratio of two multiplicative factors--the exponent of the coefficient of

235

74 X1 or X2--would be equal 1. The three-way interaction between system-level variables and each attitudinal variable is considered here; D1D2X1 and D1D2X2 reflect the interaction between the two political arrangements and each of the attitudinal variables.

Because this sort of quasi-experimental research has not been conducted before, looking at the effect of one institutional setting each time will make it more convenient to keep track of the source of inter-system differences. Therefore, the comparison of party voting will start from the two cases with different government structures but using the same electoral rules, which are the United

State and Japan after 1994, and Taiwan and Japan before 1994. That will be followed by comparison of voting behavior with different election systems but the same government structure, which are Taiwan and the United States, and Japan before and after 1994. Simply changing the value of the dummy variables can derive the models for both comparisons.

P(Y│D,X)= β0+ β1 X1+ β2 X2+β3D1+β5D1X1+β6D1X2 ---(6.3)

P(Y│D,X)= β0+ β1 X1+ β2 X2+β4D2+β7D2X1+β8D2X2 ---(6.4)

The control variables included in Equation 6.1 and Equation 6.2 are excluded from the analysis here because they are not fully comparable between countries. This exclusion may lead to attenuation of the coefficients of party attachment and candidate evaluations, so the tests that follow can be seen as conservative. Furthermore, most of them are proved to be insignificant or less

74 The exponent of the logistic coefficient is the odds ratio of the probability of an event to the probability that the event does not occur. If the multiplicative factor across different category of

236

influential than partisanship and candidate evaluations in the previous individual case studies. The association between voting choice and partisanship, and between voting choice and candidate evaluations remain significant controlling for those variables. Therefore, it is fair to exclude those control variables from the model.

Besides, keeping only significant explanatory variables fits the requirement of parsimony for a scientific model.

Rescoring the value of the dummy variable from 0 to 1, the magnitude of the main effects of partisanship and candidate evaluations should change accordingly. In line with my hypotheses of less party voting with presidential systems and with multi-member districts, it is expected that changing each dummy variable from 0 to 1 will increase the effect of candidate evaluations and decrease that of party identification. In other words, we can observe how the change in the institution causes the change of relative influence in the two determinants.

To test the equations above, it is necessary to pool the data from the four countries. The 1998 congressional election of the United States, the 1995 legislative election of Taiwan, and the 1976 and the 1996 general election of

Japan are four elections held in different combinations of the executive-legislative relations and electoral rules, so the survey data on those elections are pooled together. These elections have been chosen because they are the most recent ones for which the needed survey data could be obtained.

The dependent variable is set up as a three-category choice--voting for the conservative party, the opposition party, and abstention. Voting for the conservative party is coded as 1, voting for the opposition party 2, and abstention

the dummy variable is the same, it means an interaction between the dummy variable and the

237

3. Because there are two parties in the United States and more than two parties in

Taiwan and Japan, the multi-party systems in Taiwan and Japan are simplified as two-party systems. The conservative party refers to the Republican in the U.S., the

LDP in Japan, and the KMT in Taiwan. People who vote for other parties or independent candidates are assumed to disagree with the conservative party. For shorthand, this will sometimes be referred to below as voting for the “liberal party”. This operationalization makes the party choice in these four countries comparable, though some non-sampling error will be introduced to the dependent variable.75

Abstention is included as the comparison group to keep as many cases as possible in this study of the legislative elections. From the comparative perspective, abstention is related to the inter-system difference; the turnout rate may vary with political institutions. Duverger (1954) suggested that PR systems tend to have multi-party systems and SMSP systems favor two-party systems. It is expected that people in PR systems are more likely to turn out because they would not waste their vote when seats are allocated to parties in proportion to their vote shares. Consequentially, PR systems should have higher turnout rate. Powell

(1986) also suggested that the chief executive is selected among legislators with

PR systems, thus parties must mobilize voters everywhere. He maintained that, on the contrary, the simple-plurality system would make some constituencies less

quantitative variable does not exist. 75 When treating multi-party systems in Taiwan and Japan as two-party systems, it is assumed that independents and a variety of parties have similar characteristics and that the difference among those parties is smaller than that between them and the conservative party. The 1993 non- communist coalition government in Japan to some extent confirms this assumption; the LDP was excluded from the government but it cooperated with the JCP to regain the cabinet in 1995. The CNP and the DPP in Taiwan confronted each other on certain issues, especially the reunification issue. But they also found some common beliefs such as on clean politics.

238

competitive and create lower turnout rate than in other districts. Franklin (1996) examined turnout in twenty-nine countries, finding that the proportionality of the electoral system positively contributes to the turnout rate, with other variables controlled. He also proved Powell’s (1986) theory by showing that the elections that decide the disposition of the executive branch have higher turnout rates because people have higher motivation to turn out. For instance, U.S. midterm elections have lower turnout rates than U.S. presidential elections and national elections in Europe because executive power is not at stake in U.S. off-year elections. Including the abstention as the comparison group will return better estimates of the effects of institutions in that they affect the individual turnout too.76

The two independent variables—partisanship and candidate evaluations-- are recoded as one-dimensional measurements. Partisanship is set up as a 7-point scale ranging from strong identification with the conservative party to strong identification with the opposition party. For the American case, the scale ranges from very strong Republican to very strong Democrat. For the Taiwanese case, very strong KMT is on the one end of the scale and very strong DPP or CNP on the other end. For the Japanese case, the scale starts from very strong LDP and extends to very strong party attachment with one of the opposition parties.

Candidate evaluation for each party is initially measured by a 5-point scale, which indicates the level of positive feelings toward candidates of each party. The

76 Another possible systematic bias introduced by setting abstention as the comparison group is that abstainers may have weaker partisan strength than people who turn out to vote, particularly in the U.S. That may affect the result of our analysis by overestimating the effect of partisanship, since that people who turn out to vote possess stronger partisanship than those who decide not to turn out.

239

drawback of this coding scheme is ignoring the difference between candidates, particularly in multi-member districts. 77 To predict the probability of voting choice, I computed the relative difference between the candidate evaluations of the conservative party candidates and that of the liberal party candidates.

Although we will lose information by regarding the parties other than the conservative as the other major party, the foundation of party competition is compatible with this operationalization. First, there are very few cases who identify with two parties with the same intensity in Taiwan and Japan; people identify with one party over others despite that they claim they identify with more than one party. In this case, party attachment can be set up as an one-dimentional variable. Second, the correlation between evaluations on candidates of both parties is low, which implies that people who highly evaluate one candidate will dislike other candidates. Therefore, it is legitimate to merge two indicators as one.

Obviously, several other variables affect voting. However, these variables are not included in this chapter because of the difficulty in finding comparable measures in surveys taken in difference countries. Later work can seek comparable measures of such control variables.

77 Lumping the candidate from the same party together and coding people’s mention about them as the party’s candidate evaluations may introduce non-sampling error to the measurement. In the Japanese campaign, for instance, different factions send candidates to a given district. Viewing faction A’s candidate and faction B’s candidate as the same and summing up the number that people mention about either candidate would overestimate people’s positive feelings toward candidates of a given party. Moreover, the geographical-based personal campaign in multi-member districts makes this coding scheme problematic in that it assumes that candidates of the same party appeal to the same constituent. In short, this coding method is likely to inflate the influence of candidate evaluations. Worse, pooling the American case, which uses single-member districts and each party nominates only one candidate, and the Japanese cases and the Taiwan cases together may underestimate the influence of the electoral system on the voting pattern across different context; the coefficient for candidate evaluations is significant despite controlling for the electoral system because the Japanese and Taiwanese cases inflate the effect of candidate evaluations.

240

Viewing institutions as independent factors and identifying their effects on the practices of a political system requires the control of the system-level variables.

Because there are two system-level variables in this study, we can control for one variable each time and see whether the predicted relationship occurs due to the variable not controlled. To put it simply, the difference in one institution is expected to lead to the predicted relationship between vote choice, party identification, and candidate evaluations in the two countries with one similar institution. Therefore, the source of inter-system difference will be analyzed by breaking down the two-by-two typology as two propositions.

Proposition 6.1: Everything being equal, the propensity of party voting in the parliamentary system is higher than in the presidential system.

Proposition 6.2: Everything being equal, the propensity of party voting in single-member districts is higher than in multi-member districts.

Regarding the first proposition, the contrast between the two types of government systems is expected to distinguish the four countries in terms of party voting. Comparing the legislative election in Taiwan and the Diet election in

Japan in 1976 will show the larger impact of candidate evaluations than partisanship on vote choice in the multi-member-district election, and comparing the congressional election in the United States and the general election in Japan in

1996 will find smaller influence of candidate evaluations and partisanship in the single-plurality election.

Regarding Proposition 6.2, different electoral rules in the same government system are supposed to determine whether vote choice depends on

241

partisanship or candidate evaluations. Additionally, it is expected that party identification will give way to candidate evaluations in multi-member districts because of the intra-party competition. The elections held in Taiwan and the

United States with similar presidential systems will be compared, after which the

Japan elections in 1976 and in 1996 in the same parliamentary system will be examined. If the predicted effect of candidate evaluations and partisanship on vote choice occurs, it is safe to claim that the voting pattern responds to different electoral rules in the identical government system and that change of electoral rules causes the change in voting pattern.

The two pair-wise comparative studies essentially test the relationship between political behavior and the political system. The difference in voting behavior between the two types of districts and the two types of government systems will be displayed by analyzing the four countries with the specified control of the system-level variables. By doing so, we can decompose the effect of the inter-system difference on political behavior. When the difference of electoral rules is ignored and the difference of government structure is upheld, for instance, it is assumed that different association between vote choice and party attachment simply comes from different government structure. After observing the difference of the four cases due to different government structure, we observe the other aspect of the four cases—election systems. If the difference of the relationship between vote choice and party attachment does not occur or occurs in a way opposite to the expectations, we are able to see which aspect of the case deviates from the prediction. In other words, the pair-wise comparison allows us to trace the source of deviation by comparing the four cases in two different ways.

242

6.2.2 Data

Because there is one data set used for each of the Japanese election before and after 1994, we can use only one data set for Taiwanese and American election respectively to keep the number of observations from each case approximately equal. Including both U.S. and both Taiwanese election studied in previous chapters would unbalance the cases for the typology of Table 6.1. We select the

1995 data for the Taiwanese case and the 1998 data for the American case because both elections are the most recent ones used in the previous individual studies. For the Taiwanese case, the 1995 election was held a few months before the first popular president election, which may increase the importance of the presidential system in the estimation. The total number of observations of the pooled-data is

5337.

Before proceeding to the comparison section, there are three caveats to make here. First of all, the number of cases limits the inference generated by this finding.

The U.S. Taiwan, and Japan represent typical political arrangements of electoral systems and executive-legislative relationship, but there are actually more qualified cases not included. Second, these cases cover a certain span of time, namely the 1990s. That also limits the external validity of this study. Moreover, it ignores the likely diffusing effect among these cases. And last, the four cases have different histories of party competition. The Legislative Yuan in Taiwan reserved life-time seats for the representatives from China until 1992, which makes the

1995 election only the second popular election. Compared to the U.S., the history

243

of party competition in Taiwan is relatively short. As for Japan, there have been only two general elections after the reform. The limitation of observation, time and maturity of case, albeit inevitable due to availability of data and comparability of measurements, thereby restrain this study from generating an inference for all political systems.

Section 6.3 will be devoted to the test of Proposition 6.1, and Section 6.4 will focus on Proposition 6.2. Section 6.5 will analyze the pooled data of the four political systems. Section 6.6 will summarize the findings.

6.3 Government Structure and Voting Pattern

The difference in the executive-legislative relationship is expected to influence the relative importance of partisanship and candidate evaluations.

Because of the separation of the executive and legislative bodies in presidential systems, individual lawmakers can build policy coalitions and cooperate with the bureaucracy to obtain interests for their constituencies (Arnold, 1979, Fiorina,

1989). The president must build policy coalitions with the use of patronage to trade for the lawmaker’s cooperation (Mainwaring, 1997). Moreover, the term of representatives is fixed; there is no confidence resolution or dissolution of the cabinet due to the policy disagreement. Without collective responsibility to maintain a majority in the legislative body, lawmakers are inclined to seek personal vote (Cain, Ferejohn, and Fiorina, 1987). Furthermore, the tie between political parties and social groups in presidential systems is weaker than in other systems, which reinforces the candidate-centered campaigns (Powell, 1982). In presidential systems, therefore, political power is decentralized and outside

244

influences can intervene in policy-making, which makes party discipline hard to maintain. Consequentially, party-based voting is relatively less than in parliamentary systems.

In parliamentary systems, political parties stress party discipline to maintain party unity and to prepare for reelection in case they lose a confidence resolution. The concept of “party government,” which indicates that parties take collective responsibility for the operation of government, is in some degree an ideal type of the parliamentary system (Rose, 1974). Strong partisanship is not the prerequisite of party government; instead, it is one of the consequences of party government. It is the institution that determines the level of party unity.

The hypothesis from Proposition 1 can be formulated as the following equation.

P(Y│D,X)= α +β1X1+β2X2+β3D1+β5D1X1+β6D1 X2---(6.5) where Y=vote choice;

X1= party identification;

X2= candidate evaluations;

D1= 1 for presidential systems; 0 otherwise.

The pooled data of Taiwan and Japan after 1994, and the United States and

Japan before 1994 are used to estimate Equation 6.5 with a multi-nomial logit model. Considering the differential effect of the product terms, two different sets multi-nomial logit coefficients are calculated. Each time the value of the dummy

245

variable is redefined so that each of the two groups takes a turn (Jaccard, 2001).78

By doing so, the effect of the product terms whose magnitude is conditioned on where voting takes place can be estimated along with the main effect of the explanatory variables. Notice that the main effect of the two quantitative variables is conditioned on the nature of the government system. When the value of the government structure variable is coded as 1 for presidential systems and 0 for parliamentary systems, for instance, the reference group is parliamentary systems, and the coefficients of candidate evaluations and partisanship reflect their main effect for people who vote in parliamentary systems.79

78 According to Jaccard, Turrisi, Wan (1990), the coefficient of β1 reflects the influence of X1 on Y holding X2 constant and D1 equal zero. For the case D1 =0, the reference group is parliamentary systems and the estimate reflects the main effect of the independent variables when D1 =0. With the coding of D1 changing, the main effect of the attitudinal variables will be estimated with different reference groups. 79 See Jaccard, Turrisi, Wan (1990) for this method of multiple regression and Jaccard (2001) for logistic regression. Jaccard (2001) argued that the exponent of the coefficient can be interpreted as the predicted odds ratio for one category of the dependent variable against the other category of it. However, I found that the exponent of the coefficient, which is always positive, changes with the direction of the coefficient. When we code, say, the 7-point category partisanship from strong conservative to strong liberal, the coefficient of partisanship variable will be in the opposite sign as shown in our result. In other words, the exponent of the coefficient will change with the coding of the dependent variable or the independent variable, which makes the odds-ratio interpretation of the effect problematic.

246

Government Structure is 1=presidential system, 0=parliamentary system Conservative Vote Liberal Vote Constant .599(.102)*** 1.074(.091)*** Partisanship -.548(.068)*** .356(.065)*** Candidate Evaluations -.870(.146)*** .712(.136)*** Government Structure -.957(.124)*** -1.596(.116)*** Partisanship X Government Structure .035(.084) .072(.082) Candidate Evaluations X Government .862(.148)*** -.618(.140)*** Structure Government Structure is 0=presidential system, 1=parliamentary system Conservative Vote Liberal Vote Constant -.358(.070)*** -.521(.072)*** Partisanship -.512(.049)*** .429(.050)*** Candidate Evaluations -.007(.028) .093(.031)** Government Structure .957(.124)*** 1.596(.116)*** Partisanship X Government Structure -.035(.084) -.072(.082) Candidate Evaluations X Government -.862(.148)*** .618(.140)*** Structure

Number of Observations 2813 Chi-square 1273.97*** Pseudo R-square .2084

Data: Pooled Data of 1995 Taiwan ESC Data and 1976 Japan JABISS data Note: Entries are logistic coefficients. Standard errors are in parentheses. **:p<.01; ***:p<.001. Abstention is the comparison group.

Table 6.2: Multi-nomial logit estimates of three-choice model, Taiwan and Japan

247

Government Structure is 1=presidential system, 0=parliamentary system Conservative Vote Liberal Vote Constant .194(.108) 1.187(.086)** Partisanship -.523(.075)*** .327(.070)*** Candidate Evaluations -.699(.108)*** .387(.098)*** Government Structure -1.508(.139)*** -2.850(.137)*** Partisanship X Government Structure .216(.085)* .000(.085) Candidate Evaluations X Government .011(.130) .227(.118) Structure Government Structure is 0=presidential system, 1=parliamentary system Conservative Vote Liberal Vote Constant -1.313(.088)*** -1.662(.107)*** Partisanship -.307(.040)*** .327(.048)*** Candidate Evaluations -.710(.072)*** .614(.066)*** Government Structure 1.508(.139)*** 2.850(.137)*** Partisanship X Government Structure -.216(.085)* .000(.085) Candidate Evaluations X Government .011(.130) -.227(.118) Structure

Number of Observations 2524 Chi-square 1543.45*** Pseudo R-square .2800

Data: Pooled Data of the 1995 U.S. NES Data and the 1996 Japan JEDS data Note: Entries are logistic coefficients. Standard errors are in parentheses.**:p<.01; ***:p<.001. Comparison group is abstention.

Table 6.3: Multi-nomial logit estimates of three-choice model, the U.S. and Japan

248

First, the governmental structure dummy variable is significant in both models, which suggests that the type of government system has an effect. Of interest is the interaction between government system and the two political attitude variables—partisanship and candidate evaluations. In Table 6.2, the product term of candidate evaluations and government structure has a significant effect on the probability of vote choices versus abstention. In Table 6.3, the interaction between partisanship and government system among others is a significant predictor. The effect of government system merely goes through political attitude for the conservative vote, which is different from the result in

Table 6.2. Using the same method employed in the three previous chapters, the effects produced by the two independent variables and their interaction with the dummy variable are shown in Table 6.4.80

80 By taking 0 and 1 for the dummy variable in turn, the coefficients of the two continuous quantitative variables reflect their main effects on the probability of vote choice against abstention. For instance, the coefficient of party identification for the conservative vote in parliamentary systems is -.548. Holding the value of the candidate evaluation variable at its mean and the dummy variable and the product terms 0, the discrete change of the probability of the dependent variable is -.186.

249

Taiwan and Japan before 1994 Parliamentary(1) Presidential(2) Difference*|(2)|- |(1)| Conservative Vote Party Identification(a) -.186 -.190 .004 Candidate -.336 -.003 -.333 Evaluations(b) Difference**|(a)|-|(b)| -.180 .187 Liberal Vote Party Identification .105 .149 .044 Candidate Evaluations .242 .038 -.204 Difference -.137 .111 US and Japan after 1994 Parliamentary Presidential Difference Conservative Vote Party Identification -.226 -.092 -.134 Candidate Evaluations -.225 -.157 .068 Difference .001 -.065 Liberal Vote Party Identification .104 .077 -.027 Candidate Evaluations .091 .108 .017 Difference .013 -.031

Note: Discrete change is calculated as follows (Long, 1997). Because [p(y=m)/1- p(y=m)]=exp(Xβ'), thus p(y=m)=[exp(Xβ')/1+exp(Xβ')]. Let p1(y=m|x, xi+½si ) the predicted probability that ith variable at its mean plus one-half of its standard deviation, holding the value of the dummy variable 1 and the value of the product terms and the two attitudinal variables at their mean, and p2(y=m|x, xi-½si ) the predicted probability that ith variable at its mean minus one-half of its standard deviation, holding the value of the dummy variable 1 and the value of the product terms and the two attitudinal variables at their mean. The difference between pi1 and pi2 is the centered standard deviation change. *: Difference is obtained by subtracting the absolute value of the discrete change caused by party identification or candidate evaluations in parliamentary systems from the absolute value of the discrete change in presidential systems. **: Difference is obtained by subtracting the absolute value of the discrete change caused by party identification from the absolute value of the discrete change caused by candidate evaluations.

Table 6.4: Discrete change of probability of vote choice with dummy variable of government structure

250

The discrete change for the two variables is computed with the same method used in the previous chapters. Notice that the effect of concern is not whether the two independent variables increase or decrease the probability of voting for either party, but the extent to which a one standard-deviation difference in the value of the independent variables leads to change in the probability of vote choice. The change of the probability in the dependent variable is therefore transformed to the absolute value.

First, we examine the situation of voting for the conservative party in the pooled data of the legislative election in Taiwan and the 1976 general election of

Japan. When voting takes place in presidential systems, a one standard-deviation change around the mean in party identification will decrease the probability of voting for the conservative party by .190. When the government system is parliamentary, the effect of partisanship is virtually identical: -.186. The difference between the two systems is only -.004. Candidate evaluations, on the other hand, have much larger impact on vote choice in parliamentary systems than in presidential systems; the difference is .333, which rejects the hypothesis that candidate evaluations has a larger effect on vote choice in presidential systems than in parliamentary systems. Examining the effect of candidate evaluations and partisanship on the conservative vote choice under presidential systems, however, it is found that partisanship has a greater impact than do candidate evaluations by .187. By contrast, when people cast their ballots in parliamentary systems and the vote choice is the conservative party against abstention and other parties, candidate evaluations have a larger impact than does partisanship by .180.

251

Turning to the liberal vote, both partisanship and candidate evaluations become weaker in parliamentary systems, but the influence of candidate evaluations remains larger than partisanship. The difference is .137. In presidential systems, the main effect of candidate evaluations is .038 and that of partisanship is .149. From the perspective of the two variables, party identification has a smaller impact in parliamentary systems than in presidential systems.

Candidate evaluations, on the other hand, have much larger impact in parliamentary systems than in presidential systems. In other words, candidate evaluations have less effect in presidential systems than in parliamentary systems, when the vote choice is the liberal party against abstention and other parties.

Changing to the single-member-district election, the influence of candidate evaluations and party identification is ambiguous. When the value of the government system variable is set as 0, namely, the place that conservative voting occurs is a parliamentary system, the effect of candidate evaluations is -.225.

When voting takes place under a presidential system, the effect of candidate evaluations is -.157. In other words, the influence of candidate evaluations decreases by .068 when the setting transfers from parliamentary systems to presidential systems. When voting takes place under a presidential system and changes to a parliamentary system, on the other hand, the effect of partisanship will decrease by .134 between presidential systems and parliamentary systems.

When the vote choice is the liberal party against abstention, the effect of candidate evaluations due to change of the government structure from parliamentary systems to presidential systems is .017, but the effect of partisanship will drop by .027.

The change is not very impressive, but it is in the expected direction. Party

252

identification has greater impact on vote choice in parliamentary systems, and candidate evaluations perform better in presidential elections, which confirms

Proposition 6.2.

According to the estimates presented in Table 6.2 and Table 6.3, the interaction between partisanship and government structure is not significant for the pooled data of Taiwan and Japan before 1994, and both the product terms of partisanship and candidate evaluations with government structure are not significant for the pooled data of the U.S. and Japan after 1994. This explains why the effect of party identification remains nearly unchanged across the two government structures of Taiwan and Japan, and why the effects of both attitudinal variables are not impressive for the U.S. and Japan after 1994.

It is also apparent that candidate evaluations and partisanship do not have consistent impact on vote choice as we control for electoral systems. In the multi- member–district legislative elections held in Taiwan in 1995 and Japan in 1976, candidate evaluations appear to have larger impact on vote choice in Japan and partisanship has a larger impact in Taiwan. I suspect that candidates remain closely related to political parties among Taiwanese voters; hence candidate evaluations play a small role in voting decision. Japanese voters, on the other hand, stayed with candidate evaluations given their strong tradition of candidate- centered pattern. The discrepancy between the two forces becomes smaller as to the liberal vote, particularly for the impact of candidate evaluations across the two political systems. Although conservative candidates in Taiwan also built a personal vote like their Japanese counterparts, the ruling party in Taiwan, the

KMT, strictly controlled party endorsement at that time. Moreover, the

253

presidential system was not yet consolidated as firmly as in the United States; the strong influence of candidate evaluations in the second panel of Table 6.4 demonstrates this point perfectly.

Because the United States is a typical presidential system, the relative influence of candidate evaluations over party identification is greater than in its

Japanese counterpart. After the 1994 electoral reform, the influence of candidate evaluations still prevails among the Japanese voters, especially for the conservative vote choice. However, party identification has increased its influence due to single-member districts; the relative advantage of candidate evaluations is smaller than in its American counterpart. As for the liberal vote, the relative influence of party attachment remains strong.

The results presented in this section imply that the influence of party identification is still overwhelming in the 1995 legislative election of Taiwan.

However, the importance of candidate judgment might increase in subsequent elections. First, the conservative party has been losing its control over its candidates as party competition intensifies and the power of resource allocation is more decentralized. The divisions between parties are ambiguous in terms of ideology and interest representation, which downgrades the role of party identification.

Second, the personal vote should gain more strength as the operation of the presidential system becomes more mature in Taiwan. In early years, the president used the party machine to coordinate individual law-makers. Nowadays, the president often instructs his administration to bargain with law-makers of different parties. Political parties will merely act as the president’s campaign machine, and

254

flexible party coordination initiated by the president will weaken the importance of party labels and increase the role of candidates.81 All in all, the maturity of party competition and presidential systems may need to be taken account for the interpretation of the results.

6.4 Election Systems and Voting Pattern

Electoral competition involves candidate coordination; candidates base their decision to enter the race on their judgment of the probability of wining the election, and public expectation shaped by the electoral rules determines individual vote choices (Cox, 1997). Both SMSP and PR party-list systems tie political parties to candidates, because each party tends to or is obligated to endorse only one candidate or party list. Candidate selection and recruitment, therefore, is determined more by the national organization than the local members

(Maor, 1997). In multi-member districts, however, it is inevitable that parties run more than one candidate and candidates compete against each other as well as against the opposition (Grofman, 1999). Cox and Niou (1994) argued that large parties are prone to commit over-nomination error in the SNTV system, particularly in large districts. In Japan, however, large parties use the power of resource allocation to facilitate particularistic interests in each district (McCubbins and Rosenbluth, 1995). Political parties manipulate their coordination strategy in the electoral stage to achieve over-representation in terms of the ratio of seats to

81 The large impact of candidate evaluations in the 1976 general election of Japan sets a benchmark for multi-member-district elections. Since the same combination of political system and election system hardly appears again, the analysis here is aimed to reveal the characteristics of Japanese voting behavior against other voting patterns in different types of political arrangements.

255

votes. In Taiwan, the extraordinary party assets and loyalty generated by the authoritarian regime allowed the ruling party—the KMT—to divide a district into several “responsibility zones,” which maximizes the efficiency of vote equalization (Liu, 1999). Those coordination strategies only manifest the intense intra-party competition in multi-member districts (Lin, 1996).

The hypothesis that candidate-based voting prevails in SNTV, like Taiwan and Japan before the electoral reform, and party-based voting is relatively important in SMSP, like the U.S. and Japan after the electoral reform, can be formulated as the following equation.

P(Y│D,X)= β0+ β1 X1+ β2 X2+β4D2+β7D2X1+β8D2X2 ---(6.6) where Y=vote choice;

X1= party identification;

X2= candidate evaluations;

D2= 1 for multi-member district system; 0 otherwise.

The pooled data of Taiwan and the U.S., and Japan before and after 1994 are used to estimate Equation 6.6 with a multi-nomial logit model. Again, different logistic models are estimated with the value of the dummy variable redefined each time. Table 6.5 and Table 6.6 present the results of the two models.

256

Election System is 1=multi-member system, 0=single-member system Conservative Vote Liberal Vote Constant -1.313(.088)*** -1.662(.107)*** Partisanship -.307(.040)*** .327(.048)*** Candidate Evaluations -.710(.072)*** .614(.066)*** Election System .954(.112)*** 1.141(.129)*** Partisanship X Election System -.205(.064)*** .101(.069) Candidate Evaluations X Election .702(.078)*** -.520(.073)*** System Election System is 0=multi-member system, 1=single-member system Conservative Vote Liberal Vote Constant -.358(.070)*** .429(.050)*** Partisanship -.512(.049)*** .093(.031)** Candidate Evaluations -.007(.028) -1.141(.129)*** Election System .954(.112)*** -.101(.069) Partisanship X Election System -.205(.064)*** .520(.073)*** Candidate Evaluations X Election .702(.078)*** -.520(.073)*** System

Number of Observations 2762 Chi-square 945.71*** Pseudo R-square .1619

Note: Entries are logistic coefficients. Standard errors are in parentheses. **: p<.01; ***:p<.001. Comparison group is abstention

Table 6.5: Multi-nomial logit estimates of three-choice model, Taiwan and the United States

257

Election System is 1=multi-member system, 0=single-member system Conservative Vote Liberal Vote Constant .194(.108) 1.187(.086)*** Partisanship -.523(.075)*** .327(.070)*** Candidate Evaluations -.233(.036)*** .129(.032)*** Election System .404(.148)** -.112(.125) Partisanship X Election System -.024(.101) .029(.096) Candidate Evaluations X Election -.637(.150)*** .583(.140)*** System Election System is 0=multi-member system, 1=single-member system Conservative Vote Liberal Vote Constant .599(.102)*** 1.074(.091)*** Partisanship -.548(.068)*** .356(.065)*** Candidate Evaluations -.870(.146)*** .712(.136)*** Election System -.404(.148)** .112(.125) Partisanship X Election System .024(.101) -.029(.096) Candidate Evaluations X Election .637(.150)*** -.583(.140)*** System

Number of Observations 2575 Chi-square 1174.99*** Pseudo R-square .2310

Note: Entries are logistic coefficients. Standard errors are in parentheses. **: p<.01; ***:p<.001. Abstention is the comparison group.

Table 6.6: Multi-nomial logit estimate of three-choice model, Japan before and after 1994

Table 6.5 reveals the differential effect of the election system variable, particularly for the conservative vote. Both partisanship and candidate evaluations are significant predictors for vote choice in this circumstance, except for the conservative vote when the election system is single-member districts and the vote choice is the conservative party versus abstention and other parties. The product terms for the election in the presidential system are significant predictors, and they appear to offset the main effect of the corresponding explanatory variables,

258

particularly for candidate evaluations. The analysis of the pooled Japanese data is presented in Table 6.6. The election system has an effect on the conservative vote versus abstention and others, but it solely interacts with the way people evaluate candidates. The coefficient of the product term has the same sign as the coefficient of candidate evaluations. To assess the relative influence of the two independent variables, the coefficients are transformed to discrete change in the probability of the voting choice, which is presented in Table 6.7.

259

Taiwan and the U.S. Single- Multi- Difference* Member(1) Member(2) |(2)|-|(1)| Conservative Vote Party Identification(a) -.087 -.214 .127 Candidate -.220 -.003 -.217 Evaluations(b) Difference**|(a)|-|(b)| -.133 .211 Liberal Vote Party Identification .080 .177 .097 Candidate Evaluations .164 .041 -.123 Difference -.084 .136 Japan before and after 1994 Single-Member Multi-Member Difference Conservative Vote Party Identification -.190 -.173 -.017 Candidate Evaluations -.142 -.438 .296 Difference .048 -.265 Liberal Vote Party Identification .093 .111 .018 Candidate Evaluations .061 .360 .299 Difference .032 -.249

Note: Discrete change is calculated as follows (Long, 1997). Because [p(y=m)/1- p(y=m)]=exp(Xβ'), thus p(y=m)=[exp(Xβ')/1+exp(Xβ')]. Let p1(y=m|x, xi+½si ) the predicted probability that ith variable at its mean plus one-half of its standard deviation, holding the value of the dummy variable 1 and the value of the product terms and the two attitudinal variables at their mean, and p2(y=m|x, xi-½si ) the predicted probability that ith variable at its mean minus one-half of its standard deviation, holding the value of the dummy variable 1 and the value of the product terms and the two attitudinal variables at their mean. The difference between pi1 and pi2 is the centered standard deviation change. *: Difference is obtained by subtracting the absolute value of the discrete change caused by party identification or candidate evaluations in single-member districts from the absolute value of the discrete change in multi-member districts. **: Difference is obtained by subtracting the absolute value of the discrete change caused by party identification from the absolute value of the discrete change caused by candidate evaluations.

Table 6.7: Discrete change of probability of vote choice with dummy variable of election system

260

In the first panel of Table 6.7, Taiwan and the U.S. pooled-data is analyzed.

Candidate evaluations have smaller impact on vote choice than party identification in multi-member districts, and larger impact than party identification in single- member districts, regardless of vote choice. When the election system changes from multi-member districts to single-member districts, the impact of candidate evaluations decreases by .217, but the impact of party identification goes up by .127. When we fix the electoral system and examine the effect of the two variables, we find that party identification has great influence in multi-member districts but little impact on vote choice in single-member districts. Conversely, candidate evaluations are critical to vote choice in single-member districts and nearly irrelevant to vote choice in multi-member districts. The result suggests that party identification still exerts influence on voting behavior in Taiwan and that candidate evaluations, as a result of incumbency advantage, prevails in the United

States. The control variable, election system, merely changes the original influence of candidate evaluations, which can be seen in Table 6.6. 82 This seemingly unsatisfactory finding should be reconciled considering the evolving candidate-centered pattern of Taiwanese voters in the future; Chapter 4 has shown that candidate evaluations gradually take up the main position of voting decisions as the influence of party label decreases with party fragmentation. In the United

States, the strong impact of candidate assessment emerges in its single-member districts because of the incumbency factor and personal vote caused by the presidential system. Under the similar government system, American voters take

82 In other words, the control variable merely changes the constant of the equation, but not the slope of the variables. In this case, the estimates of the equation with the dummy variable are not different from that of the original equation.

261

candidate evaluations further than Taiwanese voters, who used to be mobilized by political parties and still vote for candidates from the same party. All in all, the effect of partisanship in single-member districts is not significant.

As for Japanese voters, the 1994 electoral reform posed a challenge to the electorate concerning their vote choice. The mixed electoral system—300 members elected in single-member districts and 200 members elected in the proportional representation districts according to their rankings on the party list— allows political parties to put candidates in both single-member districts and party list. Most candidates took dual candidacy—they ran for the SMD seats while being listed on party list (Kohno, 1997; McKean and Scheiner, 2000).83 The mixed system encourages the electorate to split their votes; the LDP won more votes in single-member districts than in PR districts (Kohno, 1997; Mair and

Sakano, 1998). Because of the dual candidacy and split-voting, the mixed electoral system may still preserve candidate-oriented voting in single-member districts. The case study presented in Chapter 5 has suggested that the importance of party label increased in the 1996 general election. The second panel of Table

6.7 shows the overwhelming influence of candidate evaluations on the conservative vote in the election before electoral reform, which is -.438. After the reform, namely the adoption of single-member districts, partisanship has the larger impact on vote choice. When the electoral rule changes from single-member districts to multi-member districts, the influence of party attachment decreases by .017 but that of candidate evaluations increases by .296. Regardless of the

83 Each candidate can run in a single-member district or multi-member district or both. When he wins the single-member-district seat, he takes it and leaves the party list. All candidates ranked

262

liberal vote or the conservative vote, the Japanese voters undoubtedly embraced a party-centered pattern after the 1994 reform. This suggests that candidate concern has been replaced by party identification, and the difference in the relative weight of party identification between the two electoral rules shows the effect of political institutions on political behavior. With the new electoral rules, the electorate would be inclined to vote along with parties regardless of the mobilization of candidates before that election. As matter of a fact, party voting has been found among the electorate (Flanagan, Kohei, Miyake, Richardson, and Watanuki, 1991); the electoral reform did not revive it but increased it.

Up to this point, it has been shown that some cases conform to the proposition but some do not. There are two ways to view this result. First, the characteristics of each individual case are not the same. Because of the evolving party competition in Taiwan, for example, Taiwanese voters were party-oriented at that time and utilized candidate-voting pattern less than what they are expected to have. When Taiwanese voters are compared with Japanese voters in 1976 and

American voters in 1998, the uncharacteristic party-oriented pattern emerged.

This explains why partisanship rather than candidate evaluations is important to

Taiwan when we control for election systems and government structures.

The second way is to look at the interaction between dummy variables and the attitudinal variables. Only candidate evaluations have a significant interaction effect with government structure, which makes their impact on vote choice greater than partisanship. As for election systems, partisanship has an interaction effect in the pooled data of Taiwan and the U.S., but not in the pooled data of Japan before

beneath him on the party’s list can move up. Some dual candidates who are defeated in single-

263

and after 1994. To answer whether the characteristic of cases or the difference between institutions leads to the results, it is necessary to pool the four cases together and control for two-way and three-way interaction between system-level and attitudinal variables..

6.5 Pooled-Data Analysis of Four Political Systems

The first research question of this project is whether or not system-level variables affect political behavior in terms of vote choice. If the answer is yes, the second central question follows: In which type of political system does party voting takes place? The two-by-two typology that differentiates political systems as four types based on electoral rules and government structure has been broken down as four pairs of comparisons, contrasting two sets of countries by their electoral systems or government structure. Finally, all four datasets are next pooled to allow us to simultaneously estimate the effect of party identification and candidate evaluations in four settings, including the two Japanese cases.

member districts could still win PR seats.

264

Panel a: Government system is 1=presidential system, 0=parliamentary system; Election system is 1=multi-member districts, 0=single-member districts Conservative Vote Liberal Vote Constant .008(.086) .903(.071)*** Partisanship -.542(.070)*** .320(.066)*** Candidate Evaluations -.704(.102)*** .408(.092)*** Election System .847(.089)*** .522(.083)*** Government -1.275(.093)*** -2.175(.086)*** Partisanship X Election System .004(.100) .020(.097) Partisanship X Government Structure .232(.080)** -.086(.078) Candidate Evaluations X Election System -.148(.185) .262(.174) Candidate Evaluations X Government .001(.123) .126(.111) Structure Partisanship X Election System X -.208(.115) .225(.116) Government Structure Candidate Evaluations X Election System .841(.198)*** -.693(.187)*** X Government Structure Panel b: Government system is 1=presidential system, 0=parliamentary system; Election system is 0=multi-member districts, 1=single-member districts Conservative Vote Liberal Vote Constant -1.267(.077)** -1.266(.076)*** Partisanship -.310(.040)*** .234(.040)*** Candidate Evaluations -.703(.071)*** .535(.061)*** Election System .847(.089)*** .522(.083)*** Government 1.275(.093)*** 2.175(.086)*** Partisanship X Election System -.203(.062)*** .246(.064)*** Partisanship X Government Structure -.232(.080)*** .086(.078) Candidate Evaluations X Election System .693(.077)*** -.431(.069)*** Candidate Evaluations X Government -.001(.123) -.126(.111) Structure Partisanship X Election System X .208(.115) -.225(.116) Government Structure Candidate Evaluations X Election System -.841(.198)*** .693(.187)*** X Government Structure Panel c: Government system is 0=presidential system, 1=parliamentary system; Election system is 1=multi-member districts, 0=single-member districts Conservative Vote Liberal Vote Constant .856(.088)*** 1.431(.080)*** Partisanship -.538(.073)*** .341(.071)*** Candidate Evaluations -.852(.156)*** .671(.147)*** Election System -.847(.089)*** -.522(.083)*** Government -1.275(.093)*** -2.175(.086)*** Partisanship X Election System -.004(.100) -.020(.097) Candidate Evaluations X Election System .023(.087) .139(.087) Partisanship X Government Structure .148(.185) -.262(.174)

265

Candidate Evaluations X Government .842(.159)*** -.567(.151)*** Structure Partisanship X Election System X .208(.115) -.225(.116) Government Structure Candidate Evaluations X Election System -.841(.198)*** .693(.187)*** X Government Structure Panel d: Government system is 0=presidential system, 1=parliamentary system; Election system is 0=multi-member districts, 1=single-member districts Conservative Vote Liberal Vote Constant -.419(.064)*** -.744(.068)*** Partisanship -.514(.048)*** .480(.051)*** Candidate Evaluations -.009(.028) .104(.031)*** Election System -.847(.089)*** -.522(.083)*** Government 1.275(.093)*** 2.175(.086)*** Partisanship X Election System .203(.062)*** -.246(.064)*** Candidate Evaluations X Election System -.023(.087) -.139(.087) Partisanship X Government Structure -.693(.077)*** .431(.069)*** Candidate Evaluations X Government -.842(.159)*** .567(.151)*** Structure Partisanship X Election System X -.208(.115) .225(.116) Government Structure Candidate Evaluations X Election System .841(.198)*** -.693(.187)*** X Government Structure

Number of Observations 5337 Chi-square 2844.78*** Pseudo R-square .2431

Note: Entries are the logit coefficients. Standard errors are in parentheses. *:p<.05, **:p<.01, ***:p<.001. Abstention is the comparison group.

Table 6.8: Multi-nomial logit estimates of three-choice model, the United States, Taiwan, and Japan before and after 1994

266

The pooled-data analysis calls attention to several points. First, both attitudinal variables are significant predictors and they have the right signs.

Second, both structural variables have an effect on the way people choose parties, which confirms the first research hypothesis that the political setting affects political behavior. However, only candidate assessment for the most part has an effect through both system-level variables; the coefficients for both interaction terms between candidate evaluations and partisanship are significant. The insignificant coefficients of the product term of partisanship with the two system- level variables make it clear that partisanship can nearly penetrate each type of political institution while the influence of candidate evaluations is conditioned on government structure and election system. Third, the interaction between candidate evaluations and government system and election system offsets the original effect of candidate evaluations. The opposite signs of the two interaction terms deflate the coefficients of candidate evaluations. To demonstrate the effects of the qualitative and quantitative variables and the product terms, the discrete change method is used, calculating the change of the probability of vote choice due to one standard deviation change around the mean of the focal variables.

Table 6.9 shows the main effect of the two quantitative variables. The difference in their effects on vote choice is displayed in Table 6.10, which sets up the frame of the typology and the four cases for easy understanding.

267

Conservative Vote Party Candidate Identification Evaluations Multi-Member District/Presidential System -.209 -.003 Multi-Member District/Parliamentary System -.104 -.223 Single-Member District/Presidential System -.157 -.236 Single-Member District/Parliamentary -.214 -.262 System Liberal Vote Party Candidate Identification Evaluations Multi-Member District/Presidential System .164 .034 Multi-Member District/Parliamentary System .062 .135 Single-Member District/Presidential System .095 .179 Single-Member District/Parliamentary .209 .138 System

Table 6.9: Discrete change of probability of vote choice, pooled data

268

Conservative Vote Presidential System Parliamentary System Multi-Member District Taiwan(.206) Japan before 1994(-.119) Single-Member District United States(-.079) Japan after 1994(-.048) Liberal Vote Presidential System Parliamentary System Multi-Member District Taiwan(.130) Japan before 1994(-.073) Single-Member District United States(-.084) Japan after 1994(.071)

Note: Parentheses are the difference in effect between candidate evaluations and partisanship. It is obtained by subtracting the absolute value of the discrete change caused by candidate evaluations from the absolute value of the discrete change caused by partisanship. The negative number therefore stands for the advantage of candidate evaluations over party identification.

Table 6.10: Comparison of candidate evaluations and partisanship, pooled data

The results presented in Table 6.9 and Table 6.10 partly confirm the expectations of the propensity of party voting in different combinations of election systems and government systems. The first half of Table 6.9 indicates the influence of partisanship over candidate evaluations in multi-member districts with presidential systems. Multi-member districts with presidential systems have a strong party-voting pattern, which is against the hypothesis set by the typology.

Both multi-member districts with parliamentary systems and single-member district with parliamentary systems have balanced party-oriented and candidate- oriented patterns. Single-member districts with parliamentary systems, however, stress the role of candidate evaluations in voting choice. All in all, two cases are off our hypothesis. The second half of Table 6.9 has the similar feature, except the case of single-member districts with parliamentary systems; party attachment outweighs candidate assessments in this system. Partisanship remains more

269

influential than candidate evaluations in multi-member districts with presidential systems, but they are nearly equal in both multi-member districts with presidential systems and single-member district with parliamentary systems.

The first panel of Table 6.10 demonstrates the effect of candidate evaluations in three out of four political systems. Taiwanese voters still take cues from parties when making their voting decisions, thus partisanship rather than candidate evaluations is strong in this country. With presidential coattail effects developing, the candidate-centered pattern is expected to prevail in Taiwan. Japan in 1976 has a candidate-centered pattern. The moderate level of party implies that its parliamentary system offsets the influence of medium-sized districts.

Considering the incumbency advantage, it is understandable that the United States exhibits stronger level of candidate voting. Compared to the United States, the

Japanese electorate in the 1996 general election looks more candidate-oriented for the LDP vote. Because of the dual candidacy and split-voting, the mixed electoral system may still preserve candidate-oriented voting in single-member districts.

The second part of Table 6.10 presents the results regarding the liberal vote. First, both the United States and Japan before the 1994 reform exhibit moderate level of candidate voting respectively. Taiwanese voters appear to be a typical party voting country, in part because of the relatively small but united opposition parties in terms of nomination. Japanese voters reveal a different voting pattern for the liberal vote. Compared to the first part of Table 6.10, a high level of party voting for the 1996 general election conforms with the typology, in part because a great portion of voters followed the defecting candidates from the

LDP to new parties. Table 5.29 (p.225) shows that NFP partisanship has more

270

influence on the probability of vote choice than candidate evaluations, which sheds light on the strong party voting in the liberal vote presented in the pooled- analysis here.

Overall, the findings for the most part confirm the hypothesis that political institutions have interaction effects with political attitudes, especially candidate evaluations. The three-way interaction between candidate evaluations and institutions actually appears across the board, which explains why candidate evaluations prevail more than expected.

Our analysis strategy echoes Shvetsova’s (1995) point that consideration of institutional factors should rest on individual case study that is aimed to take the specific characteristic of context into account. Because of the characteristics of individual cases, however, the results do not entirely fit the original predictions regarding the level of party voting across four types of political institutions. The developing political systems of Taiwan and post-reform Japan undermine the prediction of the theoretical framework, but the findings of the previous individual case studies positively account for the deviations of voting patterns in these two political systems. The assessment of the variables based on multi-nomial logit model not only confirms the influence of the system-level variables, but also reveals the inadequate explanation of institutional factors regarding individual voting patterns. All in all, the evidence shown in the three preceding sections strongly supports the assumption that the level of party voting varies with political institutions in terms of election systems and government structure.

271

6.6 Conclusion

The individual case study approach has been used to disclose how people decide their vote choices on the grounds of ideology, partisanship, issues, and campaign context. The explanation regarding why a particular political attitude prevails in the country is context-specific, if not idiosyncratic. The goal of developing a generalizable social science theory will be fulfilled only when the cumulative knowledge is cross-context.

The “most different” research design allows us to test the theory with control of certain system-level variables. The basic assumption is that political settings affect the level of a variable or its relationship with another variable, and that system-level variables account for the variation of the relationship in each setting.

The two-by-two typology developed in Chapter 2 predicts that multi-member districts with presidential systems have the weakest propensity of party voting and single-member districts with parliamentary systems have the strongest tendency for party voting. The pair-wise comparisons of four cases presented in Section 6.3 and Section 6.4 reveal that the main effect of party identification and candidate evaluations indeed vary with the setting of electoral rules and legislative-executive relationship, but Taiwan and Japan in 1996 deviate from the prediction. Section

6.5 analyzes the pooled data, showing that the three-way interaction between candidate evaluations and political institutions is significant across the four cases.

Taiwanese voters still hold party attachment tight as they turn out to vote. Japan in

1996 shows weak candidate voting for the conservative vote and moderate party voting for the liberal vote. The U.S. and pre-reform Japan show a tendency for

272

candidate voting. Nevertheless, the level of party voting in both countries is expected to lie between Japan after 1994 and Taiwan. The deviation may be caused by the strong effect of candidate evaluation across the context.

All in all, the empirical evidence shows that party voting theory— partisanship is the most effective determinant of voting behavior—does not hold at least in these four cases. The relative influence between party attachment and candidate evaluations, however, is not entirely correct as the theoretical framework expects. The results of the pooled data analysis actually conform to what the four individual case studies have found, which indicates that some characteristics of each case are part of the source of the variance that needs to be explained.

The three limitations—the number of observations, the timing of observations, and the maturity of two of the four observations—limit the inference generated by the pooled data. Considering the quasi-experimental design, this is still a useful study. At least in the four political systems studied here, inter-system differences affect the association between vote choice and partisanship. Due to the limitations, the evidence presented here may not challenge the generalizability of party voting theory. However, it is shown that we can employ both individual case studies and pooled-data analysis to study the generalization of party voting model systematically.

273

CHAPTER 7

CONCLUSION

7.1 Review of Method

The authors of The American Voter introduced a new perspective into political behavior research to study the causes of the voting act. Their integrated social-psychological model includes long-term forces--social demographic background, ideology, partisanship—and short-term forces—candidate evaluations and issue positions—and argues that partisanship has a stable impact on campaign-related attitudes and the voting act. While they examined party identification and voting behavior, they noticed the importance of context, asserting that “Certainly the number of parties competing effectively in the electoral process is a tremendously significant aspect of the context in which the electorate must reach a decision; the influence of a two-party system on American electoral behavior has been implicit in every chapter of this book.” (Campbell,

Converse, Miller, and Stokes, 1960: 552). It is not fair, from this quote, to regard party voting theory as an entirely context-free theory. The data used in The

American Voter, however, was collected solely in America. Additionally, their

274

study covers only two elections (1952, 1956).84 The generalizability of this social psychological model, which has reigned in the study of political behavior for decades, has received attention from scholars in America and abroad.

According to the revisionists of party voting theory, party identification, which orchestrates the perceptions of candidates, issues, and political parties, is influenced by short-term forces. Partisanship reflects not only psychological attachment but also evaluation of the performance of government. A major finding, the revisionists discovered, is that the relative assessment of candidates summarizes issue positions and party attachment, and that partisanship plays a role in voting choice when people cannot find any differences between candidates.

The decline of partisan strength and the increasing number of independents in the

U.S. challenge the theory of the stability of partisanship. In Europe, critics pointed out that social class may be closer to vote choice than is partisanship. Moreover, a multi-party setting undermines the effect of partisanship, because partisans of minor parties have unstable turnout and, consequently, volatile party support.

Social cleavages lead to better prediction of the change of aggregate-level voting than affective party orientation.

Some scholars have considered the linkage between institutions and political behavior when they ponder the causes of the voting act. Cain, Ferejohn, and Fiorina (1987), for instance, contend that the different government structures in the United States and Great Britain may account for the different levels of personal votes. In his research on party image, Wattenberg (1982) also notes that differences in terms of culture and political system account for the seemingly

84 In the later research conducted by the authors of The American Voter, however, more elections

275

similar pattern of party decline. Inspired by the literature of political parties, this dissertation aims to present a systematic comparative study of political behavior.

It is expected that political institutions have an impact on the association between partisanship and the voting act. And it also predicts that multi-member districts with presidential systems, such as Taiwan, have the weakest level of party voting while single-member districts with parliamentary systems, such as Japan after

1994, have the greatest level. The propensity of party voting in multi-member districts with parliamentary systems, like Japan before 1994 and single-member districts with presidential systems like the U.S., lie between these two extremes.

To test this hypothesis, two political institution variables--electoral rules and executive-legislative relationship--are set up to predict the propensity of party voting in different political systems. Using the “most different design,” this dissertation compares the voting pattern in the United States, Taiwan, and Japan under two different electoral systems. The result allows us to assess the importance of the partisanship component and the candidate evaluation component.

7.2 Review of Findings

The influence of political arrangements aside, the individual case studies on voting behavior in the United States, Taiwan, and Japan, reveal, if not confirm, the importance of candidate evaluations in legislative elections. Based on the previous literature, the incumbency advantage has played an important role in U.S. congressional elections since the 1940s (Mayhew, 1973; Fiorina, 1989). Since

are covered.

276

then, the electoral importance of party labels diminished (Jacobson, 1997).

Incumbent representatives enjoy a great deal of privileges granted by the institution, and they have higher visibility and better access to campaign money than do challengers. The notion of a personal vote applies to congressional elections because the presidential system allows individual lawmakers to wield their influence in the policy-making process. However, single-member districts impose the consideration of party labels on the electorate’s mind, thus partisanship joins candidate evaluations in determining vote decision. Chapter 3 demonstrates that candidate evaluations outweigh partisanship in voting in 1990 and 1998, particularly for the majority party in the House of Representatives.

Taiwan was undergoing a transition from a one-party authoritarian regime to a multi-party system from the late 1980s to the mid-90s. Due to SNTV with a presidential system, the candidate-centered pattern should have prevailed among

Taiwanese voters. Partisanship nevertheless remains dominant in terms of the vote decision because of intense party confrontation on democratization and unification issues. In addition, the president still relies on the party organization to exert his influence over individual legislators. The deviation of Taiwan’s case from the prediction implies the importance of political culture and the evolving party system, which mirrors the successful prediction of Japan’s case.

Japan used SNTV for its general election from 1947 to 1993. SNTV coupled with medium-sized multi-member districts creates incentives for candidates to seek a personal vote. The combination of a parliamentary system and medium-sized multi-member election districts makes it possible for the election campaign to be decentralized (Richardson, 1991). The 1994 electoral

277

reform imposed single-member-district elections on three-fifth of the lower house seats, with the remainder to be elected by proportional representation. According to the multi-nomial logit estimates presented in Chapter 5, the influence of partisanship among Japanese voters is growing. Whether partisanship gains its momentum remains a question to be answered, yet the fact that the electoral reform contributes to its strength is far from illusion. Thanks to the old personal campaign machines and remaining proportional-representation districts, however, the electorate still draws on candidate assessment to a certain degree.

The association between vote choice, partisanship, and candidate evaluations is assessed with control of other variables, including issue positions, party images, retrospective evaluation of the national economy, ideology, and social demographic variables. With these short-term and long-term forces controlled, the estimated impact of partisanship and candidate evaluations is assessed more accurately and comparably with other studies. Partisanship and candidate evaluations for the most part dominate vote choice. That implies that legislative elections lack decisive issues, national figures, and salient party images, regardless of country. This implication is important for the theory of representative democracy.

The individual case studies allow us to examine the voting pattern in each country by estimating the effects of attitudinal and demographic variables. Voters emphasize candidate evaluations because legislative elections essentially involve representation of local interests to a great degree. Moreover, legislative elections receive little media coverage, which gives incumbents more leeway to influence their constituents through personal contacts. In Japan, as a counter example,

278

national media concentrate on parties rather than candidates, which offsets the personal campaign machine in local districts. Individual candidates, for the most part, approach voters more efficiently than do political parties. Because of the endorsement system and campaign money, however, parties retain considerable influence on the electorate. Local factions in Taiwan and legislative factions in

Japan, however, tip the balance between candidates and parties in terms of campaign finance and party endorsement. In Japan, the Diet candidates with the backing of the major legislative factions can easily win the LDP nomination. The competition between candidate concerns and party labels, in this view, is inevitable. The extent to which party labels and candidate evaluations shape vote choice depends on the political settings. Only when the relative influence of candidate evaluations and party predisposition can be estimated in different political settings, can we assess how party voting theory works for legislative elections in a common setting.

Chapter 6 provides a cross-country comparison of party identification and candidate evaluations. Two system-level variables—election rules and executive- legislative relationships—are employed to create a typology of the propensity of party voting. It is expected that multi-member districts with presidential systems emphasize the candidate-centered pattern, and that single-member districts with parliamentary systems yields the greatest party voting. Multi-member districts with parliamentary systems and single-member districts with presidential systems lie between these two extremes in terms of party voting. Multi-nomial logit estimates confirm the influence of the system-level variables. The prediction of party voting generated by the coefficients, however, does not fully fit the typology.

279

The strong party-oriented voting pattern in Taiwan is to a large extent responsible for its deviation from the typology. Also, Japan in 1996 shows weak candidate voting for the conservative vote and moderate party voting for the liberal vote.

The U.S. and pre-reform Japan show tendencies of candidate voting, both of which is expected to lie between Japan after 1994 and Taiwan. The pooled data analysis shows that the three-way interaction between candidate evaluations and political institutions is significant across the four cases, which explains why candidate evaluations have an impact stronger than was expected.

All in all, the empirical evidence shows that the level of party voting varies with political institutions. The relative influence between party attachment and candidate evaluations, however, is not entirely consistent with the predictions of the theoretical framework. The results of the pooled data analysis actually conform to what the four individual case studies have found, which indicates that some characteristics of each case are part of the source of the variance that needs to be explained.

7.3 Implications

The social psychological approach postulates that the voting act is a response to various political attitudes, including party loyalty, evaluations of candidates and policy issues. Early family socialization shapes the direction and strength of party loyalty, and policy issues and candidate-related information come from the campaign. Party voting theory suggests that the assessment of candidates is largely related to personal characteristics of the candidates and very campaign-specific. The revisionists instead assert that the competing candidates

280

are actually the focus of campaign (Kelly and Mirer, 1974; Page and Jones, 1979;

Miller and Shanks, 1996). The increasing importance of feelings toward candidates in presidential elections stems from primary elections that give voters power to select their candidates and make candidates recognizable. The media coverage of candidates from primary elections and the party convention through

Election Day reinforces the saliency of perception of the candidates. Since the

1970s, the decline of party images and the increase in independents also coincide with the increasing attention to the candidates.

In legislative elections, the incumbency advantage--such as the franking privilege, transportation reimbursement, and large staffs--facilitates the visibility of incumbents. Although scholars debated whether recognition itself gives voters positive evaluations of the incumbents, they agree that party attachment and candidate evaluations in general account for most of the variations in the voting decision (Hinckley, 1981). Niemi and Weisberg (1993) maintained that it is a controversy to view congressional elections as local events or national elections.

In Chapter 3, it is also found that candidate evaluations are more influential to the voting decision than party attachment in the U S. With the strength of partisanship declining and personal vote increasing, the candidate-centered pattern prevails in the United States. Because of single-member districts and the long history of the two-party system, however, strong partisans still rely on cues provided by parties.

Why do people deviate from the party line and vote for their preferred candidates? Scholars use the social psychological approach to study how candidate evaluations factor into the voting act, arguing that candidate assessments reflect issues positions (Page and Jones, 1979). The great influence of

281

candidate evaluation on the voting decision may also come from people’s expectation that the candidate can win the election (Bartels, 1985). Instead of putting any new variable in the individual voting model, this dissertation argues that the level of party voting is predicated on institutional arrangements. While the variation in party voting is due to the short-term forces, political institutions implicitly constrain party organizations and, as a result, the way people evaluate candidates and parties. Party voting theory can apply to certain institutions, but the candidate-centered pattern prevails in some other settings. In multi-member districts with presidential systems, the candidates are more active in districts because they are faced with intra-party competition and the demands of particular interests. In single-member districts with parliamentary systems, however, the level of party voting is expected to be high because of strong party discipline and single party endorsement.

Due to limitations of data, the behavior of party organizations and the candidates in each political system is not observed in detail.85 Through a quasi- experimental research, however, the relative difference in the effect of party attachment and candidate evaluation can be seen across different contexts. Except for Taiwan, the order of party voting predicted by the typology indeed takes place.

The pooled-data analysis shows that Japan in 1996 has the strongest level of party voting and the United States and Japan in 1976 have moderate levels. Party voting still prevails in Taiwan for its short history of party competition; party organizations still have strong control over candidates and lawmakers. That not only sheds some lights on the trend of party voting in each country that this

282

dissertation covers, but also pushes the study of voting behavior forward. More empirical inquiry is needed to test the typology and the findings; however, the merit of comparative research of political behavior will become apparent in evaluating established theories.

7.4 Limitations

Considering the small number of observations, the findings and conclusions above should be taken with caution. Because only one case is included for each category of the two-by-two typology (see Chapter 2), the findings may not apply to other cases: any substitution of cases might change the results. Moreover, three out of the four cases in the pooled data are from the 1990s, which limits the generalizability of the findings. The selection of cases for this test of party voting model, however, is predicated on the theory concerning party behavior and electoral systems.

More importantly, two of the four cases—Taiwan and Japan after the electoral reform—have just undergone change in party system and electoral system. Compared to the United States and other democracies, these two regimes have not yet run elections enough times to ascertain a long-term voting pattern. In some sense, the observation based on the short history of the institution should be reconsidered a few years later.

At last, the measurement of candidate evaluations and partisanship across the four cases may pose another limitation to this study. Due to the fact that this is secondary data analysis, the way to measure people’s party attachment,

85 Regarding the discussion of how parties and candidates coordinate voters in multi-member

283

perceptions of candidates, issue positions, and other attitudes are not entirely the same in these four cases. Incomparable measurement indeed would undermine the validity of comparative research (Przeworski and Teune, 1970). Even though we can check the validity of the concepts through various tests and assure that they tap into the same attitudes, it is inevitable that different context may have entirely different attitudes. For instance, we can find similar question wording of party identification in these four datasets used in this dissertation, but what “political party” means to the mass public could be very different across these three countries. In the U.S., party attachment refers to a sort of group identification developed in the family. In Taiwan, partisanship means one holds certain belief in a given party or even party membership. Therefore, party strength measured by the set of traditional party identification questions (see Chapter 2) may not exactly depict the extent to which Taiwanese identify with their parties. The solution to this problem lies in an integrated study that includes all of the potential concepts across countries so we can observe their validity at more than one time point.

Besides, it is hard to avoid the cultural problem in comparative research.

For instance, representatives in Taiwan and Japan may get payback in the form of bribery from their constituents, and they are praised for not being corrupted. In the

U.S. context, nevertheless, “being clean” is taken for granted by most voters and representatives. Norpoth (1978) also pointed out that it is difficult to find a equivalent concept of “party identification” in Germany, because German consider party identification as voting preference. Therefore, the language problem due to culture and translation is another limitation of this study.

districts, see Lin (1996).

284

7.5 Ideas for Future Research

One of the drawbacks of this research is the small number of cases. A multicollinearity problem may occur in analysis of pooled-data (Achen, 1982).

Adding more cases to the study would be useful in solving this problem (Gujurati,

1995). According to the two dimensions of political systems, I would suggest adding Britain, Korea, and New Zealand to this study. Britain is a good case in terms of its SMSP system with parliamentary government, and Korea and New

Zealand can be studied for the consequences of changes in their electoral systems in various periods.86 Including more elections from each country in the pooled- data analysis and controlling for the timing of the election may also explain more systematic variance within a given institution and make the results less dependent on the possible idiosyncrasies of particular elections—provided comparable measurements can be located. Including more predictors for each case would provide more complete specification of the voting model. However, the survey data, if available, may not be comparable and including “less mature” earlier elections could yield biased estimates.

One of the assumptions in this dissertation is that people vote sincerely; people have a strict order of preference over the outcome of election as to which party will claim the office and vote accordingly. In some circumstances, however,

86 Korea used single-member-district for the National Assembly elections between 1948 and 1972, and two-member-district system between 1973 and 1988. Since 1988, Korea has used single- member-district system (Lee, 1999). New Zealand had used the single-member-district election system since 1914. In 1993, the people voted to abandon SMD system and adopted a two-tier PR system. The old 99-member parliament was replaced by a 120-member parliament. Among the 120 seats, 65 are elected from single-member districts and 55 from a national list. Like Japanese voters, everyone has two votes, one for the candidate in a single-member district and one for the party list candidate (Gallagher, 1998).

285

people vote against their preferences in order to either break a tie between the two candidates who are likely to win or prevent their least preferred candidate from winning, which is called “strategic voting” (Downs, 1957; Riker and Ordeshook,

1968; Cain, 1978). As mentioned in Chapter 1, voters are engaged in strategic voting before the number of effective candidates reaches an equilibrium. It has been shown that the equilibrium number of candidates is the number of seats plus one (Cox, 1997). 87 Different institutions can lead to different ideas about preference and voting behavior. In simple-majority-single-member districts, voters vote sincerely unless third parties are very competitive at large scale. In multi-member districts with SNTV, however, strategic voting is encouraged among candidates from the same party to maximize the seat share of a given party.

One of the shortcomings of this dissertation is that all of the candidates from the same party in multi-member districts are lumped together due to limitations of data, which may underestimate the influence of candidate evaluation and ignore the influence of expectation over the likelihood to win the election. If we could ask the respondents to estimate the probability that each candidate wins the election and identify individual candidates in the data, we might be able to consider the relationship between voting choice and preference in a comparative perspective.88

87 According to Cox’s game-theoretical analysis, people equalize their votes among the candidates who are expected to end up with M+1 and M+2 or desert the candidates who are expected to run in M+2 or behind. Therefore, M+1 is the equilibrium number of effective candidates. 88 Tsai and Hsu (2002) have discussed how to translate people’s perception toward candidate’s issue position to the likelihood that the candidate wins the election, and found that the expected- utility voting model based on the expectation can account for the voting decision in Taiwan. However, candidates from the same party are still assumed to share the same characteristics due to the availability of data.

286

If not for the compelling theoretical framework of the social psychological approach, the sociological approach may contribute to not only the study of voting behavior and political activity but also to the study of the development of political attitudes. Since the 1940s, scholars have employed the sociological approach to examine how social context determines public opinion and voting behavior

(Berelson, Lazarsfeld, and McPhee, 1954; Lazarsfeld, Berelson, and Gaudet,

1944). In the United States, voters exhibit a weak link between social background and voting behavior because both parties converge in ideology and claim to be a catch-all party. Nevertheless, different electoral rules and political systems give rise to different associations between social cleavages and political attitude.

Scholars have examined European party identification, revealing that social cleavages reinforce the strength of partisanship (Richardson, 1991). To consider the development of partisanship in different countries, it is plausible to view social cleavages as an important factor. This would involve the analysis of socialization and examination of the effect of certain events on cohorts, which will help estimate the effect of partisanship on voting choice.

All in all, studying the voting act involves knowledge of social psychology, micro-economics, and sociology. Political scientists, who relentlessly have investigated political behavior since the 1920s, have borrowed concepts and techniques from social psychology, micro-economics, and sociology and developed three grand approaches that are still to be followed. Granted that we have already seen a great deal of findings generated by the three approaches, it is time to give more consideration to the effect of institutional design on political behavior. With the help of these three approaches, it is no doubt that a

287

generalizable theory of voting is likely to be found in the future. This dissertation, in a sense, brings political institutions back to the study of voting behavior.

Institutional features belong in a generalizable theory of voting.

288

Bibliography

Abramowitz, Alan I., 1980. “A Comparison of Voting for U.S. Senator and Representative in 1978.” American Political Science Review 74: 633-640.

Abramowitz, Alan I., 1984. “National Issues, Strategic Politicians, and Voting Behavior in the 1980 and 1982 Congressional Elections.” American Journal of Political Science, 28:710-721.

Abramowitz, Alan, 1991. “Incumbency, Campaign Spending, and the Decline of Competition in U.S. House Elections.” Journal of Politics, 53: 34-56.

Achen, Christopher, H., 1982. Interpreting and Using Regression. Beverly Hills: Sage Publications.

Aldrich, John H., 1995. Why Parties? Chicago: The University of Chicago Press.

Allsop, Dee, and Herbert F. Weisberg, 1988. “Measuring Change in Party Identification in an Election Campaign.” American Journal of Political Science, 32: 996-1017.

Almond, Gabriel A. and Sidney Verba, 1963. The Civic Culture: Political Attitudes and Democracy in Five Nations. Boston, Mass. :Little, Brown.

Alvarez, R. Michael, and Jonathan Nagler, 1995. “Economics, Issues and the Perot Candidacy: Voter Choice in the 1992 Presidential Election.” American Journal of Political Science, 39:714-744.

Alvarez, R. Michael and Jonathan Nagler, 1998. “When Politics and Models Collide: Estimating Models of Multiparty Elections.” American Journal of Political Science, 42: 55-96.

Arnold, R. Douglas, 1979. Congress and the Bureaucracy. New Haven: Yale University Press.

Asher, Herbert, and Bradley Richardson, 1977. Comparative Voting Behavior. Washington: American Political Science Association.

Baerwald, Hans H. and Nobuo Tomita, 1977. “Japan’s 34th General Election: Cautious Change amidst Incremental Instability.” Asian Survey, 17: 221- 236.

289

Bartels, Larry, 2000. “Partisanship and Voting Behavior, 1952-1996.” American Journal of Political Science, 44: 35-50.

Bartolini, Stefano, and Peter Mair, 2001. “Challenges to Contemporary Political Parties.” In Political Parties and Democracy. Larry Diamond and Richard Gunther Baltimore (eds.), Md. : Johns Hopkins University Press.

Bawn, Kathleen, Gary W. Cox, Frances McCall Rosenbluth, 1999. “Measuring the Ties That Bind: Electoral Cohesiveness in Four Democracies.” in Elections in Japan, Korea, and Taiwan under the Single Non-Transferable Vote. Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall (eds.) Ann Arbor: The University of Michigan.

Beck, Pual A., 1979. “The Electoral Cycle and Patterns of American Politics.” British Journal of Political Science 9:129-156.

Beck, Paul A., 1986. “Choice, Context, and Consequence: Beaten and Unbeaten Paths toward a Science of Electoral Behavior” in Political Science: The Science of Politics. Herbert Weisberg (ed.). New York: Agathon Press.

Belknap, George, and Angus Campbell, 1952. “Political Party Identification and Attitudes Toward Foreign Policy.” Public Opinion Quarterly, 15:601-623.

Berelson,, Bernard, Paul Lazarsfeld, and William McPhee, 1954. Voting. Chicago: University of Chicago Press.

Blaker, Michael, 1995. “Japan in 1994: Out with the Old, in with the New?”Asian Survey, 35: 1-12.

Bloom, Harold, and H. Douglas Price, 1975. “Voter Response to Short Run Economic Conditions: The Asymmetric Effect of Prosperity and Recession.” American Political Science Review 69:1240-54.

Bond, Jon R., Gary Covington, and Richard Fleisher, 1985. “Explaining Challenger Quality in congressional Elections.” Journal of Political Science 41: 510-29.

Born, Richard, 1986. “Strategic Politicians and Unresponsive Voters.” American Political Science Review, 80: 599-612.

Bowler, Shaun, David J. Lanoue, and Paul Savoie, 1994. “Electoral Systems, Party Competition, and Strength of Partisan Attachment: Evidence from Three Countries.” The Journal of Politics, 56, 4: 991-1007.

Brody Richard A. and Benjamin I. Page, 1972. “Comment: The Assessment of Policy Voting.” American Political Science Review 66: 450-458.

Brown, Robert D., James A. Woods, 1991. “Toward a Model of Congressional Elections” The Journal of Politics, 53, 2: 454-473.

290

Bulter, David and Donald Stokes, 1969. Political Change in Britain: Forces Shaping Electoral Choice. New York: St. Martin’s Press.

Burnham, Walter. D., 1965. “The Changing Shape of the American Political Universe.” American Political Science Review, 59: 7-28.

Burnham, Walter Dean, 1970. Critical Elections and the Mainsprings of American Politics. New York: Norton.

Butler, David and Donald Stokes, 1969. Political Change in Britain. New York: St. Martin’s.

Cain, Bruce E., 1978. “Strategic Voting in Britain.” American Political Science Review, 22: 639-655.

Cain, Bruce E. and John Ferejohn, 1981. “Party Identification in the United States and Great Britain.” Comparative Political Studies 14:31-47.

Cain, Bruce, John Ferejohn, Morris Fiorina, 1987. Personal Vote. Cambridge: Harvard University Press.

Campbell, Angus, 1960. “Surge and Decline: A Study of Electoral Change.” Public Opinion Quarterly, 24.

Campbell, Angus, Philip E. Converse, Warren Miller, and Donald Stokes, 1960. The American Voter. Chicago: The University of Chicago Press.

Campbell, Angus, Philip E. Converse, Warren Miller, and Donald Stokes, 1966. Elections and the Political Order. New York: Wiley.

Cassel, Carol, 1999. “Testing the Converse Party Support in Britain.” Comparative Political Studies 32: 626-644.

Chen, Yi-yen and Hwang Li-chiu, 1992. Hsuanchiu hsinwei yu chengchi fachan (Electoral Behavior and Political Development). Taipei: Li-min wenhua Inc.

Christensen, Ray, 2000. Ending the LDP Hegemony: Party Cooperation in Japan. Honolulu: University of Hawaii Press.

Chu, Yun-han, 1993. “Hsuanmin te chengtang hsuangtse“ (Voters’ Party Choices) Paper presented on Voting Behavior, Constitutional Order, and Political Change conference.

Chu, Yun-Han, 1995. “SNTV and the Evolving Party System in Taiwan.” Chinese Political Science Review, 22:

291

Chu, Yun-han, 2001. “The Legacy of One-Party Hegemony in Taiwan.” In Political Parties and Democracy. Larry Diamond and Richard Gunther (eds.) Baltimore, Md. : Johns Hopkins University Press.

Chu, Yun-han and Tse-min Lin, 1996. “The Process of Democratic Consolidation in Taiwan: Social Cleavage, Electoral Competition, and the Emerging Party System.” In Taiwan’s Electoral Politics and Democratic Transition, Tien, Hung-mao (ed.). New York: M.E.Sharpe, Inc.

Claggett, William, William Flanigan, Nancy Zingale, 1984. “Nationalization of the American Electorate” American Political Science Review, 78, 1: 77-91.

Clarke, Harold and Marianne C. Stewart,1987. “Partisan Inconsistency and Partisan Change in Federal States: The Case of Canada.” American Journal of Political Science, 31:383-407.

Converse, Philip E., 1964. “The Nature of Belief Systems in Mass Publics.” In Ideology and Discontent, David E. Apter (ed). New York: Free Press.

Converse, Philip, 1966. “The Concept of a Normal Vote.” in Elections and the Political Order. Angus Campbell, Philip E. Converse, Warren E. Miller and Donald E. Stokes (eds.) John Wiley and Sons, Inc.

Converse, Philip E., 1969. “Of Time and Partisan Stability.” Comparative Political Studies, 2:139-171.

Converse, Philip E., 1976. The Dynamics of Party Support. Beverly Hills: Sage.

Converse, Philip E. and Georges Dupeux, 1962. “Politicization of the Electorate in France and the United States.” Public Opinion Quarterly 26:1-23.

Converse, Philip E., and Gregory B. Markus, 1979. “Plus ca change...: The New CPS Election Study Panel” The American Political Science Review, 73, 1: 32-49.

Converse, Philip E., Warren E. Miller, Jerrold G. Rusk, and Arthur C. Wolfe, 1969. “Continuity and Change in American Politics: Parties and Issues in the 1968 Election.” American Political Science Review, 63: 1083-1105.

Cover, Albert D., 1977. “One Good Term Deserves Another: The Advantage of Incumbency in Congressional Elections.” American Journal of Political Science 21: 523-41.

Cox, Gary W., 1987. The Efficient Secret: the Cabinet and the Development of Political Parties in Victorian England. New York: Cambridge University Press.

292

Cox, Gary W., 1996. “Is the Single Nontransferable Vote Superproportional? Evidence from Japan and Taiwan.” American Journal of Political Science, 40, 3: 740-755.

Cox, Gary W., 1997. Making Votes Count. Cambridge: Cambridge University Press.

Cox, Gary W., and Jonathan N. Katz, 1996. “Why Did the Incumbency Advantage in U.S. House Elections Grow?” American Journal of Political Science, 40, 2:478-497.

Cox, Gary W., and Mathew D. McCubbins, 1993. Legislative Leviathan. Berkeley : University of California Press.

Cox, Gary W., and Emerson Niou, 1994. ''Seat Bonus Under the Single Non- Transferable Vote System: Evidence from Japan and Taiwan.'' Comparative Politics, 26:221-236.

Cox, Gary W., and Frances McCall Rosenbluth, 1993. ''The Electoral Fortunes of Legislative Factions in Japan.'' American Political Science Review, 83: 577-589.

Cox, Gary W., and Frances McCall Rosenbluth, 1994. “Reducing Nomination Errors: Factional Competition and Party Strategy in Japan.” Electoral Studies, 13: 4-16.

Cox, Gary W., and Frances McCall Rosenbluth, 1995. “The Structural Determinants of Electoral Cohesiveness: England, Japan, and the United States.” In Structure and Policy in Japan and the United States. Peter F. Cowhey and Mathew D. McCubbins (eds.) Cambridge: Cambridge University Press.

Curtis, Gerald L., 1988. The Japanese Way of Politics. New York: Columbia University Press.

Curtis, Gerald, 1999. The Logic of Japanese Politics. Columbia University Press.

Dahl, Robert A., 1961. “The Behavioral Approach in Political Science: Epitaph for a Monument to a Successful Protest.” American Political Science Review, 55: 763-772.

Dalton, Russell J., Scott Flanagan, and Paul Allen Beck (eds.), 1984. Electoral Change in Advanced Industrial Democracies: Realignment or Dealignment? Princeton, N.J.: Princeton University Press.

Downs, Anthony, 1957. An Economic Theory of Democracy. New York: Harper and Row.

Duverger, Maurice, 1954. Political Parties. New York: Wiley, Science Ed.

293

Eldersveld, Samuel J., 1964. Political Parties: A Behavioral Analysis. Chicago: Rand McNally.

Eldersveld, Samuel J., 1973. “Party Identification in India in Comparative Perspective.” Comparative Political Studies 6,3:271-295.

Epstein, Leon D., 1967. Political Parties in Western Democracies. Drederick A. Praeger, Inc.

Epstein, Leon D., 1986. Political Parties in the American Mold. Madison: University of Wisconsin Press.

Erikson, Robert S., and Gerald C. Wright, 1993. “Voters, Candidates, and Issues in Congressional Elections.” In Congress: Reconsidered, 5th ed, Lawrence C. Dodd and Bruce I. Oppenheimer (eds.). Washington, D.C.: CQ Press.

Erikson, Robert S., Gerald C. Wright, Jr., and John P. McIver, 1989. “Political Parties, Public Opinion, and State Policy in the United States.” American Political Science Review, 83,3: 729-750.

Feldman, Stanley, 1982. “Economic Self-Interest and Political Behavior.” American Journal of Political Science 26: 446-466.

Feldmand, Stanley, and Alan S. Zuckerman, 1982. “Partisan Attitudes and the Vote: Moving Beyond Party Identification.” Comparative Political Studies 15,2:197-222.

Fiorina, Morris P., 1977.”The Case of the Vanishing Marginals: The Bureaucracy Did It.” American Political Science Review, 71:177-181.

Fiorina, Morris P., 1981. Retrospective Voting in American National Elections. New Haven, Conn.: Yale University Press.

Fiorina, Morris P., 1989. Congress: Keystone of the Washington Establishment. Yale University Press.

Fiorina, Morris P., 1992. Divided Government. New York: Maxwell Macmillan International.

Flanagan, Scott C., 1980. “Value Cleavages, Economic Cleavages and the Japanese Voter.” American Journal of Political Science, 24:177-206.

Flanagan, Scott C., 1991. “Value Cleavages, Contextual Influences, and the Vote.” In The Japanese Voter. Flanigan, Scott, Shinsaku Kohei, Ichiro Miyake, Bradley M. Richardson, and Joji Watanuki (eds.) New Haven: Yale University Press.

Flanigan, Scott, Shinsaku Kohei, Ichiro Miyake, Bradley M. Richardson, and Joji Watanuki, 1991. The Japanese Voter. New Haven: Yale University Press.

294

Flanigan, William H., and Nancy H. Zingale, 1974. “The Measurement of Electoral Change.” Political Methodology, 1: 49-82.

Foster, James, 1982. “Ghost-Hunting: Local Party Organization in Japan.” Asian Survey, 12: 843-857.

Franklin, Charles H., and John E. Jackson, 1983. “The Dynamics of Party Identification” The American Political Science Review, 77: 957-973.

Franklin, Mark, 1996. “Electoral Participation.” In Comparing Democracies: Elections and Voting in Comparative Perspective, Lawrence LeDuc, Richard Niemi, and Pippa Norris (eds.). New Yor: Sage.

Fukui, Haruhiro, 1970. Party in Power: the Japanese Liberal-Democrats and Policy-Making. Berkeley: University of California Press.

Fukui, Haruhiro, and Shigeko N. Fukai, 1999. “Campaigning for the Japanese Diet.” in Elections in Japan, Korea, and Taiwan under the Single Non- Transferable Vote. Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall (eds.) Ann Arbor: The University of Michigan.

Gallagher, Michael, 1998. “The Political Impact of Electoral System Change in Japan and New Zealand, 1996.” Party Politics, 4: 203-228.

Gelman, Andrew, and Gary King, 1990. “Estimating Incumbency Advantage without Bias.” American Journal of Political Science, 34: 1142-1164.

Goldberg, Arthur S., 1966. “Discerning a Causal Pattern among Data on Voting Behavior.” American Political Science Review 60, 4:913-922.

Green, Donald Philip, and Bradley Palmquist, 1990. “Of Artifacts and Partisan Instability.” American Journal of Political Science, 34, 3: 872-902.

Grofman, Bernard, 1999. “SNTV, STV, and Single-Member-District Systems: Theoretical Comparisons and Contrasts.” in Elections in Japan, Korea, and Taiwan under the Single Non-Transferable Vote. Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall (eds.) Ann Arbor: The University of Michigan.

Gujurati, Damodar, N., 1995. Basic Econometric. New York: McGraw Hill, Inc.

Gunther, Richard, and Larry Diamond, 2001. “Types and Functions of Parties.” in Political Parties and Democracy. Larry Diamond and Richard Gunther (eds.) Baltimore, Md.: Johns Hopkins University Press.

Hartwig, Frederick, William R. Jenkins, Earl M. Temchin, 1980. “Variability in Electoral Behavior: The 1960, 1968, and 1976 Elections.” American Journal of Political Science, 24,3:553-558.

295

Herrnson, Paul S., 1998. Congressional Elections: Campaigning at Home and in Washington, 2th ed. Washington D.C.: Congressional Quarterly.

Hinckley, Barbara, 1980. “The American Voter in Congressional Elections.” The American Political Science Review, 74, 3:641-650.

Hinckley, Barbara, 1981. Congressional Elections. Washington, D.C.: Congressional Quarterly Press.

Holmberg, Soren, 1994. “Party Identification Compared across the Atlantic.” In Elections at Home and Abroad. M. Kent Jennings and Thomas E. Mann (eds.) Ann Arbor: the University of Michigan Press.

Howell, Susan E., 1981. “Short Term Forces and Changing Partisanship” Political Behavior, 3, 2: 163-180.

Hrebenar Ronald J., 2000. “Rules of the Game: The Impact of the Electoral System on Political Parties.” In Japan’s New Party System. Ronald J. Hrebenar (ed.) Boulder: Westview Press.

Hsiao, Hsin-Huang Michael, 1992. “The Rise of Social Movements and Civil Protests.” In Political Change in Taiwan, Tun-jen Cheng and Stephan Haggard (eds.). Boulder: Lynne Rienner Publishers, Inc.

Hu, Fu, and You, Ying-lung, 1983. “Hsuan-min te tang-p’ai hsuan-tse“ (Partisan Choice of the Voters: An Analysis of Their Attitudes and Background). Cheng-chih hsueh-pao (Journal of Politics) , 21: 31-53.

Hu, Fu, and Chu, Yun-han, 1992."The Development of New Partisanship in Taiwan's Regime Transition." Paper presented at the 1992 annual meeting of the American Political Science Association, Chicago.

Hwang, Hsiao-Tuan, 1999. “Strategic Voting: Comparing Single-Member District and SNTV System.” Journal of Electoral Study, 1-31.

Hsu, Yung-Ming, 2001, "Making Sense of Political Learning: The Case of Forming Identities in Taiwan," Proceeding of National Science Council (C), 11:.371-389.

Hsu, Yung-ming, and Chen, Hung-chan, 2002. “Laugo hsuae basii: liwei hsuanchiu chengtang timin te yuhsiaohsin.” (An Old Dog Learns New Trick: the Effectiveness of Party Nomination in Legislative Elections) Soochow Journal of Political Science, 15: 87-121.

Hsu, Yung-ming, and Tsai, Chia-hung, 2002. “Splitting and Making Parties: Analysis of Party Reconfiguration in Taiwan.” Presented at the UNT Conference on Parties and Elections in New Democracies, Denton, 2002.

296

Inglehart, Ronald, and Avram Hochstein, 1972. “Alignment and Dealignment of the Electorate in France and the United States.” Comparative Political Studies 5,3:343-372.

Inoguchi, Takashi, 1981. “Explaining and Predicting Japanese General Elections, 1960-1980.” Journal of Japanese Studies, 7:285-318.

Jaccard, James, 2001. Interaction Effects in Logistic Regression. Thousand Oaks: Sage Publications.

Jaccard, James, Robert Turrisi, Choi K. Wan, 1990. Interaction Effects in Multiple Regression. Newbury Park: Sage Publications.

Jackson, John E., 1975. “Issues, Party Choices, and Presidential Votes.” American Journal of Political Science, 19: 161-185.

Jacobson, Gary C., 1990. The Electoral Origins of Divided Government: Competition in U.S. House Elections, 1946-1988. Boulder: Westview Press.

Jacobson, Gary C., 1994. The Politics of Congressional Elections. New York: Longman.

Jacobson, Gary C., 1997. The Politics of Congressional Elections. New York: Longman, Inc.

Jacobson, Gary C., 2001. The Politics of Congressional Elections. New York: Longman, Inc.

Jacobson, Gary C., and Samuel Kernell, 1983. Strategy and Choice in Congressional Elections. New Haven: Yale University Press.

Jennings, M. Kent, and Gregory B. Markus, 1984. “Partisan Orientations over the Long Haul: Results from the Three-Wave Political Socialization Panel Study “ American Political Science Review, 78: 1000-1018.

Kaase, Max, and Hans-Dieter Klingemann, 1994. “The Cumbersome Way to Partisan Orientations in a ‘New’ Democracy: The Case of the Former GDR.” in M. Kent Jennings and Thomas E. Mann (eds.) Elections at Home and Abroad. Ann Arbor: The University of Michigan Press.

Kaase, Max, 1976. “Party Identification and Voting Behavior in the West German Election of 1969.” In Party Identification and Beyond, ed. Ian Budge, Ivor Crewe and Dennis Farlie. London: Wiley.

Kabashima, Ikuo, 1986. “Yukensha no ideorogi”. In Nihonjin no sekyo kodo. Joji Watanuki, Ichiro Miyake, Takashi Inoguchi, and Ikuo Kabashima (eds.). Tokyo: Tokyo Daigaku Shuppankai.

297

Kabashima, Ikuo, and Yoshito Ishio, 1998. “The Instability of Party Identification Among Elibible Japanese Voters: A Seven-wave Panel Study, 1993-6.” Party Politics, 4,2:151-176.

Kalyvas, Stathis N., 1996. The Rise of Christian Democracy in Europe. Cornell University Press.

Katz, Richard S., 1997. Democracy and Elections. New York: Oxford University Press.

Kawato, Sadafumi, 1987. “Nationalization and Partisan Realignment in Congressional Elections.” American Political Science Review, 81: 1235- 1250.

Kelley, Stanley, Jr. and Thad W. Mirer, 1974. “The Simple Act of Voting.” American Political Science Review, 68: 572-591.

Kernell, Samuell, 1977. “Presidential Popularity and Negative Voting: An Alternative Explanation of the Midterm Congressional Decline of the President’s Party.” American Political Science Review 71: 44-66.

Key, V.O., Jr., 1942. Politics, Parties, and Pressure Groups. New York, Crowell.

Key, V. O., Jr., 1955. “A Theory of Critical Elections.” The Journal of Politics, 17: 3-18.

Kinder, Donald and D. Roderick Kiewiet, 1981. “Sociotropic Politics: The American Case.” British Journal of Political Science,11:129-161.

King, Gary and Andrew Gelman, 1991. “Systematic Consequences of Incumbency Advantage in U.S. House Elections.” American Journal of Political Science, 35:110-38,.

Klecka, William R., 1980. Discriminant Analysis. Beverly Hills: Sage Publications.

Klingemann, Hans-Dieter, and Martin P. Wattenberg, 1992. “Decaying versus Developing Party Systems: A Comparison of Party Images in the United States and West Germany.” British Journal of Political Science, 22:131- 149.

Kohei, Shisaku, Ichiro Miyake, and Joji Watanuki, 1991. “Issues and Voting Behavior.” In The Japanese Voter. Flanigan, Scott, Shinsaku Kohei, Ichiro Miyake, Bradley M. Richardson, and Joji Watanuki (eds.) New Haven: Yale University Press.

Kohn, Melvin L.,1987. “Cross-National Research as an Analytic Strategy: American Sociological Association, 1987 Presidential Address.” American Sociological Review, 52, 6:713-731.

298

Kohno, Masaru, 1997. “Voter Turnout and Strategic Ticket-Splitting Under Japan’s New Electoral Rules.” Asian Survey, 37:429-440.

Kramer, Gerald, 1971. “Short-Term Fluctuations in U.S. Voting Behavior.” American Political Science Review, 65:131-143.

Kramer, Gerald, 1983. “The Ecological Fallacy Revisited: Aggregate- versus Individual-Level Findings on Economics and Elections and Sociotropic Voting.” American Political Science Review 77:92-111.

Krasno, Jonathan S., and Donald Philip Green, 1988. “Preempting Quality Challengers in House Elections.” Journal of Political Science 50: 920-36.

Lacy, Dean, and Barry C. Burden, 1999. “The Vote-Stealing and Turnout Effects of Ross Perot in the 1992 U.S. Presidential Election.” American Journal of Political Science 43: 233-255.

Lazarsfeld, Paul F., Bernard Berelson and Hazel Gaudet, 1944. The People's Choice. New York: Duell, Sloan and Pearce.

LeBlanc, Hugh L., and Mary Beth Merrin, 1978. “Parties, Issues, and Candidates: Another Look at Responsible Parties.” The Western Political Quarterly, 31, 4:523-534.

LeDuc, Lawrence, 1985. “Partisan Change and Dealignment in Canada, Great Britain, and the United States.” Comparative Politics, 17,4:379-398.

LeDuc, Lawrence, Harold D. Clarke, Jane Jenson, Jon H. Pammett, 1980. “Partisanship, Voting Behavior, and Election Outcomes in Canada.” Comparative Politics, 12, 4: 401-417.

Lee, Chae-jin, 1971. “Socio-Economic Conditions and Party Politics in Japan: A Statistical Analysis of the 1969 General Election.” Journal of Politics, 33:158-179.

Lee, Sung-Chull, 1999. “Politics of Electoral Reforms and Practices: The Case of Korean SNTV Elections under the Yushin Constitutions.” In Elections in Japan, Korea, and Taiwan under the Single Non-Transferable Vote. Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall (eds.). Ann Arbor: The University of Michigan Press.

Lijphart, Arend, 1984. Democracies. New Haven: Yale University Press.

Lin, Chia-lung, 1989. “Opposition Movement in Taiwan under an Authoritarian- Clientelist Regime: A Political Explanation for the Social Base of the DPP.” Taiwan: A Radical Quarterly in Social Studies, Spring: 117-143.

Lin, Chia-lung, 1998. “Ti-fang hsuan-chu yu Kuomingtang cheng-chuan te shih- chung hwa“ (Local Election and Market-Making of Kuomingtang Regime)

299

in Lian-ang chi-tseng hsuan-chu yu cheng-chi bian-chien (Cross-Strait Local Election and Change of Political Society). Chen Ming-tong and Cheng Yung-nen (eds). Taipei: Yeh-Tan Publishers, Inc.

Lin, Chia-lung, 2000. “Taiwan’s Democratization and Party System Change: Electoral Linkage of Elite and Mass” Taiwan Political Science Review ,4:3-53.

Lin, Jih-wen, 1996. “Consequence of the Single Nontransferable Voting Rule: Comparing the Japan and Taiwan Experiences.” Ph.D. dissertation, University of California at Los Angles.

Lipset, Seymour Martin, and Stein Rokkan, 1967. Party Systems and Voter Alignments. New York: Free Press.

Liu, I-chou, 1996. “The Behavior of Taiwanese Voters in 1992: Consolidation of Partisan Ties.” In Taiwan's Electoral Politics and Democratic Transition: Riding the Third Wave. Hung-mao Tien (ed.) N.Y.: M.E. Sharpe.

Lowi, Theodore, 1979. The End of Liberalism. New York: Norton. Long, J. Scott, 1997. Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks: Sage Publications.

Long, J. Scott, 1997. Regression Models for Categorical and Limited Dependent Variables. Thousand Oaks: Sage Publications.

Lu, Alexander Ya-Li, 1992. “Political Opposition in Taiwan: The Development of the Democratic Progressive Party.” In Political Change in Taiwan,. Cheng Tun-jen and Stephan Haggard (eds). Boulder: Lynne Rienner Publishers, Inc.

MacDonald, Stuart Elaine, James W. Prothro, George Rabinowitz, Keith J. Brown, 1988.”Political Evocation and Styles of Candidate Evaluation.” Political Behavior, 10: 117-135.

McKelvey, Richard D. and Peter C. Ordeshook. 1972. “A General Theory of the Calculus of Voting.” In Mathematical Applications in Political Science, VI. J. F. Herndon and J. L. Bernd (eds.)Charlottesville: University Press of Virginia, pp. 32-78.

MacKuen, Michael B., Robert S. Erikson, James A. Stimson, 1992. “Peasants or Bankers? The American Electorate and the U.S. Economy.” American Political Science Review, 86:597-611.

Mainwaring, Scott, 1997. “Multipartism, Robust Federalism, and Presidentialism in Brazil.” In Presidentialism and Democracy in Latin America. Scott Mainwaring and Matthew Soberg Shugart (eds.) New York : Cambridge University Press.

300

Mair, Peter, 1997. Party System Change: Approaches and Interpretations. New York: Oxford University Press.

Mair, Peter, and Tomokazu Sakano, 1998. “Japanese Political Realignment in Perspective: Change or Restoration?” Party Politics, 4:177-202.

Manabe, Kazufumi, 1983. Yoron to masu komyunikeshon: sono riron to chosa. Tokyo: Keio Tsushin.

Mann, Thomas E. and Raymond E. Wolfinger, 1980. “Candidates and Parties in Congressional Elections.” American Political Science Review, 74: 617-32.

Maor, Moshe, 1997. Political Parties and Party Systems: Comparative Approaches and the British Experience. New York: Routledge.

Markus, Gregory B., and Philip E. Converse, 1979. “A Dynamic Simultaneous Equation Model of Electoral Choice” The American Political Science Review, 73, 4: 1055-1070.

Markus, Gregory, and Philip E. Converse, 1979. “A Model of Electoral Choice.” American Political Science Review, 73:1055-70.

Matthews, Donald R. and James W. Prothro, 1966. “The Concept of Party Image and Its Importance for the Southern Electorate.” In The Electoral Process, M. Kent Jennings and L. Harmon Zeigler (eds.) N.J.:Prentice-Hall.

Mayhew, David, 1973. “Congressional Elections: The Case of the Vanishing Marginals.” Policy 6: 295-318.

Mayhew, David, R., 1974. Congress: The Electoral Connection. Yale University Press.

McAllister, Ian and Anthony Mughan, 1987. “Class, Attitudes, and Electoral Politics in Britain, 1974-1983.” Comparative Political Studies, 20:47-71.

McCubbins, Mathew D., and Frances McCall Rosenbluth, 1995. “Party Provision for Personal Politics: Dividing the Vote in Japan.” In Structure and Policy in Japan and the United States. Peter F. Cowhey and Mathew D. McCubbins (eds.) Cambridge: Cambridge University Press.

McKean, Margaret, and Ethan Scheiner, 2000. “Japan’s New Electoral System: la plus ca change…” Electoral Studies,19: 447-477.

Michels, Robert, 1962. Political Parties. New Brunswick: Transaction Publishers.

Miller, Warren E., and J. Merrill Shanks, 1996. The New American Voter. Cambridge: Harvard University Press.

301

Miller, Arthur H. and Martin P. Wattenberg, 1983. “Throwing the Rascals Out: Policy and Performance Evaluations of Presidential Candidates, 1952- 1980.” American Political Science Review, 77: 359-372.

Miller, Warren E., 1976. “The Cross-National Use of Party Identification as a Stimulus to Political Inquiry.” In Party Identification and Beyond. Ian Budge, Ivor Crewe, and Dennis Farlie (eds.) London: Wiley.

Miyake, Ichiro, 1985. Seito shiji no bunseki. Tokyo: Sobunsha.

Miyake, Ichiro, 1989. Tohyo kodo. Tokyo: Tokyo Daigaku Shuppankai.

Miyake, Ichiro, 1991. “Types of Partisanship, Partisan Attitudes, and Voting Choices.” In The Japanese Voter. Flanigan, Scott, Shinsaku Kohei, Ichiro Miyake, Bradley M. Richardson, and Joji Watanuki (eds.) New Haven: Yale University Press.

Miyake, Ichiro, 1999. “Candidate Evaluation and Voting Choice under the Japanese Electoral System.” In Elections in Japan, Korea, and Taiwan under the Single Non-Transferable Vote. Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall (eds.) Ann Arbor: The University of Michigan.

Myerson, Roger, 1993. “Incentives to Cultivate Favored Minorities under Alternative Electoral Systems.” American Political Science Review, 87: 856-869.

Naikasusouridaijin, 1990. Zenkoku Seron Chosa no Genkyo. Tokyo: Naikasusouridaijin Shingishitsu.

Nie, Norman H., Sidney Verba, John R. Petrocik, 1976. The Changing American Voter. Cambridge, Mass.: Harvard University Press.

Niemi, Richard G., and Herbert F. Weisberg, 1993. Classics in Voting Behavior. D.C.: Congressional Quarterly.

Niemi, Richard G., and M. Kent Jennings, 1991. “Issues and Inheritance in the Formation of Party Identification.” American Journal of Political Science, 35: 970-988.

Niemi, Richard G., G. Bingham Powell, Jr., Harold W. Stanley, and C. Lawrence Evans, 1985. “Testing the Converse Partisanship Model with New Electorate.” Comparative Political Studies, 18: 300-322.

Niemi, Richard, C., and Herbert F. Weisberg, 1993. Classics in Voting Behavior. Washington, D.C.: Congressional Quarterly Press.

Norpoth, Helmut, 1978. “Party Identification in West Germany: Tracing an Elusive Concept.” Comparative Political Studies 11: 36-61.

302

Page, Benjamin I., and Calvin C. Jones, 1979. “Reciprocal Effects of Policy Preferences, Party Loyalties and the Vote.” American Political Science Review, 73: 1071-1089.

Page, Benjamin I., and Richard A. Brody, 1972. “Policy Voting and the Electoral Process: The Vietnam War Issue.” American Political Science Review, 66: 979-995.

Panebianco, Angelo, 1988. Political Parties: Organization and Power. New York: Cambridge University Press.

Pempel, T. J., 1990. Uncommon Democracies: The One-Party Dominant Regimes. Ithaca: Cornell University Press.

Pomper, Gerald M., 1975. Voters' Choice: Varieties of American Electoral Behavior. New York: Harper & Row.

Popkin, Samuel L., 1991. The Reasoning Voter. Chicago : University of Chicago Press.

Powell, G. Bingham, Jr., 1986. “American Voter Turnout in Comparative Perspective.” American Political Science Review, 80: 17-43.

Przeworski, Adam and Henry Teune, 1970. The Logic of Comparative Social Inquiry. New York: Wiley-Interscience.

Rabinowitz, George, James W. Prothro, William Jacoby, 1982. “Salience as a Factor in the Impact of Issues on Candidate Evaluation.” Journal of Politics, 44: 41-63.

Rae, Douglas W., 1971. The Political Consequences of Electoral Laws. New Haven: Yale University Press.

Rapoport, Ronald B., 1997. “Partisanship Change in a Candidate-Centered Era.” Journal of Politics, 59:185-199.

Reed, Steven, R., 1991. “Structure and Behavior: Extending Duverger’s Law to the Japanese Case.” British Journal of Political Science, 20: 335-356.

Reed, Steven R., and John M. Bolland, 1999. “The Fragmentation Effect of SNTV in Japan.” In Elections in Japan, Korea, and Taiwan under the Single Non- Transferable Vote. Bernard Grofman, Sung-Chull Lee, Edwin A. Winckler, and Brian Woodall (eds.). Ann Arbor: The University of Michigan Press.

RePass, David E., 1971. “Issue Salience and Party Choice.” American Political Science Review, 65: 389-400.

Richardson, Bradley M., 1967. “Japanese Local Politics: Support Mobilization and Leadership Styles.” Asian Surveys, 7: 860-875.

303

Richardson, Bradley M., 1975. “Party Loyalties and Party Silence in Japan.” Comparative Political Studies, 8: 32-57.

Richardson, Bradley M., 1977. “Stability and Change in Japanese Electoral Behavior.” Journal of Asian Studies, 35:675-93.

Richardson, Bradley M., 1986. “Japan’s Habitual Voters: Partisanship on the Emotional Periphery.” Comparative Political Studies, 19: 675-93.

Richardson, Bradley M., 1988. “Constituency Candidates vs. Parties in Japanese Voting Behavior.” American Political Science Review, 82: 705-712.

Richardson, Bradley M., 1991. “European Party Loyalties Revisited.” American Political Science Review, 85:751-775.

Richardson, Bradley M., 1997. Japanese Democracy. New Haven: Yale University Press.

Richardson, Bradley, and Scott Flanagan, 1984. Politics in Japan. Boston: Little, Brown.

Riker, William H., 1976. “The Number of Political Parties: A Reexamination of Duverger's Law.” Comparative Politics, 9:93-106.

Riker, William H., 1982. “The Two-Party System and Duverger's Law: An Essay on the History of Political Science.” American Political Science Review, 76: 753-766.

Riker, William H., and Peter Ordeshook, 1968. “A Theory of the Calculus of Voting.” American Political Science Review, 62:25-42.

Rochon, Thomas R., 1981. “Electoral Systems and the Basis of the Vote: The Case of Japan.” In Parties, Candidates, and Voters in Japan: Six Quantitative Studies. John C. Campbell (ed.). Ann Arbor: Center for Japanese Studies.

Rohde, David W., 1991. Parties and Leaders in the Postreform House. The University of Chicago Press.

Rokkan, Stein, 1996. “Cross-Cultural, Cross-Societal, and Cross-National Research.” In Comparing Nations and Cultures, Alex Inkeles and Masamichi Sasaki (eds.) New Jersey: Prentice Hall.

Rose, Richard, 1974. Politics in England. Boston: Little Brown.

Samuels, David J., 1999. “Incentives to Cultivate a Party Vote in Candidate- Centric Electoral Systems: Evidence from Brazil.” Comparative Political Studies , 32: 487-518.

304

Samuels, David J., 2002. “Presidentialized Parties: The Separation of Powers and Party Organization and Behavior.” Comparative Political Studies, 35: 461-483.

Sanders, Arthur, 1988. “The Meaning of Party Images.” The Western Political Quarterly, 41:583-599.

Sartori, Giovanni, 1976. Parties and Party Systems: A Framework for Analysis. Cambridge: Cambridge University Press.

Scalapino, Robert A., and Junnosuke Masumi, 1962. Parties and Politics in Contemporary Japan. Berkerly: California University Press.

Schattschneider, E. E., 1942. Party Government . New York, Holt, Rinehart and Winston.

Schattschneider, E.E., 1960. The Semisovereign People. New York: Longman.

Schickler, Eric, and Donald Philip Green, 1997. “The Stability of Party Identification in Western Democracies: Results from Eight Panel Surveys.” Comparative Political Studies, 30: 450-483.

Schulman, Mark A. and Gerald M. Pomper, 1975. “Variability in Electoral Behavior: Longitudinal Perspectives from Causal Modeling.” American Journal of Political Science, 19: 1-18.

Shaffer, William R., 1972. “Partisan Loyalty and the Perceptions of Party, Candidates and Issues.” The Western Political Quarterly, 25: 424-433.

Shvetsova, Olga, 1995. Electoral Institutional Design. Ph.D dissertation, California Institution of Technology.

Shyu, Huo-yan, 1996. “Kuo-chia jen-tung ho cheng-tang tou-piao hsuan- tse”(National Identity and Partisan Vote-Choices in Taiwan: Evidence from Survey Data between 1991 and 1993) Taiwanese Political Science Review ,1: 85-128.

Stockwin, J.A.A., 2000. “The Social Democratic Party (Formerly Japan Socialist Party): A Turbulent Odyssey.” In Japan’s New Party System. Ronald J. Hrebenar. Boulder: Westview Press.

Stokes, Donald E., Angus Campbell, and Warren E. Miller, 1958. “Components of Electoral Decision.” American Political Science Review, 52: 367-387.

Stokes, Donald E., 1965. “A Variance Component Model of Political Effect.” In Mathematical Applications in Political Science. John M. Claunch (ed.). Dallas: The Arnold Foundation.

305

Stokes, Donald E., 1966. “Some Dynamic Elements in Contests for the Presidency.” American Political Science Review, 60: 19-28.

Stokes, Donald E., 1967. “Parties and the Nationalization of Electoral Forces.” In The American Party Systems: Stages of Political Development. W. N. Chambers and W. D. Burnham (eds.). New York: Oxford University Press.

Taagepera, Rein, and Matthew Soberg Shugart, 1989. Seats and Votes: The Effects and Determinants of Electoral Systems. New Haven: Yale University Press. Watanuki, Joji, 1991. “Social Structure and Voting Behavior.” In The Japanese Voter. Scott C. Flanigan, Shinsaku Kohei, Ichiro Miyake, Bradley M. Richardson, and Joji Watanuki (eds.). New Haven: Yale University Press.

Thomassen, Jacques, 1976. “Party Identification as a Cross-National Concept: Its Meaning in the Netherlands.” In Party Identification and Beyond, Ian Budge, Ivor Crewe and Dennis Farlie (eds.) London: Wiley & Sons.

Tomita, Nobuo, Akira Nakamura, and Ronald J. Hrebenar, 1986. “The Liberal Democratic Party: The Ruling Party of Japan.”In The Japanese Party System: From One Party Rule to Coalition Government. Ronald J. Hrebenar (ed.) Boulder: Westview.

Trilling, Richard, J., 1976. Party Image and Electoral Behavior. New York: John Wiley & Sons.

Tsai, Chia-hung and Hsu, Yung-ming, 2002. “A Multi-Party Expected Utility Voting Model.” Presented in the 2001 Election and Democratization Survey Conference, Taipei.

Tsai, Chia-Hung, 2002. “'Policy-Making and Candidate Coordination: Toward a Party-Based Model”, working paper.

Tufte, Edward R., 1975. “Determinants of the Outcomes of Midterm Congressional Elections.” American Political Science Review, 69: 812-826.

Uslaner, Eric M. and M. Margaret Conway, 1985. “The Responsible Congressional Electorate: Watergate, the Economy, and Vote Choice in 1974.” American Political Science Review, 79: 788-803.

Verba, Sidney and Lucian W. Pye, 1978. The Citizen and Politics: A Comparative Perspective. Stamford: Greylock Publishers.

Wald, Kenneth, 1982. “Stratification and Voting Behavior: Electoral Cleavage in Britain under the Third Reform Act.” Comparative Political Studies, 15:57-83.

306

Wolinetz, Steven, B., 1990. “The Transformation of Western European Party System.” In The West European Party Systems. Peter Mair (ed.) Cambridge: Oxford University Press.

Wang, Cheng-Hwan, 1989. “Taiwan Cheng-chi chuang-hsin ho fan-dweh yun- tang”(Political Transition and Opposition Movement in Taiwan) Taiwan Social Radical Review ,2, 1: 71-116.

Wang, Fu-chang, 1994. “Tsu-chun tung-hua yu tung-yuen: Taiwan min-chung cheng-tang chih-chi chih fen-hsi”(Ethnic Assimilation and Mobilization: an Analysis of Party Support in Taiwan) Bulletin of the Institute of Ethnology, Academia Sinica , 77: 1-34. Taipei, Academia Sinica.

Watanuki, Joji, 1967. “Patterns of Politics in Present-Day Japan.” In Party Systems and Voter Alignments. Seymour M. Lipset and Stein Rokkan (eds.) New York: Free Press.

Watanuki, Joji, 1986. “Senkyo doin to kohosha yoin.” In Nihonjin no sekyo kodo. Joji Watanuki, Ichiro Miyake, Takashi Inoguchi, and Ikuo Kabashima (eds.) Tokyo: Tokyo Daigaku Shuppankai.

Watanuki, Joji, 1991. “Social Structure and Voting Behavior.” In The Japanese Voter. Flanigan, Scott, Shinsaku Kohei, Ichiro Miyake, Bradley M. Richardson, and Joji Watanuki (eds.) New Haven: Yale University Press.

Wattenberg, Martin P., 1984. The Decline of American Political Parties, 1952- 1984. Cambridge: Harvard University Press.

Wattenberg, Martin, P., 1982. “Party Identification and Party Images: A Comparison of Britain, Canada, Australia, and the United States.” Comparative Politics, 15: 23-40.

Weisberg, Herbert F., 1980. “A Multidimensional Conceptualization of Party Identification.” Political Behavior, 2:33-60.

Weisberg, Herbert F. 1998. “The Political Psychology of Party Identification.” Paper prepared for session in honor of Philip E. Converse at the Annual Meeting of the American Political Science Association.

Weisberg, Herbert F., and Jerrold G. Rusk, 1970. “Dimensions of Candidate Evaluation.” American Political Science Review, 64: 1167-1185.

Whiteley, Paul F., 1988. “The Causal Relationships between Issues, Candidate Evaluations, Party Identification, and Vote Choice--the View From "Rolling Thunder.” Journal of Politics, 50: 961-984.

Wright, John R. and Richard G. Niemi, 1983. “Perceptions of Candidates' Issue Positions.” Political Behavior, 5: 209-223.

307

Wu, Jaushieh Joseph, 1995. Taiwan’s Democratization: Forces behind the New Momentum. Hong Kong: Oxford University Press.

Wu, Neiteh, 1996. “Tzu-yu chu-yi, chu-chuen jen-tang ho Taiwane ming-chu chu- yi” (Liberalism, Ethnic Identity and Taiwanese Nationalism) Taiwanese Political Science Review ,1: 5-39.

Yiannakis, Diana E., 1981. “The Grateful Electorate: Casework and Congressional Elections.” American Journal of Political Science 25:568- 80.

You, Ying-lung, 1994. “Cheng-tang hsing-hsiang, yishihsingtai yu Taiwan hsuan- chu pien-chien” (Party Image, Ideology, and Secular Realignment in Taiwan) Bulletin of the Institute of Ethnology, Academia Sinica ,78: 61-99. Taipei, Academia Sinica.

You, Ying-lung, 1996. Ming-yi yu Taiwan cheng-chi bian-chien (Public Opinion and Changing Taiwan Politics) Taipei: Yue-Tan Press.

Zielinski, Jakub. 2000. On Political Parties in New Democracies. Ph.D. Dissertation, The University of Chicago.

308