1 Non-transferable vote
Party switching – the practice among legislators of changing affiliation from one party to another – merits both empirical and normative studies (Laver and Benoit, Reference Laver and Benoit2003: 215). Normatively, in modern democracies, legislators typically join a political party because it offers the possibility of structured collective action with copartisans, brand names, information shortcuts, and political accountabilities that voters can identify. Frequent party switching influences party function and indicates weak and uneven party loyalty and weak institutionalization of party systems (Ames, Reference Ames1995a, Reference Amesb, Reference Ames2002; Desposato, Reference Desposato1997, Reference Desposato2006a, b; Mainwaring, Reference Mainwaring1998). When a party fails to control its legislators, its platform, policies, and organization are threatened with crisis.Footnote 1 Empirically, party switching is a common political phenomenon. It not only happens in new democracies, such as Russia (White et al., Reference White, Rose and McAllister1997), Hungary (Ágh, Reference Ágh, Bowler and Katz1999), and Poland, but occurs in democracies undergoing political transition, such as the Philippines, Taiwan, India (Kamath, Reference Kamath1985), and Spain (Mershon and Heller, Reference Mershon and Heller2003; Sánchez de Dios, Reference Sánchez de Dios, Bowler, Farrell and Katz1999), and in consolidated democracies such as Japan (Laver and Kato, Reference Laver and Kato2001; Kato, Reference Kato1998; Reed and Scheiner, Reference Reed2003) and the United States (Nokken, Reference Nokken2000). The primary task of political scientists is to answer the following questions: Why do legislators defect from their parties? and What kinds of legislators are more likely to switch their party affiliation?
The literature on party switching and party organization includes a proliferation of studies on the relationship between legislators’ re-election incentives and party cohesion. Scholars generally agree that legislators switch party affiliation in order to maximize their re-election chances, and that the decision to switch is the result of rational calculus on the part of legislators. Based on the rational-choice premise, some scholars assert that legislators defect from their party because they are so popular that they do not have to rely on a party label for re-election, while others claim that legislators switch party affiliation in order to secure re-election. In this article, we investigate the two theoretical hypotheses from the perspective of legislator popularity and find that the two assertions actually contradict each other. More surprisingly, we find that the two contradictory incentives above cohabit within the single non-transferable vote (SNTV) system due to its special electoral configuration.
The remainder of this article is organized as follows. Section 1 introduces the two rational-choice explanations for party switching: transaction cost theory and risk-averse theory. We then formalize the arguments and prove that the two explanations contradict each other. In the second section, the authors briefly introduce the characteristics of SNTV and assert that due to the system's electoral setup, the two contradictory motivations for party switching coexist in the multimember district. Section 3 describes the methodology employed, including a description of the data, the measures used, and the empirical strategy. We apply a rare-events logit model to analyze party switching data from the second to the sixth Legislative Yuan in Taiwan.Footnote 2Section 4 reports the results of the empirical tests. As predicted, the results support our assumptions and show that it was not only highly popular legislators who defected from their parties; members who had only barely achieved election also switched party affiliation in order to maximize their chances of re-election. Section 5 suggests further research and draws conclusions.
2 Re-election incentives and switching
Scholars attempting to understand legislators’ behaviour base their theories on electoral incentives. They assume that all legislative members have a primary interest in getting re-elected (Mayhew, Reference Mayhew1974: 16; Mershon and Shvetsova, Reference Mershon and Shvetsova2013). In order to maximize their re-election probability legislators surrender part of their autonomy by forming political parties. Once a party is established, to maximize its seat share and political power, party leaders persuade and mobilize members of the public to support the re-election of its legislators. In short, re-election is in the interest of both party leaders and legislators, and legislators’ chances of re-election are affected by their own personal traits and the reputations of their parties (Cox and McCubbins, Reference Cox and McCubbins1993: 109–10).
2.1 Transaction cost and switching
As scholars of comparative politics assert, the extent to which a legislator has to rely on party support for re-election determines the relationship between him/her and the party leader. Suppose that the legislator has to campaign under the party's label and rely on its popularity. Without the need to appeal to the electorate, the legislator has no incentive to break ranks with the party line. Instead, because voters mainly vote on the basis of party label and the legislator only has limited connections with his constituency, party leaders can exercise ballot control over nomination to hold the legislator in the party (Carey and Shugart Reference Carey and Shugart1995; Heller and Mershon, Reference Heller and Mershon2005; Heller and Mershon, Reference Heller and Mershon2009: 5). Thus, the legislator yields, complies with party discipline, and is less likely to transgress the party line. In contrast, if the legislator's chances of re-election improve as a result of being personally well-known and liked by the electorate, he will chiefly care about his personal connections with the voters. In this case, the legislator will not only focus on providing constituency services, but will also logroll other legislators’ projects in order to bring more ‘pork’ back to his district (Bradbury and Crain, Reference Bradbury and Crain2001). Since voters mainly vote on the basis of the legislator's personal reputation instead of party affiliation, the leaders have to bank on the legislator's popularity and are less able to rein him/her in. In short, the base of electoral support from the voters he has courted can also encourage a legislator to switch party affiliation (O'Brien and Shomer, Reference O'Brien and Shomer2013: 119).
The discussion above highlights how legislators’ personal popularity influences their interactions with party leaders and can be linked with the transaction cost of party switching. Scholars assert that switching often carries an electoral cost (Mershon and Heller, Reference Mershon and Heller2003; Heller and Mershon, Reference Heller and Mershon2009). As stated above, a legislator with party affiliation is elected on the basis of a mixture of personal reputation and party support. When the legislator decides to reject party discipline and defect from his party, he cannot be certain of the result of his defection and may be seen as implicitly gambling with his career (Yoshinaka, Reference Yoshinaka2003). If the legislator perceives his personal reputation as outweighing the transaction cost – namely, that he can win re-election without relying on the party brand – he is more likely to switch. If, on the other hand, he believes that party endorsement is indispensable for his re-election, and that switching might endanger his seat, he will remain under the party's banner. For instance, in Italy and Brazil, an open-list proportional representation system (OLPR) encouraged personal voting and fostered candidate reputations and patron–client relationships (Ames, Reference Ames1995a, Reference Amesb; Chang, Reference Chang2005; Chang and Golden, Reference Chang and Golden2006; Katz, Reference Katz, Shugart and Wattenberg2001; O'Brien and Shomer, Reference O'Brien and Shomer2013: 119), resulting in frequent switching in the two countries (Mershon and Heller, Reference Mershon and Heller2003; Mainwaring and Scully, Reference Mainwaring and Scully1995; Ames, Reference Ames2002). In the 49th and 50th Brazilian Chamber of Deputies, legislators’ switching rates were 51% and 41%, respectively (Desposato, Reference Desposato2004: 6).
2.2 Risk-averse theory and switching
Other scholars have investigated the same research topic from the perspective of party support. Although one function of a party is to enhance legislators’ chances of re-election, poor performance by the government, declining electoral support for their party, and fierce competition in party primaries are all factors that might threaten legislators’ re-election and drive them to switch parties. A number of studies have found that a government's poor economic performance drives the electorate to vote against the incumbent (Feldman, Reference Feldman1984; Fiorina, Reference Fiorina1978; Kiewiet and Rivers, Reference Kiewiet and Rivers1984; Markus, Reference Markus1988; Lewis-Beck, Reference Lewis-Beck1988; Powell and Whitten, Reference Powell and Whitten1993). In this circumstance, once they perceive that social grievances might endanger their seats, legislators will switch parties to secure their re-election. Mershon and Heller (Reference Mershon and Heller2003) investigate party switching in the Spanish Congress of Deputies between 1982 and 1996 and conclude that defecting deputies were motivated by the desire to escape a bad situation that jeopardized their re-election. In Brazil, between 1986 and 1990, the poor performance of the incumbent Party of the Brazilian Democratic Movement (Partido do Movimento Democrático Brasileiro, PMDB) encouraged legislators to switch parties in order to enhance their re-election chances (Ames, Reference Ames2002: 71). Similarly, Zielinski et al. (Reference Zielinski, Slomczynski and Shabad2005) investigate defection in the Polish Sejm between the 1991 and 2001 elections and find that when the economic performance of the government was disappointing, deputies who defected from the dominant party had a higher chance of re-election than those who did not switch.Footnote 3
2.3 Contradiction between two theories
The discussion above highlights two incentives for party switching. On the one hand, transaction cost theory indicates that legislators are more likely to switch party affiliation if their personal reputation can enable them to win re-election without relying on a party label. On the other hand, the risk-averse theory finds that legislators are more likely to defect from their party if they sense that staying put might risk their chances of re-election. The two assertions above both feature legislators’ re-election incentives and rational calculus. Nevertheless, after formalizing the assertions, we find that the two theories contradict each other.
LetP(Defect = 1)denote the probability of party switching. We use v i|p to represent Legislator i’s vote share when staying in Party P, v i| ~ p is his vote share if defecting from the party, and Tm is the electoral threshold of District m. The transaction cost theory can thus be formalized as follows:
$$\begin{equation}
P\left[ {{\it Defect} = 1} \right] = {\beta _0} + {\beta _1}({v_{i|\sim p}} - {T_m}){\rm{\quad }}where{\rm{\quad }}{\beta _1}{\rm{,\ }}{T_m}{\rm{,\ }}v{\rm{ > 0}}
\end{equation}$$Equation (1) shows that the more the legislator perceives that his vote share after switching (v i| ~ p) will exceed the electoral threshold (Tm), the more likely he is to defect. Letπ denote the vote share from party loyalists – namely, the share of votes which the legislator will lose after switching party affiliation:
$$\begin{equation}
{v_{i|p}} = {v_{i|\sim p}} + \pi {\rm{\quad }}or{\rm{\quad }}{v_{i|\sim p}} = {v_{i|p}} - \pi ,{\rm{\quad }}where{\rm{\ }}{v_{i|p}} \ge \pi \ge 0
\end{equation}$$Based on equation (2), we can revise equation (1) accordingly:
$$\begin{equation}
P\left[ {{\it Defect} = 1} \right] = {\beta _0} + {\beta _1}({v_{i|\sim p}} - {T_m}) = {\beta _0} + {\beta _1}({v_{i|p}} - \pi - {T_m}){\rm{}}
\end{equation}$$Since transaction cost theory assumes a positive relationship between the switcher's perceived vote share (v i| ~ p) and the probability of his defection (P), i.e.β1 > 0, the relationship between the legislator's share of the vote (v i|p) and the probability of switching should also be positive if we control for other factors by taking the derivative of (3) with respect tov i|p. That is, the more popular a legislator is, the more likely it is that he will defect. In contrast, if he perceives that switching might endanger his re-election, the legislator will yield and toe the party line.
According to risk-averse theory, party switching stems from a sense of danger. If running for re-election under the old party banner is likely to jeopardize a legislator's electoral support, he might defect in order to improve his chances of re-election. The argument above can thus be formalized as the following equation:
$$\begin{equation}
P[ {\it Defect} = 1] = \beta _0^{\prime} + \beta _1^{\prime}(v_{i|p} - {T_m})\quad where\ \beta _1^{\prime} < 0, {T_m}, v > 0
\end{equation}$$Equation (4) states that the more a legislator's vote share falls below the electoral threshold, the more likely he is to defect from his party. When we control for other factors, it demonstrates a negative relationship between the legislator's vote share v i|p and the probability of switching, i.e. β′1< 0, which apparently contradicts the finding derived from equation (3).
Despite the contradiction between them, the transaction cost and risk-averse theories present the two stimuli that cause legislators to leave their parties. In most circumstances, the two impetuses do not coexist in one district. For example, in a single-member district, a legislator switches party affiliation either because his popularity outweighs the transaction cost of switching or because the party label might endanger his re-election. In closed-list proportional representation elections, a legislator defects only because there is little chance of his being elected from the party list. Nevertheless, due to the distinct electoral configuration of SNTV, the dual incentives for defection cohabit in multimember districts under this system.
2.4 Single non-transferable vote and party switching
The most significant feature of SNTV is its multimember constituencies. Under SNTV, each voter casts one nominal vote for a specific candidate and that vote cannot be transferred to other candidates under any circumstances (Cox and Niou, Reference Cox and Niou1994: 222). After the election, seats are allocated to the top vote-getters based on the plurality rule. Simply put, SNTV and SMD differ only in their magnitude. For instance, in the pre-reform Japanese House of Representatives, between three and five legislators were elected from each constituency (Reed and Thies, Reference Reed, Thies, Shugart and Wattenberg2001). In the pre-reform Taiwanese Legislative Yuan, the range of magnitude was between one and 17. Despite the minor difference in magnitude, the multimember district became the distinctive feature of SNTV and hence influenced the intraparty relationship as well as the relationship between legislators and their electorates.
2.5 Centrifugal political competition under SNTV
The first characteristic of SNTV is its centrifugal incentive, which refers to the incentive driving candidates to deviate away from other candidates’ positions. In an m-seat SNTV district, once a candidate gains more than 
$\frac{1}{{m + 1}} \times 100\% $ of votes, he will pass the threshold of exclusion and win a seat. The low electoral threshold thus influences the ideological positions of candidates as well as their campaign strategies. Unlike SMD which encourages the two candidates to pursue support from the median voter (Downs, Reference Downs1957), SNTV, as Cox (Reference Cox1990) suggests, encourages candidates to adopt more or less extreme positions and to avoid sharing an ideological position with others, which might result in a loss of votes and thus threaten the candidate's chance of re-election. Consequently, as the number of competitors increases in a multimember district, the centrifugal force will gradually outweigh the centripetal one and encourage the candidates to disperse across the ideological spectrum.
2.6 Intraparty competition in SNTV
This centrifugal effect has a direct impact on the intraparty relationship. Under SNTV, in order to obtain a majority in the legislature, a party must pit multiple candidates against each other and get several of them elected in each of the many multimember districts. Undernomination, overnomination, or the failure to equalize the vote all cause the loss of seats (Cox and Niou, Reference Cox and Niou1994: 222–3; Bergman et al., Reference Bergman, Shugart and Watt2013). As a result, competition among copartisans is often fiercer than it is among candidates from different parties. In order to be re-elected, legislators compete for limited votes and for passing the exclusion threshold, because hopelessly trailing candidates are often strategically abandoned by voters. Reed (Reference Reed1990) examines the electoral outcomes of postwar Japanese SNTV and finds a tendency for there to be only M + 1 candidates in each district.
This does not necessarily mean, however, that top vote-getters benefit from strategic behaviour; instead, they might also be the victims of it. Cox (Reference Cox1990: 616) formalizes strategic voting under SNTV and asserts that ‘in the multimember districts, voters who care only about the outcome of the election will strategically desert both candidates who are “too weak” and candidates who are “too strong”’. Because a highly popular candidate might steal too many votes from his copartisans and cause them to fail to be elected, in order to maximize the seat share of their party, voters transfer their votes from the popular candidates to marginal ones (Bergman et al., Reference Bergman, Shugart and Watt2013: 322). Batto (Reference Batto2008) studies the 2004 legislative election and provides empirical evidence for voters’ strategic behaviour under SNTV. He finds that due to spontaneous and strategic coordination among the electorate, in SNTV, candidates who achieved top ranking in pre-election surveys were often reduced to marginal or even to losing positions in the actual election.
Strategic voting behaviour adds more fuel to the flames of intraparty competition. In order to secure their re-election and gain more seats for the party, marginal candidates cry out for votes to be transferred from top-ranked copartisans. In reaction to this challenge, top-ranked candidates appeal to their supporters to stick with them rather than waste their votes on the hopeless marginals. Although some copartisan candidates form electoral coalitions in order to equalize votes among them, candidates who achieve only a low ranking in pre-election surveys usually start calling out for votes to secure their seats at the last minute, thus destroying the coordination of the coalition. Therefore, instead of relying on party brand names, candidates differentiate themselves from their copartisans. After being elected, most legislators join factions that may improve their chances of re-election by helping them win party endorsement, climb the career ladder through the seniority rule, and gain access to campaign funding (Cox and Rosenbluth Reference Ramseyer and Rosenbluth1996; Ramseyer and Rosenbluth, Reference Ramseyer and Rosenbluth1997: 59).
2.7 Candidate–voter relationship under SNTV
In addition to distinguishing themselves by means of extreme ideology, legislators also seek to establish personal networks in their constituencies in order to get consistent electoral support and votes and hence to spin out their political careers. In Japan, candidates enlisted help from local politicians, businessmen, and leaders of agricultural organizations in organizing and maintaining the branches of their support networks (Baerwald Reference Baerwald1986; Cox et al. Reference Cox and Thies1998; Fukui Reference Fukui1984; Ramseyer and Rosebluth Reference Ramseyer and Rosenbluth1997: 23–4; Woodall, Reference Woodall, Grofman, Lee, Winckler and Woodall1999: 30). Nevertheless, such a campaign strategy presents two problems. First of all, building and managing this kind of network of loyal electors costs a great deal of money (Cox and Thies, Reference Cox and Thies1998, Reference Cox and Thies2000; Cox et al. Reference Cox, Rosenbluth and Thies1999, Reference Cox, Rosenbluth and Thies2000). According to Woodall (Reference Woodall, Grofman, Lee, Winckler and Woodall1999: 31), the actual cost of political life for the average member of the Japanese lower house rose from ¥5 million per annum to ¥120 million in the late 1970s, and more than doubled at election time. Similarly, in Taiwan, it cost more than US$10 million to get elected as a legislator in 1992 (Winckler, Reference Winckler, Grofman, Lee, Winckler and Woodall1999: 343). In addition, even if the locals generously grant a legislator their support, this comes at a price. In return for their votes, the legislator has to provide private goods and particular favours as constituency services (Ramseyer and Rosenbluth, Reference Ramseyer and Rosenbluth1997: 20–1). Government-dispensed ‘pork’, cash, gifts, and even bureaucratic intervention, are applied to fertilize the electoral network. Legislators utilize their influence over specific areas of regulatory, budget, license issuing, and tax policy in order to woo particular groups of voters and to differentiate themselves from their copartisans. In short, SNTV encourages legislators to privatize public goods in exchange for personal votes.
2.8 Theoretical hypothesis
The discussion above links SNTV to party switching. The centrifugal incentive of SNTV, the ideological dispersion among members, and fierce intraparty competition make it difficult to maintain party discipline under this system. In addition, candidate-centered SNTV encourages legislators to focus more on cultivating their individual reputations in order to attract consistent and solid personal support instead of relying on party brand names. Moreover, the existence of multimember districts under SNTV inevitably fuels intraparty competition and drives legislators to switch party affiliation, as the major threat to re-election usually comes from within a legislator's own party. Pitting multiple candidates against each other in a constituency directly threatens an individual legislator's votes as well as his chances of re-election. Last, but not least, party malfeasance or government inefficiency might also reduce a legislator's chances of re-election.
For barely elected members, their chances of re-election are threatened by fierce competition from popular candidates and public discontent with their party. As long as they perceive that their chances of re-election will be improved by switching parties, namely,v i| ~ p > v i|p, they will switch affiliation.
Top-ranked candidates, however, might also defect from their party because of strong challenges from within the party and the low transaction cost of party switching. In multimember districts, in order to avoid wasting their votes on top vote-getters and to maximize the utility of their votes, voters spontaneously shift their support to copartisan marginal candidates. Therefore, the party label does not provide political superstars with much electoral support. Instead, their popularity among copartisans might stimulate spontaneous strategic behaviour by voters and encourage them to transfer their votes from the top-ranked candidates to the marginal ones. The ferocity of intraparty competition might cause high-ranking candidates to lose votes, even to the extent of losing their seats, to their copartisans if they stay put. By switching from his original party, a top-ranked defector can escape fierce intraparty competition and force supporters to concentrate their votes on him/her. Although he may lose votes from party loyalists by defecting, as long as personal popularity makes the switch affordable (v i| ~ p > v i|p) and the candidate is re-elected, party switching remains an alternative to be considered.
Based on the discussion above, we hypothesize a non-linear relationship between legislator popularity and party switching under SNTV. Barely elected legislators might defect from their parties because the party brand name does not provide enough votes for their re-election. If they stay put, their chances of re-election might be threatened by vote-stealing by copartisans or by poor party reputation, and the lower their share of votes, the more vulnerable they are. Party switching, therefore, provides an alternative way for these legislators to avoid this risk. By the same token, highly popular legislators also switch party to avoid intraparty competition. Knowing that they can be re-elected without the party label, these top vote-getters have looser ties to their party than other legislators. When they perceive that instead of benefitting their campaign, personal reputation might endanger their votes and seats, these top ranked legislators may also switch their affiliation in order to secure their re-election. As a result, both barely elected members and popular members all have strong incentives to defect from their party.
In the following section, we analyze the party switching data from Taiwanese SNTV legislative elections using the rare-event model. The empirical analysis fortifies our theoretical assumptions and demonstrates a V-shape relationship between a legislator's vote share and the probability of switching. That is, as the distance between a legislator's vote share and the exclusion threshold increases, the probability of his switching party also increases.
3 Data and empirical analysis
3.1 Dependent variable
In order to figure out how a legislator's share of the votes is associated with his decision whether or not to switch parties, we select as the dependent variable of this study whether a legislator switches or not. A legislator is defined as a switcher (coded as 1) if he was a member of Party A in the previous election but switched to another party or became an independent candidate in the current election. Otherwise, he is considered as having stayed put (coded as 0). The data record party switching between the second and the sixth Legislative Yuan. Since we mainly focus on party switching under SNTV, legislators elected from party lists and aboriginal candidates are excluded.Footnote 4
Table 1 shows the number of switchers in each term. In contrast to other legislative terms, there was at least a twofold increase in party switching in the fourth and the fifth terms due to the expansion of the Legislative Yuan in the former case and the first transfer of power in Taiwan in the latter. The fourth Legislative Yuan, elected in 1998, was expanded to 225 members in order to absorb members from the Taiwan provincial legislature, which had been abolished. This expansion resulted in a dramatic increase in the number of candidates, which stimulated intraparty competition within constituencies. The split in the ruling Kuomintang (KMT) and the subsequent victory of the opposition Democratic Progressive Party (DPP) in the 2000 presidential election resulted in another wave of party switching in the fifth Legislative Yuan. After the 2001 legislative election, KMT legislators could no longer bring government pork home to their constituency networks. In addition, the fact that the KMT had only been able to win 23% of the vote in the presidential election also concerned its legislators. The scarcity of government resources and the drastic decline in electoral support thus drove many KMT legislators to switch their party affiliation. In the fifth Legislative Yuan, 11 of the 19 switchers were from the KMT.
Table 1. Party switching in each legislative term
The discussion above highlights two problems regarding the data. Intuitively, the model that best fits the binary dependent variable is logistic regression or the probit model. Nevertheless, the distribution of switching presents a rare-event distribution: of the 541 cases, only 39 are labelled as switchers. Since popular statistical procedures, such as logistic regression, can sharply underestimate the probability of rare events (King and Zeng, Reference King and Zeng2001), we apply the rare-events logistic model of Tomz et al. (Reference Tomz, King and Zeng2003) to analyze the switching data. Moreover, Table 1 also shows that due to changes in the macro-level political environment, such as the expansion of the legislature and the split in the KMT, switches were independent across legislative terms but not necessarily independent within each term. To take into account concurrent switches, we cluster legislators elected in the same term.
3.2 Independent variables
3.2.1 Share of vote
Since this study mainly focuses on the relationship between a legislator's vote share and his decision to switch parties, the major explanatory variable is the legislator's vote share. To avoid the fallacy of post hoc, we apply the legislator's vote share in the previous election, i.e. v t − 1. In other words, we assume that when a legislator decides not to toe the party line, he uses the vote share acquired in the previous legislative election as the reference. This configuration has a problem, however. Due to the multimember constituencies under SNTV, the same vote share has a different meaning in each constituency. For instance, while garnering 20% of the votes in a ten-member district indicates a high degree of popularity, a candidate with the same vote share in a four-member district might only scrape in, or may even lose. To address this problem, we review the equation above and generate a cross-district indicator of legislator popularity by subtracting the exclusion threshold (
$\frac{1}{{m + 1}}$) of the legislator's constituency from his share of votes, i.e. (
${v_{t - 1}} - \frac{1}{{m + 1}}$) (et seqq. Lijphart and Gibberd Reference Lijphart and Gibberd1977; Loosemore and Hanby Reference Loosemore and Hanby1971; Rae Reference Rae1971; Rae et al. Reference Rae1972). If the vote share is greater than the electoral threshold, the indicator is positive. The indicator is negative if the legislator is below the threshold.
If transaction cost theory is tenable, we should observe a significant positive relationship between the indicator of popularity and the switching decision. If risk-averse theory is correct, there should be a negative and significant coefficient regarding the indicator. However, if, as suggested, transaction cost theory and risk-averse theory both apply under SNTV, we should observe that the indicator of legislator popularity is not significant, but its square term, i.e.
${({v_{t - 1}} - \frac{1}{{m + 1}})^2}$, is positively associated with the defection decision. That is, there is a V-shape relationship, as illustrated in Figure 1, between a legislator's vote share and party switching, which suggests that legislators are more likely to defect from their parties as their share of the vote increases with the distance from the electoral threshold.
3.2.2 Economic indicators
As stated above, according to risk-averse theory, if the electorate's dissatisfaction with the government's performance seems likely to jeopardize a legislator's re-election, he will be more likely to switch party in order to avoid being punished by the voters and to improve his chances of re-election. According to studies of economic voting and electoral accountability, the issue of most concern to citizens is the economy (Feldman, Reference Feldman1984; Fiorina, Reference Fiorina1978). Thus, we calculate the change in the unemployment rate for each legislative term and use this as the indicator for measuring economic performance. Legislators are more likely to switch parties if unemployment increases.
3.2.3 Incumbent government
Incumbency can be a double-edged sword. On the one hand, scholars assert that incumbency status provides such advantages as ideal committee positions, electoral funding, and participation in policymaking, and therefore helps legislators achieve re-election (Kato and Yamamoto, Reference Kato, Yamamoto, Heller and Mershon2009: 240). In other words, in contrast to opposition parties, the governing party should find it easier to hold on to its members. On the other hand, poor performance by the government might also be expected to damage electoral support for the ruling party. Hibbing and Alford (Reference Hibbing and Alford1981) and Fiorina (Reference Fiorina1983) find that voters are not only capable of evaluating government performance, but can also clearly identify who is responsible for the policy failure. Based on the discussion above, we control for incumbency status and code the KMT members as incumbents between the second and the fourth Legislative Yuan. Similarly, DPP legislators are labelled as incumbents in the fifth and sixth terms.
3.2.4 Legislative seniority
Nemoto et al. (2008) assert that junior, midcareer, and senior legislators have different incentives to maximize their electoral fortunes. The junior and midcareer legislators concentrate on mobilizing votes as an investment in future higher policy positions. However, unlike the juniors who maximize their re-election chances by organizing personal networks in their districts, the midcareer legislators accumulate personal influence over the policymaking process and use their current policy positions to increase their electoral security and seniority. In contrast to other legislators, electorally secure seniors concentrate on maintaining their leadership, abiding by existing rules and norms, and continuing their involvement in internal party and factional affairs. In short, unlike Mayhew's assumption which highlights their re-election incentives, legislators might have different goals at different stages in their careers. These diverse career patterns thus affect a legislator's relations with his party. Since backbenchers are more vulnerable to electoral volatility, they are more likely to defect when their constituency interests conflict with party policy (Nemoto et al., 2008). Accordingly, we calculate the number of elected terms for each legislator. To reflect the real seniority structure of the Legislative Yuan, the elected terms in the period of supplementary legislative elections are also included. For instance, Wang Jin-pyng, the president of the Legislative Yuan, was re-elected ten times between 1974 and 2004.
Despite the configuration above, it is difficult to predict the influence of the seniority rule on party switching. Taiwan's Legislative Yuan is less institutionalized and professionalized than the legislative branches of most advanced industrial democracies. In addition, legislators cannot simultaneously serve as ministers. Given the relative scarcity of government and legislative posts open to legislators, we assert that re-election remains their major goal and that seniority might not have a significant association with switching.
3.3 Control variables
We further control for legislators’ gender. Previous studies on party switching do not directly address gender difference in party switching. Nevertheless, in contrast with their male colleagues, studies show that female legislators are less ambitious (Bernstein, Reference Bernstein1986; Bledsoe and Herring, Reference Bledsoe and Herring1990; Carroll, Reference Carroll1994; Costantini, Reference Costantini1990; Lowless and Fox, Reference Lowless and Fox2010; Kirkpatrick, Reference Kirkpatrick1976; Fulton et al., Reference Fulton, Maestas, Maisel and Stone2006). Therefore, in theory, female legislators should be less likely to defect. In addition, given the finding of Nemoto et al. (2008), which shows a positive relationship between a legislator's age and rebellion, we also control for age.
4 Results
The statistical work proceeds in two steps. Table 2 shows how the data correspond with transaction cost theory and risk-averse theory, both of which assert a linear relationship between legislator popularity and party switching. As demonstrated in model 1, the empirical data do not significantly support either transaction cost or risk-averse theory. The indicator for measuring legislator popularity, the same as the number of terms the legislator has been elected, has a negative, but insignificant, association with party switching. Although change in the unemployment rate, as predicted by risk-averse theory, is positively associated with switching, the macro-level explanatory variable only provides an interpretation for the variation in legislator defection across different terms, not for variation in defection within each term.
Table 2. Analysis of party switching
Notes: (a) *p<0.1, ** p<0.05, *** p<0.01.
(b) Numbers in the parentheses are standard errors.
We further examine our hypothesis, which is that both transaction cost theory and risk-averse theory capture the incentives for SNTV legislators to switch party affiliation. Model 2 shows that the square term of the popularity indicator is significantly positive (β = 20.1, t = 1.98), presenting a V-shape relationship between legislator popularity and party switching: as the distance between a legislator's vote share and the electoral exclusion threshold increases, the likelihood of him/her switching parties also increases (see Figure 1). While the empirical analysis indicates that highly popular legislators are more likely to defect, because, for them, personal reputation can easily outweigh the loss of party votes but staying put may result in loss of votes to marginal copartisans, it also proves that barely elected legislators prefer to switch because intraparty competition threatens their re-election. In contrast to the two kinds of members mentioned above, legislators whose vote share is around the electoral threshold prefer to stay put. Given the combination of personal reputation and party brand name, their chances of re-election are secure and they have no incentive to defect from the party. Furthermore, because their personal reputation is not sufficient to ensure their re-election without relying on the party label, they prefer to keep running under the party banner.
Model 2 also demonstrates that poor economic performance remains significantly associated with the defection rate. When the increase in the unemployment rate is 1% above average, the predicted probability of party switching increases from 6.51% to 12.61%. As stated above, this seems to support risk-averse theory's assertion that legislators defect in order to avoid punishment by the electorate. To investigate whether legislators belonging to the incumbent party are more likely to switch party affiliation than their opposition colleagues under poor economic circumstances, we further multiply the indicator variable of government incumbency by the two economic indicators to create interactive variables. According to risk-averse theory, these variables should be negatively associated with party switching. Model 3 incorporates the interactive variables but fails to support the assertion above. Moreover, the unemployment rate becomes less significant while the coefficient of the square term of legislator popularity remains the same. In sum, model 2 fits our theoretical assumptions better and provides more robust and consistent analysis.
5 Conclusion and discussion
Party switching not only challenges party discipline, but also has an impact on the party system as a whole. This study investigates the relationship between legislator popularity and party switching behaviour from the perspective of rational choice. Although transaction cost theory and risk-averse theory both emphasize legislators’ rationality and re-election incentives, we find that the two assertions repudiate each other. While transaction cost theory suggests that popular legislators are more likely to defect, risk-averse theory indicates that legislators switch party affiliation when they perceive their chances of re-election to be under threat. Given the specific electoral configuration of SNTV, the two diverse incentives for party switching cohabit under this system. The existence of multimember districts not only encourages legislators to cultivate personal votes instead of relying on a party label, but also fuels intraparty competition and endangers the re-election chances of both the barely elected and highly popular members. We further test our theoretical assumption against party switching in the Taiwanese Legislative Yuan. The empirical analysis supports our hypothesis and shows that in contrast to other legislators, highly popular legislators and barely elected legislators were more likely to defect from their parties.
This study provides at least three contributions to the literature of party switching. First of all, it highlights various considerations behind legislators’ party switching. Despite the contradiction between the two theories, our empirical tests show that transaction cost theory and risk-averse theory both capture considerations behind legislator defection. Second, like previous studies of party switching, our study links party switching with rational choice. Although legislators may have a variety of reasons for defecting, achieving re-election remained at the top of their list of priorities, which is not surprising given the vast amount of funds they must invest in elections and their political sensitivity. Hence, while highly popular legislators calculate the transaction cost of switching party affiliation, barely elected legislators estimate the advantages of staying put. Last but not least, we show how electoral configuration shapes political behaviour and hence impacts intraparty politics. Under SNTV with its multimember districts, majority-seeking parties pit multiple candidates against each other in constituencies in order to maximize their share of seats, while legislators cultivate their personal reputations in order to differentiate themselves from their copartisans. As a result, SNTV not only encourages highly popular legislators to discard their party labels, but also drives barely elected legislators to seek to escape from intraparty competition.
As stated at the outset, party switching impacts on intraparty politics as well as party systems, and definitely merits further study. Ideological incongruence, the allocation of government posts, the lash of party discipline, etc. – all of these might drive legislators to stop toeing the party line. In this study, we demonstrate how SNTV provides incentives for accumulating personal votes and fuels intraparty competition, thus shaping a specific pattern of party switching. This pattern might also be observed in an open-list proportional representation system, which also encourages personal votes and competition for higher ranking on party lists. Moreover, while the logic behind party switching concerns us here, we admit that which party the defectors switch to is also an interesting topic. For instance, Desposato and Scheiner (Reference Desposato and Scheiner2008) compare party switching in Brazil and Japan and find that the centralization or decentralization of resource control determines switchers’ party choice. Their theory can be applied to examine party switching in Taiwan. However, other topics will have to await further research.
About the authors
Alex Chang is an assistant research fellow of the Institute of Political Science, Academia Sinica, Taiwan. He received his Ph.D. from University of Iowa in 2007. One of his research interests is party–legislator relationships, especially focusing on party switching. In addition to party switching, he also introduces new quantitative methods to social science and applies them to cross-national studies. His publications appear in prominent journals such as Journal of Democracy, Issues and Studies, Taiwan Journal of Democracy, and Journal of Contemporary China.
Yen-chen Tang received her Ph.D. from National Taiwan University in 2014 and now serves as a lecturer in the Department of Political Science. Her research interests include electoral systems and party politics, electoral behaviour, comparative democratization, and political communication. Her publication appears in Journal of US–China Public Administration.
