2. Personal votes, electoral rules, and legislative representation

As Pitkin (Reference Pitkin1967) depicts, political representation requires individual representatives to act in and to be responsive to the interest of the represented. Depending on whose interest to represent, legislators need to balance their resource and develop different styles of representation in order to serve their particular or general constituencies. Among a variety of potential determinants, electoral institutions have long been considered to be one critical factor that structures the style of democratic representation (Mayhew, Reference Mayhew1974; Fenno, Reference Fenno1978; Cain et al., Reference Cain, Ferejohn and Fiorina1987; Carey and Shugart, Reference Carey and Shugart1995; Samuels, Reference Samuels1999; Crisp et al., Reference Crisp, Escobar-Lemmon, Jones, Jones and Taylor-Robinson2004 ). Specifically, institutional settings determine the available information shortcuts (i.e., partisan labels vs. individual characteristics) that voters may employ to cast their votes (Shugart et al., Reference Shugart, Valdini and Suominen2005) and thus further condition the incentives for legislative representatives to cultivate either their own personal brand or a collective party reputation (e.g., Carey and Shugart, Reference Carey and Shugart1995; Carey, Reference Carey2009; Colomer, Reference Colomer2013; Karvonen et al., Reference Karvonen, Atmor, Carey, Enten, Hazan, Lutz, Margetts, Marsh, Massicotte and Rahat2011; Bräuninger et al., Reference Bräuninger, Brunner and Däubler2012). Consequently, legislative representation under this literature has been largely seen as a zero-sum game between strategies of pursuing the ‘personal vote’ or the ‘party vote’.

Previous research has demonstrated that personal vote-seeking incentives often present in systems such as open-list PR and Single Non-Transferable Voting (SNTV) systems, where voters have the opportunity to disturb party list (i.e., open-list PR) or voters are allowed to cast votes for individual candidates (i.e., SNTV). On the contrary, when voters can only cast votes for political parties, rather than individual candidates (e.g., closed list PR), personal vote-seeking incentives are weaker and representatives are motivated to construct a collective party brand. Moreover, incentives for legislators to cultivate a personal vote increase with district magnitude in open list systems but decrease with district magnitude when lists are closed (Carey and Shugart, Reference Carey and Shugart1995; André and Depauw, Reference André and Depauw2014).

Under electoral systems that promote personal vote-seeking, abundant empirical evidence has demonstrated that legislators are motivated to signal their supporters a strong constituency-representative connection through a variety of activities that help them promote their personal reputation. For instance, legislators may initiate specific bills that benefit their local communities, commit to casework and constituency service, and vote against their party position in floor on policies (e.g., Cox and Thies, Reference Cox and Thies1998; Grofman et al., Reference Grofman, Lee, Winckler and Woodall1999; Crisp et al., Reference Crisp, Escobar-Lemmon, Jones, Jones and Taylor-Robinson2004; Carey, Reference Carey2007; Lundberg, Reference Lundberg2007; Depauw and Martin, Reference Depauw, Martin, Giannetti and Benoit2009).

A special attention in this vibrant literature has been paid to mixed-member electoral systems, as they allow a speculation of legislators elected under different rules within a single country holding confounding factors at the country level. Theoretically, it is expected that legislators elected under party-centric rules are more party-oriented and tend to show greater party loyalty than legislators elected from individual-centric rules that promote personal vote-seeking incentives. If this pattern is observed, it is often said that the ‘best of both worlds’ (Shugart and Wattenberg, Reference Shugart and Wattenberg2001) is realized in the legislature. A considerable number of research efforts have been made to examine this impact of electoral rules on representative behavior in mixed-member electoral systems around the world (for a comprehensive review, see Batto, Reference Batto2012).

Interestingly, this tension between the interest of individual legislators and the interest of political parties has implicitly assumed that personal vote-seeking incentives are all alike since they are all considered to be harmful to a collective party brand. Yet, the literature has largely overlooked which type of voters these personal vote-seeking legislators attempt to communicate with. Recognizing this missing piece in the literature, Batto (Reference Batto2009) differentiates personal votes into two categories: ‘extra-party’ personal votes and ‘intra-party’ personal votes. Specifically, the former is a vote cast by voters who cross party lines. These voters tend to rely on the characteristics of individual candidates, instead of party affiliation, to determine their votes. On the contrary, the latter refers to those voters who are generally attracted by appeals that are compatible with party platforms. Implying by this distinction, a candidate must convince voters that his or her position is different from the overall party platform in order to cultivate ‘extra-party personal votes’ or signal to the voters that he or she is a hardcore partisan to win ‘intra-party personal votes’.

In the next section, we adopt Batto's distinction of intra- and extra-party personal votes and develop a new theoretical expectation of varying representation patterns among legislators elected within the same rule.

3. Intra- and extra-party personal votes and legislative representation in Taiwan under the SNTV period (1992–2008)

A consensus in the prior literature suggests that personal vote-seeking incentives embedded in electoral rules tend to motivate legislators to deviate from their party positions in roll call voting. However, if we look closely, there are actually two behavioral patterns of voting against the party. In one situation, a legislator may vote against her party and side with other parties. In a two-party system, this deviation can be interpreted in a way that the legislator stands on a more moderate position, as he/she seems to be closer to the other party than to his/her own party. In the other situation, the legislator may refuse to follow his/her party's guide to collaborating with the other party, which makes the legislator appear to be more extreme than his/her own party. Nevertheless, existent research has largely overlooked this directional nature of representation. A question then arises: what makes legislators under the same electoral rule be more moderate or more extreme?

We consider the Legislative Yuan in Taiwan under the period of SNTV system in MMDs an ideal case to investigate this directional motivation of personal votes. In particular, the SNTV-MMDs system is known to produce strong incentives for candidates to build their personal reputation. Existing research on Taiwan has demonstrated that candidates in elections work extremely hard to be personally well known and liked by voters in order to get the entrance ticket to the legislature (e.g., Cox and Niou, Reference Cox and Niou1994; Swindle, Reference Swindle2002; Batto, Reference Batto2008; Johnson and Hoyo, Reference Johnson and Hoyo2012). In addition, the SNTV system in Taiwan also motivates politicians to pursue either an ‘extra-party’ personal vote or an ‘intra-party’ personal vote in elections (Batto, Reference Batto2009). Last but not least, multi-party competition in Taiwan can be largely considered the competition between two political camps, which allows us to explore our expectation about directional legislative representation (Fell, Reference Fell2005, Reference Fell2018; Huang, Reference Huang2005; Hsieh, Reference Hsieh, Paolino and Meernik2008; Achen and Wang, Reference Achen and Wang2017).

The party system in Taiwan before 2000 was actually close to a two-party system with two major parties, the Kuomingtang (KMT) and the Democratic Progress Party (DPP), and a small party named the New Party (NP), which allied with the KMT on major issues. Although a multiparty system emerged in Taiwan after 2000, it largely functioned as a two party-bloc system, in which the KMT formed an alliance with the NP and the People First Party while the DPP formed the other alliance with the Taiwan Solidarity Union. These two party-blocs are known as the ‘Pan-blue’ and the ‘Pan-green’ camps, leading by the KMT and the DPP, respectively.

Political competition in Taiwan is largely explained by a single factor that centers on the relationship between Taiwan and China (Fell, Reference Fell2005, Reference Fell2018; Huang, Reference Huang2005; Hsieh, Reference Hsieh, Paolino and Meernik2008; Achen and Wang, Reference Achen and Wang2017). The Pan-blue camp stands for a closer relationship with China, both economically and culturally. It champions the Chinese identity, and eventually the unification with the Mainland China. On the contrary, the Pan-green camp favors policies that help differentiate Taiwan from China. It promotes the Taiwanese identity and advocates independence from China. Overall, it is argued that the confrontation between these two party blocs has been intensified after the presidential election in 2000.

Moreover, electoral competitiveness at the district level clearly plays an important role in structuring the campaign strategies of political parties in both camps. Specifically, both camps have to adjust their campaign strategies by nominating fewer (more) candidates in their weak (strong) districts. Yet, they still could not have all their nominees elected given the intra-party competition resulted from the SNTV system. In the 2001 legislative election, for instance, the Pan-blue coalition nominated four candidates in Pingtung County (five nominees in the previous election), one of the strongholds of the Pan-green camp, and had only two of them elected to the legislature.Footnote ³ In contrast, the Pan-green camp nominated five candidates and had four elected. While electoral competitiveness has demonstrated us its impact on campaign strategies as well as electoral outcomes, we suggest that it also conditions legislators’ incentives of pursuing a different type of personal votes (i.e., intra- versus extra-party personal votes) and therefore further imposes an effect on legislators’ position taking strategies.

We contend that legislators tend to be intra-party personal vote seekers and are more likely to deviate from the party toward a more extreme direction if they are elected from districts where the electoral strength of their party is weaker. Our rationale is twofold. First, in weak districts of a political party, it is always costly for candidates to seek support from voters attached to other parties. For these candidates, lowering one's partisan profile by engaging in cross-party collaboration in the legislature may not necessarily be appreciated by voters cross party lines. Most importantly, doing so risks losing one's own partisan supporters since partisans are less supportive to a bipartisan image and may consider such behavior as ‘selling out’ one's own party (Hibbing and Theiss-Morse, Reference Hibbing and Theiss-Morse2002; Harbridge and Malhotra, Reference Harbridge and Malhotra2011). In such case, an ideal strategy is to behave as an extreme partisan in the legislature and serve sub-constituencies to earn their electoral support (Ames, Reference Ames1995; Griffin and Newman, Reference Griffin and Newman2005; Miller, Reference Miller1999; Ellis, Reference Ellis2013). That is, legislators should appeal to partisan voters in districts where their party is relatively weak. Thus, we expect that legislators elected from their party's weak districts tend to take a more extreme position than their party does. Doing so allows them to act as if they are the protector of the party since party label is much more valuable in these regions.

On the contrary, legislative members from their party's strong districts should take a more moderate position than their party since partisan advantage (reflected in better electoral performance) in the district provides a better position for them to downplay their own partisan image so that they may solicit support from voters across party lines.Footnote ⁴ To put it differently, legislators from MMDs where their party enjoys greater electoral strength tend to be ‘extra-party’ personal vote seekers and they are more likely to deviate from the party toward a moderate direction.

Indeed, parties in their strongholds may nominate more candidates, which could potentially encourage legislators to seek for intra-party personal votes in such districts since the presence of additional candidates may foster a redistribution of partisan voters. However, we argue that this may not be the case. Implied the vote division literature (e.g., McCubbins and Rosenbluth, Reference McCubbins, Rosenbluth, Cowhey and McCubbins1995; Cox, Reference Cox1996; Tsai, Reference Tsai2005), incumbent legislators have more resource than newly nominated candidates to establish and maintain personal support networks among party loyalists, particularly, those who are associated with the government party. They do so through providing constituents with particularistic policy benefits and better constituency service, among other things. In other words, being an incumbent has an advantage in keeping a stable connection with existing supporters (also see e.g., Hayama, Reference Hayama1992; Reed, Reference Reed, Somit, Wildenmann and Boll1994). This stable connection should always allow incumbent legislators of the strong party to get a greater share of votes from the partisan foundation than their newly nominated co-partisans if there is any. In other words, the presence of new candidates under the same partisan label may not necessarily induce the intra-party personal vote-seeking behavior (i.e., be more extreme) of incumbent legislators.Footnote ⁵

Putting our theoretical expectation in Taiwan's context, we expect that Pan-blue and Pan-green legislators from their weak districts to be more ‘blue’ or ‘green’ (i.e., more extreme than the camp median), whereas legislators from their strongholds to be less ‘blue’ or ‘green’. This discussion leads to our major hypothesis:

4. Data and research design

This section describes the data, measurement of key variables, and the regression model. Our hypothesis concerns roll call voting data of district-elected legislators. To examine our argument, we obtained this information from the Center for Legislative Studies at the Soochow University in Taiwan. This dataset includes all roll call voting records from the 2nd to the 6th term of the Legislative Yuan in Taiwan, which covers the entire period of the SNTV-MMD system.Footnote ⁶

To capture the nature of legislative representation in a directional manner, we rely on the spatial model of roll call voting (Enelow and Hinich, Reference Enelow and Hinich1984; Poole and Rosenthal, Reference Poole and Rosenthal1997; Clinton et al., Reference Clinton, Jackman and Rivers2004a) and apply the one-dimensional W-NOMINATE model on roll call voting data in each legislative term. This scaling method allows us to capture the position taking strategies of legislators by transforming roll call voting data into ideal points of legislators in a geometric space. The fitness of this model is sufficiently large, with the scores of Aggregate Proportional Reduction of Error in general higher than 90 (the 4th term is 88.9). It suggests that the estimated ideal points of legislators explain more than 96% of legislative decision-making in all legislative terms.

The overall pattern of legislators’ geometric positions along the main Pan-blue vs Pan-green cleavage is presented in Figure 1. In general, legislators from the Pan-blue camp are distributed along the right-hand side of the dimension, while the Pan-green legislators are distributed along the left-hand side. Note in this figure, we only include legislators elected from the nominal tier (i.e., MMDs). Clearly, there is a significant intra-party/camp variance, which is useful for our interest in this study. Pan-blue legislators located at far-right positions than other bloc members are considered to be more extreme (or more ‘blue’). Same interpretation can also be applied to Pan-green members. Locating in the far-left space suggests that Pan-green legislators have the least records of voting with Pan-blue legislators. In contrast, representatives placed in centric areas are considered to be moderate as they involve in cross-camp voting more frequently than other legislators from their own camp.

Figure 1. The Distribution of W-NOMINATE Score in the Legislative Yuan.

To generate our dependent variable, we rely on the estimated positions of legislators. We first identify the median position for each party bloc in each legislative term and then calculate the directional distance from each legislator to the bloc that the legislator is associated with by subtracting the bloc median from the ideal point of the legislator. Since the estimated positions of all legislators from the Pan-green camp are negative values, we first multiply these values by −1 and then perform the subtraction. In this way, the extremeness/moderateness of a Pan-green legislator is comparable with a Pan-blue legislator. This variable can be summarized as Distance_ij = P_i–M_j, where P_i represents the ideal point of a legislator i and M_j indicates the median position of a party-bloc j. It therefore captures to which direction (i.e., extreme or moderate) a legislator is deviating from the median position of the party bloc he or she is affiliated with. Since being extreme in the W-NOMINATE score indicates that a legislator had rarely voted with the other camp, we take this as a proxy of the extent to which the legislator is a hardcore partisan.Footnote ⁷ In other words, this measure reveals the partisan strength of the legislative representative – how ‘blue’ a Pan-blue representative is or how ‘green’ a Pan-green legislator stands. In our measure, a positive value suggests that a legislator takes a more extreme position (i.e., deep blue or deep green) than his/her party-bloc, while a negative value represents a more moderate position (i.e., light blue or light green) of the representative than his/her party bloc.

Our main explanatory variable is the electoral margin of the legislator's camp in the district where s/he was elected. To measure this variable, we first obtain the full electoral results from the official National Electoral Commission in Taiwan, calculate the vote shares of the pan-blue and pan-green camps, and then subtract the camp's vote share of the legislator from the other camp's vote share. Greater values indicate a better electoral performance (and therefore more competitive) of a camp in a district and we expect this variable to be negatively correlated with our dependent variable.

We also include several control variables. The first three control variables capture the institutional context in the legislature, including the legislative majority, seniority, and executive-legislative relation. We include a binary variable capturing if the legislator is from the Pan-blue camp, which has always been the legislative majority in our time period. This is coded as ‘1’ if a Pan-blue camp member and zero for otherwise. We also include seniority, which measures the number of terms that the legislator had served in the Legislative Yuan. The last one is a variable that captures whether the legislator is associated with the governing coalition. It is coded as ‘1’ if the legislator's party-bloc holds the executive branch and zero for otherwise. These three variables are considered to account for the potential impact exerted by institutional settings on legislative behavior (e.g., Crisp et al., Reference Crisp, Escobar-Lemmon, Jones, Jones and Taylor-Robinson2004).

The other four control variables, we include, reflect the legislator's district characteristics beyond our main explanatory variable of electoral margin. First, the logged value of district magnitudeFootnote ⁸ is included in order to isolate its independent impact on representatives’ position taking strategies (e.g., Cox, Reference Cox1997; Bertelli and Richardson, Reference Bertelli and Richardson2008). Second, we further control for the number of candidates from the same bloc in the district where the legislator was elected in the previous election. This variable captures the potential influence of intra-bloc competition on the legislator's position taking strategy (e.g., Cox and Thies, Reference Cox and Thies1998). Third, we include a variable indicating the partisan control of the local government (i.e. the city/county government). Finally, we also include two regional variables (i.e., north and south) in order to account for what some dubbed ‘Southern Politics’ in Taiwan (Lee and Hsu, Reference Lee and Hsu2002; Hsu and Lin, Reference Hsu and Lin2009).

As our dependent variable ranges from negative to positive values, we employ a classic OLS regression with the legislative term fixed effects. To reiterate, if our theoretical expectation holds, we should observe a negative coefficient (and statistically significant) of the main explanatory variable that measures the electoral competitiveness of parties at the district level.

5. Empirical results

Table 1 presents the estimated results from our OLS regression. In Model 1, we examine a simple model that only includes the main explanatory variable as well as the legislative term fixed effects. In Model 2, we perform a full model by including all other control variables.

Table 1. OLS model of directional legislative representation

t statistics in parentheses.

*P < 0.05, **P < 0.01, ***P < 0.001.

As one may immediately discover, the major finding from Table 1 is consistent with our theoretical expectation. With the estimated coefficient of Electoral Margin being negative and statistically significant, it suggests that legislators tend to deviate toward a moderate direction from their camp median if they are elected from districts where their camp performs better than the other camp. On the contrary, legislators from the camp that performs poorly in districts tend to adopt a different strategy by moving toward a more extreme direction from the camp median. The evidence here suggests that legislators indeed face extra- and intra-party personal vote-seeking incentives and therefore behave differently in the Legislative Yuan. Put it simply, deep blue legislators come from green districts, and vice versa.Footnote ⁹ Our main hypothesis clearly obtains empirical support.

In addition to our primary findings, Model 2 also reveals some interesting findings regarding our control variables. First, legislators in the blue camp are relatively moderate than their counterparts in the green camp, as the estimated coefficient for a Pan-blue variable is negative and statistically significant. We consider this finding relevant to the main argument of the partisan theories of legislative organization (Rohde, Reference Rohde1991; Cox and McCubbins, Reference Cox and McCubbins1993, Reference Cox and McCubbins2005; Aldrich, Reference Aldrich1995). In this literature, the legislative policy outcome is made through a series of agenda control, which is structurally favorable for the legislative majority. The roll call voting decision is the final stage of legislative agenda setting, and legislators only face a selected set of bills to take their position. Compared with the opposition green camp, most of the blue legislators do not have to be extreme given this condition.

Second, district magnitude clearly shapes the position taking strategies of legislators. The negative and statistically significant coefficient of District Magnitude suggests that legislators elected from a larger district tend to take a more moderate position than their party in order to solicit voters across party lines. This finding is also consistent with Kirkland (Reference Kirkland2012) who demonstrates that a larger district motivates legislators to be more collaborative with legislators from other parties. Doing so allow legislators to cultivate a better personal image and gain support from moderate (i.e., less partisan) or independent voters. In this case, there are little intra-party personal vote-seeking incentives to be extreme.

Finally, our results also show that the number of candidates from the same party-bloc running in the previous election exerts an impact on legislators’ position taking strategies. More precisely, as the degree of intra-bloc competition increases, legislators tend to take a more extreme position. We interpret this as evidence that legislators under intense intra-bloc competition are more likely to target and signal to partisan supporters that they are the legitimate representative of the party/bloc.

6. Conclusion and discussion

Since Carey and Shugart (Reference Carey and Shugart1995), how personal vote incentives affect legislators’ strategies to cultivate a personal reputation has been a major interest of students who study legislative representation. One critical behavioral consequence of seeking for personal votes is that legislative members in the legislature tend to deviate from party lines. Yet, to which direction a legislator would deviate has surprisingly received little attention in the current literature. In this paper, we fill this void by extending the logic of extra- and intra-party personal votes (Batto, Reference Batto2009) to the explanation of legislators’ position taking strategies in Taiwan.

Based on the political context of the Legislative Yuan under the SNTV period, we draw a theoretical hypothesis of directional forces of personal votes and incumbent legislators’ position taking strategies. We claim that the disparity of electoral strength across districts motivates legislators either to strengthen their connections with their party loyalists or to reach across the aisle to gain potential supports from moderate or even less partisan voters who are affiliated with the other camp. We expect legislators elected from a district where their party is electorally less competitive tend to take an extreme position than their co-partisans in other districts.

To examine our claim, we rely on the ideal points of legislators by estimating the comprehensive roll call voting records from the Taiwan Legislative Yuan in the period of 1992–2008, provided by the Center for Legislative Studies at the Soochow University in Taiwan. Our roll call voting analysis, in particular, through an empirical application of spatial modeling, allows us to examine our theoretical interest, which is difficult to capture using other measures such as the Rice index. Using our dependent variable, we find robust evidence for our theoretical conjecture.

Theoretically, this paper contributes to the literature by demonstrating that personal vote-seeking incentives generated by the SNTV system can yield more than one behavioral consequences of legislative representatives. Even elected under the same electoral arrangement, legislators do not behave similarly as they face different personal vote incentives. Our case study of Taiwan's SNTV system – a system that promotes individual representation – presents that electoral strength at the district level actually encourages partisan representation (or partisan extremism). This finding should shed some lights for future studies on the intertwining nature between district-level factors, electoral systems, and legislative representation. Particularly, future research should go down to the district level to explore the potential influence of district features on legislative representation.

Substantively, this paper provides valuable information to improve the scholarly understanding of Taiwan politics. Specifically, describing Taiwan politics in a colorful manner, namely blue vs green has been a common practice and it is relatively easy to identify the ‘color’ of a politician in Taiwan. However, figuring out whether the politician is a hardcore or moderate partisan (or deeper/lighter blue/green) is a more difficult task as it requires detailed knowledge about the politician and the party (s)he is affiliated with. Similar to Western examples that calibrate politicians’ ideological positions using roll call voting analysis (e.g., Clinton et al., Reference Clinton, Jackman and Rivers2004b), we locate the ideological positions of legislators in Taiwan under the SNTV period. By making use of the empirical data we offer in this paper, future research may advance our understanding of how democracy functions in Taiwan to another level.