Substantive representation of women and policy-vote trade-offs: does supporting women’s issue bills decrease a legislator’s chance of reelection?

ABSTRACT The paper investigates how parliamentary efforts to represent the interests of female electorates influence the legislators’ re-election chances. Taiwan is chosen as the case study and, for empirical analysis, I utilise an original bill co-sponsorship dataset that consists of roughly 400,000 cosponsors for all bills submitted between 1992 and 2016. The findings, based on regression analyses, show that making more legislative effort on women’s issues – by prioritising them over other issues – results in electoral losses, and this negative effect is more pronounced among female legislators. The paper contributes to the gender politics literature by theorising and testing a hitherto underexplored relationship between two representational processes: how the substantive representation women by female legislators affects their descriptive representation. It also contributes to legislative and electoral studies by demonstrating that legislators’ policy-vote trade-offs are policy-sensitive and gendered, thus calling for a more nuanced approach to be taken in future research.


Introduction
It has been long noted that legislators' active policy initiatives while in office has a positive connection with elections, specifically by increasing their chances of reelection (Mayhew, 1974). However, empirical evidence also reveals that specialising in a particular policy area can be an electorally risky strategysince it can in effect neglect or deviate from the needs and preferences of voters who are not the direct beneficiaries of that specialised policy (Carey, 2008;Spoon, 2011). The paper builds on this policy-vote trade-offs thesis, but goes further by focusing on the policy commitment to women's issues by raising the following two questions: Does a legislator's commitment to women's issues affect him or her individually by having a negative or positive impact on future reelection chances? If so, is there any difference between male and female legislators?
By addressing these two questions, the paper fills an important gap in the existing gender politics literature. In the light of different types of representation noted by Pitkin (1967), students of gender politics thus far have extensively examined the effect of descriptive representation (women's presence in the legislature) on substantive representation (representation of female voters' interests). That is, a wide range of empirical works have demonstrated that the greater presence of women in the legislature actually makes a substantial difference to the commitment of more women-friendly policies. For instance, using a variety of measures capturing legislators' representation activitieslegislative sponsorship and voting, legislative speeches, plenary and committee debatesthe gender politics scholarship has repeatedly shown that, in both old and new democracies, women are more likely than men to support issues promoting women's interests (such as gender equity, daycare programmes, accessible abortion, domestic violence reduction) (Bäck et al., 2014;Barnes, 2016;Clayton et al., 2017;Vega & Firestone, 1995).
However, over time, students of gender politics have noted that the relationships between the two representation types are complex and contingent, and thus increasingly ask questions such as who specifically acts for women and at which specific periods (Childs & Lovenduski, 2013). Going further, the latest scholarship concerning women's substantive representation focuses on the extent to which women's diverse representational needs are met or how represented women feel about their representation (Celis & Childs, 2020). Echoing a more nuanced and holistic approach taken in the research concerning substantive representation of women, the paper here contributes to the scholarship explicitly theorising and testing the relationship between descriptive and substantive relationship in an inversed manner: namely the effect of a female legislator's substantive representation of women on her future presence as an elected politician.
Furthermore, the paper also contributes to legislative and electoral studies by adding much-needed nuance to the existing debates on policy-vote tradeoffs. First, it tests the policy-specificity of electoral consequences by focusing on policy issues directly concerning the interests of female voters and then compares that to the total amount of policy commitment. Considering that women's issues are known to be distant from the policy issues that are at the heart of power politics (Heath et al., 2005), the paper investigates the electoral consequences of legislators' commitment to non-mainstream issues. Secondly, the paper takes an actor-sensitive approach by comparing the electoral effect of legislative commitment by legislator's gender.
To test the policy and actor-specific nature of policy-vote trade-offs, the paper takes Taiwan as its case. Although understudied in the gender politics literature so far, the island nation has seen a dramatic increase in both women's descriptive and substantive representation in the past few decades: women made up 38 per cent of the legislature in 2016, and Taiwan ranked as the eighth-most gender-equal country in the world according to the United Nations Development Programme's Gender Inequality Index (GII) in 2017. In view of this, whether there is a virtuous cycle between legislators' acting for women's interests and, as a result, having greater electoral prospects merits close empirical scrutiny. For empirical verification, I utilise an original bill co-sponsorship record since Taiwan's democratic transition in 1992 until the 2016 general election. The findings based on regressions show that the more legislators commit to women's issue legislation, the less likely they are to be reelected; and this effect was particularly clear among female legislators.
The paper proceeds as follows: First, it introduces relevant debates and expectations about the electoral consequences of commitment to women's issues vis-à-vis total amount of policy commitment, and how these consequences are likely to be gendered. Second, it explains why Taiwan is chosen as the case study, how women's issues are defined in light of existing studies, and why the bill co-sponsorship record is selected as the proxy for legislators' substantive representation effort. Third, based on logistic regressions, the electoral consequence of commitment to women's issues is tested by examining their effect on reelection chances. Finally, the concluding section summarises key findings and identifies potential avenues for future research.

Women's issue and policy-vote trade-offs
Starting from Mayhew's (1974) seminal work on 'electoral connection', students of legislative studies have long noted the electoral incentives associated with legislators' policy efforts. That is, politicians' desire to be reelected in futurethrough which they can continue to pursue power, prestige, and incomehas been known to motivate them to take policy-related initiatives while in office. Among other things, legislative accomplishments on important policy issues can be an effective means of achieving their electoral goals, as it can serve as clear evidence of one's credentials, commitment, and achievements (Mayhew, 1974). Specifically, from the perspective of politicians, showcasing accomplishments matters because the electorate oftentimes cast their votes based on their retrospective judgement of a politician's past performance or prospectively use it as a proxy for politician's intrinsic qualities pertinent to the future representative role (Anzia & Berry, 2011).
In light of the electoral connection thesis, ceteris paribus, it can be concluded that legislators' active engagement with policy issues will in general raise their reelection odds. However, when it comes to specialising in a particular policy area, empirical evidence drawn from both party and individual politician levels suggests that policy commitment can be an electorally risky strategy. Supporting the potential for policy-vote trade-offs, research based on European niche parties points to the importance of balancing between policy specialisation and vote maximisation (Spoon, 2011); that is, niche parties should have sufficiently different policy profiles from old, large parties while, at the same time, not being too distant from potential voters. For individual politicians, given that every legislator has only finite resources and time, committing oneself to the pursuit of particular policy goals largely geared towards a specific group can often mean paying less attention to other ones; and this can be electorally detrimental. As noted in the idea of 'competing principals' (Carey, 2008), a politician elected in a particular district often should go beyond reflecting the interests of one's local constituency and be responsive to others to maximise the reelection prospect. In many occasions, incumbents ought to work for the benefit of their affiliated party to be nominated as candidate or actively represent general voters or organised interests to increase the total vote share.
The paper builds on the policy-vote trade-offs literature, but goes further in taking a policy and actor-specific approach. As for the policy-specificity part, I focus on the electoral implications of commitment to not just any policies but rather ones related specifically to women's issues. There have been several valuable contributions in the gender politics scholarship that examine the potential trade-offs resulting from the specialisation in women's issues. For instance, Franceschet and Piscopo (2008) demonstrate that quotaelected female legislators' emphasis on women's issues can come at the cost of losing the necessary status to legislate successfully. Caminotti and Piscopo (2019) investigate meanwhile whether being a 'critical actor' on women's issues affects one's access to both marginal and prestigious posts. Despite the progress made, to the best of my knowledge, no work to date has explicitly tested policy-vote trade-offs in a granular manner by directly examining the relationship between legislators' amount of substantive representation on women's issues and the degree of change in their subsequent reelection chances. Drawing from gender, legislative, and electoral studies, the paper here aims to fill this gap.
At first glance, women's issues might come across as a policy field where politicians can benefit electorally. After all, female voters make up a large portion of the electorate in most countries and, at the same time, gender equality has become a symbol of modernity carrying positive connotations (Krook, 2007). As a result, even for men, focusing on tasks conventionally reserved for women, such as childcare, increasingly tends to be less stigmatised and in certain cases accepted as a new form of masculinity too (Johansson & Klinth, 2008).
Recent progress in gender equality notwithstanding, I contend that the policy commitment to women's issues will be electorally detrimental to politicians due to the substitution effect. First, promoting women's issues can result in being electorally marginalised, since it is likely to result in them having less access to the powerful networking or resource-mobilising opportunities that are concentrated in the areas focusing on more topics other than women's issuesfor example economy or foreign affairs committees within the legislature (Heath et al., 2005). In other words, high levels of commitment to women's issues can come at the cost of legislators missing out on the resources necessary to reach wider electorates. In light of this, it is not surprising that male legislators often hold gender equality-promoting attitudes in word but avoid supporting legislative measures promoting women in deed (Schwindt-Bayer, 2006). Moreover, findings in the gender and media literature notes that women's issues tend to be a 'ghetto' (Larson & Andrade, 2005;Poindexter et al., 2010)namely, forming only a small portion of media coverage in its entirety. Considering that the media tends to be the primary venue for demonstrating legislators' performance to voters (Barabas & Jerit, 2009;Lyengar & Kinder, 1987), the dearth of coverage dedicated to women's issues effectively means a lack of important credit-claiming or position-taking opportunities for those legislators who are concerned with women's issues. Beyond the media arena, the marginalised nature of women's issues have been reported in various other institutional contexts too (e.g. Annesley, 2010;Escobar-Lemmon & Taylor-Robinson, 2016). All considered, the following hypothesis can be formulated: H1: The more legislators commit to women's issues (over other issues), the less likely they are to be reelected.
Moreover, in regards to the actor-specific part, I expect that the electoral disadvantage associated with commitment to women's issues will be particularly pronounced for female legislators. To be clear, some of the recent findings based on election campaigns indicate that direct voter bias towards female candidates does not exist anymore (Dolan, 2014;Hayes & Lawless, 2016). However, direct hostility that women face by virtue of their gender is not the only way female politicians can be discriminated against. Often known as 'double bind' bias (Teele et al., 2018), women can be disadvantaged in politics if the qualifications and attributes valued in getting elected or re-elected create larger burdens for women than men. I contend that the double bind bias exists for women in the legislative arena since the aforementioned substitution effect pertinent to a commitment to women's issues will be particularly large for female legislators. The related logic can be elaborated as follows.
In highly gendered institutions like the legislature that can be characterised by a deeply embedded culture of masculinity (Lovenduski, 2005), women have been known to face difficulties in securing key leadership positions, campaign funding, or strategic information necessary to further their re-election prospects (Atkinson & Windett, 2019). To offset this disadvantage, female legislators can network with male legislators who often occupy key positions and have ready access to important resources in politics (Bjarnegård, 2013). In this sense, female legislators' commitment to women's issues will take away opportunities to network with male legislators, since women's issues are largely known as being a 'feminine' policy area (Lawless, 2004;Shim, 2020) where we see disproportionately high levels of female presence. The substitution effect arising from a women's issue commitment would be smaller for male legislators because, known as 'possessing homosocial capital', male legislators build trust more easily with the maledominant leadership circle by virtue of shared gender (Bjarnegård, 2013) In light of this, it is not surprising to observe ambitious female politicians trying to follow in male politicians' footsteps both in networking style and in policy substance. For instance, evidence from Latin American countries demonstrates that women are not discriminated in cabinet appointments due to their gender per se, but that those women who retain their elected position or successfully climb the political ladder resemble men in their background and credentials (Escobar-Lemmon & Taylor-Robinson, 2016). Moreover, female legislators trying to approach the core of power are motivated to become 'surrogate men' or 'honorary men' by erasing their womanhood and distancing themselves from women's issues (Escobar-Lemmon & Taylor-Robinson, 2016). Similarly, empirical evidence gathered from South Korea demonstrates that female legislators often reap career benefits by actively networking with powerful male legislators (often leading to them being called 'sunflowers') and that they become senior politicians (often called 'queen bees') by dissociating themselves from their gender (Lee, 2017). In addition to the legislature level, female legislators' commitment to women's issues is expected to be disadvantageous at the voter level as well. Again, the substitution effect is key here. Research shows that female politicians tend to utilise formal and informal networks, for example local or alumni ones, to attract voters to a much lesser extent compared to their male counterparts (Lee, 2017). Added to this, female legislators tend to be in the position of having to juggle not only between public duties but also between public and private ones too. For instance, the latest survey results indicate (Kage et al., 2019) the supply-side barriers on female legislators due to their expected private obligations outside their full-time jobsuch as care for the young and/or elderly. Because fulfilling more private obligations, ceteris paribus, means having less total time for public duties than male legislators, female ones' extra legislative commitment to non-mainstream public policies like women's issues can come at the higher cost of neglecting other public tasks imperative to improving their reelection chances, such as engaging in time-consuming constituency service that requires constant face-to-face interactions (Miura et al., 2018).
All in all, due to the substitution effect, female legislators are expected to face further electoral disadvantages when they focus on women's issues, which leads to the following hypothesis: H2. The more female legislators prioritize women's issues (over other issues), the less likely they are to be reelected compared to male legislators.
To validate the policy-specific nature of policy-vote trade-offs, the paper here also examines the electoral consequence as a result of a legislator's total amount of policy commitment. On this, the related expectation derives from the 'electoral connection' thesis (Mayhew, 1974) which predicts that higher levels of total policy commitment by a legislator lead to higher chances of getting reelected. Although the higher amount of total policy commitment will go hand-in-hand with the higher re-election prospect, we can also assume a gendered pattern due to the double-standard biasthe use of different criteria, formula, and weights for different groups in assessing an individual's productivity (Bertrand & Duflo, 2017). In other words, there will be a performance premium demanded of women in politics. Supporting this idea, a wide range of empirical evidence shows that women who win elections tend to be more skilled, qualified, and better-performing than men. For instance, research shows that, compared to male legislators, female ones tend to bring more federal funding to their districts (Anzia & Berry, 2011) and receive better-quality ratings (Fulton et al., 2006). Furthermore, they not only sponsor and co-sponsor more legislation (Anzia & Berry, 2011) but also tend to score higher in their legislative effectiveness by making proposed legislation go further in the legislation process (Volden et al., 2013).
Moreover, existing empirical evidencealthough few and far betweenshows that the potential of increasing electoral chances as a result of overall policy commitment can be much more limited for female legislators. For instance, based on the rates of challenger emergence in Congressional races, Atkinson and Windett (2019) demonstrate that female legislators can combat gender stereotypes by demonstrating their policy performance. However, to offset their electoral disadvantage what is needed is a substantially higher amount of available time and resources in lawmakingquantitatively introducing more bills that are, simultaneously, qualitatively diverse in nature. In view of the positive electoral effect resulting from the higher levels of total policy commitment in the literature alongside its gendered effect, two further hypotheses can be derived: H3. The higher legislators' total policy commitment level is, the more likely they are to be reelected.
H4. The higher male legislators' total policy commitment level is, the more likely they are to be reelected compared to female legislators Case selection, women's issues and bill cosponsorship data For empirical verification, I have chosen Taiwan as my case study here and will examine the time period between 1992 and 2016. Although the islandnation's legislative politics on women's issues have not received much attention in the existing gender politics literature to date, Taiwan serves as an ideal case to test the aforementioned hypotheses for the following reasons.
First, the presidential constitutional structure in Taiwan renders legislators an important political actor meriting close analytical focus. Established research on executive-legislative relations, such as Tsebelis (2002), demonstrates that there tends to be executive dominance over the legislature on agenda-setting in parliamentary democracies, and vice versa in presidential democracies. In this sense, Taiwan is highly comparable to the previous works this paper draws onsince most of them are based on presidential democracies in the Americas, where members of the legislature play a significant role in shaping the legislative agenda concerning women.
Second, Taiwan is a country where women's descriptive and substantive representation have seen dramatic progress in a relatively short period of time. From its first democratic election in 1992 to the most recent election in 2016, female legislators have substantially increased their presence, moving from 10 to 38 per cent representation. While increasing the proportion of females in the legislature, Taiwan also adopted a mandatory gender quotawhich obliges parties to fill 50 per cent of the party-tier list with female candidatesafter its 2008 general election as well as elected its first female president, Tsai Ing-wen, in the 2016 presidential one. Moreover, many of the elected politicians have successfully pushed for legislative measures whose impact can be felt by a large number of female electorates, such as making divorce and parenting law more women-friendly, protecting mothers' right to breastfeed in public areas by setting up relevant public facilities, or extending the number of days for parental leave. Reflecting the progress made on various fronts, Taiwan was, as noted, ranked eighth in the world in 2017 according to the GII. In this sense, Taiwan is an ideal case to examine whether there exists a virtuous cycle between legislators' acting for women's interests and, as a result, having greater electoral prospects.
Third, the environment in which Taiwanese politicians operate incentivises politicians to maximise their policy performance. In his landmark study, Mayhew (1974) noted the importance of electoral insecurity as one of the key motivations pushing legislators to take policy initiatives. With respect to this, vote security is not guaranteed for a large proportion of incumbents in Taiwan. A clear manifestation of this is, for candidates running for single-member districts, approximately 60 per cent of those winning seats received less than 55 per cent of the vote. 1 Combined with this, politicians are constantly exposed to Taiwan's highly liberal media 2 and their individual-level policy performance is evaluated on a regular basis by monitoring NGOs such as Citizen Congress Watch; the evaluation (which often includes the number of legislative proposals that legislators have made) frequently results in a ranking list of best-and worst-performing legislators publicised in the media. Legislators in Taiwan are cognizant of such external legislative-performance evaluations (Shim, 2016) and are inclined to publicise the positive evaluations they receive via their personal webpages and numerous social media channels.
To evaluate the electoral consequences of Taiwanese legislators' substantive representation efforts regarding women's issues, we first need to clarify what 'women's issues' are. Here, I define them as ones that, for either social or biological reasons, affect women disproportionately more compared to men, or address a condition in which women are particularly more disadvantaged vis-à-vis men (Celis, 2008). Applying this definition, the following categories clearly correspond to women's issues in Taiwan as well as in many other developed democracies: abortion rights, maternity leave, subsidies for mothers, family violence protection, workplace/employment discrimination prevention, female career advancement, maternal healthcare, and gender equality promotion. Beyond this, I also included topics related to the long-term care of the elderly, childcare, and work-family balance as women's issues because women are disproportionately more affected by them in Taiwanwhere women have often been expected to look after the elderly and children at the expense of their careers (Sung & Pascall, 2014).
Oftentimes, women's issues have been approached rather broadly by equating them with policy areas where the topics of concern relate to care, compassion, and civil rightssuch as health, education, welfare, and minorities' rights (e.g. Atkinson & Windett, 2019). These issues are excluded in this paper since including them would make women's issues indistinguishable from 'soft issues' (Bäck & Debus, 2016), many of which do not affect women disproportionately or have masculine characters in the Taiwanese contexts. Health insurance or pension matters, for example, are 'welfare' issues but, at the same time, key 'economic issues' too in Taiwan given the large proportion it takes from the national budget (Shim, 2016); and economics issues are often categorised as 'men's issues' along with military, security, crime, and foreign policy ones (e.g. Hayes & Lawless, 2015).
Having defined what constitutes women's issues, what is left is to select an appropriate data set capturing legislators' related substantive representation activities. Students of gender politics have noted that these representation activities can take various forms, ranging from participating in policy debates and speeches (Clayton et al., 2017;Piscopo, 2011) to introducing bills and voting (Barnes, 2016;Vega & Firestone, 1995). Among other things, I choose here legislators' bill cosponsoring related to women's issues as a key measure of their substantive representation effort in view of the following two reasons.
First, the bill cosponsorship data is used because the most substantial and meaningful changes for female voters that occurred over the past three decades in Taiwan have taken the form of legislationfor instance, laws on parental leave and financial penalties for workplace discrimination. This coincides with the existing view that it is legislation more than any other activity that, either directly or indirectly, influences the lives of female citizens (Celis, 2008).
Second, cosponsoring bills is an indication of politicians' policy preferences and, oftentimes, policy expertise. Proposing a piece of legislation is an opportunity to express support for a certain issue through which a legislator can send signals to voters or fellow party members (Campbell, 1982;Mayhew, 1974); echoing this, evidence from both the US and the Latin American countries has repeatedly confirmed that cosponsorship content and patterns reflect election-induced position-taking (Balla & Nemacheck, 2000;Crisp et al., 2004). Moreover, cosponsorship is a serious and selective legislative initiative in many cases. Cosponsoring a bill implies being responsible for the potential consequences of legislation and, owing to its easy public accessibility, it often becomes a source of criticism or praise from monitoring organisations. The selectivity of cosponsorship has led scholars to using it as a proxy measure for policy specialisation (e.g. Gilligan & Krehbiel, 1997). Analytical advantages from employing bill sponsorship to measure policy preferences and specialisation apply to the Taiwanese case as well. Research shows that, as a presidential democracy, individual legislators there have a lot of discretion in sponsoring and cosponsoring particular legislation but, at the same time, are quite selective about what to support (Shim, 2016(Shim, , 2020. According to the Taiwanese Constitution, both the legislative and executive branches can propose bills. Legislative branch proposed bills can be distinguished between those submitted by individual legislators and those emerging through party caucuses. Although it is not required for executive bills or party-caucus bills, individual-legislator-initiated bills must have the support (i.e. bill co-sponsorship) of 15 or more members of the Legislative Yuan (the threshold was 33 before 2008). 3 Since the primary interest of the paper lies in analysing legislators' cosponsorship record in particular legislative sessions and their electoral result one session afterward, it utilises election records from the first democratic general election in 1992 up to the latest election in 2016 and legislative records from the second legislative session in 1992 up to the eighth legislative session in 2016. During this period, in light of the definition adopted in this paper, a total of 261 women's issue bills (henceforth women bills) were submitted by legislators (see Appendix A for the major examples of women bills and their distribution in Taiwan). Reading the title and bill summaries, it was clear that not all women's issues in Taiwan have a clear feminist orientationbut that they nevertheless try to promote women's interests by providing benefits or removing barriers (and no bill had an explicit anti-feminist orientation). Some 221 bills makes up roughly 2 per cent of the total number of legislator bills -14,567 4but, at the same time, more than 80 per cent of all submitted women bills (see Appendix B for the coding rules and a coding example). These 261 women bills were supported by 5074 cosponsors, while the remaining 14,306 non-women were supported by 381,168 cosponsors.

Empirical results
I test the aforementioned hypotheses by employing logistic regressions while holding other confounding variables constant. In total, there are three explanatory variableslegislators' policy prioritisation of women's issues, legislators' total level of policy commitment, and legislators' genderused to predict the outcome variable 'reelection success'.
Here the 'policy prioritization of women's issue' is measured as the proportion of co-sponsored women bills out of the total number of bills that a legislator co-sponsored during a particular legislative session (continuous variable, designed to capture the 'substitution effect'). The 'total levels of policy commitment' is measured as the number of all cosponsored bills by a legislator within a particular legislative session. 5 And 'reelection success' is measured as whether a particular legislator was reelected in the next general election (not elected 0, reelected 1). Model 1a tests H1 by examining the effect of cosponsored women bill number on a legislator's reelection chance. To investigate H2which expects that H1 is gendered -Model 1b adds an interaction term between legislators' gender (male 0, female 1) and cosponsored women bill number. In a similar fashion, Model 1c tests the effect of all cosponsored bill numbers (continuous) on a legislator's reelection odds to test H3; a legislator's gender and all cosponsored bill number is added as an interaction term in Model 1d to test the gendered effect predicted in H4.
The total number of observations nears 400,000, because I separate each into the unit of 'cosponsor ID-bill ID' in order to explicitly cope with individual and bill-variant confounders in the analysis at the same time (see Appendix C for the illustration of the data structure). With this extended unit form, the subsequent regression analyses directly align with the policy and actor-specific approach this paper takes. Although the observations are confined by cosponsored bills, the data includes all elected legislators during the period of observation since everyone at least cosponsored one bill. 6 Reflecting the data structure, control variables in all models include both 'legislator-specific' and 'bill-specific' factors. For the legislator-specific factors that can potentially affect reelection chances in Taiwan, I included a legislator's elected tier (multi-member district tier 0, single-member district tier 1, party tier 2), elected terms (continuous), party affiliation (multinomial variable), 7 presidential-party affiliation (others 0, presidential party 1) in combination with the latest annual GDP growth rate prior to the general election (continuous). In the case of bill-specific confounders, I included legislation initiative type (amendment 0, enactment 1) and the timing of bill submission distinguished by legislative session (multinomial variable) as well as electoral cycle (others 0, election period 1). 8 The rationale behind controlling for each variable draws from legislative and electoral studies, and can be explained briefly as follows. First, for the elected tier, Taiwan has constantly had a two-tier electoral system wherein legislators can be elected through either the district or party list; however, the district tier shifted from multi-member districts to single-member districts after the 2006 electoral reform. Previous findings note that singlemember districts offer better incumbency advantages than multi-member districts do (e.g. Berry et al., 2000), therefore these two types of districttier legislators are distinguished. Besides, the party-tier legislators are also separated out, since in Taiwan the two major partiesthe Chinese Nationalist Party (KMT) and Democratic Progressive Party (DPP)largely did not employ re-nomination practices for part-tier candidates; such legislators seeking reelection had to search for a new district for the next election, which is substantially more difficult than it is for incumbent district-tier legislators. Second, elected terms are controlled for following the expectation that more political experience tends to have a positive effect on the likelihood of reelection (Pereira & Renno, 2003). Third, in light of the evidence demonstrating that electoral support tends to differ substantially between parties (Kenny & Verge, 2016), legislators' party affiliations are distinguished. Fourth, based on the extant research showing that a better national economy is electorally favourable for presidential-party candidates (Hibbing & Alford, 1981), an interaction term combining a legislator's presidential-party affiliation status and the national GDP growth rate prior to the election is included.
Fifth, legislative initiative type is included since enacting a billmaking a bill from scratchis, in general, substantially more time-consuming and politically tricky than amending a bill whose change is based on existing legislation. Therefore, I expect the electoral impact will be higher because it usually takes more efforts on the legislator's side and intended legislative changes are more significant and salient for enacting rather than amending a bill. Sixth, a legislative session dummy 9 is included to account for the effect of session-specific factors, or the over-time trend affecting legislators' reelection chances. Seventh, electoral cycle is considered since it has been long noted that voters reward incumbents for their positive policy records closer to the next election particularly more so than for other time periods (Nordhaus, 1975). 10 Appendix H includes the summary statistics and sources for key variables; for all models, the numbers included in parentheses indicate robust standard errors.
To begin with, the results in Model 1a included in Table 1 show that higher legislative commitment to women's issues is negatively related to a legislator's future reelection success (statistically significant at the 1 per cent level). This supports H1. Moving from the lowest to the highest proportion of cosponsored women bills bills -0 and 17 per cent respectively the reelection success rate drops quite substantially from 45.67 to 29.9 per cent (Figure 1  Note: ***p < 0.01; **p < 0.05; *p < 0.1.

left). From other included variables, what is interesting
is that being a female (rather than being a male) legislator makes an individual more likely to be reelected; although it is not the key concern of this paper, this rather counterintuitive pattern might be attributed to the fact that in Taiwan female legislators tend to come from 'political families' 11significantly more so compared to their male counterparts there (Huang, 2019). 12 Model 1b tests whether and to what extent legislators' commitment to women's issues have gendered electoral consequences. Confirming H2, the interaction between women bill proportion and legislator's gender is statistically significant and negative (at the 1 per cent level), demonstrating that the negative electoral effect of prioritising more women bills is particularly strong for female legislators. Specifically, the predicted probability of getting re-elected drops much more for female legislators (from 48.2 to 23.6 per cent) than male legislators (45.08 to 31.94 percent). Model 1c shifts the explanatory variable from women bills to all bills cosponsored by Taiwanese legislators. In line with H3, the findings show a clear positive electoral connection to general policy commitmentsmoving from the lowest to the highest numbers of cosponsored bills, a legislator's chances of reelection increases more than 20 per cent: from the lowest 36.7 per cent to highest 58.6 per cent. As illustrated in Figure 1 (right), the electoral consequences of committing to policy issues overall can be clearly contrasted to prioritisation of women's issues (Figure 1 left). Finally, Model 1d tests the expected gendered effect of more general policy commitments; it Note: The predicted probability of all logistic regression models in the paper are calculated based on the average-adjusted prediction method. For example, in calculating the predicted probability of re-election chance difference between the lowest and the highest proportion of women bill co-sponsorship -0 and 17 per cent respectivelyin Figure 1 (left), the model sets women bill proportion = 0 in every observation in the dataset and calculates a predicted probability for each observation as if it were women bill proportion = 0 per cent but otherwise the same as it originally was. Then it averages those predictions and outputs the average. The same process repeats for women bill proportion = 17 per cent; the minimum and maximum predicted probability of re-election rate I included in the text, i.e. 45.67 and 29.9 per cent respectively are based on this calculation.
demonstrates that the potential electoral gains hereof are substantially less if a legislator is a woman rather than a man. Namely, the predicted probability of being electorally successful increases by 27 per cent for male legislators (from 35.33 to 62.08 per cent) but only by 6 per cent for female ones (from 42.13 to 48.4 per cent). This clearly supports H4.
So far, the paper has defined reelection success as whether a particular legislator in a specific legislative session was subsequently reelected. However, to examine if the positive or negative effects of legislative commitment (both all bills and women bills) that can be observed in the previous models are to be found at the candidate-selection stage (which is more under party-level influence) and/or at the legislator-selection stage (which is more under voter-level influence) further regressions were ran in Appendix D. Specifically, I ran separate regressions by dividing the dependent variable into two stageswhether a particular legislator in a specific legislative session was subsequently running as a candidate (Model 2) and whether the candidate was reelected (Model 3). Judging by the fact that the key results are identical (in terms of both direction and statistical significance) to the main model, the trade-offs occur at both stages.
All results presented in this paper are based on the cosponsors of submitted bills. However, ultimately, what gets rewarded or punished more for reelection can be only those cosponsored bills that actually brought real legislative changes. 13 Therefore, Appendix E includes results based only on the number of cosponsored bills that led to successful legislative outcome, i.e. becoming a law. Third, each bill varies in terms of the amount of budget attached, the degree of media saliency, and the scope of voter coverage all of which can potentially affect the legislator's effectiveness in credit-claiming for re-election. Considering this, Appendix F includes regressions with bill-fixed effect. All these robustness tests resulted in identical findings to the main model. Finally, considering that women's seat share in Taiwan passed the 30 per cent from the 8th legislative sessionoften known as the threshold of 'critical mass' (Dahlerup, 1988)separate regressions were conducted in Appendix G on H1 and H2 with this period and the time prior to that. The results show that although the electoral disadvantage of prioritising women's issues per se have weakened it still manifests in a strongly gendered form.

Concluding remarks
Drawing insights from the previous research in gender, electoral and legislative studies, the paper has examined whether the amount of legislative commitment that legislators give to women's issues comes at the cost of their electoral success. Using an original bill cosponsorship dataset and electoral records from Taiwan, the paper has demonstrated that legislators' policy-vote trade-offs are negative for women's issues and furthermore vary according to legislators' gender, calling for a more policy and actor-specific approach in the analysis hereof. This sheds light on a hitherto understudied aspect of the relationship between the substantive and descriptive representation of women. And, by doing so, it clearly adds to the extant gender politics literature which touches on the career implications of representing women as an elected politician (e.g. Caminotti & Piscopo, 2019).
Taiwan's remarkable strides in women's descriptive and substantive representation notwithstanding, the results show that prioritising women's issues over other issues came at steep electoral costs for such legislatorsand particularly so for female ones. These policy-vote trade-offs are concerning for numerous reasons. First, they hamper the accumulation of legislative expertise on women's issues by increasing the dedication costs of pertinent policymaking. Second, for critical actors promoting women's interests, getting elected as a legislator and working in the legislature will be considered no more than a temporary stepping stone for other career paths. Third, as long as the key to maintaining power lies in assimilating to masculine culture and prioritising (stereotyped) masculine issues, we cannot expect a re-gendering of state institutions and processes to occur though high levels of women's descriptive representation.
Building on this paper, future research can pursue the following avenues. First, to add further depth to the policy-specificity of policy-vote trade-offs; the degree of trade-offs across various policy areas could be explicitly compared side-by-side. Particularly relevant areas of comparison will be between more mainstream policy areas related to hard issues like economy, construction, or diplomacy and rather marginalised soft issue areas such as gender, human rights, or the environment.
Second, beyond the substitution effect, other potential mechanisms behind the policy-vote trade-offs can be tested. Among others, another promising mechanism is the 'negative branding' effecti.e. strong emphasis on women's issues can be stigmatising. For instance, legislators can be labelled as feminists and that can be detrimental since many feminist platforms tend to challenge entrenched interestssuch as patriarchal, religious, and class onesand demand the equal distribution of economic and political power (Htun & Weldon, 2010). High levels of commitment on women's issues can also lead to the pigeonholing of a legislator as 'a representative of a specific social group' (Lee, 2017); this branding can be problematic if she wants to be nominated as a representative of a particular local district.
Third, although not dealt with explicitly in this paper, the conditions underlying both district-and party-tier legislators in Taiwan coming into office took various forms (e.g. different district sizes at the district tier, or different seat allocation methodsbased on the party's seat share in the district tier or direct votes by the publicat the party tier) throughout the period of observation. Future works could explicitly theorise and test then how legislators' policy-vote trade-offs manifest under specific institutional arrangements.

Notes
1. Author's own calculation, based on data provided by the Election Study Center at National Chengchi University https://esc.nccu.edu.tw/PageDoc/Detail?fid= 7802&id=6964 2. For a population of 23 million people, the country has seven 24-h news stations (compared to three each in the US and the United Kingdom) further to about 2500 newspaper publishers. Moreover, Taiwan is ranked as Asia's freest media according to the World Press Freedom Index (2015). 3. Although the number varies over time, there were approximately 225 elected legislators before 2005, which has halved to 113 since the 2008 general election. For further details on Taiwan's legislative process see https://www.ly.gov.tw/ EngPages/Detail.aspx?nodeid=335&pid=43232 4. Cosponsorship records are based on the Legislative Yuan Legislation Search (https://lis.ly.gov.tw/lydbc/lydbkmout). 5. The number of a legislator's cosponsored bills within a legislative session ranges from 0 to 65 for women's issues and from 1 to 1913 for all bills. 6. However, there are some legislators (22 per cent) who did not cosponsor any women bills. For these legislators, the number of cosponsored ones is thus included as '0'. 7. During the period of observation, less than 1 per cent of elected legislators were elected as independents; instead of removing them, they are treated as a separate category here. 8. The 'election period' includes both general and presidential elections, and is calculated as a 'year before an election day' to 'the election day'. In Taiwan , general elections were held in 1992, 1995, 1998, 2001, 2004, 2008, 2012 and 2016, while presidential elections were held in 1996, 2000, 2004, 2008, 2012 and 2016. After the 1997 constitutional reform, Taiwan clearly moved from a presidential to a semi-presidential democracy in which the legislature can technically be dissolved. However, due to high coordination hurdles between parties and personal costs of running elections, the Legislative Yuan has not been dissolved so far and general elections have always been held at the scheduled time (Wu & Tsai, 2011). 9. Between 1992 and 2016, Taiwan held seven legislative sessions with three-year intervals until 2008 (2, 3, 4, 5, 6 sessions) and four-year intervals thereafter (7, 8 sessions). 10. Due to a shortage of data, other factors known to affect legislators' reelection chances are left out in the model. Among other things, they are available legislative resources: often defined as the level of legislative professionalism (Berry et al., 2000), personal reasons like a legislator's health or family issues (Hines & Napier, 1985), or the assumption of leadership roles (Pereira & Renno, 2003). 11. This term refers to a legislator having at least one family member who is a politician, usually one's father or brother, but occasionally mother, father-in law, or uncle. 12. Other noteworthy patterns are that legislators elected from the multi-member district tier are less likely to be reelected in the next election than in the single-member district-tier, while more likely to be elected than their party-tier counterparts. 13. The extant gender politics literature separates policy promise and policy delivering by distinguishing between the substantive representation of women in the legislation process and in legislation outcome (Franceschet & Piscopo, 2008).

Disclosure statement
No potential conflict of interest was reported by the author(s).

Notes on contributor
Jaemin Shim is an assistant professor at Department of Government and International Studies at Hong Kong Baptist University. His primary research interests lie in democratic representation, comparative welfare states, gender and legislative politics. His works have appeared or are forthcoming in international journals including Democratization, Parliamentary Affairs, European Political Science, and Journal of Women, Politics and Policy.