This is the accepted version of the journal article: Font, Nuria. «Competing principals and non-vote decisions in the European Parliament». Parliamentary Affairs, Vol. 73 Núm. 1 (january 2020), p. 166-185. DOI 10.1093/pa/gsy043 This version is available at https://ddd.uab.cat/record/257639 under the terms of the license 1 Competing Principals and Non-Vote Decisions in the European Parliament1 Nuria Font Universitat Autònoma de Barcelona Summary This paper analyses the effects of competing principals on legislators’ decisions not to vote in the European Parliament. We argue that Members of the European Parliament (MEPs) are likely to decide not to vote to avoid defecting from either the national party or the European political group when both political principals disagree. Moreover, the paper demonstrates that competing demands between principals interact with the expected closeness of a vote. MEPs are more likely to opt for not voting when they have few chances to influence the vote outcome and are torn between the two main principals. Based on a novel dataset on individual votes in the 2009-2014 term, this paper demonstrates that competing demands moderate the effect of the expected vote closeness on non- vote decisions and highlights the need to incorporate this type of legislative non-response in future research. 1 This manuscript is the accepted version published at Parliamentary Affairs 73(1): 166–185. https://doi.org/10.1093/pa/gsy043 2 Keywords: competing principals, European Parliament, legislative behaviour, not-voting, representation 3 Literature on legislative voting behaviour has abundantly addressed the question of why legislators defect from their political principals when facing a vote (Carey 2007, Close and Gherghina 2018, Costello and Thomson 2016, Faas 2003, Hix 2004, Kam 2009, Klüver and Spoon 2013, Meserve et al. 2017, Sieberer 2006). Other studies have investigated legislative non-response, be it by abstentions or by not casting a vote (Cohen and Noll 1991, Jones 2003, Mühlböck and Yordanova 2017, Noury 2004, Poole and Rosenthal 1997, Rosas and Shomer 2008, Rosas et al. 2015, Rothenberg and Sanders 2002). However, the question of whether legislators strategically choose not to vote while being in the chamber when faced with competing principals has been underexplored. This study addresses the issue of whether Members of the European Parliament (MEPs) decide not voting despite being present in the chamber (Present but Not Voting, PNV)i to avoid defecting from either the national party or the European political group (EPG) when both principals disagree. We focus on the EP as, contrary to most legislatures in Western democracies, recorded voting data in the EP distinguishes PNV from abstention, which is a type of explicit vote, and absence. This makes the EP a unique case for the study of PNV as a distinct, observable and measurable type of legislative behaviour. Comparative research on legislative behaviour has commonly assumed that abstentions (or absences) are not random. Scholars frequently conclude that the expected closeness of a vote, and hence the chances that a legislator is pivotal in the vote outcome, is an important determinant of abstentions (Cohen and Noll 1991, Jones 2003, Noury 2004, Poole and Rosenthal 1997, Rothenberg and Sanders 2002). In the EP context, theoretical accounts for legislative participation and PNV indicate that MEPs’ capacity to influence the voting results is the main motivation behind their decision to vote (Hix et al. 2007, Noury 2004). In particular, PNV decreases when the margin 4 of victory on a vote is expected to be narrow and when legislation is adopted through co-decision, by which the European Parliament and the Council are co-legislators (Noury 2004). However, the question of whether PNV in the EP may serve legislators to avoid defection when facing loyalty dilemmas between the two main principals − the national party and the EPG − has received scant attention. To fill this gap, this study partly builds on the ‘competing principals’ approach to vote unity (Carey 2007, Meserve et al. 2017) and non-response (Jones 2003, Mühlböck and Yordanova 2017, Rosas and Shomer 2008, Rosas et al. 2015). The analysis demonstrates that MEPs are more likely to choose PNV when faced with competing pressures from principals to which they are accountable. More importantly, the analysis nuances the standard account in the literature by demonstrating that competing demands between the two main principals interact with the expected closeness of a vote. MEPs are more inclined to decide not to vote when they have few chances to be pivotal in the vote outcome and are torn between the two main principals. However, when the vote is expected to be close, MEPs have fewer incentives to choose PNV, irrespective of whether the principals disagree or not. The analysis aims to contribute to the literature on non-response in the EP by arguing that dissent between political principals moderates the effect of the expected vote closeness on PNV. The analysis corroborates the idea that PNV is largely driven by strategic calculations and highlights the need to avoid ignoring this type of non-response in future research. A further concern of this study is whether PNV in the EP should be differentiated from explicit abstention. In addition to the fact that the former involves not voting, while the latter involves casting a vote, if certain MEPs choose PNV and others choose explicit abstention, it is plausible to assume that the motivations behind such choices may not necessarily be the same. This study performs robustness 5 checks generally suggesting that the accounts for PNV and abstentions are not outstandingly different. Following previous studies on non-response in the EP (Mühlböck and Yordanova 2017), this analysis focuses on votes adopting legislation under the standard legislative procedure (hereinafter, co-decision) during the 2009-2014 legislature and discards other types of procedures. The main reason is that co-decision has become the standard legislative procedure in the EP following the entry into effect of the Lisbon Treaty in 2009. We assume that final co-decision votes are the most important, as they serve to adopt most legislation. At the same time, final votes reflect whether MEPs side with the political principals’ overall position on a legislative act. By analysing the Seventh term, the study benefits from the 2009 change in the EP Rules of Procedure, according to which the Parliament shall take all final votes by roll call.ii Hence, the study avoids the selection bias problem allegedly associated with vote request in previous parliamentary terms (Carrubba et al. 2006, Yordanova and Mühlböck 2015). The article is structured as follows. The first section presents a brief description of the different voting options in the EP and the incidence of PNV. Then, the theoretical debates on which this study is built and the derived hypotheses are presented. Next, the data and operationalisation of the variables are described. The following section presents the results of the empirical analysis, which is followed by a discussion of the results and a conclusion. 6 1. Voting and not voting in the EP Comparative research on legislative behaviour has commonly ignored abstentions and absences, either because abstentions may not exist or because abstentions and absences constitute a single category in many legislative chambers (Mühlböck and Yordanova 2017). As a result, abstentions (and/or absences) are commonly coded as missing values in roll-call data analysis (Rosas et al. 2015). The voting system in the EP allows for distinguishing between different types of behaviour. When it is time for voting, MEPs may explicitly vote ‘For’, ‘Against’, or ‘Abstain’; they may register attendance but not vote (PNV)iii; or they may be absent from the chamber, for instance, because of illness or because they are elsewhere on a mission. This study focuses on PNV as a distinct type of non-response in the EP. The incidence of PNV in the EP is small but non- negligible. According to our data, on average seven percent of MEPs chose PNV in votes adopting legislation under co-decision during the 2009-2014 period. Note that the occurrence of PNV is twice that of abstentions and, for the three groups whose members most frequently are in the winning coalition, PNV is considerably higher than votes against. Table 1 displays the share of different types of behaviour by MEPs attending plenary sessions by EPGs.iv Table 1 7 2. Competing pressures and non-vote decisions Comparative research on legislative behaviour has commonly assumed that not voting is not random, but instead largely driven by legislators’ strategic motivations (Cohen and Noll 1991, Jones 2003, Poole and Rosenthal 1997, Rothenberg and Sanders 2002, Rosas et al. 2015). In the context of the EP, Noury (2004) finds that MEPs have a higher probability of choosing not voting when they have more chances to play a pivotal role in the voting results. Most studies on legislative turnout in the EP are in line with this finding and show that participation on a vote is more likely to occur when there is more uncertainty about the results and legislators are more likely to influence the policy outcome (Hix et al. 2007). A recent contribution finds that electoral institutions moderate the effect of career ambition on legislative participation in the EP (Høyland et al. 2018). The specialized literature also provides EU-specific accounts for non-participation. MEPs’ turnout in legislative votes increases when the EP has more power to influence policymaking (Hix et al. 2007, Scully 1997). That is, not-voting is lower when the legal act is adopted under co-decision, where the EP co-legislates with the Council on an equal footing, as compared to consultation. Indeed, as the EP has gained legislative powers, not-voting has considerably decreased (Noury 2004) and, consistently, turnout has increased. According to data from Votewatch, participation in roll-call votes was 80.6% in 2004, 84% in 2009 and 88.7% in 2014.v Previous research has contributed to better understand the motivations behind MEPs’ decision to vote when they attend plenary sessions. However, the question of whether PNV in the EP is a purposeful response by legislators faced with competing demands from their main party principals has been overlooked in the literature. The competing principals’ theory proposes that when 8 multiple principals control resources to influence legislators’ votes, disagreement between principals reduces party unity (Carey 2007). The logic of competing principals may also affect abstentions and absences. A recent study by Rosas et al. (2015) simulates models of non-response and concludes that models assuming random missingness generate biased inferences in estimated ideal points compared to models incorporating the logic of competing principals. Building on the ‘competing principals’ theoretical accounts for party unity (Carey 2007) and non-response (Longley 2003, Mühlböck and Yordanova 2017, Rosas et al. 2015), this paper argues that MEPs are more likely to opt for not voting when facing cross-pressures from the national party and the EPG. For the elaboration of our argument, we share the widely accepted assumption in the literature that MEPs are agents of two principals: the national party and the EPG (Hix 2002, 2004). Either principal disposes of mechanisms to monitor legislators’ activity and hold them accountable. National parties control candidates’ selection and career prospects, whereas transnational groups control the distribution of policy visibility and the assignment of institutional positions among MEPs. Several studies indicate that when MEPs are faced with confronted pressures from both principals, they are more likely to follow the instructions of the national party (Faas 2003, Hix 2004, Klüver and Spoon 2013, Trumm 2015). A recent contribution by Mühlböck and Yordanova (2017) also shows that MEPs may strategically vote abstention when torn between competing principals, namely the national party and the EPG. Their study implicitly assumes that MEPs are aware of the existence of dissent between the two main principals prior to the vote. We share this assumption in the understanding that, even though MEPs may not receive voting instructions from the national party and the EPG for all voting occasions, it is highly likely that before the vote they receive information on the party/group positions on legislative dossiers from key actors, for 9 instance national party delegation leaders, shadow rapporteurs or group coordinators. We check whether competing pressures between the national party and the EPG affect PNV. We hypothesize that PNV in the EP is higher when the national party and the EPG disagree on the direction of a particular vote. MEPs subject to competing pressures may prefer to refrain from voting in order to avoid position-taking and hence displease one of the two main principals. That is, legislators may strategically opt for PNV as a cost-minimizing strategy when facing a loyalty dilemma. Ultimately, MEPs facing contradictory pressures from the national party and the EPG could plausibly argue that they involuntarily missed a vote while indeed did not vote intentionally as a way to avoid potential sanctions derived from defecting one of the two party principals. Hence, we propose: H1. MEPs are more likely to choose PNV when the national party and the EPG disagree than when the national party and the EPG agree The literature on voting behaviour has demonstrated that the electoral rules by which candidates are elected influence legislative party unity (Carey 2007, Cordero and Coller 2015, Däubler and Hix 2017, Hazan and Rahat 2010, Hix 2004, Itzkovitch-Malka and Hazan 2016). The general argument is that in closed lists systems, legislators are more inclined to satisfy the national party because party leadership controls key resources critical to legislators, including career prospects for re-election. By contrast, in open list systems, where candidates compete with other candidates for voter support, legislators have incentives to cultivate a personal vote to increase their chances for re-election (Carey 2007). As for the EP, current research also demonstrates that MEP’s defection is affected by electoral rules (Faas 2003, Finke 2016, Hix 2004). National parties are more likely to impose vote discipline on MEPs elected under closed and blocked lists (Hix 2004). Hence, MEPs narrowly scrutinized by party leaders have incentives to follow party instructions 10 (Hix 2004). In contrast, MEPs elected under open or unblocked lists are less constrained by party leadership and have more incentives to build a personal profile to be re-elected, even though voters are scarcely concerned with EU policy-making (Finke 2016). Contributing to this literature, Høyland et al. (2018) find that legislative participation in plenary voting in EP5 is higher for legislators who want to continue in the EP if they originate from party-centered than candidate- centered electoral systems. On this basis, we investigate whether the list system affects PNV. We hypothesize that MEPs elected under closed lists are less likely to choose PNV as they have more incentives to please national party leaders as active legislators that perform their legislative duties. Furthermore, we also hypothesize that the effect of the list system on PNV is affected by the presence of competing pressures between the national party and the EPG. When both principals disagree, the national party would value that its members in the EP support the national party position. This would lead MEPs selected under closed rather than open list systems to avoid displeasing the national party and, hence, to vote rather than not voting. We propose: H2. MEPs elected under closed lists are less likely to choose PNV than MEPs elected under open lists H3. MEPs elected under closed lists are less likely to choose PNV than MEPs elected under open lists when the national party and the EPG disagree Finally, comparative research on legislative non-response demonstrates that legislators’ greater perception of having an impact on the outcome, typically occurring when the expected margin of victory on a vote is narrow, increases parliamentary turnout (Cohen and Noll 1991, Rothenberg and Sanders 2002). In the EP, Noury (2004) shows that MEPs are less likely to choose PNV when they have a greater chance to play a pivotal role in determining the vote outcome. While Noury’s 11 (2004) analysis is probably the most relevant contribution to the study of PNV in the EP, it takes the aggregate number of PNV as a dependent variable. We check the standard prediction in the literature by performing an individual-level analysis and expect to confirm that legislators’ chances to influence the outcome reduce PNV. We follow well-established research on legislative behaviour in assuming that MEPs may anticipate conflict on a vote and, as a result, their chances to be pivotal (Hix et al. 2007). The assumption is based on the idea that prior to the final vote, MEPs are normally informed by national party and group leaders of the degree of conflict and the composition of the winning majority on a given legislative dossier. If legislators can anticipate the divisiveness of a vote, they can also assess their chances to be influential. Turning to our argument, we purport to step forward in the literature and test whether the effect of the expected vote closeness on PNV is affected by the presence of competing pressures between the national party and the EPG. We hypothesize that when the expected margin of victory is high, MEPs are less inclined to cast a vote if the two main principals disagree than if the two of them agree. With few options to influence the outcome, MEPs torn between cross-pressures may prefer missing a vote as a way to minimize potential sanctions of being disloyal to one of the two principals. However, as the expected margin of victory is narrower and MEPs increase their chances to influence the outcome, the differences between agreement and disagreement tend to be reduced. We propose: 12 H4. MEPs are less likely to choose PNV when the vote is expected to be close than when not expected to be close H5. When the vote is not expected to be close, MEPs are more likely to choose PNV when the national party and the EPG disagree than when the national party and the EPG agree. However, as the expected vote closeness increases, differences in the probability of PNV between principals’ agreement and disagreement are reduced 3. Data and methods We base our analysis on a novel dataset including individual votes adopting legislative acts under co-decision during EP7. The votes correspond to 395 legislative acts, which represent more than 95 percent of the entire population. Following previous studies on non-response in the EP, we focus on final votes adopting legislation and discard votes on amendments (see for instance Mühlböck and Yordanova 2017). We assume that final votes are the most relevant as they serve to adopt legislation. In addition, final votes reflect MEPs’ position on a legislative act as a whole and thus whether a legislator sides with the political principals. Besides passage votes, legislative proposals voted and passed in first reading during the 7th term and later withdrawn by the Commission are considered. We perform several robustness checks by running models including 37 additional votes on dossiers withdrawn further on. We discard legislative votes in first reading that failed as they represent less than one percent of the total votes, according to data cross-checked from Votewatch and the Legislative Observatory of the European Parliament. 13 General data on legislative dossiers were coded taking information from the Legislative Observatory of the European Parliament while data on individual votes were retrieved from Votewatch. Following several restrictions with the data (see below), the analysis was performed with roughly 200,000 observations. Our dependent variable, PNV, is a binary variable indicating whether an MEP did not vote despite being present (=1) or voted (=0). Absences are coded as missing cases. Since much of the literature on legislative non-response has treated PNV and abstentions similarly (see for instance Rosas et al. 2015), we perform robustness checks by taking PNV or explicit abstentions as a dependent variable. We operationalise the independent variables as follows. For competing demands, we first follow Hix (2004) by defining national party and group positions as the voting behaviour of the majority of their respective members. Partly based on previous operationalisations of competing demands (Mühlböck and Yordanova 2017), we create a binary variable indicating whether the majority of MEPs from the national party and the majority of those from the EPG vote differently (vote differently=1; vote the same=0). We exclude MEPs from national parties with less than three members and those not attached to any group. For the calculation, ties on the majoritarian position of national parties, which represent less than two percent of the observations, are coded as missing cases. For electoral list system, we create a binary variable that takes the value one when the list system is closed and blocked, and zero otherwise. Following Hix et al. (2007), we take vote closeness as a proxy for legislators’ expected chances of influencing the outcome. It is calculated by subtracting the number of votes against from the number of votes in favour. Then the difference is divided by the addition of votes in favour and against, and the result is subtracted from one. The measure 14 ranges from zero to one, where zero indicates no votes against and one reflects an equally divided vote (see Noury 2004). Since the study focuses on votes passed, there are no observations with a value of one. We introduce a series of controls. Based on the idea that MEPs are more likely to defect from the EPG when the national party is in government (Costello and Thomson 2016, Klüver and Spoon 2013), we control for whether MEPs’ national parties are in government at the time of the vote (in government=1; otherwise=0). Data is extracted from the ParlGov database.vi While MEPs are subject to higher pressures when their party is governing, such pressures might on occasion result in not-voting. Given that MEPs from parties with a high presence in the EP behave differently from those with a low seat share (Finke 2016, Klüver and Spoon 2013), we add a control for the national party delegation size, which is operationalised as the number of seats of national parties in the EP. While debatable, our first expectation is that large parties decrease PNV. Following Finke (2015), we also control for whether an MEP’s group provides the rapporteur. The expectation is that PNV decreases under such condition. Based on the idea that salience affects party defections in the EP (Klüver and Spoon 2013), we control for the salience of the legislative dossier. We follow Mühlböck and Yordanova (2017) and employ a measure of media salience indicating whether the legislative dossier received at least three mentions in Agence Europe during the year of the vote. As MEPs’ concerns should be higher for relevant dossiers, we expect that salience incentivises turnout and decreases PNV. We also add controls for MEP’s gender, age, and terms served in the EP. Fixed effects for EPGs and for EP committees are controlled for. The Supplementary Table S1 summarizes the descriptive statistics of the dependent, independent, and control variables. 15 4. Results Our dataset is hierarchically structured, as vote choices are nested within MEPs who in turn belong to parties, which are members of EPGs. To avoid violating the assumption of independence of observations, we perform a multi-level model including MEPs and national party levels. Given the small number of EPGs, they are included as fixed effects. As the dependent variable is dichotomous, we perform a multi-level logistic regression. The results are reported as odd ratios and are presented in Table 2. We estimate five models. Model 1 includes the three main explanatory variables proposed in this study. Model 2 adds the controls. Based on Model 2, we calculate the marginal effects of the independent variables on the probability PNV, holding continuous variables at their means and binary variables at their modes (the latter take the value 0). Model 3 incorporates two interactions: national party-EPG disagreement is interacted with the list system as well as with the expected vote closeness. Model 4 adds fixed effects for EPGs, taking The Greens/EFA as a reference category. Finally, Model 5 includes committee dummies, not reported for reasons of lack of space. Table 2 The results provide strong support for H1 on competing demands between the national party and the EPG, as expected. Disagreement between the two political principals attains statistical significance across models. The odds ratio indicates that the effect of principals’ dissent on PNV 16 is positive and high. That is, the presence of competing principals significantly and substantially increases PNV in final co-decision votes. In other words, legislators in the EP are considerably more likely to opt for not voting when they face contradictions between the national party and the EPG than when not. The predicted probability of PNV when the two principals agree is four per cent and increases up to 21 per cent when they disagree (see Table 3). Contrary to our expectations, H2 on the effects of the electoral list systems on PNV is not supported. As regards H3, the interaction between competing principals and the electoral list system attains statistical significance in all models. For a better assessment of the effect of the interaction, and based on Model 3, we plot the predicted probability of PNV for electoral rules when the national party and the EPG disagree, holding continuous variables at their means and binary variables at their mode. As shown in Figure 1, given disagreement between principals, the predicted probability of PNV is 27 percent when MEPs are elected under open lists and is reduced in six percentage points when they are elected under closed lists. While, given disagreement, MEPs elected under closed lists as compared to open lists would be less likely to choose PNV, H3 cannot be clearly confirmed as confidence intervals slightly overlap. That is, the probability of PNV when the two principals hold conflicting positions is not significantly different for MEPs elected under closed and blocked lists as compared to lists allowing voters expressing preferences. Regarding H4, the anticipated closeness of a vote significantly decreases the probability of PNV in all models, corroborating the standard prediction in the literature. Beyond this confirmation, the results indicate that the expected vote closeness interacts with party-group dissent, as expected in H5. The interaction between closeness and party-EPG disagreement attains statistical significance 17 across models. To better illustrate the interaction effect between dissent and the expected closeness on PNV, we simulate the predicted probabilities. Based on Model 3, Figure 2 plots the predicted probability of PNV at different levels of the expected closeness when the national party and the EPG disagree, holding all other variables at their mean (continuous variables) or their modes (binary variables). The effect of the expected closeness of a vote on PNV is affected by the presence of contradictions between the national party and the EPG. The probability of PNV rises substantially when closeness is at its minimum value if the two principals disagree. At the minimum level of closeness, the predicted probability of PNV is roughly 34 percent when the two principals disagree and five percent when they agree. That is, the predicted PNV increases by 29 percentage points for MEPs facing contradictory pressures. As the expected closeness increases, the predicted probability of PNV when both principals disagree is steadily reduced and ends up converging at three percent at the maximum level of closeness. Between the minimum and maximum values of vote closeness, the predicted probability of PNV decreases by 31 percentage points when the principals disagree and by two percentage points when they agree. Figure 1 Figure 2 18 The main results are mostly maintained when introducing controls. MEPs whose party is in national government at the time of the vote are significantly more likely to choose PNV, although the odds ratio shows that the effect is very weak. The results suggest that higher pressure on MEPs when national parties are governing translates into slightly higher levels of PNV. Neither does national delegation size nor rapporteurship impact on PNV. We cannot confirm that MEPs are significantly less likely to avoid voting when the legislative dossier is relevant. While the odds ratio in Models 1 to 4 suggests that salience significantly reduces PNV, it approximates one and the variable does not attain statistical significance in Model 5. The results remain almost unchanged when adding controls for gender, age and terms served in the EP. Gender (female) slightly decreases PNV and the terms served increase it. Note that the size of the effect of the latter is not low. The predicted probability of PNV between the minimum and maximum levels of seniority moves from four to eight percent. Based on Model 2, Table 3 reports the predicted probability of PNV for the independent variables and controls that reached statistical significance in all models, at their minimum and maximum values. Table 3 The results do not change when fixed effects for EPG are included. As shown in Models 4 and 5, no significant effects are found for political groups, except for EFD. Members from EFD have a significantly lower probability of avoiding a vote than members from The Greens/EFA, which is our reference category. The main results remain stable when adding fixed effects by responsible 19 committee, except for salience, as noted earlier. PNV significantly increases when legislative dossiers deal with budgetary issues, international trade, legal affairs or civil liberties. On the contrary, PNV decreases for policy issues dealing with economic issues, internal market and consumer protection, transport and regional development. We perform several robustness checks. We first replicate the analysis taking final votes as well as votes on legislative proposals in first reading that were further withdrawn by the Commission, covering 227,732 individual votes on 432 legislative dossiers. The results remain virtually unaffected. In light of previous studies on non-response treating abstentions and not-voting similarly, we perform robustness checks incorporating abstentions in the analysis. We run the analysis by using a dependent variable taking the value one if the MEP explicitly abstained or was present but did not vote, and zero if the MEP voted in favour or against. That is, we treat both types of non- response together, as opposed to votes in favour or against. The results are reported in Table 4. In broad terms, the results are similar to those obtained when the dependent variable is PNV. The effect of principals’ dissent on Abstentions/PNV is positive, high and statistically significant in all models. Note that the odds ratio indicates that the effect is higher when the dependent variable is PNV or abstaining than when it is just PNV. The results also show that the effect of both interactions on the probability of Abstention/PNV and on the probability of just PNV is also similar. Still, some differences are also observed. When the rapporteur is from the same group as the MEP, the probability of abstaining or choosing PNV significantly decreases. As shown earlier, this effect is not found when the dependent variable is PNV. Finally, when introducing fixed effects for EPGs, group variations are detected. Members from EPP, S&D, ALDE, and EFD are 20 significantly less likely to opt for PNV or abstention, whereas ECR and GUE-NGL are more likely to do so, taking The Greens-EFA as a reference category. The results slightly differ from those obtained when PNV alone is the dependent variable, where no significant effects for groups were found, except for EFD. Overall, while some differences between PNV and Abstentions/PNV are observed, the results generally suggest that the accounts for PNV and abstentions are not outstandingly different. Table 4 5. Discussion and conclusion This study adopts a ‘competing principals’ approach to better understand the motivations behind MEPs’ decision not to vote despite attending plenary sessions. Our results are robust in demonstrating that competing principals is a strong determinant of PNV. The study finds support for the argument that legislators in the EP are likely to strategically refrain from voting when facing conflicting demands from the main principals to which they are accountable. When the national party and the European political group disagree on the direction of a vote, legislators are substantially more likely to choose not voting. By not casting a vote, they avoid displeasing either the national party or the EPG, and hence minimize the potential costs derived from defection. Moreover, our basic argument is that competing principals interact with vote closeness on PNV. Legislators in the EP are more likely to opt for not voting when they have fewer chances to 21 influence the outcome if the two main principals disagree, whereas the effect of disagreement decreases as the anticipated closeness of a vote increases. In other words, MEPs have more incentives to decide not voting when they have few chances to influence the outcome and the two principals hold different positions on a given legislative act. However, they are more likely to cast a vote when their chances to influence the vote outcome are high, irrespective of whether the two main principals disagree or not. The study also identifies non-strategic drivers of PNV and shows that seniority increases the probability of not casting a vote. One possible interpretation is that, as senior MEPs frequently hold leading positions in the EP, they devote more time to perform institutional activities and, as a result, miss a higher number of votes. Note, however, that the effect of seniority on PNV is considerably smaller than the effect of competing principals. Overall, the findings suggest that the drivers of PNV are to a much greater extent strategic. Turning to our basic argument, the results nuance previous research on the determinants of non- response in the EP by demonstrating that the effect of the expected closeness of a vote on PNV is conditioned by the presence of competing principals. The findings have implications for the study of legislative behaviour and party unity in contexts of competing principals. Most studies have commonly addressed the question of party unity in the EP as whether, and under what conditions, MEPs follow national parties or EPGs if contradictions occur between the two main principals. Existing related research highlights that MEPs confronted with competing demands tend to follow the national party (Hix 2004, Faas 2003), or, as a recent study reveals, to vote abstention (Mühlböck and Yordanova 2017). We demonstrate that under conditions of competing pressures, legislators may also decide not casting a vote. That is, PNV is one of the possible choices by MEPs 22 facing dissent between principals. In light of these results, we claim that much of the research agenda on legislative behaviour in general and on party unity in particular would benefit from not ignoring PNV. If MEPs decide not voting to avoid the costs of vote defection when faced with contradictions between the two main principals, incorporating PNV into research that addresses how legislators behave when national parties and EPGs disagree may contribute to better understanding the phenomena of party unity and vote discipline in the EP. The findings have methodological implications for the study of legislative behaviour in the EP and suggest that the accounts for PNV are not markedly different from those for explicit abstention. If this holds, abstentions and PNV could be treated together in certain types of research designs. Still, exploring the reasons why MEPs may choose between one of the two forms of non-response when facing conflicting pressures deserves further in-depth analysis. Arguably, in contexts of multiple principals, intentional not-voting could be the least-costly strategy as compared to explicit abstention in the EP when legislators face contradictory demands from their party principals. Alternatively, legislators faced with competing demands could consider the cost of voting ‘Abstention’ as lower than the cost of eluding a vote. Research on non-response could also be extended to different types of votes. Since previous studies demonstrate that final votes are slightly more cohesive than amendment votes in EP7 (Hix, Noury and Roland 2013), a future research agenda would also benefit from investigating whether conflicting pressures between party principals have a similar impact on PNV in amendment votes. Finally, our analysis has implications in terms of democratic representation. As legislators must be responsive to multiple principals who often hold discrepant positions, intentional not-voting may serve to address representational dilemmas without displeasing political principals and 23 without undermining the democratic mandate. At the same time, PNV may also result in representation gaps. If intentional not-voting is systematic, holding legislators accountable to the public might be problematic. This issue could be particularly critical in the EU context, where claims about the growing need for transparency and democratic policymaking, as well as concerns about the disconnection between citizens and politicians, dominate public debates. References Carey, J.M. (2007), ‘Competing Principals, Political Institutions and party Unity in Legislative Voting’, American Journal of Political Science 51(1): 92-107. Carrubba, C.J., M. Gabel, L. Murrah, R. Clough, E. Montgomery and R. Schambach (2006), ‘Off the Record: Unrecorded Legislative Votes, Selection Bias and Roll-Call Vote Analysis’, British Journal of Political Science 36(4): 691-704. Close, C. and S. Gherghina (2018), ‘Introduction: Towards a Better Understanding of Parliamentary Unity’, Parliamentary Affairs. Online 22 February 2018. Cohen, I.R. and R.G. Noll (1991), ‘How to vote, whether to vote: Strategies for voting and abstaining on congressional roll calls’, Political Behavior 13: 97-127. Cordero, G. and X. Coller (2015), ‘Cohesion and Candidate Selection in Parliamentary Groups’, Parliamentary Affairs 68(3): 592-615. 24 Costello, R. and R. Thomson (2016), ‘Bicameralism, nationality and party cohesion in the European Parliament’, Party Politics 22(6): 773-783. Däubler, T. and S. Hix (2017), ‘Ballot structure, list flexibility and policy representation’, Journal of European Public Policy. Published online 17 August 2017. Faas, T. (2003), ‘To defect or not to defect? National, institutional and party group pressures on MEPs and their consequences for party group cohesion in the European Parliament’, European Journal of Political Research 42(6):841-886. Finke, D. (2015), ‘Why do EPGs call the roll?’, Party Politics 21(5): 750-762. Finke, D. (2016), ‘The burden of authorship: how agenda-setting and electoral rules shape legislative behaviour’, Journal of European Public Policy 23(4): 604-623. Hazan, R.Y. and G. Rahat (2006), ‘The Influence of Candidate Selection Methods on Legislatures and Legislators: Theoretical Propositions, Methodological Suggestions and Empirical Evidence’, The Journal of Legislative Studies 12: 366-385. Hix, S. (2002), ‘Parliamentary behavior with two principals: Preferences, parties, and voting in the European Parliament’, American Journal of Political Science 46(3): 688-698. Hix, S. (2004), ‘Electoral Institutions and Legislative Behavior’, World Politics 56:194-223. Hix, S., A. Noury and G. Roland (2007), Democratic Politics in the European Parliament, Cambridge: Cambridge University Press. 25 Hix, S., A. Noury and G. Roland (2018), ‘Is there a selection bias in roll call votes? Evidence from the European Parliament2, Public Choice. Online 10 March 2018. Høyland, B., S. Hobolt and S. Hix (2017), ‘Career Ambitions and Legislative Participation: The Moderating Effect of Electoral Institutions’, British Journal of Political Science. Online 21 March 2017. Itzkovitch-Malka, R. and R. Y. Hazan (2016), ‘Unpacking Party Unity: The Combined Effects of Electoral Systems and Candidate Selection Methods on Legislative Attitudes and Behavioural Norms’, Political Studies 65(2): 452-474. Jones, D.R. (2003), ‘Position Taking and Position Avoidance in the U.S. Senate’, The Journal of Politics 65(3): 851-863. Kam, C.J. (2009), Party Discipline and Parliamentary Politics, New York: Cambridge University Press. Klüver, H. and J-J. Spoon (2013), ‘Bringing salience back in: Explaining voting defection in the European Parliament’, Party Politics 21(4):553-564. Longley, N. (2003), ‘Modeling the Legislator as an Agent for the Party: The Effects of Strict Party Discipline on Legislator Voting Behavior’, Contemporary Economic Policy 21 (4): 490–499. Meserve, S., J. Robbins and F. Thames (2017), ‘Multiple Principals and Legislative Cohesion’, Legislative Studies Quarterly 42(4): 515-548. 26 Mühlböck, M. and N. Yordanova (2017), ‘When legislators choose not to decide: Abstentions in the European Parliament’, European Union Politics. Published online 30 March 2017. Noury, A.G. (2004), ‘Abstention in daylight: Strategic calculus of voting in the European Parliament’, Public Choice 212: 179-211. Poole, K.T. and H. Rosenthal (1997), Congress: A political-economic history of roll call voting, Oxford: Oxford University Press. Rosas, G. and Y. Shomer (2008), ‘Models of Nonresponse in Legislative Politics’, Legislative Studies Quarterly 33(4):573–602 Rosas, G., Y. Shomer and S.R. Haptonstahl (2015), ‘No News is News: Noningorable Nonresponse in Roll-Call Data Analysis’, American Journal of Political Science 59(2): 511- 528. Rothenberg, L.S. and M.S. Sanders (2002), ‘Modelling Legislator Decision Making. A Historical Perspective’, American Politics Research 30(3): 235-264. Scully, M.R. (1997), ‘Policy influence and participation in the European Parliament’, Legislative Studies Quarterly 12: 233-252. Sieberer, U. (2006), ‘Party Unity in Parliamentary Democracies: A Comparative Analysis’, The Journal of Legislative Studies 12: 150-178. Trumm, S. (2015), ‘Voting Procedures and Parliamentary Representation in the European Parliament’, Journal of Common Market Studies 53: 1126-1142. 27 Yordanova, N. and M. Mühlböck (2015), ‘Tracing the selection bias in roll call votes: Party group cohesion in the European Parliament’, European Political Science Review 7(3): 373-399. Table 1 Distribution of votes (‘For’, ‘Against’, ‘Abstain’) and PNV in final passage votes, by EPG and total, EP7 (%) EPG* For Against Abstain PNV EPP 92,2 1.16 0,45 6,33 S&D 88,2 3.4 1.2 7.2 ALDE 90,3 1.4 0.9 7.4 The Greens/EFA 74.7 14.7 3.9 6.7 ECR 73.8 7.4 10 8,8 GUE-NGL 48,4 24.7 19.7 7.2 EFD 50.4 29.7 11.5 8.4 NA 43 34.8 15.1 7.1 Total 82.7 6.7 3.5 7.1 Source: own, based on data from Votewatch Note: N=274,411 (Absence excluded) * EPP (European popular Party), S&D (Progressive Alliance of Socialists and Democrats), ALDE (Alliance of Liberals and Democrats for Europe), Greens/EFA (The Greens-European Free Alliance), ECR (European Conservative and Reformists), GUE-NGL (European United Left-Nordic Green Left), EFD (Europe of Freedom and Democracy), NA (Non-attached) 28 Table 2 Multi-level logistic regression of the probability of PNV, EP7 M1 M2 M3 M4 M5 Party-EPG disagreement 5.62*** 5.62*** 10.58*** 11.22*** 10.59*** (0.2) (0.2) (0.64) (0.68) (0.65) List system 1.12 1.08 1.12 1.12 1.12 (0.12) (0.11) (0.12) (0.11) (0.11) Party-EPG disagreement * list system 0.66*** 0.65*** 0.64*** (0.05) (0.05) (0.05) Closeness 0.44*** 0.45*** 0.63*** 0.63*** 0.53*** (0.03) (0.03) (0.04) (0.04) (0.04) Party-EPG disagreement * closeness 0.11*** 0.11*** 0.12*** (0.02) (0.02) (0.02) Government 1.07*** 1.07** 1.06** 1.07** (0.03) (0.03) (0.03) (0.03) National party size 1.01 1.01 1.01 1.01 (0.01) (0.01) (0.01) (0.01) Group providing rapporteur 1 0.99 0.99 0.99 (0.02) (0.02) (0.02) (0.02) Salience 0.92*** 0.91*** 0.91*** 0.98 (0.03) (0.03) (0.03) (0.03) Gender 0.83*** 0.83*** 0.8*** 0.8*** (0.06) (0.06) (0.06) (0.06) Age 1 1 1 1 (0.004) (0.004) (0.004) (0.004) Terms served 1.13*** 1.13*** 1.12*** 1.12*** (0.04) (0.04) (0.04) (0.04) EPG (ref. = The Greens/EFA) EPP 0.8 0.8 (0.16) (0.16) S&D 0.95 0.94 (0.18) (0.18) ALDE 0.92 0.91 (0.2) (0.2) ECR 1.01 1.01 (0.23) (0.23) GUE-NGL 0.65 0.66 (0.18) (0.18) EFD 0.31*** 0.33*** (0.07) (0.08) Committee fixed effects YES Constant 0.05*** 0.04*** 0.04*** 0.05*** 0.05*** (0.004) (0.01) (0.01) (0.01) (0.01) Random effects Party 0.09 0.09 0.09 0.05 0.05 (0.04) (0.04) (0.04) (0.03) (0.03) MEP 0.67 0.64 0.64 0.66 0.67 (0.05) (0.05) (0.05) (0.05) (0.05) Log likelihood -46838.12 -46817.704 -46722.512 -46679.594 -46293.054 N 208,049 208,049 208,049 208,049 208,049 Parties / MEPs 75/663 75/663 75/663 75/663 75/663 *p < 0.1; **p < 0.05; ***p < 0.01. Entries are odds ratios able 1. 29 Table 3 Predicted probability of PNV Minimum Maximum % of change Disagreement 0.04 0.21 367 (0.04; 0.05) (0.18; 0.23) Closeness 0.05 0.02 -54 (0.04; 0.06) (0.02; 0.03) Government 0.045 0.048 7 (0.04; 0.05) (0.04; 0.06) Gender (female) 0.044 0.037 -17 (0.04; 0.05) (0.03; 0.04) Terms served 0.04 0.08 98 (0.04; 0.05) (0.05; 0.11) 95% CIs in parentheses. Based on Model 2. Figure 1 Predicted probability of PNV for principals’ disagreement and list system (95% CIs). Based on Model 3. 30 Figure 2 Predicted probability of PNV for principals’ disagreement at different levels of vote closeness (95% CIs). Based on Model 3. Notes i We employ Noury’s (2004) acronym. ii Rule 166 in 2009; Rule 179 in 2014. iii It may be the case that an MEP, once registered attendance, performs activities outside the building and misses the votes during that period. As this type of behaviour is registered as PNV, we ignore its actual incidence. In this respect, it is important to note that MEPs are incentivized to participate in final votes as allowances are reduced by a half if they participate in less than 50% of roll-call votes in a given session. iv Results in Table 1 include MEPs having the chance to cast a vote, that is, those having registered attendance. When absent MEPs (those not having registered attendance) are included in the calculation, the percentage of PNV moves from 7.1 to 6.6. v The data correspond to the first six months of the legislatures. http://www.votewatch.eu vi Döring, H. and P. Manow. 2018. Parliaments and governments database (ParlGov): Information on parties, elections and cabinets in modern democracies. http://www.parlgov.org/