Explaining stability and change in legislative party affiliations in the Italian Parliament ( )

Explaining stability and change in legislative party affiliations in the Italian Parliament (1996-2011) Luca Pinto (Università di Bologna) Daniela Giannetti (Università di Bologna) Aldo Di Virgilio (Università di Bologna) Paper prepared to be delivered at the ECPR Joint Sessions of Workshops 2013, Workshop 25: Understanding Parliamentarians: Individual Goals and Behavior in European Legislatures, Mainz, 11-16 March 2013. First draft. Please, do not quote without the authors permission. For contacts: Department of Political and Social Sciences Alma Mater Studiorum - Università di Bologna Strada Maggiore, 45 40125 - Bologna (Italy) E-mail: luca.pinto@unibo.it daniela.giannetti@unibo.it aldo.divirgilio@unibo.it Abstract: Party affiliation is generally considered as a long-lasting choice. This tends to be particularly true for MPs, whose career normally evolves within a single party. Therefore, legislative party switching is relatively rare in most democratic systems. However, this is not the case for the Italian Parliament where party switching is a common phenomenon. Consequently, it is not surprising that the Italian case attracted the attention of several scholars, who focused in particular on Legislature XIII (1996-2001) when party switching had a strong impact on the making and breaking of governments. Using a new dataset with information on the duration of MPs party affiliation in the Italian Lower Chamber from 1996 to 2011, when the Berlusconi government was replaced by a non-partisan administration, this paper aims to explain what makes some affiliation more stable than others and why some legislators are more loyal to their parties, while others decide to leave and switch (sometimes more than once) legislative group. Focusing on Italy as a case study, this paper will test a set of general explanations about the determinants of party switching using survival data. 1

Introduction Political parties are central actors in democratic regimes. This is particularly true in parliamentary democracies, where parties exert a strong control over candidates nominations, administer almost the totality of funds for the electoral campaign, and constitute the building block for governments, whose life inevitably depends on their support. Parties aggregate social interests and organize electoral competition. They serve both as a mechanism to reduce information costs for voters and as an instrument to recruit political elites, providing a key link between voters themselves and their representatives. Moreover, political parties guarantee the functioning of legislative assemblies. Their size and their positions on salient issue dimensions ultimately define the decision-making structure and the content of resource-allocation choices. In this way, parties build a link between the legislative process and policy outputs by grouping legislators under common labels (Aldrich 1995). These labels give voters information about how candidates are likely to vote across a large number of policy issues and represent the main vehicle under which politicians can enter in public office and influence policy making (Snyder and Ting 2002). Despite the importance of parties, situations were legislators who do not remain loyal to the partisan labels under which they were elected are more frequent than expected. Research shows that in the last two decades party switching has occurred regularly both in new democracies or weakly institutionalized party systems, such as Brazil (Desposato 2006), Poland (McMenamin and Gwiazda 2011), Russia (Mershon and Shvetsova 2008; Mershon 2008); and in established democracies facing institutional changes and electoral realignment, such as Italy (Giannetti and Laver 2001; Heller and Mershon 2005, 2008, 2009c; Mershon and Shvetsova 2008; Mershon 2008) and Japan (Cox and Rosenbluth 1995; Kato 1998; Kato and Yamamoto 2009; Reed and Scheiner 2003). Besides, the phenomenon is not limited only to new democracies or weakly institutionalized party systems, as O Brien and Shomer (2013, 114-115) have shown: out of 20 established democracies included in their study, 14 include at least one party which has encountered some sort of switching, while, among the 145 unique parties considered in the study, about 40% experienced at least one legislative party switching. Several scholars have focused on the reasons explaining why legislators switch parties. Our approach builds on their findings, exploring not only why, but also when or, better, at which point of the affiliation history of an MP to a given legislative group party switching happens. This means introducing time in our analysis, in order to understand if and how the risk of switching changes as long as MPs remain in their parliamentary parties. Thus, the longitudinal dimension allows us to revise the classic relationship between legislators policy- and office-seeking 2

motivations behind the choice of switching party in a more realistic and accurate way, opening new research questions: what affects the stability of party affiliations? What influences the risk of party switching? Do determinants of party switching affect the risk of abandoning a party in the same way across the affiliation history of the MPs? To answer these questions we concentrate on an in-depth single-country study spanning over four Italian legislatures (1996-2011), using a new dataset which registers MPs party defections and their timing in each legislature. The Italian Parliament is one of the most studied cases within the literature, as legislative party switching became an important phenomenon especially in Legislature XIII (1996-2001), when almost one out of four Italian legislators changed their parliamentary party affiliation during their mandate. After 2001 legislative party switching declined substantially, while during Legislature XVI (2008-2013) it increased dramatically once again. Thus, this study allows us to explore the determinants of this instability from the perspective of the individual legislator. The argument and evidence presented in this paper are structured as follows. The second section outlines a number of theoretical expectations regarding stability and change in parliamentary party affiliation. The importance of the Italian case is highlighted in the third section. In the fourth one, we present original data and methods used in the analysis. The fifth section presents some of the main results. The implications for the analysis of the Italian case and more generally for a better understanding of party switching are discussed in the concluding section. Stability and change in parliamentary party affiliations According to McMenamin and Gwiazda (2011), the existence of legislative parties lies on a basic transaction: autonomy in exchange for membership in a stable and identifiable party label. This understanding of legislative parties derives from the game-theoretic tradition of presenting institutions as a basic result of transactions (Greif and Laitin 2004). When legislators join a parliamentary group they accept to sacrificing part of their independence for the sake of the collective action (Aldrich 1995). This is because a successful cooperation serves both party leaders and rank-and-file members in creating and maintaining a strong and valuable label. For the former, a strong label increases the influence in coalition negotiations and in policy making; for the latter, it represents a necessary condition for realizing their ambitions (Heller and Mershon 2008, 2009b). Building on Müller and Strøm s (1999) classification of political goals, ambition has been modeled by scholars in terms of electoral, office and policy seeking motivations. Legislators constantly evaluate all the possible strategies to maximize electoral support, to take advantage of the benefits 3

of retaining office, and to influence policy outcomes according to their ideological views. Therefore, to stay in a party with a strong label represents a strategy that MPs adopt to see their ambitions fulfilled. However, in trading individual autonomy for the benefits of strong labels, legislators face a classic collective action problem: on the one hand, they need to guarantee the integrity of the party label as a collective good in order to fulfill their wishes; on the other, the efforts required to maintain it can become an obstacle for the pursue of personal ambitions (Heller and Mershon 2008). When legislators realize that they cannot fulfill their desires in a given party, they may consider changing their affiliation. Thus, under vote-seeking assumption an MP may switch party to enhance her possibility to be re-elected (McMenamin and Gwiazda 2011; Reed and Scheiner 2003). Instead, office-seeking assumptions rely on the idea that a legislator may change party affiliation to advance her position within the government and relevant parliamentary bodies (Desposato 2006). Finally, when the policy-seeking assumption applies, an MP may switch from a party that frustrated her policy preferences to another party that better supports her favorite policies (Desposato 2006; Heller and Mershon 2005, 2008; Reed and Scheiner 2003). This dynamics become clearer by adopting a longitudinal approach and analyzing MPs party affiliations across time. This is because the relationship between legislative parties and their members is mainly marked by a series of repeated interactions resulting in a sequence of votes on policy proposals versus the status quo and, as the number of interactions increases, it becomes more likely that from time to time party line will conflict with the preferences of some of its members on some issue (Heller and Mershon 2008). 1 Therefore, in the short term, the benefits of the collective action aimed at obtaining a strong party label are associated with positive returns for both party leaders and members, producing stability in party affiliation; progressively, however, the tensions accumulated in subsequent legislative voting can generate a negative feedback, undermining the very basis on which party affiliation rests: to allow politicians to achieve goals they could not realize alone (Aldrich 1995; Greif and Laitin 2004). Specifically, strong party labels are initially attractive because they are perceived as a powerful vehicle to realize MPs political goals, but as time passes the costs imposed to members to preserve their integrity can frustrate these ambitions enough to lead them to change party affiliation. 1 A similar trend has been identified by Warwick (1992) regarding the stability of coalition governments in Western Europe. Cabinets face an higher risk of termination the longer they remain in power because of the accumulation of policy delays and disagreements within the government, which results in changing patterns of loyalty and support among coalition partners. 4

In sum, ambitious politicians need parties to realize their desires, which in turn can be better fulfilled in strong labels. As a consequence, all the elements that initially make a label stronger and attractive are expected to better help MPs to reach their goals, strengthening in this way the stability of party affiliations. Among these elements, a primary role is played by the degree of party unity, the size of the legislative group and its governing status. However, we argue that in a longitudinal perspective those elements that initially enforce cooperation between legislators and, therefore, reinforce the attractiveness of the party, should become, in the long run, obstacles to the fulfilment of MPs ambitions, thus reducing the perceived value of the label and the worthiness of their membership. The repeated conflicts that might follow over time make in fact unity more costly and cooperation more difficult to achieve, especially in larger and governing parties. Conversely, those elements like ideological divergences that originally constituted a constrain to cooperation should increase their impact over time, undermining increasingly the stability of MPs affiliations. Unity Unity, understood as the capacity to act in unison (Bowler et. al. 1999; Hazan 2006; Sieberer 2006), should initially decrease the likelihood of switching by strengthening the party label. The reasons are twofold: first, unity places parties in a stronger position compared to competitors both in government formation and in policy bargaining (Laver and Shepsle 1990); second, cohesiveness increases the coherence of the party label and the identifiability of its policy platform, making simpler for voters to distinguish among different parties (Heller and Mershon 2005, 2009b; Snyder and Ting 2002). In such a party, MPs aspirations can have greater chances to be fulfilled and party members should not have incentives to leave. However, it is likely that any MP over time will find that her goals will conflict with those of her party on some issue. In these cases, in order to achieve unity, party leaders need to enforce some mechanism able to ensure members compliance to party policy positions. When this happens, the threat of punishment should induce rebel legislators to subordinate their ideals to those of their party. Discipline is therefore needed in order to preserve party unity and it succeeds when party leaders can credibly interfere with MPs ambitions, questioning for instance their re-election, blocking their advancement in party hierarchies or excluding them from the distribution of relevant offices like cabinet portfolios or committee chairs. 2 In a longitudinal perspective, when leaders are capable to enforce and maintain a high level of 2 The power of party leaders to interfere with legislators re-election varies according electoral rules and candidate selection processes (Sieberer 2006). While all the party members can enjoy of the value of the party label, only few can benefit from parliamentary and governmental offices. These are examples of rival and exclusive (i.e. private) goods and the access to these benefits depends to the extent of the control of party leaders (Desposato 2006). 5

unity, legislators are more frequently forced to toe the party line, sacrificing their own policy preferences in order to avoid the costs of punishment. However, in the long run, the recurrence of these sacrifices can seriously frustrate the ambitions of party members, thus reducing the benefit that they could expect from their affiliation to the party. According to this line of reasoning, we can formulate the following hypothesis: 3 H1: Unity should decrease the risk of party switching in the short term, but this effect should decline across time, increasing the risk of switching in the long run. Size and governing status The other elements that initially help to make a label stronger and more valuable depend on the allocation of the resources in the Parliament. All things being equal, larger parties should have more political influence, more cabinet positions and better committee assignments. Furthermore, legislators should be more comfortable in parties that win many seats, since they can offer them a higher chance to be re-elected. Similarly, governing parties should have larger potential for patronage and, once in government, they should be more able to offer legislators leadership positions and the benefits related with these roles. For these reasons, both larger parties and governing parties have better chances to satisfy MPs ambitions (Desposato 2006; Heller and Mershon 2008; O Brien and Shomer 2013). However, across time, these characteristics may be expected to influence also the efforts needed to maintain the integrity of party labels and to adhere to the party line (Sieberer 2006). Larger parties are likely to be more ideologically heterogeneous, grouping together legislators with different preferences. As a consequence, in a situation like this the incentives to act cohesively should derive mainly from discipline; making membership less worthwhile, as the instances of conflict increases over time. In the same way, it is expected that parties that have the responsibility to govern cannot avoid taking controversial decisions; and they are often forced to make compromises, especially when they share power within a coalition. As time passes, this situation can exacerbate the efforts to maintain a strong label, making party membership less valuable for legislators. Thus, we can formulate two additional hypotheses: 3 Heller and Mershon (2008, 915) were the first who highlight that «discipline and loyalty or party unity [ ] should create an incentive for legislators to switch parties. Our hypothesis does not contradict this statement, but it refines it by considering the history of party affiliation. Our approach recalls the argument of Kato (1998), based on an original interpretation of the classic exit, voice and loyalty model developed by Hirschman (1970). When leaders have a complete control of private benefits derived from party affiliation, including those relating to the prospects of reelection, legislators should tend to vote according the party line in subsequent legislative voting in order to do not put at risk their career, limiting in that way their capacity of expressing dissent (voice). Thus, MPs may recur to an extreme act of disloyalty as party switching (exit) in order to express their disaffection (O Brien and Shomer 2013). 6

H2: Larger parties should decrease the risk of party switching in the short term, but this effect should decline across time, increasing the risk of switching in the long run. H3: The risk of switching should be lower for legislators when their party is in government, but this effect should decline across time, increasing the risk of switching in the long run. Ideological divergences Policy divergences are considered by the literature one of the major determinants of party switching (Desposato 2006; Heller and Mershon 2005, 2008; Reed and Scheiner 2003). The effects of both time and policy divergences should follow the same logic described above: over time, instances of potential conflict increase and the constraints to maintain the integrity of the party label start to produce negative feedbacks, making membership less worthwhile; similarly, ideological distance increases the sacrifice required in order to maintain cooperation, therefore generating even higher negative returns. In fact, as time passes the number of instances in which individual preferences conflict with those of their leader should be more frequent for those MPs who are distant from their party policy position. Moreover, in each instance, the costs that those legislators must bear in order to toe the party line are higher than that of other members due to the ideological distance that separates them from their leader. For this reason it is plausible to expect that the risk of switching associated with policy divergence should increase over time, as the chances for conflict increase. This leads us to hypothesize that: H4: Policy divergence should increase the risk of changing party affiliation, and this effect should magnify over time. The Italian case: 1996-2011 We choose to study the length of the duration of parliamentary party affiliation and the occurrence of party switching in the Italian Chamber of Deputies between 1996 and 2011. The Italian case is particularly suited for testing our theoretical framework because of the high rate at which Italian MPs changed their legislative party affiliation during this period. Compared to other established democracies, legislative party switching in Italy is a quite common phenomenon (see Heller and Mershon 2009a, 11). As Table 1 shows, in each legislature a number of deputies comprised between 7

6,5% and 19,5% changed their parliamentary group during the term, and someone more than once, since the absolute number of switches is higher than the MPs who actually change party affiliation 4. [Table 1 about here] Italy represents a good choice also because of the large variations that have occurred within and between the legislatures under examinations, which result in a shift in the identity, size, policy positions and organization of parties, and consequently in different levels of incentives to switch for the MPs. The four legislatures experienced perfect government alternation. Legislature XIII (1996-2001), which shows the highest rate of switching, was characterized by the victory of a centre-left coalition. This coalition was replaced by a centre-right government at the beginning of Legislature XIV (2001-2006), which in turn presents the lowest rate of party affiliation change. The same government alternation scheme was reproduced in the following two legislatures: Legislature XV (2006-2008) was dominated by a centre-left coalition, while Legislature XVI (2008-2013, although here observed only until the end of 2011) experienced again a centre-right government, replaced by a non partisan one in November 2011. Both legislatures were characterized by an increasing trend in party switching. Moreover, within each term, there have been a number of different cabinets, each with a different party composition. Another potential source of variation is the change of the electoral system in 2005, when the mixedmember majoritarian electoral rule in force since 1993 was replaced by a bonus-adjusted proportional system. Despite their differences, both electoral systems encouraged the formation of electoral cartels, which functioned rather like highly undisciplined and factionalised large parties when entering in the Parliament. This was due primarily to permissive legislative rules for the formation of parliamentary parties. As a consequence, the electoral and the legislative party system remained quite distinct and the degree to which legislative parties could split created a highly dynamic environment (Giannetti and Laver 2001). This change in the electoral rules was accompanied by a partial renovation of the party system. In 1996, when parties electoral coordination fostered the formation of two pre-electoral coalitions contesting the general election, the Italian party system was quasi-atomized. At the beginning of Legislature XIII, the Effective Number of Parliamentary Parties (Laakso and Taagepera 1979) was 6.3 as the two largest parties gained 42% of seats. In 2008 parties adopted electoral strategies that led to the creation of two new large political formations the Democratic Party (Pd) and the Freedom s People Party (Pdl) to contest the general election. As a consequence, at the beginning 4 Our figure for Legislature XIII differs from the one reported in Heller and Mershon (2009a, 4). This is probably due to the stricter rules that we adopt in order to count party switching. 8

of Legislature XVI the parliamentary party system s level of fragmentation decreased dramatically. In 2008 the ENPP was 3.1 and the two largest parties gained 79% of seats (Di Virgilio and Kato 2011). Despite this, in Legislature XVI the same fissiparous logic observed in previous legislatures prevailed. Regardless of all these changes and variations, the internal rules of the Chamber of Deputies suspected to directly affect party switching have not changed substantially between 1996 and 2011. Parliamentary groups require at least twenty MPs to exist. Falling below this threshold, the group is disbanded and its members are enrolled ex officio among the independents in the mixed group (gruppo misto). Here MPs can organize themselves in sub-groups, which were introduced only from legislature XIII onwards. 5 In order to be formally recognised, sub-groups must comply with less stringent criteria than parliamentary groups, but once formed they can have access to parliamentary resources as their counterparts. The fact that parliamentary rules remained largely constant in the period under examination should avoid biases in our conclusions. Choosing to study the period 1996-2011 allows us also to exclude another source of bias from our analysis. The party switching activity in the previous period, i.e. the one that immediately followed the 1992 corruption scandals, could be simply a product of the decomposition of the Italian party system in the light of the transition from Christian Democratic dominance to bipolar competition (Verzichelli 1996). Data and Methods In their analysis on the Italian legislature XIII, Heller and Mershon (2005, 542) identified as switchers all the MPs who changed their parliamentary party affiliation, as reported in the Chamber of Deputies official website. This way of coding led the authors to define party switching as an umbrella label for any recorded change in party affiliation on the part of a politician holding [...] elective office (Heller and Mershon 2009a, 8). We believe that the focus on the formal record is too broad. In fact, a recorded change may be the product of events that cannot be classified as switching. This is the case, for example, of MPs belonging to a party whose number of seats falls below the minimum threshold required in order to form a parliamentary party group, who are forced to join the mixed group. This is also the case of members of a party that merged into another one. Here MPs affiliation change is not the product of a disagreement with their party position, but a simple adaptation to the new-born party configuration. Therefore, we propose to define party 5 According to the new regulation approved by the Chamber between September and November 1997 (Art. 14, par. 5; Art. 15, par. 2, 3; Art. 15-bis), sub-groups in the mixed group have to be formed at least by 10 legislators. Smaller groups are allowed, providing that they prove to represent an existing party. The final decision about these cases is left to the President of the Chamber. 9

switching as any recorded change in party affiliation on the part of a politician holding elective office which is not the product of parliamentary rules or of an act of coordination between parties. Measuring the duration of party affiliation According to the definition given above, we recorded survival data on MPs party defection and their timing in each legislature under examination. To track MPs movement we used a new web source developed by the Chamber of Deputies. 6 Compared to older digital records, which track only the moves between the main parliamentary groups, this new source is characterized by greater completeness and precision, since it facilitates to record all the MPs movements between the subgroups in the mixed group. In each legislature, MPs are observed from the day in which they choose their parliamentary party (group or sub-group) for the first time to the last day of life of the legislative body. Then, for each legislator, we track the duration and the timing of her changes in group affiliation. Each change constitutes an event, while the group affiliation of each MP in each legislature (if she serves for more than one term) represents an observation. 7 The duration of this affiliation is our outcome variable. For clarifying purpose, a legislator who served on for only one term and never changed affiliation has only one record starting from 0 to the end of the legislature, and then she is censored. On the contrary, an MP who for example left his party on day 500 has two records with start-stop times set at 0-500 and 0-the end of the term. At the end of the first period she experienced an event, while at the end of the second one she is right censored. Independent variables Our first hypothesis posits that the level of party unity should influence the risk of switching for its members. In the short run affiliations should be more stable in high unified parties, but, as a legislator survives in the legislative group, time should erode this stability. Unity can be the product of shared preferences, and in this case it is usually referred as cohesion. Yet, even like-minded individuals often disagree on something. In this situation unity is sustained by discipline. As a consequence, unity is observably indistinguishable from cohesion and discipline (when it is 6 The Historical Portal of the Chamber of Deputies (http://storia.camera.it/) and the Open-Data list (http://dati.camera.it/en/) offer a complete catalogue of digital data and documents of all previous legislatures, from legislature I of the Kingdom of Sardinia to legislature XV of the Republic. For Legislature (XVI), we use the website of the Chamber of Deputies (http://www.camera.it/). 7 In the case a legislator serves on in more than one term, we restart the clock at the beginning of each legislature. We focus exclusively on party switching during the legislative term, excluding cases of MPs switching parties between legislatures. 10

successful) (Bowler et. al. 1999; Hazan 2006; Sieberer 2006). Here, we consider unity as the observable degree to which members of a legislative group behave in a similar way in legislative voting patterns. 8 To compute the level of party unity, we first estimate MPs revealed ideal points, applying W-Nominate scaling method (Poole and Rosenthal 1997; for an application to the Italian case see Curini and Zucchini 2010) on all the final passage roll-call votes for bills in each legislature. 9 We fit a one-dimensional behavioral space that can be interpreted as the classical leftright dimension. 10 Second, for each legislative party we obtain the Inter-Quartile Range (IQR), which is the difference between the third and the first quartile of the distribution of the ideal points of all the MPs who have been affiliated to that party during the legislature. 11 Therefore, party IQR can be perceived as a proxy of the observable divergences in the revealed behavior of legislators belonging to the same group. For exemplificative purpose, we plot the party IQR for Legislature XVI in figure 1. Within this figure the smaller boxes indicate that the majority of legislators in a party are closely clustered. Finally, unity is calculated as 2-IQR: this is the difference between the maximum theoretical observable IQR (i.e. 2) and the actual Inter-Quartile range. 12 [Figure 1 about here] Our second and third hypotheses pertain to the classical office seeking tradition. We created a variable called party size, i.e. the share of seats of the parliamentary party to which a legislator belongs. The variable executive coalition traces the governing status of parties, coding as 1 a governing party, 0 otherwise. In the short run, affiliations should be more stable in larger and 8 Unity is usually captured by the Rice Index (Sieberer 2006). While Rice scores highlight the quantitative aspect of the concept the size of the dissident group, our measure underline more the qualitative side of the problem, taking into consideration also the original preferences that at the end, together with discipline, have generated MPs voting behavior. Nominate is not without its limits. In particular, according to Rosenthal and Voeten (2004), the application in legislative assemblies other than the Congress, could violate the random distribution of the error terms due to the different level of party pressures to which legislators are subjected. However, proven that Nominate is applied on a sufficient number of votes and subjects, its estimates result highly correlated with those of other scaling techniques (Carroll et al. 2009). On the other hand, also the Rice Index suffers of numerous drawbacks (see Sieberer 2006) so that O Brien and Shomer (2013, 135) recommend the use of Nominate-like, or Item Response Theory based, measures of party cohesion. 9 As suggested by Poole and Rosenthal (1997), we include in the analysis only roll calls with a 2.5 percent in the minority and MPs who have casted at least 20 votes. This results in a limited number of missing values in our analysis. Nominate allows only a dichotomous behavior: yes or nay. However, in the Chamber of Deputies there are at least two more alternatives: abstention and absence. Given the strategic implications implicit in both these options (see Curini and Zucchini 2010 and Heller and Mershon 2008 for a detailed discussion), we recode abstention as yes if the bill was approved by the floor; absence as nay when almost entirely a parliamentary group was absent and the remaining members voted nay. 10 A one-dimensional behavioral space classifies correctly at least 95% of the votes in our sample. Fitting a second dimension resulted in an improvement in classification accuracy of about 2% in each legislature. 11 IQR is computed by party and legislatures. For each group, the score takes into account the contribution of all the legislators switchers included that have been affiliated to the party, even for a short time period. Roll-call votes are one of the ways parties can demonstrate their relative policy preferences and the coherence of their label clearly to citizen. As temporary members, switchers contributed to shape both. 12 Nominate constraints ideal points to lie between the range -1, 1 (Poole and Rosenthal 1997). 11

governing parties. However, in the long run the effect should decline and change, increasing the risk of switching. Size and governing status are time-varying variables, in the sense that they can change their value throughout the history of MPs affiliations. 13 To accommodate their time-varying nature, we split our dataset into sub-periods. This produces a huge increase in the number of observations. Our final hypothesis focuses on the role of policy divergences in shaping the risk of changing party affiliation. Larger the ideological distance between an MP and her leader, higher should be the risk of switching as a legislator survives in her legislative group. To obtain the ideological distance, we employ the estimations of MPs locations derived from the application of Nominate. 14 Then we use the revealed ideal points of all the MPs who have been affiliated to a parliamentary group during the legislature to generate party policy positions, which we compute as the position of the median legislator (the white line in figure 1). 15 The use of this measure relies on the assumption that parties are not unitary actors but clusters of legislators, functioning as aggregators of different policy positions (Giannetti and Laver 2009). Finally, we compute ideological distance as the absolute distance between an MP s location in the behavioral space and her party policy position. Finally, we include a series of control variables derived from the existing literature on party switching. First, we control for the effects of the electoral rules on the incentives to switch party. For this reason we include a dummy variable coded 1 if a legislator was elected in single member districts (only in Legislatures XIII and XIV), while 0 if in PR closed-lists (the proportional quota in Legislatures XIII and XIV plus all the legislators in Legislatures XV and XVI). Candidates elected in single member district can rely not only on party label, but also on personal support. For this reason, they should be more prone to change party affiliation without fearing to be not re-elected (Ferrara 2004; Heller and Mershon 2005). The second conjecture is that more experienced deputies should be less likely to change party. The rationale behind this proposition is that MPs who have not yet socialized into their role have less bounds and also less to lose than more experienced legislators (Heller and Mershon 2005; Desposato 2006). Therefore, we include a variable called newcomer, which distinguishes legislators at their first political mandate from those who have already served on in the Chamber of Deputies (or in the Senate). Third, we control for the level of 13 The size of a legislative group updates at each switching. 14 We treat MPs ideal points as constant throughout the entire legislature, regardless of any switching. In this way, ideal points reflect a mixture of sincere preferences, strategic behavior and party pressures. We are aware that this choice is problematic, but the pre/post strategy (i.e. splitting the voting records according to legislators party affiliations; see Rosenthal and Voeten 2004) results in a too high number of missing cases. However, the choice to include all the legislators that have been affiliated to a given group in a legislature switchers included in order to compute IQR and party positions should mitigate the possible bias. In this way switchers contribute in determining unity and party position multiple times, both in the sending and the receiving party. 15 Nominate derived scores for the main parties in Legislature XIV and XVI correlate highly (r=.95 and.93 respectively) with the general left-right scale of the 2004 and 2008 waves of expert surveys (Benoit and Laver 2006; Curini and Iacus 2008). 12

education, distinguishing MPs that have a masters degree, a medical specialization or a PhD (university education) from the others. Education could be interpreted as a personal resource for a politician, but regarding this variable we do not have any clear expectations. Finally, we include a dummy variable indicating whether or not an MP belongs to the independents in the Mixed Group (independent). Regarding this point there are contrasting expectations. On the one hand, the absence of discipline should lower the number of switching within this sub-group; on the other hand, the lack of unity and coordination among its members should push independents to switch in search of a party label capable of better fulfill legislators ambitions (Heller and Mershon 2005, 2008). Model In order to test our hypotheses we use a Cox proportional hazards regression model. The aim of the model is to estimate how the covariates affect the hazard rate h(t x j ) of an event for the jth subject, which is the probability that a failure event occurs (or a party affiliation breaks up) at time t, conditional on the fact that the subject has remained in the party until t (Box-Steffensmeier and Jones 2004, 48). The hazard rate is defined in the following way: h()=h () The hazard rate has two components. The first is the baseline hazard, h () which indicates the underlying probability of changing party affiliation over time when the vector of all covariates is zero. The second is the set of covariates that we hypothesized systematically to affect the timing of our event of interest. Cox regression has several advantages in comparison with other modeling strategies previously used in empirical works on party switching (mainly binary regression models). First, it takes seriously time into account. In particular, it has the advantage to make no assumptions about the shape of the baseline hazard. This means that the baseline hazard could increase with time and then decrease and so on, without being constrained to a specific distribution function like in parametric duration models (Box-Steffensmeier and Jones 2004, 21). 16 Second, it can easily handle timedependent covariates, as suggested by our hypotheses (Box-Steffensmeier et al. 2003, Licht 2011). Finally, switching is an event that may occur repeatedly for the same legislator. This could create event dependence, which means that subsequent events are likely to be influenced by the previous one, making them substantially different. Cox regression deals with event dependence through 16 Party switching was studied using parametric survival models by McMenamin and Gwiazda (2011). 13

stratification. In this case, the baseline hazard is allowed to vary by different risk pools (Box- Steffensmeier and Zorn 2002). 17 Results We conduct our empirical analysis on an original dataset comprising 3,138 party affiliations, which represent our unit of analysis. These affiliations are clustered within the 1,653 legislators that have served on in at least one of the four legislatures included in our study. Each affiliation is divided in sub-periods in order to take into account time-varying independent variables. This has increased the number of observations to more than 200,000. 18 In order to test our hypotheses we need a three-step strategy. First, we require to estimate a proportional-hazard (PH) Cox model, including all the covariates described above. Second, since one of our main expectation is that the effect of party unity, size, executive coalition and policy distance on the risk of switching changes over time, we have to assess if proportional-hazard assumption (PHA) is violated for these variables. Finally, if violated, we should accommodate for this violation estimating a NPH Cox model and interpret the results. Table 2 reports the outcome of this process. Both PH and NPH models include dummies to take into account the effect of different legislatures and allow for stratification by event number to check for event dependence. We report unstandardised coefficients: a positive (negative) sign inform us if a variable increases (decreases) the hazard or risk of switching for MPs, reducing (increasing) consequently affiliation stability. [Table 2 about here] A key feature of our hypotheses is the expectation that the effect of the four main variables, which we have identified to affect the stability of party affiliation, change over time. For this reason, we estimate a PH Cox model (first column of table 2). Cox regression analysis conventionally presupposes what is called the Proportional Hazards Assumption (PHA). This means that the the magnitudes of the effects of covariates on the duration of a state remain proportional across the life of the process (Box-Steffensmeier et al. 2003, 34). A conventional strategy for assessing the violation of the PHA in the Cox model is to use the scaled Schoenfeld residuals test (Box- Steffensmeier and Zorn 2001). In our case, table 3 provides residual-based diagnostics for the 17 Variance-corrected models for repeated events differ in how they count time at risk for an event. In our case the clock restarts after each failure, estimating a model in gap-time (Box-Steffensmeier and Zorn 2002). Statistical techniques that treat repeated events to be independent could produce biased and inconsistent results, such as research designs that consider only the first event as representative of all the others. 18 The number of observations is quite meaningless in survival analysis. What really matters is the number of subjects at risk (3138 affiliations) and the total time at risk (Box-Steffensmeier and Jones 2004). 14

possibility of non-proportionality among the covariates. The global test shows strong evidence (when p-value is lower than.05) that there is a violation of PHA in the model, while covariatespecific tests reveal potential violations of PHA for the degree of unity, party size, governing status and ideological distance. The standard solution to accommodate for NPH consists in interacting these variables with some function of time, most commonly the natural-log (Box-Steffensmeier et al. 2003, Licht 2011). The results are provided in the second column of table 2. As it is possible to see, the constitutive terms of our variables of interest is always negative and significant, while the interactions are always positive and significant, with the relevant exception of the coefficients of party size, which result indistinguishable from zero. This gives evidence that the time trend reverses the initial effect. Furthermore, the NPH model presents a lower value of AIC and BIC and a likelihood-ratio test supports our choice to include time interactions (chi-square 101.73, p<.01). [Table 3 about here] After having assessed that the effect of our covariates changes during MPs affiliation history, we need to understand if this variation moves in the predicted manner. Following Licht (2011), we present substantive interpretations of NPH in a graphical way. 19 As noted above, H1 states that members of a unified party should face a lower risk of switching in the short run. However, as a legislator remains or survives in a legislative group, unity should begin to erode affiliation stability, thereby increasing the risk of switching in the long run. Figure 2 charts the first difference of a hypothetical change in party unity. The solid black line represents the average percentage change in the hazard of switching party for an increase from low unity to the highest degree of unity observed in our sample. The dashed lines are 95% confidence intervals. 20 When the confidence intervals are both below (above) the zero line, this suggests that our covariate negatively (positively) impact the risk of party switching. After a month of affiliation, members of a highly unified party face a probability of switching 63% lower than legislators affiliated to parties that show a low degree of unity. As predicted, this difference declines as the length of party affiliation increases, reducing to 22% after about nine months. Although after this threshold the effect of unity on the risk of switching becomes positive, contrary to our predictions its impact is no longer statistically significant. Thus, our analysis supports the first part of our hypothesis. In the first months of their affiliation MPs perceive unity as a powerful mechanism to strengthen their party and, at the same time, an instrument to feed their ambitions. However, the longer a legislator remains in their group, the more the constraints and sacrifices required to act in unison become explicit. As a consequence, 19 Time-interactions should be treated as common interactions terms in multivariate analysis (see Brambor et al. 2006). 20 The plots are constructed using parameter estimates for NPH model from table 2. The confidence intervals are calculated via simulation using 1,000 draws from the estimated coefficient vector and variance-covariance matrix. 15

the contribution that unity gives to the stability of party affiliations decreases with time and after less than one year it lose completely its effect. [Figure 2 about here] H2 and H3 pertain to the allocation of resources in the Parliament. All the things being equal, legislators affiliated to larger and governing parties should face a lower risk of switching in the short term. However, in the long run their impact should decline and change, increasing the risk of switching. Our second expectation is not supported by our data. Both the constitutive and the interaction terms of party size are never significant. This result contrasts with that of PH Cox model, where an increase in the share of seats of a legislative group corresponds to a lower risk of switching. However, it should be noted that violation of PHA can lead to false inferences about a variable s substantive and statistical significance (Box-Steffensmeier et al. 2003). Moving to our third hypothesis, figure 3 plots the effect that belonging to governing parties have on the probability to switch. As predicted, executive coalition membership has initially a positive impact on the stability of party affiliations. After about two months, legislators that are also members of governing parties face a risk of switching 33% lower than colleagues in opposition parties. As a result, in this short period only 1% of our affiliations break up when in government, suggesting a strong negative effect of this variable on the risk of changing party. Yet, this rate rapidly decreases, and after two months it is no longer significant. 21 After about two years, time subverts the original effect, bringing the curve into positive territory for over the 60% of affiliations that have survived beyond this date. At day 861 of their membership history, the risk of switching for MPs in governing parties is 24% higher than for legislators in opposition parties. The positive effect continues to grow over time, reaching 50% at the longest observed affiliation time. This supports our expectation about the impact of executive coalition membership on party switching. Initially, being part of governing parties is a powerful mechanism to enforce cooperation and to encourage legislators to maintain party affiliation. However, governing soon ceases to be a benefit, indeed it starts to become a problem for the stability of affiliations in the long run. [Figure 3 about here] Our last hypothesis (H4) states that the ideological distance between MPs and their party should decrease the stability of party affiliations in an increasing way across time. Figure 4 charts the hypothetical effect that one step away from the party median produces on the probability of switching. Surprisingly, policy divergences are not significant in shaping the risk of changing 21 This rapid decline of the effect of executive coalitions could essentially depends by the fact that the benefit associated to government are localized in short period of time (Mershon and Shvetsova 2008; Mershon 2008). 16

legislative group for about the first two months of party affiliations. Thus, this period stands out as a sort of genuine honeymoon between MPs and their parties. However, this truce is destined to disappear soon. After four months, each step increases the risk of switching by 2%. The positive effect continues to increase until a figure of 6% at the longest observed affiliation time. These results are consistent with our expectations. After few months ideological divergences become a powerful constraint for legislators, undermining increasingly the stability of MPs affiliations. [Figure 4 about here] Among the control variables only having a university degree or being an independent (in the mixedgroup) affects positively and in a significant way the hazard of changing party affiliation. MPs with a degree have a 60% higher risk of switching than their less educated colleagues, while the stability of the affiliations within the independents in the mixed group is 84% lower than for those MPs with recognizable party labels. As a final step in our analysis, figure 5 plots the estimated cumulative hazard functions by event number of the NPH model in table 2. The figure provides us with two important pieces of information. First, the risk of having a new switching increases after having experienced one, given that the cumulative hazard functions for each stratum are distinct and those associated to events 2 to 7 are higher than that describing event one. This is strong evidence of event dependence. Second, the cumulative hazard function resembles the baseline hazard, which is the underlying risk of changing party affiliation associated only with time trends. In sum, figure 5 shows that the risk of changing party increases with the length of affiliation, independently of the effect of the covariates. This evidence supports our initial conjecture that the level of conflict within a party increases the longer a legislator remains with a party. [Figure 5 about here] Conclusions Focusing directly on the history of Italian MPs party affiliation over a fifteen year period, this case study has explored a number of general questions of interest to students of legislative behavior. Specifically, the evidence presented in this study provides some insight into (a) What affects the stability of party affiliations? (b) What influences the risk of party switching? (c) Do the determinants of party switching affect the risk of abandoning a party in the same way across the affiliation history of the MPs? 17