The mandate divide revisited: Electoral incentives and parliamentary party unity in mixed-member electoral systems

(1)

The Mandate Divide

Revisited

Electoral incentives and parliamentary party

unity in mixed-member electoral systems

Bachelorscriptie Politicologie

Michael Souer (s1864041)

Politicologie

Eerste lezer: Dr. T.A. Mickler

Tweede lezer: Dr. S.P. Otjes

(2)

2

Abstract

Parliamentary party unity is formed, or defied, by a multitude of incentives and considerations of individual representatives, shaped in part by the institutional framework in which they operate. One aspect of this framework is the electoral system. In order to better understand its effect on parliamentary voting behaviour, this study researches the effect of the different modes of election on parliamentary voting behaviour in the mixed-member electoral system of the German Bundestag in all legislative periods up to the seventeenth. The results paint a mixed picture. Linear and logistic regression analyses show additional support for a "reverse mandate divide" than for one consistent with the theoretical expectation, and the linear regression analyses show this reverse mandate divide to have existed in all legislative periods from the seventh onward. There is also some support for an effect of the closeness of the district race in the most recent legislative period. This finding may suggest that a representative's electoral safety is more influential in the decision making process than the mode of election, although new hypotheses arise with these findings.

Introduction

Political parties have long been an inherent feature of European democracies (Schumpeter, 1942, p. 1). But while their existence as such has been accepted as an essential part of representative democracy, different conceptions of representative democracy attribute vastly different roles to parties, in particular in relation to their representatives. Some argue that parties, when attributed a stronger role, can overcome collective action problems and as such are able to achieve more than individual representatives, and that representatives carrying a party label are more easily recognised by voters, allowing voters "to make more informed judgements about how they will behave once elected" (Müller, 2000, p. 309; pp. 313-315). Others, on the other hand, claim that the preponderant role of political parties "undermine[s] the democratic chain of delegation" by minimising the influence of formal political actors, such as representatives (De Winter & Dumont, 2006, p. 958) and that, in accordance with Michel's Iron Law of Oligarchy, this process leads to the "concentration of power in the hands of party leaderships" who are not themselves democratically accountable (p. 972).

The controversial nature of the position of parties vis á vis their individual legislators is particularly evident in the discussion on parliamentary party unity. Party discipline can be seen as "a necessary condition for the proper functioning of a democratic regime" (Heidar, 2016, p. 260), or as a prerequisite for effective government accountability by some, and as an

(4)

4

obstacle for free deliberation in the parliamentary arena by others (Setälä, 2009, p. 160). Parliamentary voting behaviour is shaped by a complex set of often conflicting incentives, so whether one is interested in strengthening or loosening party unity, it is important to understand the institutional arrangements which create these incentives. In particular, the electoral system plays a significant role in the shaping of parliamentary voting behaviour as it determines on whom the representative dependents for his re-election and thus to whom he is ultimately accountable.

One would theoretically expect party unity to be greater in electoral systems where the party is only collectively accountable, as representatives individually accountable to a constituency may have an electoral incentive to deviate from the party line when they believe that doing so will improve their personal popularity among their constituents and therefore increase their chance of re-election (Carey, 2008, pp. 7-8). This difference can be studied in mixed-member electoral systems, in which different tiers of representatives face different electoral incentives. The proposed study will therefore research whether a "mandate divide" in parliamentary voting behaviour can be observed in a mixed-member electoral system. Such research provides insight not only in the mixed-member systems themselves, but also in the incentives created by the different modes of election of which they are comprised. It can therefore help in outlining how different institutional arrangements shape parliamentary voting behaviour in a more general sense.

For these reasons, research will be conducted on the basis of the following research question: "How does the mode of election affect parliamentary party unity in mixed-member electoral systems?" The answer to this question will contribute to determining which institutional arrangements are desirable from the point of view of different conceptions of representative democracy. The question will be answered using a longitudinal case study of the German Bundestag. The longitudinal nature of the study may show how the expected effects have developed since the institution's inception. Contrary to the theoretical expectation, this study shows some support for the existence of a "reverse mandate divide", where list representatives are in fact more likely to deviate from the party line than district representatives. The study also shows some support for a historical development.

For the purpose of answering the research question, it is important first to define and conceptualise the dependent variable of this research, party unity, as well as the independent variables. The following section will therefore outline the theoretical framework of this thesis.

(5)

5

Theoretic framework

Parliamentary party unity

Parliamentary party unity has been defined as "the extent to which parties vote in unison" (Norpoth, 1976, p. 1157). Thus, to know how party unity arises requires an understanding of the decision-making process of the individual legislator preceding the casting of a vote. This decision-making process is shaped in many ways by the institutional context; "[t]he institutional environment in which parties operate is widely held to affect their voting unity" (Carey, 2008, p. 125). Under the assumptions of rational choice institutionalism, representatives can have different motivations for casting their votes in a parliament. Three distinct motivations have been identified. Firstly, a representative will seek to be re-elected (vote-seeking). Secondly, a representative will want to further their parliamentary career (office-seeking). Thirdly, a representative will want to shape policy in line with his personal convictions (policy-seeking) (Sieberer, 2010, p. 485). These motivations have a complex relationship to the concept of party unity. If the party line conflicts with the personal convictions of the representative, he may deviate from the party line out of conviction. By contrast, if the party line conflicts with what the representative perceives to be the wishes of his constituents, he may deviate from the party line to increase his chances of being re-elected. Any deviation from the party line, however, will damage the advancement of the representative's parliamentary career, which is typically dependent on his parliamentary party. These rational choice explanations of party unity may be supplemented by other institutionalist theories; the rational choice that a representative makes may be limited by the norms governing the institution in which he operates. Party unity has become the norm in European parliamentary democracies, and as such, representatives may feel that it is appropriate in their role to be loyal to their party in their voting behaviour. They may adhere to the party line "out of a 'logic of appropriateness' rather than 'a logic of consequentiality', because this is part of the role conception that they have internalized" (Andeweg & Thomassen, 2010, p. 657). These rational choice and normative institutionalist factors come together in a complex, sequential decision-making process of representatives, as described by Van Vonno (2019, p. 3). If a representative has no particular opinion on a vote, he will simply take the cue of his party's specialist in the particular policy area, thus resulting in party unity as a result of the "division of labour" within parliamentary parties. If he agrees with the party line, what results is party unity as a result of cohesion of opinion. If the representative disagrees with the party line, he might still adhere to it if his loyalty to the party overrides his

(6)

6

disagreement. Even if it does not, the threat of sanctions outweigh his disagreement, and he is disciplined into obedience. Only when that fails will the representative make the choice to deviate from the party line (Van Vonno, 2019, p. 3).

Party unity, then, is the aggregate of individual choices of representatives within each parliamentary party. Each representative goes through a decision-making process in which both rational choice and normative institutionalist and other factors play a role. While this is a convincing theoretical explanation of party unity, it remains unclear as to how electoral incentives fit into this process. The following section therefore discusses the theoretical expectations of the electoral effect on party unity and the empirical support for this relation in previous research.

Electoral incentives

Electoral systems have an effect on the position of individual representatives because they differ in their extent to which representatives are individually accountable to voters or, conversely, the extent to which parties are exclusively collectively accountable (Carey & Shugart, 1995, pp. 417-418). The two extremes represent two competing conceptions of the relation between parties and their individual representatives (Carey, 2008, pp. 7-8). They would theoretically influence the aforementioned motivations of representatives, and thus play a role in shaping their legislative voting behaviour. The degree to which a representative is individually accountable to a constituency determines whether deviating from the party line would help or harm his chances of re-election (Traber et al., 2014, pp. 193-194).

In the context of individual accountability, representatives may have a strategic incentive to deviate from the party line when it conflicts with the wishes of their constituents in order to "cultivate [the] personal support among the electorate" necessary for their re-election (Hix, 2004, pp. 196-197). Research from the United Kingdom shows that, to some extent, voters reward individual MPs for rebellion in line with their personal convictions (Vivyan & Wagner, 2012, p. 255). Indeed, empirical research from Canada shows that MPs in competitive districts are considerably more likely to have casted a dissenting vote (Schaufele, 2013, p. 23), thus showing that this constituency-level accountability influences MPs' voting behaviour, too.

Conversely, in the context of exclusively collective accountability, no such incentive exists, as representatives' chances of re-election depend solely on the success of the party as a whole (Hix, 2004, pp. 196-197). Furthermore, a majoritarian system incentivises voting behaviour that is closest to the mean of the constituency's voters. Indeed, a study of

(7)

7

Switzerland shows that members elected by proportional representation at average deviated more strongly from the mean position of voters in their constituency than members in single-seat constituencies (Hug & Martin, 2012, pp. 196-197). As the mean position of voters differs between constituencies, different representatives within the same parliamentary party face different electoral incentives, which may promote less homogenous voting behaviour. In conclusion, both in theory and in empirical research, open party list PR is associated with a higher level of parliamentary party unity, while closed list and majoritarian systems are associated with a lower level of parliamentary party unity. This brings us to the case of mixed electoral systems combining these two opposing formulas.

Mixed-member electoral systems are systems which typically combine these two opposing electoral formulas, typically first-past-the-post and closed-list PR. These two component formulas can function parallel to each other, in order to produce a result which lies somewhere in the middle between distributed in such a way so as to produce an overall result which is proportional (Massicotte & Blais, 1999, p. 353). It stands to reason that the effects that different electoral systems have on representatives' voting behaviour also apply to the different modes of election in mixed electoral systems. Theoretically, one would expect district representatives to be more likely to deviate from the party line as they may have an incentive to do so when it may create the personal popularity necessary for their re-election (Ohmura, 2014, p. 71). Indeed, at the time of the introduction of MMP in New Zealand and Scotland in the 1990s, some feared that the system would produce two "classes" of representatives, in which the "second class", those elected from party lists, would have little incentive to perform constituency service because its re-election depends not on the constituents but on the party (Lindberg, 2006, p. 60).

In conclusion, there is sufficient theoretical ground to make the assumption that district representatives are more likely to deviate from the party line than list representatives. Empirical support for this "mandate divide", however, has been mixed. A comparative study of Russia, Ukraine and Hungary between 1994 and 2002 shows no statistically significant mandate divide in all but the Russian Duma (Thames, 2005, p. 295). Another study of the Ukrainian Rada confirms this absence of the mandate divide, but also finds that, contrary to the theoretical expectation, constituency representatives representing safe seats were more likely to deviate from the party line than those representing marginal seats (Herron, 2002, p. 374). Additional research of the Russian State Duma shows more support for the existence of a mandate divide in Russia, which is more pronounced in the second and third Dumas than in the first (Kunicova & Remington, 2008, 568). A similar study of the first Duma also found a

(8)

8

small, statistically significant mandate divide, and found the same counterintuitive difference between constituency representatives in safe seats and in marginal seats as in Ukraine (Haspel et al., 1998, p. 428). Data from the Italian Chamber of Deputies between 2001 and 2002 shows the opposite of the theoretical expectation, with constituency representatives more responsive to party discipline than list representatives (Ferrara, 2006, p. 22). However, data from Lithuania between 2000 and 2006 does show a significant mandate divide (Clark et al., 2008, p. 330).

Studies on the German Bundestag have shown equally inconclusive results. The prevalence of dual candidacy in Germany makes a simple difference between the two groups of representatives more contentious; list representatives who seek election in a constituency will also have an incentive to look after local interests and personal popularity (Becher & Sieberer, 2008, p. 304), functioning as de facto "shadow representatives" for the constituency (Cain et al., 1987, p. 220). Some have resolved this "contamination issue" by additionally considering the likelihood of district mandate representatives to be re-elected in the following election, and of list representatives to be elected in a district in the following election. Previous research based on data from 1983 to 1994 shows that while a simple divide in party unity between the two groups of representatives based on mode of election cannot be found, the likelihood of (re-)election in a district proves to be a greater determinant of deviating from the party line (Ohmura, 2014, p. 80). Another study using data from 1990 to 1994, however, does show a simple mandate divide, with constituency representatives as much as 20% more likely than list representatives to deviate from the party line (Stratmann, 2006, pp. 123-124). More recent research studying roll-call votes between 2009 and 2013, on the other hand, shows neither a statistically significant mandate divide, nor a statistically significant relation between margin of victory of constituency representatives and the propensity to deviate from the party line (Zittel & Nyhuis, 2019, p. 419-420).

This body of empirical research shows little unambiguous support for a mandate divide, even though it may be expected based on the theory. The first hypothesis will based on the latter, but previous empirical research does not suggest that the outcome will be positive. The first hypothesis tested in this study is as follows:

Hypothesis 1: List representatives are less likely to deviate from the party line than constituency representatives.

(9)

9

Besides empirical support, there are also theoretical reasons to suspect that the margin of victory of constituency representatives, or the margin of defeat of dual candidacy list representatives, affects parliamentary voting behaviour. District representatives who were elected by a large margin face a relatively small chance of defeat in the next election, while dual candidate list representatives who were defeated by a large margin face a relatively small chance of victory. They may therefore have little incentive to risk defiance of the party line for popularity among the constituents compared to representatives who won or lost by a small margin. The aforementioned study of the Canadian House of Commons supports this hypothesis (Schaufele, 2013, p. 23). However, as discussed, empirical research from Russia and Ukraine shows support for the contrary. The addition of this factor to the model may be able to explain some of the variance in deviant voting behaviour even across mandate types. The second hypothesis is therefore as follows:

Hypothesis 2: Representatives with a closer district race are more likely to deviate from the party line than representatives with a district race that was less close.

As for a last hypothesis, it can be assumed that if the previous two hypotheses are supported by the research, it would have taken time for these effects of the different modes of election to manifest, something supported by the inconclusive international empirical research. In 1949, the Bundestag was established as a new institution, elected in the first free election in more than sixteen years. In normative institutionalist terms, its inaugural members would still be exploring the rules, boundaries and incentives of this new system, and no "logic of appropriateness" would have existed yet at that point. One might therefore expect the first years after 1949 to show less of a mandate divide or a "majority effect" than more recent years. Thus, the last hypothesis is as follows:

Hypothesis 3: More recent legislative periods display stronger effects of mandate type and closeness of the district race than earlier legislative periods.

Now that the variables have been conceptualised and placed in a theoretical framework, it is important to operationalise in relation to the selected case and data. Therefore, the following section will be dedicated to discussing the methodology of the research.

(10)

10

Methodology

Research design and case selection

The proposed research will be a diachronic case study; that is to say, a study of a single case of the phenomenon which this paper seeks to explain, as it developed through time (Gerring, 2004, p. 343). The reason for using the method of a case study is twofold. First, it is a method suitable for the relatively small-scale nature of this paper. Second, there is a multitude of differences between countries that might explain different levels of party unity, and as such, a comparative analysis becomes difficult. Studying different modes of election within one country is more reliable than studying different electoral systems among different countries.

The case selected for this study is the German Bundestag. There are a great number of countries, as well as sub-national units, which currently use any sort of mixed-member electoral system, or have done so in the past, but the Bundestag has been selected using a number of criteria.

First, the electoral effect on voting behaviour which this paper seeks to study can only be observed when these elections are direct and free. This criterion eliminates instances of indirect mixed-member elections, such as the French Senate (since 1948) (Massicotte & Blais, 1999, p. 348). It also eliminates instances of mixed-member elections in countries not classified as at least flawed democracies in the EIU's Democracy Index, such as Madagascar (1960–1972 and 1977–1992), Zimbabwe (1980–1985) and Cameroon (since 1988).

Second, the case must be one in which this system has been continuously used for a long time period. In normative institutionalist terms, it takes time for norms and incentives within an institution to take shape. As such, a mixed-member system only recently adopted might not show the theoretically expected phenomena yet. Many countries today using a mixed electoral system have implemented this system only relatively recently, most prominently in a wave in the 1990s, such as Taiwan, New Zealand and Scotland. Additionally, there are instances where such a system was introduced before the 1990s, but repealed again only briefly later, such as Greece, Madagascar and Zimbabwe. There are six potential cases which have used a mixed electoral system for at least ten subsequent elections. Iceland has used such a system for 15 subsequent elections from 1920 to 1959. The German state of North Rhine-Westphalia has used one in 17 elections since 1946, and Germany itself in 19 since 1949. Mexico has used one in 12 since 1963, Ecuador in 16 since 1978, and the

(11)

11

Japanese House of Councillors in 12 since 1982 (Massicotte & Blais, 1999). All these cases also satisfy the first condition.

Of these remaining cases, this study uses the case of the German Bundestag. Germany is a well-established democracy marked by a long and continuous tradition of MMP. The

Bundestag has been elected using the system since 1949, therefore representing the most

long-standing use of the system. This means that political actors have long grown accustomed to the system, and possible long-term effects have long had the ability to manifest themselves. Another benefit of this long-standing tradition of MMP is the ready availability of comprehensive data over a longer period. The data to be used in the research is Sieberer et al.'s 2018 dataset of all roll-call votes in the Bundestag between 1949 and 2013. For the purpose of longitudinal research, the data used in this research will be roll-call votes from all seventeen legislative periods included in the datasets.

Roll-call votes are a strategically convenient subject of empirical research because they "allow individual legislators to position take and provide all observers with a clear signal of not only where they as individual members stand but also whether they are standing together as part of a relatively cohesive partisan bloc" (Crisp & Driscoll, 2012, p. 68), and are thus suitable as indicators of representatives' willingness to deviate from the party line under specific conditions.

Operationalisation

The research question will be answered, and the hypotheses tested, through a number of statistical analyses. In this section, the concepts used in the research question and the hypotheses will be operationalised.

The dependent variable is the proportion of votes cast in deviance of the party line. The party line is, in turn, as per the dataset, defined as "the absolute majority position within the party group in the roll-call vote" (Sieberer, 2018b, p. 29). Deviating from this, then, is understood as casting a vote in any way other than the party line, which includes abstention. The data makes a distinction between two types of deviation. Strong deviation means voting "yes" when the majority position is "no" or vice verse, while weak deviation means abstaining in contrast to the majority position. The conception of deviancy as including abstention means that, both strong and weak deviation count as deviation and no distinction is made between these two for the purpose of this study. Absence, either excused or unexcused, is treated as a missing value, as it does not count as voting behaviour, and cannot be understood as defiance of the party line. The percentage of votes cast in line with, or in opposition to, the party line

(12)

12

are calculated in the Voting Behavior dataset for each representative in each legislative period, and subsequently added to the MP Characteristics dataset as a new variable.

Because of the nature of the data, there is a problematic number of outliers which are positively skewed to one side; the mean deviant voting behaviour, in all legislative terms, is relatively close to 0, and as such there are no outliers on the lower end of the model, but outliers on the upper end can deviate strongly from the model, with mean deviant voting behaviour over 30% observable in most legislative terms, and even over 60% in the first and third. In order to make the data more suitable for a linear regression analysis, the dependent variable has been changed to logarithmic form. While this does not eliminate all outliers, it brings them in a slightly more agreeable range and tempers their skew. Moreover, it should be noted that this measure eliminates cases with a deviant voting behaviour of 0, and thus leads to a smaller sample size.

Because of these downsides of the linear regression, a logistic regression using a binominal variable is conducted in addition to the linear regression. This binominal variable will make a distinction between the value "no deviancy" for those cases with a deviant voting behaviour of 0%, and "deviancy" for those with a deviant voting behaviour of more than 0%.

The first hypothesis tests the effect of the mandate type on the dependent variable. This variable is present in the dataset under the name mandate. The previously discussed prevalence of dual candidacy in Germany complicates the simple mandate divide as list representatives who seek election in a constituency might face electoral incentives similar to those of constituency representatives. Therefore, the variable dual candidacy (dualcand) will also be included in the model (Sieberer, 2018).

The second hypothesis tests the effect of the closeness of the constituency race for constituency and dual candidate representatives on the dependent variable. This variable,

closeness_district, already exists in the MP Characteristics dataset (Sieberer, 2018),

expressing the proportion of the vote of the winning candidate minus that of the second candidate. Thus, the lower the value of this variable, the closer the district. The hypothesis would therefore expect a negative effect of this variable on the independent variable.

The linear and logistic regressions are performed for each of the seventeen legislative terms included in the dataset. The first two hypotheses concern primarily the most recent legislative period, but the third hypothesis will be tested by analysing whether any historical trend can be deduced from the results for all previous legislative periods.

(13)

13

Analysis

Data

As Figure 1 shows, mean deviant voting behaviour has fluctuated throughout the history of the Bundestag, starting at a high of 6,03% (SD = 9,564) in the first legislative period before dropping to a mean consistently below 3% between the sixth and the eleventh. German reunification saw a brief spike in the mean deviant voting behaviour in the twelfth legislative period, after which it dropped again, ending at a mean of 1,75% (SD = 2,948) in the most recent legislative period included in the data. A possible explanation of the peaks at the first and twelfth legislative periods is the unfamiliarity of all or a large group of political actors with the dominant customs and mechanisms of party unity, fitting the normative institutionalist hypothesis. Another possible explanation, however, would be that the specific historical circumstances of these periods necessitated a larger number of controversial legislation about the future of the country, leading to a larger number of defectors.

(14)

14

As Figure 2 shows, political parties have differed substantially in their mean deviant voting behaviour in the Bundestag's existence. The Social Democratic Party of Germany (SDP) displays the lowest mean deviant behaviour, 1,94% (SD = 3,560). Both the Christian Democratic Union (CDU) and the Christian Social Union (CSU) have a slightly higher mean, just below the 3%. The Left (Linke) and Alliance 90/The Greens (Grüne) have a mean just above the 3%, while the highest mean deviant voting behaviour can be observed among members of the Free Democratic Party (FDP): 3,90% (SD = 5,548). It should be noted that Alliance 90/The Greens and The Left only entered the Bundestag in the 1980s and '90s, respectively.

One might be able to explain the relatively high level of deviancy among Alliance 90/The Greens and The Left by their relative lack of government participation; the former has been in government in only two legislative periods, while the latter has never been in government. However, in the included legislative periods, no party has been in government more than the FDP, precisely the party with the highest level of deviancy.

(15)

15

When the historical development of the mean deviant voting behaviour is shown separately for list and district mandates, as displayed in Figure 3, what can be observed defies the theoretical expectation. In almost all legislative periods, list representatives were roughly equally likely or in fact more likely to deviate from the party line than district representatives; only in the third and fifth legislative periods show a clear mandate divide in accordance with the theoretical expectation. This shows that throughout most of the Bundestag's history, there has been a reverse mandate divide. A point of interest is that by far the greatest reverse mandate divide can be observed in the first legislative period, in which district representatives show a mean deviant voting behaviour of 4,42% (SD = 7,194), while list representatives show a mean as high as 8,03% (SD = 11,602), almost twice as high. But the most recent legislative period does not escape the reverse mandate divide either; list representatives show a mean deviant voting behaviour of 2,20% (SD = 2,893) while for district representatives this is only 1,25% (SD = 2,851).

(16)

16 Results

The results of the linear regression analyses for each legislative period are shown in Table 1, while the results of the logistic regression analyses are shown in Table 2. Despite the logarithmic scale of the dependent variable in the linear regressions, certain legislative periods display a problematic number of outliers, and the same problem applies to certain legislative periods in the logistic regressions. For this reason, cases with a standardised residual below -1.96 or above -1.96, as per Field (2013, p. 306), are removed from the analysis in all but seven legislative periods. No outliers were removed from the second, sixth, ninth, twelfth, fourteenth and sixteenth legislative periods because doing so would distort the results. Additionally, in the thirteenth period, all representatives who never voted had a standardised residual lower than -1.96. Therefore, an alternative range of acceptable outlier of -3.29 to 3.29 was used for this period.

Besides outliers, problematic degrees of heteroscedasticity can be observed in the linear regressions of a few legislative periods, particularly in those with a small number of cases. Most legislative periods, however, show a degree of heteroscedasticity that forms no cause of concern. All other assumptions of linear regression are met. In no legislative period did the model suffer from influential cases; in both linear and logistic regressions, the maximum Cook's distance fell far short of 1 in each (see Field, 2013, p. 306). Multicolinearity is similarly absent from the linear regressions in all legislative periods: in no legislative period does the mean VIF rise above 1.06, in none does the tolerance statistic drop below 0.94 (see Field, 2013, p. 325). Finally, the errors in each linear regression model are sufficiently independent, with Durbin-Watson scores ranging from 1.77 in the fifteenth period to 2.15 in the second (see Field, 2013, p. 311).

In the following section, these results will be discussed in relation to the previously formulated hypotheses.

(17)

Table 1. Linear regression models of deviant voting behaviour per legislative period. 1 2 3 4 5 6 7 8 9 (Constant) -1.73 (0.72) -2.74 (0.23) -2.87 (0.20) -2.40 (0.21) -2.66 (0.13) -3.31 (0.30) -3.31 (0.21) -3.63 (0.22) -3.37 (0.36) Mandate type -0.64*** (0.11) -0.06 (0.11) 0.00 (0.09) -0.16 (0.09) 0.01 (0.05) -0.07 (0.15) 0.04 (0.09) -0.07 (0.09) -0.09 (0.12) Dual candidate -0.13 (0.11) -0.25* (0.10) -0.10 (0.09) -0.11 (0.08) -0.01 (0.05) 0.03 (0.15) -0.11 (0.09) 0.14 (0.09) 0.19 (0.14) Closeness of district race 1.09** (0.40) -0.60 (0.31) -0.13 (0.26) -0.02 (0.27) -0.21 (0.20) -0.35 (0.75) 1.15** (0.35) -0.31 (0.34) 0.77 (0.43) R2 0.13 0.03 0.01 0.02 0.00 0.01 0.05 0.03 0.12 Adj. R2 0.12 0.02 -0.01 0.01 -0.01 -0.04 0.04 0.01 0.06 N 270 421 197 272 284 56 251 169 45 10 11 12 13 14 15 16 17 (Constant) -3.72 (0.29) -3.15 (0.24) -3.74 (0.15) -3.49 (0.23) -2.45 (0.21) -3.92 (0.19) -3.41 (0.26) -3.52 (0.36) Mandate type -0.39** (0.11) -0.49*** (0.08) -0.13* (0.05) -0.40*** (0.07) -0.28*** (0.06) -0.27*** (0.06) -0.18* (0.08) -0.52*** (0.10) Dual candidate 0.26* (0.13) -0.23* (0.10) -0.44*** (0.06) 0.24* (0.09) -0.47*** (0.09) -0.06 (0.08) -0.02 (0.11) 0.02 (0.15) Closeness of district race -0.51 (0.39) -0.50 (0.33) -0.11 (0.22) -0.02 (0.31) -0.36 (0.33) 0.19 (0.23) -0.13 (0.38) -0.41 (0.49) R2 0.08 0.10 0.10 0.08 0.07 0.10 0.01 0.08 Adj. R2 0.07 0.09 0.09 0.08 0.06 0.09 0.01 0.08 N 233 348 622 600 629 218 464 370

(18)

18 Table 2. Logistic regression models of deviant voting behaviour per legislative period.

1 2 3 4 5 6 7 8 9 (Constant) 1.88 (0.53) 2.34 (0.50) -0.63 (0.43) 1.26 (0.43) 0.73 (0.44) 0.45 (0.70) 2.16 (0.46) 0.23 (0.50) -2.87 (0.96) Mandate type -0.41 (0.22) 0.31 (0.23) 0.77*** (0.20) -0.34 (0.19) 0.29 (0.18) -1.18** (0.36) -0.25 (0.19) -0.42* (0.20) -0.30 (0.33) Dual candidate -0.65** (0.21) -1.10*** (0.24) -0.68*** (0.57) -0.59** (0.18) -0.54** (0.19) -0.60 (0.32) -1.22*** (0.20) -0.23 (0.20) 0.43 (0.38) Closeness of district race 2.28* (1.00) 1.86* (0.77) -0.09 (0.57) 1.12 (0.62) -0.39 (0.67) -0.44 (1.42) 1.18 (0.82) 0.01 (0.76) 1.97 (1.26) -2LL 590.9 536.8 684.9 767.7 727.6 352.4 706.4 675.2 326.5

Cox and Snell's R2 0.03 0.06 0.06 0.03 0.02 0.03 0.07 0.01 0.01

Nagelkerke R2 0.04 0.09 0.09 0.03 0.03 0.07 0.09 0.01 0.02 N 458 546 549 564 538 554 538 544 534 10 11 12 13 14 15 16 17 (Constant) 0.17 (0.47) -3.38 (0.40) 1.34 (1.00) -2.20 (1.30) 4.69 (1.18) -0.67 (0.55) 4.38 (1.27) 4.89 (0.91) Mandate type -1.00*** (0.19) 0.34 (0.19) -0.00 (0.45) 2.43*** (0.64) -0.35 (0.29) 0.17 (0.17) -0.07 (0.40) -1.36*** (0.25) Dual candidate 0.50* (0.20) 1.62*** (0.18) 0.97* (0.44) 0.98 (0.53) -0.93 (0.54) -0.05 (0.23) -1.03 (0.55) -0.36 (0.32) Closeness of district race 0.62 (0.67) 5.00*** (0.86) 2.60 (2.11) 2.07 (2.08) 0.81 (1.57) 0.03 (0.72) -0.87 (1.47) -7.71*** (1.11) -2LL 734.1 769.3 202.4 244.2 370.5 826.4 330.1 519.3

Cox and Snell's R2 0.07 0.22 0.01 0.04 0.01 0.00 0.01 0.16

Nagelkerke R2 0.10 0.30 0.04 0.13 0.02 0.00 0.02 0.23

N 569 682 697 664 698 624 637 522

(19)

Discussion

The first thing to note about the results displayed in both tables is the poor predictive capacity on the models. In Table 1, while the independent variables can account for 13% of the variance in the dependent variable in the first legislative period, the R2 does not rise above the 0.05 in the seven succeeding period, and even after that, no period displays an R2 higher than 0.12. The model for the most recent legislative period can account for 8% of the variance in the mean deviant voting behaviour. Meanwhile, Table 2 shows similarly weak models, with a Nagelkerke R2 below 0.1 in all but four legislative periods. Because the results of the models themselves are quite different in both types of analyses, they will now be discussed separately before drawing general conclusions.

The effect of the independent variable mandate type on deviancy as shown by the linear regression analyses correspond to those previously displayed in Figure 3. The first legislative period shows a statistically significant, relatively strong reverse mandate divide, the greatest effect observed in any legislative period. The subsequent periods show no statistically significant effect of the mandate type. All legislative periods from the tenth onward show a statistically significant reverse mandate divide again. The reverse mandate divide observed in the first legislative period is in fact stronger than any other.

The general trend shown by these results, then, is that of a lack of any statistically significant effect of the mandate type until the ninth legislative period, and a statistically significant reverse mandate divide from the tenth onward. This trend obviously defies the theoretical expectation formulated in the first hypothesis. No legislative period shows a statistically significant mandate divide in the direction as theoretically expected; in fact, since the 1980s, list representatives have consistently shown a higher likelihood of deviance than district representatives. Yet this historical development does support the third hypothesis. In general, the apparent effect of the electoral incentive can be observed in later legislative periods, but not in earlier ones, which might reflect a crystallisation of the norms created by the electoral framework. However, the fact that the strongest reverse mandate divide can be observed in the first legislative period complicates this conclusion and raises new questions.

The second independent variable, dual candidacy, shows a rather volatile image. The second legislative period shows a positive effect, after which no statistically significant effect can be observed for some periods. The tenth through fourteenth legislative periods do display a statistically significant effect again, but the direction of this effect constantly shifts, making it difficult to identify any particular trend. Similar to dual candidacy, the third independent

(20)

20

variable, the closeness of the district race, does not show a clear effect on deviancy throughout the Bundestag's history, with only a positive relationship observable in first and seventh legislative periods. All other legislative periods show no statistically significant effect. These linear regression analyses therefore do not support the second hypothesis.

The results of the logistic regression analyses differ substantially from those of the linear regression analyses. The significant reverse mandate divide observed in the first legislative period in the linear regression is strangely absent in the logistic regression. In succeeding legislative periods, a statistically significant effect only pops up sporadically. A positive effect can be observed in the third legislative period, negative effects in the sixth, eighth and tenth, and a positive effect again in the thirteenth. The most recent legislative period, like in the linear regression, shows a statistically significant, negative effect in the logistic regression. Thus, while the logistic regression enforced the linear regression in the existence of a reverse mandate divide in the most recent legislative period, little can be deduced from the preceding periods; the effects which can be observed are inconsequent and seemingly lacking of a general trend. The dual candidate variable shows statistically significant negative effects in the first through the fifth, as well as the seventh, legislative periods, indicating that dual candidates are less likely to deviate from the party line than representatives who were not dual candidates. However, the tenth, eleventh and twelfth periods show a positive effect. No statistically significant effect can be observed in the most recent legislative period.

Similar to the linear regressions, the logistic regressions show few statistically significant effects of the closeness of the district race on deviancy. The first and second legislative periods show a relatively weak, positive effect, while the eleventh period shows a significantly stronger positive effect. The most recent legislative period, meanwhile, shows an even stronger negative effect. This suggests that, in line with the second hypothesis, closer district races are associated with higher rates of deviance, although this only applies to the most recent legislative period. This contrasts with the linear regressions, in which the seventeenth legislative period shows no statistically significant effect. Much like the mandate type, closeness of district race does not a clear historical development throughout the seventeen legislative periods, and the logistic regressions can thus, unlike the linear regressions, not be said to support the third hypothesis.

In conclusion, both types of analyses show different results in certain respects, but a number of general conclusions can be drawn on the basis of both types of analyses. Both of them show support for the existence of a reverse mandate divide in the most recent legislative

(21)

21

period, contrary to the theoretical expectation expressed in the first hypothesis. Moreover, neither of them display support for an effect of the closeness of the district race on deviancy, either in the most recent legislative period or on a longer term. Additionally, the linear regressions show support for a reverse long-term mandate divide that started in the tenth legislative period, although the greatest effect can be observed in the first. Finally, the logistic regressions show support for a negative effect of the closeness of the district race in the seventeenth legislative period. These results offer no support the first hypothesis, which stated that district representatives would be more likely to deviate from the party line than list representatives; the data supports an effect in the opposite direction, at least in the most recent legislative period.

The results, as outlined above, are as inconclusive as the preceding empirical research into the existence of the mandate divide. Despite the clear theoretical expectation that this mandate divide is the logical consequence of the different electoral incentives, the phenomenon cannot be said to have materialised. Much of the trouble with observing a mandate divide in previous empirical research has been attributed to the contamination issue; list representatives who contest elections in districts as well face electoral incentives similar to those already elected in districts. However, dual candidacy has not been found to affect deviancy in any consistent fashion. Making a distinction specifically between dual candidate list representatives and non-dual candidate list representatives shows further support for this position. List representatives who had also stood unsuccessfully in a district had a mean deviant voting behaviour of 2,97% (SD = 5,51), almost identical to that of list representatives who had not, 2,95% (SD = 4,81), both being slightly higher than that of district representatives, who displayed a mean of 2,49 (SD = 4,05). Those representatives which should in theory be "contaminated" by a district race nevertheless show a level of deviancy nearly identical to those who are not, which suggests that the contamination issue does not play a significant role.

The failure of empirical findings, this one included, to meet the theoretical expectation necessitates a revaluation of the latter. A possible explanation for the reverse mandate divide is that district members are generally more electorally secure in their seat. After all, even if they fail to be re-elected in their district, the prevalence of dual candidacy means that most still have a "second line of defence" in the form of a position on the party list. This is supported by the data, as displayed in Figure 4; district representatives are more electorally secure than dual candidate list representatives, who are in turn more electorally secure than their non-dual candidate counterparts. The result of this may be that dual candidate list

(22)

22

representatives have a greater incentive to deviate from the party line in order to gain personal popularity in the district which he contests. District representatives, on the other hand, may instead be in the position to advance their parliamentary career by abiding by the party line, even when doing so is unpopular in their district, because their re-election is less uncertain. The positive relationship between closeness of the district race and deviancy further supports this hypothesis, although additional research could serve to test it.

Figure 4. Overall electoral safety by mandate type in all legislative periods.

The inclusion of the margin of victory in a district, or the closeness of the district race, has been another attempt to address the dual candidacy issue. This study has shown some evidence that, in the most recent legislative period, a closer district race is positively associated with a higher rate of defiance of the party line, consistent with the theoretical expectation. This finding, combined with the failure to observe a mandate divide, might tell us that it is not necessarily the mode of election which shapes the electoral incentive for deviant voting behaviour, but rather the electoral safety of representatives. What is meant by the electoral incentive for deviant voting behaviour, after all, is that a representative may vote

(23)

23

defiantly in order to further his chances of re-election. For representatives whose re-election is relatively certain, then, this end may not be a first priority, and the role that this incentive plays in determining his parliamentary voting behaviour may be small compared to other incentives.

The longitudinal nature of the analyses has resulted in findings showing some cautious support for the theoretical expectation that the effects of the independent variables would be more pronounced in more recent legislative periods because of a maturing of the norms of behaviour and electoral incentives. Of course, this latter explanation cannot be attributed to the lesser pronunciation of the effects on the basis of the empirical research, and there may be other explanations for this phenomenon. Furthermore, it should be noted that the reverse mandate divide was in fact more pronounced in the first legislative period than in any other, which raises additional questions.

Conclusion

The purpose of this study is to add to the empirical research on the effects of the mode of election on parliamentary voting behaviour in mixed-member electoral systems, which in turn may tell us something about different electoral formulas in general, and the effect they may have on the phenomenon of party unity. This research takes the form of a case study of the German Bundestag, which has used a mixed-member electoral system consistently for longer than any other national legislature in the world, and whose political actors have therefore presumably long grown accustomed to its incentives. In order to test whether this historical factor plays a role, all available legislative terms have been subjected to analysis. For reasons relating to the peculiarities of the available data, this analysis consisted of both linear and a logistic regression models.

The results have been consistent with previous research in their lack of decisive support for the existence of a mandate divide as presupposed on theoretical grounds; in fact, the results show support for a mandate divide in the opposite direction in the most recent legislative period, as well as limited support for such a mandate divide in all but one periods since the seventh. Interestingly, the strongest reverse mandate divide can be observed in the very first period. These finding prompts additional questions and hypotheses to be answered and tested in future research. A better predictor of the dependent variable appears to be the closeness of the district race, an increase in which is associated with a higher rate of deviant voting behaviour in the most recent legislative period, although there is little support for the existence of this effect in earlier periods.

(24)

24

These results show a very mixed image of the expected effects, consistent with earlier empirical research. A conclusion that may be drawn from them is that, at least in the specific case of Germany, the mere mode of election is less influential in determining parliamentary voting behaviour than a representative's certainty of re-election. However, the prevalence of dual candidacy means that we still have little understanding of how this process works in modes of election characterised by collective accountability, unaffected by the electoral incentives of individual accountability. Here, again, lies opportunity for future research.

Turning back to the model of representatives' sequential decision-making process of Van Vonno (2019), the findings of this study appear difficult to fit into this model. The existing steps in the model each show a mechanism by which a representative would opt for voting with the party group; only if all these mechanisms have failed, will the representative opt for dissent. Dissent is treated as the default position of a representative, as all mechanisms described pull him away from this position. However, there are not only mechanisms pulling a representative away from dissent, but also institutional incentives pushing them toward it; the electoral circumstance, as can be observed, shapes one such incentive. These electoral incentives are complex, and previous empirical research has shown mixed results, something this study does not change. It appears that any attempt of finding answers only serves to raise additional questions.

The picture which theory and empirical research paint of the decision making process perhaps most resembles a multitude of different incentives for and against party unity, each pushing and pulling the representative with different degrees of force. This includes both incentives for rationally calculated ends (votes, office or policy) as well as a logic of appropriateness. The force of these different exogenous forces, furthermore, may depend in part on the personality of the particular representative. A more ambitious representative, for example, may be more prone to voting behaviour furthering his desire for office, while a more principled or opinionated representative may award greater priority to his personal policy positions, and a more loyal representative may be more inclined to abide by the dominant norms of his parliamentary party. In conclusion, to limit ourselves to a single approach is to discard necessary pieces of the puzzle of understanding party unity.

(25)

25

Literature

Andeweg, R. & Thomassen, J. (2010). Pathways to party unity: Sanctions, loyalty,

homogeneity and division of labour in the Dutch parliament. Party Politics, 17(5), 655-672.

Becher, M. & Sieberer, U. (2008). Discipline, Electoral Rules and Defection in the Bundestag, 1983-94. German Politics, 17(3), 293-304.

Cain, B., Ferejohn, J. & Fiorina, M. (1987). The Personal Vote: Constituency Service and Electoral Independence. Cambridge, MA: Harvard University Press.

Carey, J. M. (2008). Legislative Voting and Accountability. New York, NY: Cambridge University Press.

Carey, .J. M. & Shugart, M. S. (1995). Incentives to Cultivate a Personal Vote: a Rank Ordering of Electoral Formulas. Electoral Studies, 14(4), 417-439.

Clark, T. D., Martinaitis, Ž. & Dilba, R. (2008). Electoral Mandate and Party Cohesion: Does it Matter in Lithuania? Journal of Communist Studies and Transition Politics, 24(3), 317-337.

Crisp, B. F. & Driscoll, A. (2012). The Strategic Use of Legislative Voting Procedures.

Legislative Studies Quarterly, 37(1), 67-97.

De Winter, L. & Dumont, P. (2006). Do Belgian parties undermine the democratic chain of delegation? West European Politics, 29(5), 957-976.

Ferrara, F. (2006). Frogs, mice and mixed electoral institutions: party discipline in Italy's XIV Chamber of Deputies. The Journal of Legislative Studies, 10(4), 10-31.

Field, A. (2013). Discovering Statistics Using IBM SPSS Statistics. London: SAGE Publications Ltd.

Gerring, J. (2004). What Is a Case Study and What Is It Good for? The American Political

Science Review, 98(2), 341-354.

Haspel, M., Remington, T. F., Smith, S. S. (1998). Electoral Institutions and Party Cohesion in the Russian Duma. The Journal of Politics, 60(2), 417-439.

Herron, E. S. (2002). Electoral Influences on Legislative Behavior in Mixed-Member Systems: Evidence from Ukraine's Verkhovna Rada. Legislative Studies Quarterly,

27(3), 361-382.

Hix, S. (2014). Electoral Institutions and Legislative Behavior: Explaining Voting Defection in the European Parliament. World Politics, 56(2), 194-223.

Hug, S. & Martin, D. (2012). How electoral systems affect MPs' positions. Electoral Studies,

(26)

26

Kunicova, J. & Remington, T. F. (2008). Mandates, Parties and Dissent: Effect of Electoral Rules on Parliamentary Party Cohesion in the Russian State Duma, 1994–2003. Party

Politics, 14(5), 555-574.

Lindberg, T. C. (2006). Second-Class Representatives? Mixed-Member Proportional Representation in Britain. Parliamentary Affairs, 59(1), 60-77.

Massicotte, L. (1999). Mixed electoral systems: a conceptual and empirical study. Electoral

Studies, 18(3), pp. 341-366.

Massicotte, L. & Blais, A. (1999). Mixed electoral systems: a conceptual and empirical survey. Electoral Studies, 18(3), 341-366.

Müller, W. (2000). Political parties in parliamentary democracies: Making delegation and accountability work. European Journal of Political Research, 37(3), 309-333. Norpoth, H. Explaining Party Cohesion in Congress: The Case of Shared Policy Attitudes.

The American Political Science Review, 70(4), 1156-1171.

Ohmura, T. (2014). When your name is on the list, it is time to party: the candidacy divide in a mixed-member proportional system. Representation, 50(1), 69-82.

Schaufele, B. (2013). Dissent in Parliament as Reputation Building. (University of Ottawa Working Paper No. 1303E).

Schumpeter, J. (1942). Capitalism, Socialism and Democracy. New York City, NY: Harper & Brothers.

Setälä, M. (2009). Representative Democracy. In M. Setälä & T. Schiller (Eds.), Referendums

and Representative Democracy (pp. 151-162). Abingdon-on-Thames: Routledge.

Sieberer, U. (2010). Behavioral consequences of mixed electoral systems: Deviating voting behavior of district and list MPs in the German Bundestag. Electoral Studies, 29(2), 484-496.

Sieberer, U. et al. (2018). Roll-Call Votes in the German Bundestag: A New Dataset, 1949– 2013. British Journal of Political Science.

Sieberer, U. et al. (2018b). The Bundestag Roll Call Vote Data (BTVote), 1949-2013: Codebook.

Stratmann, T. (2006). Party-Line Voting and Committee Assignments in the German Mixed-Member System. In R. D. Congleton & B. Swedenborg (Eds.), Democratic

Constitutional Design and Public Policy (pp. 111-130). Cambridge, MA: The MIT

Press.

Thames, F. C. (2005). A House Divided: Party Strength and the Mandate Divide in Hungary, Russia, and Ukraine. Comparative Political Studies, 38(3), 282-303.

(27)

27

Traber, D., Hug, S. & Sciarini, P. (2014). Party Unity in the Swiss Parliament: The Electoral Connection. The Journal of Legislative Studies, 20(2), 193-215.

Van Vonno, C. M. C. (2019). Achieving party unity in the Netherlands: Representatives' sequential decision-making mechanisms at three levels of Dutch government. Party

Politics, 20(10), 1-15.

Vivyan, N. & Wagner, M. (2012). Do voters reward rebellion? The electoral accountability of MPs in Britain. European Journal of Political Research, 51(2), 235-264.

Zittel, T. & Nyhuis, D. (2019). Two Faces of Party Unity: Roll-Call Behaviour and Vote Explanations in the German Bundestag. Parliamentary Affairs, 72(2), 406-424.

(28)

28

Appendix

Dataset

* Encoding: UTF-8. SET DECIMAL=DOT.

DATA LIST FILE= "(folder directory)/voting_behavior.txt" free (",") ENCODING="Locale"

/ mp_id * lastname (A32) firstname (A37) party_det ppg * party_text (A9) elecper office_spell vote_id vote_date * vote_beh (F8.0) vote_deviate (F8.0) . VARIABLE LABELS mp_id "mp_id" lastname "lastname" firstname "firstname" party_det "party_det" ppg "ppg" party_text "party_text" elecper "elecper" office_spell "office_spell" vote_id "vote_id" vote_date "vote_date" vote_beh "vote_beh" vote_deviate "vote_deviate" . VALUE LABELS / vote_beh 1 "excused absence" 2 "yes" 3 "no" 4 "abstain" 5 "unexcused absent" 6 "invalid vote"

7 "voting behavior not/wrongly protocolled" / vote_deviate 1 "excused absence" 2 "no deviation" 3 "strong deviation" 4 "weak deviation" 5 "unexcused absent" 6 "invalid vote"

7 "voting behavior not/wrongly protocolled"

8 "no party line due to tie within the party group"

9 "no party line because no absolute majority within the party group" .

VARIABLE LEVEL mp_id, party_det, ppg, elecper, office_spell, vote_id (scale).

EXECUTE.

STRING mp_id2 (A200).

COMPUTE mp_id2=CONCAT(string(mp_id,f6),lastname,string(elecper,f2)). EXECUTE.

(29)

29 select if vote_deviate ne 1. select if vote_deviate ne 5. select if vote_deviate ne 6. select if vote_deviate ne 7. select if vote_deviate ne 8. select if vote_deviate ne 9. execute.

RECODE vote_deviate (2 = 0) (ELSE = 1) into deviations. execute.

AGGREGATE

/OUTFILE=* MODE=ADDVARIABLES OVERWRITE = YES /BREAK=mp_id2

/sum_of_deviations=SUM(deviations) /nr_votes_participated=N.

freq vote_deviate.

SORT CASES BY mp_id2(A). MATCH FILES /FILE=* /BY mp_id2 /FIRST=PrimaryFirst /LAST=PrimaryLast. DO IF (PrimaryFirst). COMPUTE MatchSequence=1-PrimaryLast. ELSE. COMPUTE MatchSequence=MatchSequence+1. END IF. LEAVE MatchSequence. FORMATS MatchSequence (f7). COMPUTE InDupGrp=MatchSequence>0. SORT CASES InDupGrp(D).

MATCH FILES /FILE=*

/DROP=PrimaryFirst InDupGrp MatchSequence.

VARIABLE LABELS PrimaryLast 'Indicator of each last matching case as Primary'.

VALUE LABELS PrimaryLast 0 'Duplicate Case' 1 'Primary Case'. VARIABLE LEVEL PrimaryLast (ORDINAL).

FREQUENCIES VARIABLES=PrimaryLast. EXECUTE.

Select if PrimaryLast = 1. execute.

COMPUTE deviations_proportion = sum_of_deviations/nr_votes_participated. execute.

SORT CASES by mp_id2.

SAVE OUTFILE='(folder directory)\VB.sav' /COMPRESSED.

********************* SET DECIMAL=DOT.

(30)

30

ENCODING="Locale"

/ mp_id * lastname (A32) firstname (A37) elecper * gender (F8.0) year_birth date_birth * date_birth_text (A10) mandate_start

mandate_end office_spell spell_start spell_end * party_elec (F8.0) party_elecdet (F8.0) mandate (F8.0) mandate_detailed (F8.0) dualcand (F8.0) district_id districtvote closeness_district * list (F8.0) listpos listpos_total elecsafe_district elecsafe_list

elecsafe_overall * minister (F8.0) junminister (F8.0) parlpres (F8.0) commchair (F8.0) ppgchair (F8.0) whip (F8.0) partyid_cmp partyid_ches partyid_bl partyid_parlgov partyid_parlgov2

. VARIABLE LABELS mp_id "mp_id" lastname "lastname" firstname "firstname" elecper "elecper" gender "gender" year_birth "year_birth" date_birth "date_birth" date_birth_text "date_birth_text" mandate_start "mandate_start" mandate_end "mandate_end" office_spell "office_spell" spell_start "spell_start" spell_end "spell_end" party_elec "party_elec" party_elecdet "party_elecdet" mandate "mandate" mandate_detailed "mandate_detailed" dualcand "dualcand" district_id "district_id" districtvote "districtvote" closeness_district "closeness_district" list "list" listpos "listpos" listpos_total "listpos_total" elecsafe_district "elecsafe_district" elecsafe_list "elecsafe_list" elecsafe_overall "elecsafe_overall" minister "minister" junminister "junminister" parlpres "parlpres" commchair "commchair" ppgchair "ppgchair" whip "whip" partyid_cmp "partyid_cmp" partyid_ches "partyid_ches" partyid_bl "partyid_bl" partyid_parlgov "partyid_parlgov" partyid_parlgov2 "partyid_parlgov2" . VALUE LABELS / gender 1 "female" 2 "male" / party_elec 1 "SPD"

(31)

31 2 "CDU" 3 "CSU" 4 "FDP" 5 "Greens" 6 "Left Party/PDS" 7 "other party" 8 "unaffiliated" / party_elecdet

1 "SPD (Social Democratic Party of Germany)" 2 "CDU (Christian Democratic Union)"

3 "CSU (Christian Social Union)" 4 "FDP (Free Democratic Party) " 5 "Greens"

6 "Left/PDS (The Left, previously Party of Democratic Socialism)" 7 "GB/BHE (All-German Block/Party of Displaced Persons)"

8 "BP (Bavarian Party)"

9 "DA/FVP (Free People's Party)" 10 "DP (German Party)"

11 "National Right/DRP (German Reich Party)" 12 "FU (Federalist Union)"

13 "KPD (Communist Party of Germany)" 14 "WAV (Economic Reconstruction League)" 15 "Z (Center Party)"

16 "SSW (Southern Schleswig Voter Committee)" 17 "CVP (Christian People's Party (Saarland))"

18 "DPS (Democratic Party Saar (liberal party in Saarland))" 19 "unaffiliated" / mandate 1 "list mandate" 2 "district mandate" / mandate_detailed 1 "district mandate"

2 "list mandate (at time of election)" 3 "elected by GDR parliament (only EP 11)" 4 "replacement mandate from list"

5 "by-election (only EP 1)" / dualcand 1 "no" 2 "yes" 3 "not applicable" / list 1 "Nordrhein-Westfalen" 2 "Bayern" 3 "Baden-Württemberg" 4 "Niedersachsen" 5 "Hessen" 6 "Sachsen" 7 "Rheinland-Pfalz" 8 "Berlin" 9 "Schleswig-Holstein" 10 "Brandenburg" 11 "Sachsen-Anhalt" 12 "Thüringen" 13 "Hamburg" 14 "Mecklenburg-Vorpommern" 15 "Saarland"

(32)

32 16 "Bremen" 17 "Baden (only EP 1)" 18 "Württemberg-Baden (only EP 1)" 19 "Württemberg-Hohenzollern (only EP 1)" / minister 1 "no" 2 "yes" 3 "not applicable" / junminister 1 "no" 2 "yes" 3 "not applicable" / parlpres 1 "no" 2 "yes" 3 "not applicable" / commchair 1 "no" 2 "yes" 3 "not applicable" / ppgchair 1 "no" 2 "yes" 3 "not applicable" / whip 1 "no" 2 "yes" 3 "not applicable" .

VARIABLE LEVEL mp_id, elecper, year_birth, office_spell, district_id, districtvote,

closeness_district, listpos, listpos_total, elecsafe_district, elecsafe_list, elecsafe_overall, partyid_cmp, partyid_ches, partyid_bl, partyid_parlgov, partyid_parlgov2

(scale). EXECUTE.

STRING mp_id2 (A200).

COMPUTE mp_id2=CONCAT(string(mp_id,f6),lastname,string(elecper,f2)). EXECUTE.

* Identify Duplicate Cases. SORT CASES BY mp_id2(A). MATCH FILES /FILE=* /BY mp_id2 /FIRST=PrimaryFirst /LAST=PrimaryLast. DO IF (PrimaryFirst). COMPUTE MatchSequence=1-PrimaryLast. ELSE. COMPUTE MatchSequence=MatchSequence+1. END IF. LEAVE MatchSequence.

(33)

33

FORMATS MatchSequence (f7).

COMPUTE InDupGrp=MatchSequence>0. SORT CASES InDupGrp(D).

MATCH FILES /FILE=*

/DROP=PrimaryFirst InDupGrp MatchSequence.

VARIABLE LABELS PrimaryLast 'Indicator of each last matching case as Primary'.

VALUE LABELS PrimaryLast 0 'Duplicate Case' 1 'Primary Case'. VARIABLE LEVEL PrimaryLast (ORDINAL).

FREQUENCIES VARIABLES=PrimaryLast. EXECUTE.

select if PrimaryLast = 1. EXECUTE.

SORT CASES by mp_id2.

SAVE OUTFILE='(folder directory)\MPC.sav' /COMPRESSED.

MATCH FILES /FILE='(folder directory)\VB.sav' /FILE='(folder directory)\MPC.sav'

/RENAME (elecper firstname PrimaryLast lastname office_spell = d0 d1 d2 d3 d4)

/BY mp_id2

/DROP= d0 d1 d2 d3 d4. EXECUTE.

SAVE OUTFILE='(folder directory)\FINAL.sav' /COMPRESSED.

Linear regression

* Encoding: UTF-8. USE ALL.

COMPUTE filter_$=(elecper = 1).

VARIABLE LABELS filter_$ 'elecper = 1 (FILTER)'. VALUE LABELS filter_$ 0 'Not Selected' 2 'Selected'. FORMATS filter_$ (f1.0).

FILTER BY filter_$. EXECUTE.

REGRESSION

/DESCRIPTIVES MEAN STDDEV CORR SIG N /MISSING LISTWISE

/STATISTICS COEFF OUTS R ANOVA TOL /CRITERIA=PIN(.05) POUT(.10)

/NOORIGIN

/DEPENDENT dev_log

/METHOD=ENTER mandate dualcand closeness_district /SAVE ZRESID.

EXECUTE.

RECODE ZRE_1 (-1.96 thru 1.96=1) (ELSE=0) INTO outliers. VARIABLE LABELS outliers 'Outliers'.

EXECUTE.

DATASET ACTIVATE DataSet1. USE ALL.

(34)

34

VARIABLE LABELS filter_$ 'elecper = 1 & outliers = 1 (FILTER)'. VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.

FORMATS filter_$ (f1.0). FILTER BY filter_$. EXECUTE.

REGRESSION

/DESCRIPTIVES MEAN STDDEV CORR SIG N /MISSING LISTWISE

/STATISTICS COEFF OUTS R ANOVA TOL /CRITERIA=PIN(.05) POUT(.10)

/NOORIGIN

/DEPENDENT dev_log

/METHOD=ENTER mandate dualcand closeness_district /SCATTERPLOT=(*ZRESID ,*ZPRED)

/RESIDUALS DURBIN NORMPROB(ZRESID) /SAVE COOK ZRESID.

EXECUTE.

FREQUENCIES VARIABLES=ZRE_2 COO_1 /ORDER=ANALYSIS.

Logistic regression

* Encoding: UTF-8. USE ALL.

COMPUTE filter_$=(elecper = 1).

VARIABLE LABELS filter_$ 'elecper = 1 (FILTER)'. VALUE LABELS filter_$ 0 'Not Selected' 2 'Selected'. FORMATS filter_$ (f1.0).

FILTER BY filter_$. EXECUTE.

LOGISTIC REGRESSION VARIABLES dev_bin

/METHOD=ENTER mandate dualcand closeness_district /SAVE=ZRESID

/CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5). EXECUTE.

RECODE ZRE_1 (-1.96 thru 1.96=1) (ELSE=0) INTO outliers. VARIABLE LABELS outliers 'Outliers'.

EXECUTE.

DATASET ACTIVATE DataSet1. USE ALL.

COMPUTE filter_$=(elecper = 1 & outliers = 1).

VARIABLE LABELS filter_$ 'elecper = 1 & outliers = 1 (FILTER)'. VALUE LABELS filter_$ 0 'Not Selected' 1 'Selected'.

FORMATS filter_$ (f1.0). FILTER BY filter_$. EXECUTE.

LOGISTIC REGRESSION VARIABLES dev_bin

/METHOD=ENTER mandate dualcand closeness_district /SAVE=COOK ZRESID

/CRITERIA=PIN(0.05) POUT(0.10) ITERATE(20) CUT(0.5). DATASET ACTIVATE DataSet1.

FREQUENCIES VARIABLES=COO_2 ZRE_2 /ORDER=ANALYSIS.

The mandate divide revisited: Electoral incentives and parliamentary party unity in mixed-member electoral systems