Leadership Competition and Disagreement at Party National Congresses

Abstract

Theories often explain intraparty competition based on electoral conditions and intraparty rules. This article further opens this black box by considering intraparty statements of preferences. In particular, it predicts that intraparty preference heterogeneity increases after electoral losses, but that candidates deviating from the party’s median receive fewer intraparty votes. Party members grant candidates greater leeway to accommodate competing policy demands when in government. The study tests the hypotheses using a new database of party congress speeches from Germany and France, and uses automated text classification to estimate speakers’ relative preferences. The results demonstrate that speeches at party meetings provide valuable insights into actors’ preferences and intraparty politics. The article finds evidence of a complex relationship between the governing context, the economy and intraparty disagreement.

Theories of party politics often make strong assumptions about the relationship between parties’ behavior and the preferences of intraparty actors. For example, scholars frequently assume that parties act as if they are unitary actors, and that the party leader represents the median preferences of the party’s membership. Despite substantial theoretical development and a number of detailed case studies, few cross-national analyses of intraparty politics consider the role of intraparty preferences. ¹ This absence is striking. Intraparty politics and party preferences hold implications for a large range of political processes such as election campaigns, legislative politics and coalition governance. ²

Building on these studies, we develop a theory of intraparty preferences and party leader selection by considering experiences in government and intraparty electoral rules. Broadly, we theorize that a party’s electoral context influences its internal preference diversity. We then argue that candidates’ statements of preferences influence their intraparty electoral success. Candidates who express preferences closer to the party’s ideological center attract more votes than more extreme candidates.

To empirically test hypotheses from our theory, we create a new dataset of intraparty actor preferences from their statements at party national congresses. By focusing on speeches at intraparty meetings, we begin to break open the black box of intraparty politics. Despite evidence that parties act as if they are internally divided in parliament, few studies seek to directly, quantitatively analyze actors’ preferences outside this arena. ³ Historically, intraparty actors’ preferences have proven complicated to measure. Limited data has created a major hurdle in testing theories of intraparty politics. Like recent research studying political documents, ⁴ we overcome this hurdle by using automated text analysis to measure the relative location of intraparty actors’ statements of preferences by analyzing their speeches at parties’ congresses. We apply the scaling method WORDFISH ⁵ to estimate actors’ ideological positions and the distance between actors within the party. This method allows us to compare actors’ relative statements of preferences across multiple political settings. We then use these estimates to study the relationship between intraparty disagreement and experiences in government and elections. In a second analysis, we predict party leadership elections at national congresses using the relative location of speakers’ preferences.

The results of our analysis are consistent with an explanation of intraparty politics that is focused on the party’s internal rules and electoral context. Likewise, the results indicate that our measures provide meaningful estimates of the location of intraparty actors’ preferences that can be adapted to fit numerous research goals. More broadly, we find evidence of a complex relationship between parties’ electoral performance, experiences in government and intraparty division.

In the following section, we discuss previous studies of intraparty politics and leadership selection before considering previous approaches to studying actors’ preferences. We then describe the role of party congresses in political party behavior and outline our empirically testable hypotheses that predict two dependent variables: intraparty division and party leadership selection. Following a discussion of our data collection and approach, we illustrate differences in parties’ disagreement over time and show how the positions of individual party actors are linked to leadership selections at the parties’ national congresses. The results from both sets of analyses provide important insights into the study of political party behavior and offer a new approach to help scholars understand and measure intraparty actors’ statements of preferences.

PERSPECTIVES ON INTRAPARTY POLITICS

Researchers show that political parties’ preferences are important for understanding electoral behavior and government outcomes. There is little consensus about the intraparty process that yields parties’ statements of preferences such as election manifestos. Scholars have previously considered intraparty politics from spatial or organizational perspectives.

Many studies from a spatial perspective assume that party leaders represent the median preferences within the party, or that the leaders’ preferences exemplify the broader organization’s goals. ⁶ The relative location of a party’s statement of preferences is frequently considered to be strategically chosen or selectively emphasized to maximize the votes it receives in an election. ⁷ The studies argue that parties selectively emphasize their preferences to attract voters, who in return select parties that they expect will be most likely to implement their policy goals. ⁸

Spatial theories frequently rely on the assumption that parties act as if they are unitary actors. This assumption requires that the intraparty organizational process generating parties’ campaign messages does not have a direct impact on their exact location. These theories treat party organizations as black boxes in which their leaders and statements of preferences are chosen for strategic electoral or policy-motivated reasons. ⁹ While spatial theories do not uniformly require parties to behave as unitary actors, ¹⁰ there is clear evidence that parties’ elected representatives act counter to the parties’ goals in numerous settings. ¹¹ Members of parliament frequently exhibit diverse opinions that do not correspond to the party leaders’ preferences on roll-call votes ¹² and speeches in parliament. ¹³ Party leaders use parliamentary procedures and the benefits of office to maintain party unity in parliament, ¹⁴ although members also have opportunities to publicly dissent from the party line. ¹⁵

Few studies, however, directly analyze intraparty politics and their influence on parties’ preferences, although empirical research of parliamentary behavior demonstrates that intraparty divisions frequently constrain party leaders. ¹⁶ Party members regularly deviate from the preferences of the party’s leadership in election campaigns and in their behavior in parliament. ¹⁷ Scholars assume that party leaders generally receive support from large majorities of intraparty supporters, but that internal party competitors may be able to mobilize the support of groups with diverse political interests within the party. Modern catch-all parties frequently assemble diverse political supporters, each of whom would prefer their policy goals to be the party’s priority. ¹⁸

Scholars of party organizations instead study intraparty politics through the lens of internal institutional rules. These studies show that intraparty organizational characteristics – such as the method of leadership selection, changes in party memberships and change in factional dominance – influence parties’ statements of preferences. ¹⁹ For example, Harmel and Janda theorize that party leadership selection and factional dominance influence parties’ preferences. ²⁰ In contexts that lead to changes in leadership and the dominant faction, such as large electoral losses or exogenous shocks, the party’s message also likely changes. ²¹ Kitschelt adds that experiences in government and electoral results influence the composition of parties’ membership, and that this eventually impacts the long-term direction of the leadership. ²² However, dynamics in numerous democracies have caused parties to rely less on their memberships for running electoral campaigns, which has made parties become more hierarchically organized. ²³ Furthermore, parties that give greater influence to party members and voters in leadership selection attract more candidates for office, but fewer intraparty candidates receive most of the votes. ²⁴

Both spatial and organizational perspectives have given us substantial insights into preferences and party politics. Little research from either perspective, however, directly studies the causes and consequences of intraparty preference heterogeneity. In the following sections, we propose a theory linking the range of intraparty preferences to their electoral context and consider the implications of intraparty preferences for party leadership selection.

THE CAUSES OF INTRAPARTY HETEROGENEITY

Intraparty preference heterogeneity holds important implications for a range of political outcomes. Studying its causes will allow us to better understand the decision-making process leading to parties’ election and policy behavior. In general, we propose that the party’s electoral and governmental performance influences intraparty politics.

Like previous research on party change, ²⁵ we expect that there are numerous factors that likely influence intraparty politics and the information party leaders have about the distribution of preferences within the party. Scholars have theorized that major events, such as elections or losing government positions, will alter parties’ internal composition and rules. Kitschelt suggests that experience in government and previous electoral results influence the types of activists, supporters and politicians that join parties. ²⁶ Similarly, Przeworski and Sprague find that socialist parties’ internal organizational structures became more hierarchical as they gained experience in office. ²⁷ Harmel and Janda add that major electoral defeats provide the opportunity for changes in factional dominance. ²⁸

Building on these studies, we predict that electoral success and experience in government influence intraparty politics. Parties that are electorally successful likely attract more pragmatic supporters who hope to benefit from the party’s access to government. ²⁹ So long as the party’s leadership maintains electoral support, intraparty competitors will bandwagon or link themselves to the leaders’ reputations in order to curry favor with the leader and groups within the party. Due to this pragmatic bandwagon behavior, intraparty disagreements are likely to be limited when the leadership attracts sufficient electoral support for the party.

However, competition between groups within the party increases when one faction or group sees an opportunity to expand its influence. Following electoral defeat, intraparty groups blame the dominant party leaders and factions. ³⁰ To clearly distinguish themselves from the previous leader’s reputation, challengers arise within the party who offer distinct alternatives to the current leader’s preferences. Large electoral defeats lead groups to challenge the party leaders’ preferences to differentiate themselves from the party’s previous direction. This logic leads us to our first hypothesis.

Hypothesis 1 Electoral losses increase intraparty disagreement.

Furthermore, we add that parties’ experience in government influences intraparty heterogeneity. The longer a party stays in office, the higher the likelihood that a world event or crisis occurs that forces leaders to make unpopular policy decisions. ³¹ Parties with long-term successes in government become filled with actors who have more disparate and pragmatic policy goals, ³² but they stick with the party leader because their primary interest is to stay in power. Furthermore, economic conditions and coalition governance encourage compromises on a diverse set of policies. ³³ Perceptions of government accountability and competence may influence the intraparty distribution of preferences and factional dominance. Negative evaluations of the party’s leadership also encourage challenges to the party’s direction.

Like studies of economic voting, we assume that voters hold parties accountable for the economy. ³⁴ As the most salient topic and ideological cleavage in most modern democracies, public perceptions of the economy provide a clear indicator of the government’s success. We argue that party members and intraparty challengers also use the economy to evaluate the party’s competencies and popularity. Challengers are unlikely to distinguish themselves from the current leadership when the party is perceived to be performing well. ³⁵ On the other hand, intraparty groups will express their discontent with the party’s direction when the public perceives the party as incompetent or unaccountable.

The effect of perceptions of competence depends on parties’ position in government or the opposition. We predict that party members hold their own leaders to a different standard when they control the cabinet. To govern effectively, leaders require flexibility to cope with changing world events. ³⁶ As long as the party maintains access to government, party leaders can emphasize their policy successes and the rewards of office. Leaders can also blame deviations from intraparty preferences on voter demands. ³⁷ Evidence suggests that party members hold realistic expectations about government leaders’ ability to enact uncompromised policies; government participation forces them to respond to voters, adhere to budgetary constraints and compromise with coalition partners to stay in office. ³⁸ Furthermore, members of governmental parties place additional value on supporting the party leader, since their continued participation in government demands internal support. ³⁹

Altogether, this implies that the ability to challenge the party’s leadership is limited in government. Unlike voters, party members value intraparty unity and support the leadership when it is in government. We predict that party members support their leadership despite poor economic conditions. This leads to fewer speakers proposing alternate deviations from the party leadership’s goals and less internal disagreement.

With few governing rewards for leaders to counterbalance negative perceptions, members of opposition parties become less forgiving of broad perceptions of incompetence. Studies of issue competition show that voters perceive parties’ competencies as connected. When the economy performs well, voters perceive opposition parties as less competent and government parties as more competent. ⁴⁰ Perceptions of incompetent or unpopular leaders create the opportunity for entrepreneurial party members to influence the party’s policy direction. By expressing alternate prescriptions, party members hope to draw the party’s policies in their preferred direction. Party members might express competing strategies for future policy success, such as incremental versus pure policy approaches, given uncertainty about future electoral successes. ⁴¹ Therefore, the range of preferences expressed at national meetings likely increases for opposition parties when the economy performs strongly. A weak economy causes the range of preferences at the party meeting to shrink, as the party leadership benefits from improved voter perceptions. We summarize this logic in the following hypotheses.

Hypothesis 2a: In the opposition, positive economic performance increases intraparty disagreement.

Hypothesis 2b: In government, poor economic performance does not increase intraparty disagreement.

In general, we predict that intraparty disagreement increases when the party is perceived to have failed, but that parties rally around party leaders when they are in government. In particular, we predict that perceptions of incompetence among party members develop when a party has just lost an election, or for opposition parties when the sitting government benefits from a strong economy. Building on this approach, we argue in the next section that the outcome of party leadership elections depends on these statements of preferences as well as their broader election context.

THE CONSEQUENCES OF DIVERGENT PREFERENCES

We theorize that the causes of intraparty preference heterogeneity are linked to parties’ experience in government. The consequences of divergent intraparty preferences also have important implications for parties’ leaderships. Building on both the spatial and organizational perspectives, we theorize that political parties are dynamic organizations. As such, the parties’ leaderships depend on the support of intraparty groups to stay in power. ⁴² Recent reforms that increased intraparty democratization in the party’s leadership selection process, however, decreased the degree of internal competition. More candidates compete, but fewer perform well. ⁴³ Like parliaments and other organizations, we expect that the internal distribution of preferences influences internal votes for the party’s leadership and outputs.

Spatial theories assume that party leaders will hold preferences somewhere close to the party’s median position. ⁴⁴ We expect that the rules for selecting leaders likely influence which leaders are chosen, as well as leaders’ incentives for making statements of preferences. ⁴⁵ Although the rules for selecting party leaders differ across parties, we predict that elections for the party’s top leadership position will reflect competitive dynamics much like Downs’ expectations for plurality-based elections. ⁴⁶ From this perspective, party leaders who are selected using plurality-based election rules are likely somewhere close to the party’s median ideological position.

In most parties, there can only be one top party leader; the candidate with the most votes wins. However, there are prominent examples in which parties divide over multiple candidates. For example, in the French Parti Socialiste in the early 1990s, three candidates each received approximately one-quarter of the party’s vote. ⁴⁷ A single faction or group may dominate parties for an extended period of time, but changes to the distribution of intraparty factions may increase uncertainty over which faction gains the party’s support. ⁴⁸ In addition to a single top leader, many parties also have multi-tier structures that include committees of various sizes that assist the elected party leader.

We argue that candidates for the party leadership signal the relative location of their preferences to build support within the party. ⁴⁹ Applying this logic to intraparty politics, two candidates under plurality election rules will emphasize preferences that are close to the median intra-party voter’s position. In this context, the candidate who gains support from the median voter wins the election. Candidates for the party leadership will signal that their preferences are close to the median party member’s position.

Rational leadership candidates would select the median position if they only valued winning intraparty elections. Numerous reasons lead candidates to express statements of preferences that diverge from the median intraparty voter. Candidates face uncertainty about the exact distribution of preferences and the location of the median intraparty voter. Furthermore, candidates rely on internal party networks and factions that seek to pull their preferences away from the median position. For example, Ceron shows that party factions in Italy constrain party leaders’ ability to dominate the party’s position. ⁵⁰ Disagreements over the party’s broad strategy also encourage party leaders to deviate from the intraparty median position. Election-minded leaders seeking to avoid the appearance of ‘U-turns’ or ‘flip-flops’ in the future may try to balance their statements in anticipation of a general election. This tactic would cause candidates to emphasize policies that fall somewhere between the party and the electorate’s preferences. Party members also disagree over whether to emphasize pure policy goals or more incremental reforms. ⁵¹ These disagreements manifest themselves in party leadership candidates seeking to distinguish themselves from their competition based on their policy statements. ⁵²

Factors that more broadly influence parties’ short- and long-term electoral strategies will also manifest as intraparty disagreements. Elected officials and prominent party members have pre-existing reputations. ⁵³ These candidates face difficulties in responding to changes in public opinion or the positions of the party’s electorate without appearing at odds with their reputation. ⁵⁴ In this context, key votes for the party leadership give the party’s membership the opportunity to select between candidates with historical policy reputations and new competitors. Party member votes in this context serve as the way in which parties negotiate difficult questions of policy reform and strategy in response to changing political conditions.

Assuming that party members at congresses vote for the candidate with preferences closest to their own, we predict that candidates with more extreme preferences gain less support from intraparty groups. Candidates closer to the ideological center are more likely to attract a broad range of support. Like elections in other arenas, alternate election rules might lead candidates to hold more diverse preferences. ⁵⁵ In our next hypothesis, we predict that the relative distance from the ideological center influences the amount of votes a candidate receives.

Hypothesis 3: Candidates farther from the ideological center of the party receive less support than ideologically central candidates.

Although we hypothesize that ideological distance generally matters for party leadership elections, we also predict that the party’s government status moderates the effect of candidates’ statements of preference. As we predict for intraparty heterogeneity in Hypothesis 2, the consequences of deviating from the median are different for government and opposition parties. Government leaders are forced to develop policies in response to rapidly changing world events. ⁵⁶ In this context, party leaders will emphasize their success in implementing policies and the benefits of controlling government institutions more generally. Following from a principal-agent perspective, leaders can also use voter preferences to explain policy deviations from their stated preferences, as if party members and voter groups act as competing principals for controlling leader behavior. ⁵⁷ Intraparty groups, therefore, will place greater emphasis on their policy records, even if they deviate from the party’s median.

In contrast to government party leaders, those in the opposition have fewer external demands to deviate from the party median’s preferences. Instead, they face greater incentives from within the party to adhere to the median position. On the one hand, opposition leaders and candidates are freed from external constraints on their statements of preferences. Unlike governing parties, which demand compromise, flexibility, and intraparty unity to effectively govern, opposition parties do not have a concurrent policy record and reputation that they must address in their policy statements. On the other hand, party members demand greater ideological purity in this context because there are fewer external incentives to govern effectively or responsibly. ⁵⁸ Without obvious policy compromises and rewards from participation in government, members will be less forgiving of deviations from their preferences when the party is in the opposition. Instead, they primarily choose to support candidates espousing preferences most similar to their own. As we hypothesize in Hypothesis 2, this increased competition emerges as different intraparty groups express a wider range of preferences. Consequently, uncertainty about the distribution of intraparty preferences also increases. Therefore, changes in party leaders and broad preferences occur as the median intraparty voter is revealed and chooses between diverse competitors in the opposition. More broadly, intraparty demands for specific policy goals may therefore explain broad differences in parties’ electoral campaign tactics when they are in the government versus the opposition. ⁵⁹ Based on this logic, we predict in our final hypothesis that ideological proximity to the party median matters more for candidates from an opposition party than for those from a party in government.

Hypothesis 4 Government participation moderates the effect of candidate statements of preference. Ideological distance from the party center decreases support for candidates in opposition parties more than for government parties.

In summary, we argue that the causes and consequences of intraparty preferences are driven by parties’ electoral context and governmental role. More broadly, we think that intraparty factions and party members use speeches at party congresses to signal their internal strength and the location of their ideal preferences. Individual party members speak to signal their preferences and therefore pull leaders’ statements of preferences toward their own. Although in some countries party leaders control who speaks at party congresses, party delegates from France and Germany are free to participate at the party meetings because the party leadership seeks to avoid excluding supportive groups. ⁶⁰ Thus diverse groups within the party send representatives to speak on their behalf at party congresses. ⁶¹ These speeches closely approximate the delegates’ preferences at these meetings and the preferences of the party membership more broadly. In the next section, we describe our new dataset and method of deriving positions from intraparty speech.

DATA AND METHODS

To test our hypotheses, we collect data on intraparty leadership elections and speeches from the national congresses of four parties in France and Germany: the Parti Socialiste (PS) and Union pour un Movement Populaire (UMP) in France and the Christlich Demokratische Union (CDU) and Sozialdemokratische Partei Deutschlands (SPD) in Germany. These parties provide a difficult first test of our approach because of their similarities and differences. For example, all four parties have faced historical divisions, have experience in government and the opposition, are hierarchically organized and include strong intraparty factions. ⁶² Likewise, for each of the parties, the primary authority for selecting candidates is the party membership. ⁶³ These similarities mean that our results are likely to hold for parties with comparable organizations and histories elsewhere.

Despite these similarities, the parties also have a large number of differences, which allow us to test our hypotheses using the logic of a most different systems research design. ⁶⁴ In particular, the CDU and UMP tend to hold more conservative ideologies, whereas the SPD and PS both hold social democratic values. The ideologies of the conservative and social democratic parties diverge and represent different historical constituencies. ⁶⁵ Broadly, the French and German parties also compete in largely different electoral and institutional frameworks. Germany is a federal parliamentary system with a mixed-member proportional election rule, while France is a unitary, semi-presidential system with a two-round run-off voting rule. ⁶⁶ Therefore, if we find evidence that our approach fits our theory for these parties, then we can be somewhat confident that the differences between the systems do not contradict our theoretical approach.

To construct a measure of ideological distance, we require measures of the locations of actors’ statements of preferences within the party. While there are numerous methods of measuring parties’ and representatives’ preferences based on their behavior in government, there are few options available for measuring the preferences of intraparty actors separately from the policy process. ⁶⁷ Instead, we use a new data source that offers estimates of actors’ preferences at parties’ national meetings.

To construct our measure, we collected transcripts of parties’ national congresses from the parties’ websites and using the data archives of the Konrad Adenauer Foundation (KAS) and the Jean Jaurès Fondation (JJF). ⁶⁸ The KAS provides transcripts for all CDU congresses since 1950, although only the transcripts from 1990 onwards are easily machine readable. The merger of the East German CDU with the West German CDU in 1990 also provides a meaningful time point at which to begin our analysis. Transcripts for the SPD are only available online since 2002. For the PS, the JJF archive contains machine-readable transcripts from its first party congress in 1969 until 2000. Transcripts for the UMP are only available in their entirety for the 2004 congress. This results in 453 speeches at the national meeting for the CDU and 166 speeches for the SPD. We include forty-nine observations for party leadership candidate votes from 1969 to 2000 for the PS and three observations from 2004 for the UMP. ⁶⁹

We downloaded the transcripts from the foundations’ and parties’ websites and converted them into plain text files. ⁷⁰ We then created separate files for every speech given by delegates at each party congress. To ensure that the speeches convey the ideological preferences of the individual speaker, we removed all interjections and all speeches announcing other speakers or discussing procedural rules of the party congress. Our sample contains 1,649 speeches from the CDU covering twenty-three national party congresses from 1990–2011, 831 speeches from the SPD covering thirteen congresses from 2002–13, 1,138 speeches from the PS covering twenty-one congresses from 1969–2000 and speeches from the UMP covering one congress in 2004.

We then used WORDFISH ⁷¹ to estimate the party delegates’ ideological positions. WORDFISH is a statistical scaling model that draws on word frequencies to estimate policy positions in text documents. ⁷² The underlying assumption of the model is that the count y of words j in a document i is distributed according to a Poisson distribution:

$$y_{{ij}} \!\sim\!{\rm Poisson}\,(\lambda _{{ij}} ).$$

The parameter λ is the mean and the variance of the distribution, and takes the following functional form:

$$\lambda _{{ij}} \,=\,\exp ({\rm \alpha }_{i} \,{\plus}\,{\rm \psi }_{j} \,{\plus}\,{\rm \beta }_{j} \,{\asterisk}\,{\rm \omega }_{i} ),$$

where α is a set of text fixed effects controlling for the length of the document, ψ is a set of word fixed effects controlling for words that are generally used more frequently than others, β is an estimate of the word-specific weight capturing the importance of word j in discriminating between policy positions, and ω is the estimate of speaker i’s policy position. We are interested in the latter two parameters. To identify the model, both α and the mean of all speaker positions are set to 0 and the standard deviation is set to 1.

In contrast to other scaling approaches such as WORDSCORES, ⁷³ WORDFISH does not require reference documents with known positions on predefined policy dimensions. It only requires that the documents used in the analysis reflect the authors’ policy position on a single dimension. We estimate separate models for each party and party congress using all meaningful speeches as data. We therefore expect that the dimension underlying the data will correspond to a basic left-right policy dimension. ⁷⁴ Before we scale the documents, we prepare the data by removing stop words, numbers, punctuation and words that appear in nearly all of the documents. We also apply the German and French Porter stemming algorithm to reduce words to their word stems.

After applying the model, WORDFISH returns position estimates for each document text. Figure 1 displays the positions of some prominent members of the PS from the 2000 national party congress. The black dots are the position estimates of the speakers listed on the y-axis and the lines around the dots are 95 per cent confidence intervals. The distribution of estimates suggests that the positions of these speakers were quite divided during the party congress. The range of positions shown goes from Lionel Jospin on the right to Jean-Luc Mélenchon on the very left of the dimension. The other delegates have positions near (or just to the left or right of) the ideological center.

Fig. 1 Positions of PS speakers

Taken at face value, we argue that these estimates appear internally valid. In particular, we might expect that the speeches given by the party chiefs (François Hollande, Pierre Mauroy and Laurent Fabius) should be relatively close, as they all supported the motion proposed by Prime Minister Jospin. Similarly, we would expect that Henri Emmanuelli and Christian Bataille would be ranked to the left of Jospin’s supporters. Emmanuelli and Bataille strongly criticized the prime minister’s business-friendly social policies and founded a faction entitled ‘Democracy and Equality’ (Démocratie – Égalité) that called for more anti-capitalist policies. Similarly, an accurate measure of policy preferences would also rank Jean-Luc Mélenchon further to the left of Emmanuelli and Bataille, as he represented the most extreme faction, the Socialist Left (Gauche Socialiste). This faction also opposed Jospin and demanded radical reforms to counter the threats of globalization. ⁷⁵

Perhaps unsurprisingly, Emmanuelli and Mélenchon later founded a common faction called the New World (Nouveau Monde). Furthermore, Mélenchon left the PS in 2008 to found the Parti de Gauche in response to Ségolène Royal’s victory at the 2008 party leadership election. The positions of the speakers illustrated in Figure 1 correspond closely to the speakers’ respective factions. Jospin et al. are all close to the center or just to the right of it. Emmanuelli and Bataille are slightly left of the center and Mélenchon is farther to the left.

Like the example presented in Figure 1, we estimate separate WORDFISH models for each national congress. We then test our first and second hypotheses using a measure of intraparty disagreement based on the WORDFISH estimates. In particular, to determine whether intraparty disagreement increases after parties suffer electoral losses for government positions, we calculate the total variance of actors’ preferences for each national congress. We then use this variance as the dependent variable in our second analysis.

The primary independent variables predicting intraparty heterogeneity are electoral success, government experience and economic conditions. We operationalize electoral success as the change in the percentage vote the party received in the following national parliamentary elections. We also include the percentage vote for presidential elections in France (treating this variable as zero for German parties). Similarly, we operationalize experience in government using a dummy variable indicating whether the party is part of the governing cabinet or controls the presidency in France (treating this variable as zero for German parties). We then include an interaction of government incumbency with the change in annual GDP growth from the OECD. ⁷⁶

As our second dependent variable, we use the percentage vote for party leadership candidates at party national congresses. For the CDU and SPD, we use the total percentage of votes that each party leadership candidate received during a national congress. The CDU leadership consists of the party leader, general secretary, four deputy leaders, seven party chairs and twenty-six extended board members. The SPD’s leadership follows a similar structure, including the party leader, general secretary, five deputy leaders and forty-two extended board members. While elections for the top leadership positions of both parties are frequently uncontested votes of affirmation, elections for the extended board frequently face greater contestation. Like the deputy leaders and party chairs, the extended board members are elected using a Block Vote, or at-large plurality, system in which each voter can cast votes for as many candidates as there are positions available. ⁷⁷ This system potentially allows a majority group, even if it is barely a majority, to control 100 per cent of the positions. Carey and Shugart explain that this form of voting encourages candidates in national elections to develop their personal reputation, rather than the faction’s reputation, as the district magnitude increases. ⁷⁸

Because the French parties, the PS and the UMP, vote for a slate of candidates or a ‘motion’, rather than for a candidate, we use the total percentage of votes that each motion received during a national congress. Each motion includes a series of policy proposals for the platform and a list of leaders. For each motion, we identify the main initiator and future party leader and add the total percentage of votes that the motion received. ⁷⁹

We use the percentage of votes for the candidates for the German parties and for each motion submitted in the French parties as the dependent variable in our second set of analyses. We then use our WORDFISH estimates to create a distance measure by calculating the absolute distance between the position of each candidate (initiator) and the median position of speakers at each party congress. ⁸⁰ We also include some controls that have been found to be important in studies of electoral politics. We include dummy variables for whether the speaker has held the office before for the German parties and a dummy variable for each type of leadership position. We also include a dummy variable for the candidates’ gender to account for the party’s gender quota requiring that one-third of the board is female. We then include a dummy variable for the PS and UMP if the motion is supported by the current party leader.

ANALYSIS I: INTRAPARTY DISAGREEMENT

Our first hypotheses predict that intraparty disagreement increases with electoral losses and parties’ experience in the opposition. The results of our analysis indicate support for these hypotheses, although the total number of party congresses limits our ability to perform extensive regression analysis for any individual party. As a preliminary demonstration of our theory, we present graphical evidence for each party.

Figure 2 demonstrates the variance of speakers at the CDU’s national congresses. While the CDU was in opposition and faced decreasing levels of disagreement from 1998–2005, the levels of disagreement immediately increased following its electoral victory that year. However, disagreement decreased in the party congress in 2008 before returning to higher levels after the general election in 2009. Furthermore, the overall trend is somewhat static prior to 1998 and increases after 2005, while the trend for disagreement decreases from 1998–2005.

Fig. 2 Intraparty disagreement (CDU)

As Figure 3 shows, the SPD follows a similar post-election trend as the CDU. Prior to the 2009 election, the SPD participated in coalitions with the Green Party (prior to 2005) and then with the CDU. The SPD’s exclusion from government after eleven years in 2009 was followed by a large increase in disagreement, as it sought to wrestle with its new situation as an opposition party. Observers suggested at this time that the SPD would have to ‘reinvent itself’ by ‘resolving bitter internal disputes between economic reforms… and the party’s powerful left wing’. ⁸¹

Fig. 3 Intraparty disagreement (SPD)

The graphical results for the PS illustrated in Figure 4 also support the theory. The amount of disagreement for the PS varied substantially prior to the party’s 1981 victory. ⁸² In contrast to the CDU, disagreement sharply decreased between the party congress prior to the election and the extraordinary party congress immediately following its electoral victory in 1981. Over the following decades, the PS controlled both t./he presidency and the prime minister for long periods of time. Throughout this period, the PS’ party congresses exhibit a general trend of increased disagreement.

Fig. 4 Intraparty disagreement (PS)

To directly test the first set of hypotheses, we predict the amount of variance at party national congresses based on the parties’ experience in government. We first conduct separate analyses before testing the relationship in a combined model. Despite the small number of observations (24, 12 and 19), our analyses yield evidence that is consistent with our theory. ⁸³

In the first hypothesis, we predict that intraparty disagreement increases after an electoral loss. Despite the descriptive evidence, Table 1 shows only weak support for this hypothesis. For the German parties and the combined analysis in Table 1, having lost control of government leads to an increase in disagreement, but the coefficient is only significant for the SPD in Model 2. The coefficient is in the wrong direction for the PS. The coefficient for the change in the percent parliamentary vote the parties receive is in the correct direction for the CDU, the PS and in the Combined model, but is in the wrong direction (and significant) for the SPD. The coefficient for losing control of the presidency in France is only in the correct direction in Model 8, but never reaches statistical significance. Overall, there is at best weak evidence that losing an election leads to increased disagreement based on these estimates (Hypothesis 1). This might reflect Harmel and Janda’s perspective that only major electoral losses lead to broad intraparty change. ⁸⁴

Table 1 Disagreement at Party National Congresses

Note: Model 8 pools the observations from Model 1, 2 and 3, but also includes the one UMP party congress from 2004. Excluding the UMP leads to nearly identical results. Standard errors in parentheses.

⁺ p<0.10, *p<0.05, **p<0.01, ***p<0.001.

We demonstrate somewhat stronger evidence in support of Hypothesis 2 in Table 1. In particular, we argue that the economy influences parties’ internal politics, particularly when they are in the opposition (Hypothesis 2a). The coefficients for the interaction of change in percent GDP growth and the party’s government status are in the expected direction for the SPD, PS and the Combined models, and statistically significant for the SPD and the Combined test. Importantly, the constitutive term for change in percent GDP growth is positive in these models; opposition parties face increased disagreement under stronger economic growth. The magnitude of the coefficient for the interaction suggests that government parties may decrease their internal disagreement as the economy grows, but the combined effect does not reach statistical significance. This evidence is consistent with our prediction for government parties in Hypothesis 2b.

Figure 5 illustrates the effect of economic growth. As the rate of growth increases for opposition parties, intraparty disagreement increases. An increase of one standard deviation in percent GDP growth increases disagreement by 0.255 or a 16 per cent increase in the mean level of disagreement. This increase likely reflects infighting over control of the party when the leaders expect the party to perform poorly. Disagreement arises when intraparty groups believe the current leadership is performing poorly. In summary, these results suggest that electoral success may cause parties to act more coherently, but that opposition parties divide when they perceive their current leadership to have failed. This evidence is consistent with Hypothesis 2.

Fig. 5 Predicted effect of the change in percent GDP growth on intraparty disagreement

Note: the solid line is the median predicted intraparty disagreement. The dashed lines are the 95 per cent confidence intervals based on simulations using 1,000 draws from the estimated variance-covariance matrix from Model 4 in Table 1. Values for the independent variables are set such that the predicted effects are for an opposition party that did not change its vote or government status in the last election. The dotted line in is the distribution of change in GDP growth rate and refers to the right-hand axis.

ANALYSIS II: IDEOLOGICAL PROXIMITY AND LEADERSHIP SELECTION

According to Hypotheses 3 and 4, candidates for the party leadership will gain more votes from intraparty groups when they are more ideologically central, but ideological distance is less important if the party is in government. To test these hypotheses, we run separate regression analyses using an ordinary least squares regression with robust standard errors for each party first, and then in a combined model with random effects that we present in Table 2. ⁸⁵

Table 2 Percent Vote in Intraparty Elections

Note: robust standard errors in parentheses.

⁺ p<0.10, *p<0.05, **p<0.01, ***p<0.001.

The results in Table 2 indicate some support for the theory. In particular, ideological distance is in the correct direction for most of the models. Consistent with Hypothesis 3, we find that greater distance from the party’s ideological center will lead candidates to receive a smaller percentage of the vote. In particular, the coefficients are negative and significant for the SPD and the PS, as well as in the full sample in Model 8. Furthermore, the effect of ideological distance depends on the party’s status in government, as Hypothesis 4 predicts. Consistent with our theory, the coefficient for the interaction of government status and distance is positive in three of the models and is significant for the SPD and the full sample. This indicates that ideological distance has a different effect for parties in government. Participation in government causes the effect of distance to disappear, as the combined coefficients are no longer significant. As Hypothesis 2 predicts, candidates in government parties are insulated from the negative effects of their statements.

Figure 6 presents this effect graphically for candidates in the SPD when it is in the opposition. It shows that more ideologically distant candidates gain fewer votes than more central candidates. In particular, a decrease of one standard deviation in ideological distance increases the candidate’s vote by nearly 4 per cent for candidates in the SPD when it is in the opposition.

Fig. 6 Predicted effect of ideological distance on percentage candidate vote in the SPD

Note: the solid line is the median predicted change in the percentage vote a candidate receives. The dashed lines are the 90 per cent confidence intervals based on simulations using 1,000 draws from the estimated variance-covariance matrix from Model 6 in Table 2. Values for the independent variables are set such that the predicted effects are for a first-time female candidate running for a position on the SPD Extended Board who is not a member of Parliament when the party is in opposition. The dotted line is the density of candidate distances and refers to the right-hand axis.

The control variables in the first set of analyses are mostly in the predicted directions. Men hold an advantage in the PS, but not in the German parties. The difference between these parties may be unsurprising, since France only adopted quota laws in 2000. ⁸⁶ Incumbent candidates earn more support than non-incumbents. Members of parliament also benefit from their positions in office in the PS and the CDU, but not the SPD, although the coefficient is never significant. Finally, the dummy variables for the position type also suggest that candidates for lower-level positions in the German parties gain a smaller percentage of votes than candidates for the top leadership position.

CONCLUSIONS

The results of our analysis support our hypotheses on intraparty politics. Using speeches at party national congresses to measure the preferences of intraparty actors, we find that party leaders with more divergent preferences attract less support from intraparty elections in the SPD and the PS, but more support in the CDU. However, party leaders are generally isolated or protected from their statements when the party participates in government. These results hold up in a pooled analysis with additional data from the UMP. Parties’ experiences in government also influence the range of disagreement, although there is at best limited evidence that electoral losses increase intraparty disagreements. Instead, disagreements arise most strongly when opposition parties are perceived as incompetent or unaccountable, such as when economic growth increases.

Our analysis provides supportive evidence for numerous studies of party politics. In particular, while party leaders’ preferences may be closer to the party’s median position, intraparty preferences vary greatly and systematically, depending on the party’s experience in government. As mounting research shows, ⁸⁷ scholars should be wary of assuming that this intraparty preference variation does not influence the party’s election and governmental behavior. Intraparty heterogeneity likely influences the degree of cohesion between the party’s leaders and their ability to negotiate with potential coalition partners or on policy agreements. Our analysis of intraparty disagreement also provides researchers with a mechanism to study the effect of rules on parties’ parliamentary discipline by indicating the extent to which members act cohesively at the intraparty level. By systematically studying intraparty elections, we show one way in which intraparty groups may be able to resolve their disputes, through votes at national congresses.

Despite the small number of party congresses and the large ideological and institutional differences between the German and French parties, our results demonstrate that speeches at party national congresses provide useful information about the relative location of actors’ preferences. Our approach provides a more direct means of measuring actors’ preferences based on their own statements in a setting that is distinct from their behavior in office. However, we caution scholars against making overly large generalizations from our limited sample. Our case selection included parties competing in widely different institutional settings, but there are also a number of similarities in the German and French parties, which limits the external validity of our study. We are uncertain whether parties with different organizations, or which have never participated in government, act in similar ways. Also, alternate national or intraparty electoral rules might lead to different outcomes. We expect that future analyses would benefit greatly from the systematic collection and analysis of party congresses in a comparative, cross-national framework. By analyzing intraparty politics, we hope to gain deeper insights into the political process. Using these estimates, we hope that scholars of parliamentary behavior and public policy will be able to improve their ability to predict a wide range of behaviors and outcomes, such as coalition formation and termination, the number and type of public policies, and government oversight behaviors, without the fear that their estimates are endogenous to the processes they are seeking to study.