Leaders, Factions and The Game of Intra-Party Politics PDF

2 Factional preferences in Italy,
France and Germany
Factional politics: the Italian case

The life of Italian parties has always been characterized by a high level of
internal factionalism. As an example, inside the DC several subgroups were
active, including Dorotei, Nuove Cronache, Base, Forze Nuove and many others;
Sinistra Socialista and Autonomia coexisted and fought to prevail inside the PSI;
Presenza Liberale and Libertà Nuova animated the internal debates of the Italian
Liberal Party (PLI). Overall, in post-war Italy one could have found a number of
factions labeled as Rinnovamento, Riscossa, Iniziativa, Democrazia, Impegno
and Unità. These names, followed by an adjective (socialist, liberal, republican)
to identify their party affiliation, were equally spread across parties.
Factions have long been crucial for Italian politics. They were steadily men-
tioned in the everyday political chronicles and played a role in the legislative
arena. Many bill proposals were rejected due to factional vetoes. Many govern-
ments resigned due to intra-party disagreements and the fate of others heavily
depended on the outcome of party congresses (Giannetti 2010; Mershon 1994).
Factions were indeed able to promote or to end governments and political for-
mulas; it is not by chance that the formation of the first center-left government
coalition in the 1960s was postponed until DC’s and PSI’s factions that sup-
ported such an alliance became strong enough to overcome the resistance of
internal minorities. Indeed, the center-left Fanfani IV cabinet was appointed with
the decisive support of the PSI (February 1962) only after the Aldo Moro, leader
of a left-of-center DC faction, won the party congress (January 1962) gathering
support on a congress motion favorable to this center-left alliance.
Beside the DC, the PSI was factionalized as well. Since its foundation in the
nineteenth century the party has been composed of several subgroups, and its
far-left factions (communists and maximalists) were challenged by more
moderate reformists groups. In view of that, this party was repeatedly weakened
by the breakaways of moderate and radical wings. During the First Italian
Republic (1946–1994), almost all the relevant parties (including the
social–democrats, the liberals, the republicans and even the post-fascist Italian
Social Movement, MSI) were composed of organized factions (Lombardo 1972)
and were continuously affected by factional quarrels and compromises.
Factional preferences 29
Only the Italian Communist Party (PCI) was apparently cohesive and non-
factionalized. This was probably the consequence of strict and undemocratic
internal rules. The party was organized according to the principles of “demo-
cratic centralism,” combining strong party discipline in its external behavior
with (theoretically) wide freedom of thought in closed-door internal debates.
Notwithstanding this, at the end of the 1960s some intra-party groups formed,
though party rules did not allow them to stabilize and organize as stable and
recognized factions. Only one faction, called Manifesto, tried to create a stable
factional organization but it was expelled from the party in 1969; others factions,
i.e., the right-wing Miglioristi and the left-wing Cossuttiani, survived as latent
entities.
After 1994, a set of institutional changes led to the Italian Second Republic,
with a renewed and fluid party system characterized by huge party instability. In
such context factionalism, although softened, seemed to persist. All the main
actors that characterized the Second Republic are internally divided. Some of
them are openly factionalized while in others internal subgroups are more latent:
this happened, for instance, inside personal parties (Calise 2010) such as Forza
Italia (FI), the Northern League (LN) or Italy of the Values (IDV).
Inside the far-left Communist Refoundation Party (PRC) several Trotskyists
factions fought against the party mainstream. The right-wing National Alliance
(AN) was divided into three groups: a liberal–democratic faction, close to the
party leader Gianfranco Fini (Nuova Alleanza), was challenged by a group loyal
to the leader of the center-right coalition, Silvio Berlusconi (Destra Protagoni-
sta), and by a third far-right faction more sympathetic with pro-State positions
on the economic dimension (Destra Sociale). The center-left DS has been
affected, since its birth, by factional struggles promoted by an internal left (Cor-
rentone) and a right-wing group (Liberal); the mainstream faction, in turn, was
divided between Walter Veltroni’s partisans (Veltroniani) and the supporters of
Massimo D’Alema (Dalemiani).
Both the AN and the Democrats of the Left (DS) are no longer alive as they
merged with other groups to form new parties. However, their factions survived
and brought their pre-existing factional conflicts inside these new parties. The
DS merged with The Daisy (DL) into the PD. Two left-wing DS factions that
opposed this merger decided to break away. Despite this, the new party was
highly factionalized and the conflict between Veltroniani and Dalemiani con-
tinued inside it. The first PD leader, Veltroni, appointed in 2007 by a wide
margin and with the support of Dalemiani too, has been under attack by
D’Alema lieutenants since the beginning and was forced to resign quite soon. In
2009 the new party leader, Pierluigi Bersani, belonging to the Dalemiani faction
Riformisti E Democratici (Reformists and Democrats, RED), defeated the
incumbent Dario Franceschini (supported by Veltroni). After the congress,
Franceschini’s faction Area Democratica re-approached Bersani in order to
enhance party unity. Disappointed by this shift, Veltroni rallied his followers and
built a new organized liberal–democratic faction, Movimento Democratico
(Democratic Movement, MoDem), that overtly opposed the mainstream group.
30 Factional preferences
Right-wing factions managed to take revenge in the following party congress
(2013), when the moderate Matteo Renzi was appointed as the new party leader
(Ceron 2017).
Something similar happened to the AN. In the 2002 congress, factional con-
flict worsened and the party leader Fini changed the rules to preserve party unity.
Fini decreased the level of democracy (with a formal ban on factional grouping),
though at the same time he built intra-party routines based on compromise (e.g.,
all the factions retained ministers in the Berlusconi II Cabinet). Later on, a few
months before the fusion involving the AN and FI (which created the PDL in
2008–2009), a far-right splinter group coming from Destra Sociale (D-Destra)
broke away from the AN. In turn, the conflict between Fini’s partisans (Fare
Futuro/Finiani) and Berlusconi’s fans (Berluscones) that characterized the
internal life of the AN continued inside the PDL.
Inside the new party, however, reaching a compromise was harder. The agree-
ment set during the fusion assigned 70 percent of all spoils (in the party body
and public office) to former FI members; former AN factions had to divide the
remaining 30 percent and Fini’s followers only retained around 10 percent of all
posts.1 Contrary to FI, the AN was a heavily organized party, able to mobilize
members and activists. For this reason, Finiani subgroups were able to gain
more than 10 percent of approval among PDL members.2 As a consequence, Fini
demanded for a new congress in which each faction could have been weighted
according to its share of members’ votes. This was also a strategy to neutralize
the overriding power of Berlusconi, who was ruling the party alone. This fac-
tional strife was openly unveiled during a party meeting in April 2010, when
Fini claimed for more consensual intra-party dynamics. However, the main-
stream’s reply was to explicitly forbid any factionalism. The conflict worsened
in July when it became clear that public dissent was no longer accepted by the
leadership (as if the party was ruled through democratic centralism). The execu-
tive party branch lowered the degree of internal democracy stating that the leader
of the internal minority Fini was unfit to be a PDL member, given that his public
statements on several policy issues (justice, immigration, devolution, social
policy and civil rights) were considered to lay far away from the PDL platform
and outside the boundaries of its ideology. The executive party branch also
imposed sanctions on three members of Finiani. Such sanctions immediately
triggered the exit of the Finiani faction from the party.
This summary of factional struggles confirms that Italy can be considered a
benchmark to study the politics of faction (Zuckerman 1979), and the politics of
fission (Ceron 2015a), in terms of factional cooperation and conflict.
In this regard, the Italian case is particularly promising: the huge number of
relevant parties (around ten in each Legislature) and the high number of factions
(around three in each factionalized party) allow a wide number of observations
to be collected and a large-N statistical analysis to be performed.
Furthermore, the striking cabinet instability rate (60 governments in 64 years)
allows us to track reshaping in alliances, party policy positions and allocation of
office payoffs almost year by year; as such, the outcome of party congresses can
be put in relation with coalition bargaining: to give an example, the lapse of time
between one-party congress and the subsequent process of negotiation over
cabinet formation is very narrow, i.e., seven months on average. Analogously,
the wide number of general elections (17) held between 1946 and 2008 under
four different electoral systems provides a lot of information about electoral
payoffs and allows to test the effect of changes in the electoral laws.
On top of that, the Italian case can be considered as a kind of political labora-
tory suitable to generate and test hypotheses (Giannetti & Grofman 2011) thanks
to its multiple adjustments in electoral rules. In more detail, the electoral system
was a pure open list proportional representation (PR) from 1946 to 1992, except
in 1953 when a majority prize (two-thirds of seats) was theoretically granted to
the coalition able to gain a majority of votes. Elections were held under a mixed
system (75 percent plurality and 25 percent closed list PR) from 1994 to 2001.
Finally, in 2006, 2008 and 2013 the electoral system was a closed list PR with
thresholds and a majority prize for the coalition that won a plurality of votes;
this system, however, was replaced before the 2018 election by a new mixed-PR
system.
Interestingly, the Italian case is particularly suitable also due to the avail-
ability of a large amount of documents related to the internal life of Italian
parties and specifically to party factions (Ceron 2012). This element is crucial as
these data can be exploited to fill one of the major gaps in the literature on party
politics, i.e., the lack of information on the policy preferences of factions.
Although several authors pointed to the importance of collecting data about
factions’ preferences (Budge et al. 2010; Giannetti & Benoit 2009; Laver &
Benoit 2003), intra-party politics is often a sort of “invisible politics” that takes
place in the shadow (Sartori 1976). Consequently, finding information can be a
difficult task.
The present book contributes to filling this gap. Here I will present a dataset
related to the Italian case that covers almost 70 years of history, providing
information on intra-party rules and on the size and position of factions from
1946 to 2010. Overall, 254 motions presented by factions in 83 congresses of 18
parties will been analyzed.
Beside Italy, however, the book will also present data on intra-party politics
in France and Germany.
Beyond Italy: intra-party politics in France and Germany

For a long time, Italy has been considered as a deviant case in the realm of polit-
ical science, due to its extreme levels of cabinet instability and to the comprom-
ises required by oversized coalition governments, to its excessive political
fragmentation (with a large number of relevant parliamentary parties and fre-
quent reshuffles of the party system, particularly during the Second Republic),
and to the huge amount of factionalism, intra-party divisions and party splits or
parliamentary switching. Actually, these features have never been a peculiarity
of the Italian political system only. In fact, other countries shared similar traits.
Just to provide some examples, in Belgium and Finland one can find similar
levels of cabinet instability; in Denmark and the Netherlands strong party system
fragmentation and frequent party fissions (Ibenskas 2017) can be noticed; in
France too one can observe instability of the party system, with frequent changes
of party labels often linked with splits and mergers, as well as party switching,
which is also quite common in Greece and Switzerland (Volpi 2019).
More interestingly, recent real-world political events have shown that some
elements that were considered as peculiar to Italy can be now extended and
attributed to almost all the other European democracies.
Let us think, for instance, to the recent episodes of government instability or
to the long-lasting post-electoral negotiations involving rival parties in countries
such as Spain, Greece, Germany or the United Kingdom, which in the past were
considered as fairly stable or not used to coalition governments and political
compromises.
Analogously, the instability of the party system widens across Europe as con-
firmed by the recent splits in Spain and Greece, as well as by the rise of new
parties in many other countries. Along this vein, the topic of factionalism gained
relevance and intra-party divisions publicly emerged in Germany and France
(see below) but also in the United Kingdom, especially (but not solely) with
respect to the debate on “Brexit” (see the discussion in Chapter 1).
Precisely because Italy is not an outlier in terms of factionalism and it is not
the only country in which intra-party politics affects the political system, the
present book will compare the internal dynamics of political parties in two addi-
tional countries, i.e., France and Germany.
So far, I argued that Italy looks like a political laboratory (Giannetti &
Grofman 2011) and, due to its multiple institutional reforms, it represents a
quasi-experimental context that allows us to perform a comparison between very
different institutional settings. While holding constant other country-level fea-
tures, I can in fact assess whether my hypotheses hold both in the First Italian
Republic, characterized by a proportional electoral system and a relatively stable
party system (features that are similar, for instance, to the German case), and in
the Second Italian Republic, when the electoral system also includes majoritar-
ian elements and the party system becomes much more fluid (traits shared by the
French context too).
From this perspective, it is immediately clear why France and Germany can
be profitably compared with the Italian case. Nevertheless, other features typical
of these two countries provide additional reasons to include them in the analysis.
Let us start from France. With respect to factional politics, the French case
seems particularly suitable and it is perfectly comparable with the Italian one.
Indeed, intra-party politics in these two countries takes a very similar shape for a
number of reasons, including the fact that many parties in both countries hold con-
tested congresses to select the leadership and to set the party line, with several
intra-party factions that compete against each other by signing policy documents,
i.e., party congress motions or similar documents (sometimes shorter texts, such as
the profession de foi written by candidates running for the party leadership).
In this regard, as far as intra-party politics is concerned, the case of the Parti
Socialiste (French Socialist Party, PS) is particularly intriguing as its factional
structure and the shape of internal dynamics closely match those of many Italian
parties. The PS has been highly factionalized since its birth (Clift & McDaniel
2017; Ceron & Greene 2019). The ancestor of the PS, the French Section of the
Workers’ International (SFIO), was composed of competitive factions since 1907,
when the party leader Jean Jaurès institutionalized internal factionalism, establish-
ing that the party body had to be selected through PR of all factions, according to
the votes received by each factional motion. Although officially forbidden, factions
punctuated the history of the PS even in more recent years, starting from the
Epinay congress (1971). The party faced factional conflict and splits, as well as
attempts to cooperate producing syntheses (sincere or insincere) between the diver-
gent stances of all factions. To sum up, factions and factional leaders (ranging from
Jean Jaurès and Jules Guesde, to François Mitterrand and Michel Rocard, and
including leaders of internal minorities such as Jean-Pierre Chevènement or Jean-
Luc Mélenchon) have played a crucial role in the history of the PS.
Indeed, from 1971 until 2018, with the only exception of four congresses in
which there was no vote on congress motions and another three in which perfect
unity can be observed (a unique motion was presented and supported by all sub-
groups jointly), in the remaining 18 congresses (72 percent of times) intra-party
competition involved a minimum of two and a maximum of seven factions that
presented alternative congress motions.
Historically, the most long-lasting division inside the PS was the one between
the supporters of Mitterrand (mainstream) and the left-wing faction CÉRÉS (Centre
d’études, de recherches et d’éducation socialiste (Center for Socialist Studies,
Research and Education)), headed by Chevènement; however, at times, one noticed
either splits inside the mainstream or alliances between the mainstream and CÉRÉS
to form a majority. Furthermore, other factions closely tied to a prominent PS
leader played a role: there were small groups like the Mollétistes (partisans of Guy
Mollet) or the Poperénistes (Jean Poperen), as well as larger factions including
Rocardiens (Rocard), Fabiusiens (Laurent Fabius), or the supporters of Pierre
Mauroy. In recent years, the main cleavage involved the moderate supporters of
François Hollande, advocating more liberal–democratic positions, and the left-
wing faction Gauche Socialiste (affected by internal splits and mergers based on
the willingness to reach or not a compromise with the mainstream of the party).
Overall, the majority faction has often tried to synthesize different positions
to enlarge its support beyond the actual share of votes won in the party con-
gresses. Despite impressive episodes of internal conflict, as in 1990 and 2005
when highly divisive congresses were held, the PS party leaders have usually
attempted to foster unity (Ceron & Greene 2019). Even in the aftermath of the
extraordinary intra-party strife in 1990, rival factions agreed on ruling the party
jointly to restore unity.
The PS is not the only factionalized party though. On the left side of the
political spectrum, the congresses of the Green Party (Les Verts) were usually
contested by multiple factions that presented competing motions and there are
subgroups even inside the Parti Communiste Français (French Communist
Party, PCF ), particularly after the abolishment of democratic centralism when
public dissent became allowed (a pattern that is very similar to the case of the
PCI in Italy). The internal life of the Bayrou’s MoDem looks instead more
similar to that of Italian “personal parties,” with a more unitary environment in
which some divisions emerged when the congresses of the MoDem forefather
party (the Union for French Democracy, UDF ) decided to break the alliance with
the Gaullists and to adopt the new party name.
In turn, the Gaullists were internally factionalized too. Factions played a role
both inside the Union pour un Mouvement Populaire (Union for a Popular
Movement, PCF ) and in the ancestor party, Rally for the Republic (RPR). Active
since 1962, under different labels, the RPR officially recognized internal factions
only in 1989; internal disputes involved the mainstream faction (headed by
Jacques Chirac), some groups of liberal, moderate or reformer partisans (usually
pro-Europe, such as the Balladuriens/Conservateurs libéraux, followers of
Édouard Balladur), and a right-wing sovereignist Eurosceptic faction that gath-
ered the followers of Philippe Séguin (the Séguinistes/Gaullistes sociaux) and
Charles Pasqua (Gaullistes orthodoxe). As far as European matters are con-
cerned, in the 1990s the RPR split twice. Due to factional divisions the party did
not take a position on the 1992 referendum concerning the Maastricht Treaty:
Chirac and other party leaders campaigned in support of the Treaty, while
Séguin and Pasqua opposed it. Later, two Gaullists’ lists ran in the 1999 Euro-
pean election when the sovereignist movement led by Villiers and Pasqua (13.06
percent) won more votes than the official RPR list (12.82 percent) and broke
away after the election.
Factional conflicts continued inside the UMP, where the mainstream group
(headed by Nicolas Sarkozy and the Sarkozystes) was opposed by right-wing
factions, such as Debout la République (Nicolas Dupont-Aignan), that broke
away in 2007, or Droite Libre (Rachid Kaci). Internal divisions blew up even
more in 2012 when factionalism was officially authorized and, for the first time,
six organized factions contested the congress presenting competing motions.
This competition involved, among others, the right-wing La Droite Populaire,
the liberal–democratic France Moderne et Humaniste, as well as the “anti-
faction” faction La Boîte à Idées. These six subgroups, however, were also inter-
nally divided. In 2012 the leadership selection process generated an extremely
tied race between two candidates, Jean-François Copé and François Fillon, and
produced an impressive public fight (as intense as in the 1990 PS congress). The
first official results, showing a tiny margin of 98 votes in favor of Copé (50.03
percent versus 49.97 percent), were in fact rejected by Fillon who claimed to
have won the election (50.01 percent). When the party commission confirmed
the success of Copé (50.28 percent), Fillon immediately created a parliamentary
party group of UMP dissidents. Finally, three weeks later, the two candidates
reached an agreement to rule the party jointly and restore internal unity. Never-
theless, the divisions between the supporters of different UMP leaders, such as
Fillon, Copé, Sarkozy, Alain Juppé or Bruno Le Maire kept playing a role in the
internal life of the Gaullist Party.
Remarkably, as confirmed by this short summary of French factional politics,
France and Italy are also similar with respect to additional intra-party dynamics.
Concerning party splits and party fissions, in the post-war period, among Euro-
pean countries, Italy is the country with the highest number of splits, while
France ranks second, with a comparable number of party fissions that occurred
over time (one can find, respectively, no less than 17 splits in France and 23
main splits in Italy, along with additional minor events; see Ibenskas 2017).
Even when looking at party switching in the parliamentary arena, one notices
that Italy is the European country with the highest number of parliamentary
switches (Volpi 2019), followed by France that reaches almost the same levels
of switching: from 1946 to 2013 one can find an average of 7.88 percent
switches in the Italian Parliament and 7.58 percent in France. As such, intra-
party divisions are – at least in magnitude – very similar in these two countries
and produce wide effects on the party system as a whole.
Finally, in these two countries there is also a similar trend toward the direct
election of party leaders, often by means of (open) primary elections, particularly
in recent years. Accordingly, these changes in intra-party rules (see Chapter 3)
can be profitably put in relation to factional conflicts.
To sum up, for the French case I collected and analyzed 154 textual docu-
ments (congress motions or similar documents, such as the already mentioned
profession de foi) that allow the estimation of the policy positions of 151 party
factions related to 39 party congresses held by five French parties in 45 years,
between 1971 and 2016.
Turning to Germany, the selection of this case refers to other criteria. Fac-
tionalism and intra-party politics here are somehow different from the French
and Italian cases and the same applies to internal rules. To start with, parties tend
to present themselves as monolithic blocs and, apparently, they are less affected
by splits. Contrary to what has just been discussed with respect to Italy and
France, in Germany the average rate of switchers is six times lower and amounts
to 1.25 percent in the post-war period; analogously one can only account for
two-party fissions in the same amount of time.
Internal divisions tend to be hidden, especially during party congresses.
Factionalism manifests itself elsewhere, for instance, within the parliamentary
party group where the strength of each faction is measured according to the
number of MPs that decide to affiliate with one or another subgroup.
Accordingly, during the party congress there is no open competition between
rival factions to get a majority around a congress motion that outlines the party’s
ideological line. Such different dynamics usually end up in unitary congresses in
which the party leader is elected by overwhelming majorities if not by unan-
imity. For instance, when focusing on the two main German parties, the CDU
and the Social–Democratic Party of Germany (SPD), one notices that on average
party leaders won 90.6 percent of support. In the 41 congresses held by these
parties in the last 35 years, between 1983 and 2018, the party leader got a vote
share lower than 80 percent only five times. Remarkably this occurred three
times in the most recent years, i.e., in 2015 and 2018, for the SPD, and in 2018
for the CDU.
In the SPD, Sigmar Gabriel (2015) and Andrea Nahles (2018) were selected
respectively by the 74 percent and the 66.35 percent of delegates (the latter
represents the second worse results ever for a SPD leader).
In turn, the CDU approached its 2018 party congress showing a spike of intra-
party divisions (see the discussion in Chapters 5 and 6). The leadership of
Angela Merkel faced resolute internal disagreements. For the first time since
1973, the CDU/CSU parliamentary party group rejected the candidate proposed
by the CDU leader as the new party whip (i.e., the chief of party’s back-
benchers). By a narrow margin (52.7 percent), rebel MPs selected an alternative
candidate that pledged to be more independent from Merkel’s viewpoints and to
renew the party. This division is even more troublesome given that such a wide
number of rebel MPs (125) would be sufficient to defy the majority retained by
the government coalition between the CDU and the SPD.
Even more interestingly, for the first time in recent years, inside the CDU a
group of party activists has been building a truly organized party faction to chal-
lenge the moderate turn imposed by the leadership of Merkel and to restore a
more conservative ideology inside the party, changing the line accordingly. As a
consequence, in December 2018, the CDU held the first contested leadership
race since 1971. The outcome was extremely tied: the pro-Merkel candidate
Annegret Kramp-Karrenbauer was appointed with a very narrow margin (51.7
percent), raising doubts on the future unity of the party.
While, from many perspectives, the German case looks like a different scen-
ario, suitable to evaluate whether the theory presented in the book holds even in
a radically different intra-party setting, these recent events seem to suggest that
intra-party politics in Germany is changing a lot. Exactly like German politics
and the German party systems have been changing in the last decades, with an
increasing fragmentation, instability and more complicated post-electoral pro-
cesses of coalition building, one can expect a rise in intra-party conflicts and, in
a near future, it could be possible to observe stronger similarities between the
structure of factional politics in Germany compared to what already happens in
Italy and France.
So far, however, the German case still represents a further stringent test suit-
able to evaluate the implications of my theory in a rather different scenario.
Taking advantage of this, I investigate the logic of intra-party dynamics in more
depth, evaluating what happens to parties in which the internal conflict takes
other and different shapes rather than factionalism in party congresses. Accord-
ingly, I analyze the German case focusing on non-factional intra-party organiza-
tions or on particular factional subgroups called kreis (circles) or wings.
While in Germany middle-sized parties such as the liberals or the left are also
internally divided, to streamline the analysis I will focus on the two main
German parties, i.e., the CDU and SPD. These two parties differ from each other
in terms of the shape of internal subgroups. Inside the SPD, in fact, there exist
(formal or informal) organized factions. Although these factions do not really (or
not always) compete during the party congress, they are actually long-lasting,
stable and well-structured intra-party groups with their own ideology and their
own organization. This includes membership fees, hierarchies, meetings, maga-
zines, merchandising and updated websites containing information and docu-
ments drafted to express the viewpoints of the faction itself. These documents
are usually called position papers and contain ideas and contributions written to
affect the policy manifesto of the SPD.
These organized factions are really able to influence the direction of the party
(Debus & Bräuninger 2009). In this regard, factionalism inside the SPD repli-
cates a pattern that was already observed elsewhere. Its historical roots date back
to the 1970s. At that time, intra-party politics took the shape of a fierce internal
fight between ideologically rival factions and two main wings were created. In
1966 young left-oriented SPD members gave birth to the Frankfurter Kreis
(Frankfurt Circle, FK), which was particularly influential between 1982 and
1998. Conversely, moderate members retaining more liberal–democratic views
in 1974 coalesced into the right-wing Seeheimer Kreis (Seeheim Circle, SK),
which was particularly strong in the 1970s and after 1998 (Debus & Bräuninger
2009).
Nowadays this division still exists, though the intra-party complexity inside the
SPD has widened. On the left side, the Demokratische Linke 21 (Forum of the
Democratic Left 21, DL21)is the heir of the FK. Founded in 2000 to advocate a left
turn of the SPD, it is so structured that it has been defined as “a party inside the
party.”3 This faction was also headed by the current SPD leader Andrea Nahles. In
spite of her left-wing origins, in the 2018 congress Nahles was challenged by a
rival candidate, Simone Lange, who held more left-wing positions and after the
congress launched a new cross-party movement (Aufstehen/Stand Up) that aimed
to strengthen the left-wing stances in German politics.
Inside the SPD, the Parlamentarische Linke (Parliamentary Left of the SPD
parliamentary group, PL) is another left-wing faction; active since 1980 as the
heir of the Leverkusen Kreis (created in 1972), this group supports a Keynesian
approach to economic policies. The Parlamentarische Linke is the strongest
faction in the SPD parliamentary party group; in the new Bundestag elected in
2017 it represents 76 MPs out of 153 (49.67 percent) and its size was similar in
the previous parliament when the PL was still the first subgroup, supported by
86 MPs out of 193 (44.55 percent).
On the right side of the party, the conservative Seeheimer Kreis is still active
and in the previous Bundestag it was the second largest group with 70 MPs (36.26
percent) affiliated to it. For a long time, it could have been considered as the main-
stream SPD group, given that many prominent politicians and many SPD leaders
including Martin Schultz, Sigmar Gabriel, Thomas Oppermann, Peer Steinbrück
and Frank-Walter Steinmeier were considered close to the SK. This faction also
played a crucial role in supporting Gerhard Schröder’s Agenda 2010 welfare
reform. This reform was also supported by another moderate faction, the Netzwerk
Berlin (Berlin Network , NB). Created in 1999 by young party members, this
faction retains pragmatic and reformist positions. At the beginning factional affili-
ation to the NB was allowed on a generational basis only. Later this group
expanded its membership and it now accounts for around 25 percent of the parlia-
mentary group, though some of its members (including Gabriel, Steinbrück,
Steinmeier and Oppermann) are cross-affiliated to the SK faction.
While the internal life of the SPD is quite similar to that of French and Italian
parties, things are different inside the CDU. Until recent years, there were no
organized factions in this party (see the discussion in Chapter 1). Internal associ-
ations, however, exist and can play a role. The most important groups are the
lobbies of employees and employers (Debus & Bräuninger 2009).4 Founded in
1945, the Christlich-Demokratische Arbeitnehmerschaft (Christian Democratic
Employees, CDA) represents the interests of Christian–democratic workers and
labor groups; relying on the Catholic Social Doctrine, the CDA supports welfare
state issues and traditional family values. Conversely, the Mittelstandsvereini-
gung (Middle Class and Business Association, MIT) promotes the interests of
the middle class, as well as those of small and middle-sized companies or free-
lancers. Active since 1956, it is stronger and more influential than the CDA
(Debus & Bräuninger 2009). Although these two groups are not proper factions,
they still retain contrasting positions. Accordingly, they present their own posi-
tion papers and policy programs (Grundsatzprogramm) as well as enacting deci-
sions and resolutions (Beschluss) approved during the national conferences held
by the CDA and MIT. As such, I will exploit these documents to get some
insights on the different viewpoints existing inside the CDU and to provide a
measure of internal polarization over time.
Overall, by analyzing intra-party documents released by factions or intra-
party associations, I have been able to estimate the policy position of five sub-
groups at 33 different points in time; 20 of these estimates refer to subgroups
inside the CDU, while 13 are SPD factions. These data have been used to map,
year by year, the internal structure of these two parties over the last 33 years
(from 1985 to 2018).
Party congress, the arena where factional bargaining

takes place
To assess intra-party equilibriums and dynamics I need to gather data about the
factional structure of parties. In particular, I need to know the number, the
strength and, most importantly, the preferences of each faction. There are
different sources of information on factionalism; leadership strength can be
evaluated by looking at cabinets (based on the number and type of portfolios
received by each faction) or legislatures (based on parliamentary memberships,
i.e., the number of MPs affiliated to each faction), even though the most suitable
indicator is probably “the voting strength of individual factions on party
decisional bodies such as party congresses (using motion votes as indicators of
factional strength) or on party executives (using individual factions’ seat shares)”
Boucek (2009: 467).
Analogously, scholars can determine policy positions of subgroups looking at
a party body or analyzing their behavior in the legislative arena. Although some
authors showed that it is possible to estimate factional preferences through roll-
call votes (Rosenthal & Voeten 2004; Spirling & Quinn 2010), it has been
argued that “content analysis on texts drafted by […] intra-party groups seem to
be the best choice to identify their respective preferences, in particular if the
research question deals with changes of the positions of political actors over
time.” (Bäck et al. 2016; Benoit et al. 2009: 443). In a congress motion each
faction is (almost) completely free to present its view about the party platform;
in this context the discipline (Laver 1999) only plays a limited role in bounding
the expression of factions’ preferences so that texts should be less influenced,
compared to roll-call votes (Spirling & McLean 2007). In addition, not all votes
are roll-call votes, thus they do not provide a complete picture of voting behav-
ior. The decision to call a roll-call vote may also be strategic and biased accord-
ing to the issues or the group that request it (Carrubba et al. 2006; Giannetti &
Benoit 2009). Roll-call analysis is strictly related to voting behavior, which pro-
vides actors with a limited set of options: “yes,” “no” or “abstention,”5 while
there could be a variety of preferences behind a similar voting behavior: actors
may say “yes” or “no” for very different (and sometimes opposite) reasons. Text
analysis helps scholars to discover and highlight the actual positions of political
actors by providing a broader set of information about their preferences. For all
these reasons I will analyze factions’ positions through text analysis focusing on
party congresses.6
Scholars suggest referring to party congresses as a source of data (Boucek
2009; Giannetti & Benoit 2009; Giannetti & Laver 2009; Mershon 2001). Indeed
“Many gains in understanding intra-party politics may derive from analysis of
internal party organization, such as formal and informal factions” (Giannetti &
Benoit 2009: 235) in particular by examining “debates at party congresses
between different party factions” (Giannetti & Benoit 2009: 234). This is par-
ticularly true for Italy and France, but the party congress outcomes and the
debates held therein are also crucial to investigate intra-party politics in other
countries, including the UK and Germany (Greene & Haber 2016, 2017).
Internal debates can be more suitable if compared to parliamentary speeches, as
these can be subjected to party discipline and “may not reflect the true distribu-
tion of preferences” (Proksch & Slapin 2012: 520).
The party congress represents the key moment in a party’s internal life, like
general elections are the crucial starting point for representative democracy.
Indeed, “For factions, the closest approximation to that sort of competitive arena
is internal party elections, since it is there that factions organize teams of candid-
ates and appeal to people enjoying the right to vote for one team or another”
(Mershon 2001: 561). During the party congress those subgroups willing to
organize as a faction can present a list of candidates to the party body. Often this
list is attached to a “motion,” i.e., an omnibus policy document that aims to
shape party strategy and ideology. Motions set out factions’ “opposing views on
the ideological direction of the party” (Giannetti & Laver 2009: 154). After a
public debate, party congress’ delegates vote on the policy cornerstones and
establish the new party line that the leadership should pursue. They appoint
members in the party executive and the party body (committees in charge of
running the party); elect (directly or indirectly) the party leader; adjust and ratify
the party statute; set the new party line and strategy casting a vote on a motion
that commits the leadership to some general policy statements until the next con-
gress. In fact, “party delegates vote on the policy principals in an annual confer-
ence whereas the balance of power between faction, that is, the proportion of
votes that each faction receives, is translated into policy recommendations”
(Levy 2004: 251). During the competition for gaining members’ support, the
party splits in many subgroups. The party congress, however, is not only a
source of division. Instead, this is also the arena in which the party tries to reach
its unity through a compromise between internal factions. If the bargaining is not
successful, the party can more likely incur a breakup. Indeed, many fissions take
place during the congress or soon after.
During the party congress, politicians who want to distinguish themselves from
the party mainstream can exploit such visibility toward party members and voters
in order to show that they retain different views and policy positions. For these
reasons party congresses can be used systematically to map the factional structure
of parties (Giannetti & Laver 2009). Data drawn from congresses provide us with
objective criteria to analyze intra-party divisions. Information that flows from offi-
cial sources is the only way to overcome scholars’ personal beliefs about the actual
number of factions inside parties. In addition, these data often provide a measure of
the strength of each faction, based on the number of congress votes gained by each
list or the number of members elected in the party body (when data on members’
votes are lacking). Finally, party congress debates, based on textual documents
produced by the party leadership or by rival factions, also allow assessment (by
means of a content analysis) of the policy preferences of different subgroups; in
particular, recent developments in the field of quantitative text analysis can be
exploited to estimate factions’ ideal points by analyzing congress’ motions.
For these methodological reasons, and given the need to identify intra-party
subgroups according to objective criteria, in the Italian case (and in France too) I
consider as a faction each party subgroup that contests the congress presenting
its own list of candidates attached to a policy motion (Giannetti & Laver 2009).
Accordingly, I will be dealing with divided parties, excluding from the analysis
political actors such as the “personal parties,” as in these parties the leader exerts
a strong influence on members and activists (e.g., Raniolo 2006). Even when
these parties are internally polarized, such divisions might not evolve into organ-
ized factions given that those parties are, by definition, “parties without factions”
(Lombardo 1972). Indeed, they held congresses very rarely and these meetings
are almost never contested (being merely conventions devoted to celebrate
internal unity and to show the charisma of the party leadership).
Similarly, unitary congresses, i.e., congresses in which there is no competi-
tion between factions and the delegates approve a common unitary motion, are
excluded from the analysis. Sometimes minority factions might refuse to present
their motion and their list for various reasons; they might agree with the main-
stream or they do not feel strong enough to defy the leadership. Indeed, the
game–theoretic model proved that the existence of internal dissenters is not a
sufficient condition for the emergence of voice (i.e., minority motions). This
does not happen only in the presence of strong and autonomous leaders but also
when the party is ruled through inter-factional compromise.
Such agreements in fact could be reached either after the congress (if this
happens, one would observe internal conflict) or before. In the latter case, I
would not be able to measure intra-party division and conflict (even if it exists)
because disagreements have been already solved: the unitary congress motion
(as well as the division of party body’s spoils) expresses a common equilibrium
on which all subgroups agreed.7
Concerning France, the data collection process has been quite similar to Italy,
even though, to maximize the number of observations, I relaxed some of the cri-
teria discussed so far (see below for details). Conversely, in the German case, as
I already noticed, things are slightly different and I do not revert to party con-
gresses as a source of data on the positions and the strength of rival factions.
Nevertheless, the outcome of German party congresses will still be taken into
account. In the German case I resort to party congresses not as a source of data
to catch the main independent variables (heterogeneity, polarization and policy
positions) but as an arena that produces political outcomes; textual data coming
from external sources will be put in relation with such congressional outcomes.
This will be done, in particular, with respect to the leadership’s contest and dura-
tion and to manifesto documents drafted by the party.
Data collection on Italy, France and Germany

For the Italian case, I gathered motions from the official proceedings related to
national congresses published by parties, and from parties’ official newspapers
or magazines.8 Motions related to the most recent congresses have been down-
loaded from the official parties’ websites.9 I tried to collect data on parties’ fac-
tional structures analyzing the whole population of party congresses held by
Italian parties (approximately 300). However, some of them (roughly 50 percent)
should be excluded for the methodological reasons discussed above. Unfortu-
nately, due to a lack of data sources I was not able to find information on the
whole population of contested congresses. After the collapse of the Italian party
system in 1992–1994, many of the existing parties disappeared and their archives
disappeared as well; moreover, issues of some parties’ newspapers are not
always available. I have been able to find information on approximately 50
percent of the total number of contested congresses. Overall I gathered 254
motions related to 83 congresses of 18 parties (see Appendix 2 for details on the
estimated policy positions). Almost 200 of them (nested in 64 congresses)
concern the First Republic while the remaining data (19 congresses with 56
motions) belong to the Second Republic. On average the number of motions per
congress is three. Table 2.1 shows more details about the motions gathered.
Table 2.1 Italian data: list of parties, congresses and motions included in the analysis
Party Total Contested Analyzed % on Total % on Contested Missing Motions (Total) Motions (Avg.)
AN 3 1 1 33 100 0 4 4
DC 18 13 11 61 85 2 41 3.73
DS 4 4 4 100 100 0 12 3
FV 18 16 2 11 13 14 6 3
MSI 18 7 5 28 71 2 20 4
NPSI 6 2 2 33 100 0 4 2
PCI 16 3 3 19 100 0 8 2.67
PD 3 3 1 33 33 2 3 3
PDA 2 2 1 50 50 1 3 3
PDCI 5 1 1 20 100 0 2 2
PLI 18 11 11 61 100 0 35 3.18
PRC 8 6 6 75 100 0 20 3.33
PRI 22 15 11 50 73 4 25 2.27
PSDI 24 21 9 38 43 12 25 2.77
PSI 24 12 12 50 100 0 38 3.17
PSIUP 4 1 1 25 100 0 3 3
PSOC 2 1 1 50 100 0 3 3
UDC 3 1 1 33 100 0 2 2
Total 198 120 83 42 69 37 254 3.06
For several parties (AN, PD, Action Party (PDA), Party of Italian Commu-
nists (PDCI), Socialist Party (PSOC), PSIUP, Union of Christian and Center
Democrats (UDC)) I found data concerning only one congress. Conversely the
maximum number of congresses analyzed pertains to the PSI (12 congresses)
and to the DC, PLI and PRI (with 11 congresses each). I found also nine PSDI
congresses and six related to the PRC.10
The present dataset contains more information on some parties than on others:
there are 38 motions nested in 12 PSI congresses but only two motions presented
in the unique contested congress held by the PDCI and UDC. This feature
however does not affect the analysis (see for instance Chapter 6). Column 7 in
Table 2.1 provides an estimate of the number of missing congresses, considering
only the contested (or presumably contested) ones. Among the parties included
in the analysis there are 37 missing cases whereas the (estimated) number of
contested congresses amounts to 120. The percentage of missing cases, then, is
around 30 percent.
With a few exceptions, the dataset covers the entire set of contested con-
gresses for a large number of parties. There are multiple missing congresses only
for two parties: PSDI and the Greens. The PSDI merged its archive with the PSI
(after 1966) but some data get lost; the Greens hold a National Assembly
approximately once a year and their internal life is particularly tangled so that it
was not possible to gather data on several old congresses. Interestingly, I
managed to account for the large part of internal debates within the DC, PLI,
Italian Republican Party (PRI) and PSI and I collected almost all the data on the
DS and PRC.
For the French case, I relied on online sources only. Several motions have
been downloaded directly from the official parties’ websites, sometimes these
documents were retrieved using the Web Archive and sometimes they were
indexed on Wikipedia. In turn, Wikipedia or the website France-Politique
(www.france-politique.fr/) also provided additional information on the outcome
of the congresses. Beside these sources, for the PS, the Socialists Archives
(www.archives-socialistes.fr/) were particularly useful as they provide a full
picture of the PS congresses from 1939 to 2005;11 they report speeches, motions
and any other information related to the congresses, referring directly to the offi-
cial party’s newspaper.
Overall 154 textual documents have been gathered to estimate the policy
positions of 151 party factions in 39 party congresses held between 1971 and
2016 (see Appendix 2 for details on the estimated policy positions). However,
with the exception of PS, data on other parties are related mostly to years
after 2000.
In one case (Fillon, 2012 UMP congress) two different documents have been
used in the analysis (the short profession de foi and a longer text outlining his
policy project). Furthermore, two textual documents related to the splits of the
PS have been included to anchor the results of the analysis: a statement released
by Marc Dolez and Jean-Luc Mélenchon that announced their choice to break
away from the PS and a policy document of the splinter group La Gauche
Moderne (2007), the movement created by Jean-Marie Bockel after his split.
Notice that, to maximize the number of observations, in the 2006 congress of the
UDF–MoDem, the speech of the leader of the dissidents has been used as a
proxy for the position of the internal minority.
Analogously, to maximize the number of cases in the dataset, I also included in
the analysis the unitary motions that have been presented in a few party congresses
(in detail, the 1984 congress of the Greens, the 2000 congress of the PCF, the 2008
and 2010 congresses of the MoDem, and the PS congresses held in 1976, 1981,
1987 and 1991), with the support of all intra-party factions and subgroups. By
doing that, and considering these documents as the factional compromise reached
before the congress, I managed to increase the number of observations comparing
such inter-factional agreement with the outcomes observed in terms of party policy
positions, payoffs allocation, party splits or leadership duration. For similar
reasons, information on contested party congresses has been considered, when
easily available, even if rival factions did not present competing motions or if these
motions were missing. As such, the total number of rows in the final dataset
increases up to 166 and this allows us to perform a more coherent and complete
test of hypotheses, especially with respect to the internal life of the PS.
I managed to gather motions related to 20 out of 24 PS congresses (83.3
percent) held between 1971 and 2015; I lack data on policy positions only for
the period 1992–1994, when the PS held three congresses in four years. I also
grasped the entire internal life of the UMP (analyzing all the four congresses
held in 2002, 2004, 2012 and 2014 and including also documents related to the
2016 primary election) and the MoDem (I have data on three congresses, from
2006 to 2010; though the first one is the last congress of the UDF, in which the
creation of the MoDem was approved). Unfortunately, I have no data on ancestor
Gaullist or centrist parties (notice that the Gaullist parties officially recognized
internal factions only in 1989 so the number of missing contested congresses is
very low). Conversely, there is only scant information on the internal life of the
PCF and the Greens. The PCF started to hold some contested congresses after
1994, when the rule of democratic centralism was abolished (see Chapter 3);
therefore, the number of missing contested congresses related to it is almost null.
For the German case, first and foremost I relied on documents collected and
analyzed by Debus and Bräuninger (2009) that have been kindly provided by the
authors. Furthermore, these documents were integrated with updated policy
papers issued in recent years by the main intra-party subgroup such as the CDA,
MIT, SK, DL21 or NB. These documents were retrieved from the official web-
sites of SPD factions and CDU associations.12 In some cases, multiple docu-
ments have been used.
While the book focuses only on CDU and SPD, in the textual analysis I also
included documents related to the factions of the Liberal Democratic Party
(FDP) or electoral manifestos of the Party of Democratic Socialism (PDS) that
are helpful to anchor the analysis (as in Debus & Bräuninger 2009), as well as
documents of associations of employees and employers within the SPD, though
these will not be considered in the empirical analyses.
Overall, 62 texts have been analyzed to estimate policy positions of 33 sub-
groups (see Appendix 2 for details on the estimated policy positions). These
estimates have been used to detect changes in factions’ positions and polariza-
tion inside the CDU and the SPD over time, from 1985 to 2018. Data have been
imputed in order to produce yearly estimates; the past estimate has been used
when there was no updated information on the policy preference of an internal
subgroup.
In the next section I illustrate how all these textual documents, concerning
Italy, France and Germany, have been analyzed in order to get factions’ policy
positions.
From words to fact(ion)s: estimating factions positions using

Wordfish
To extract factions’ policy positions from congress motions or policy papers I
need to apply some sort of text/content analysis technique. In political science,
one of the oldest methods of analysis consists in the hand-coding of textual
documents, following strategies similar to that of the Comparative Manifesto
Project (CMP). The CMP analyzes electoral manifestos presented by parties
through manual coding (Budge et al. 2001). For each manifesto, human coders
divide the text into a number of quasi-sentences and assign each of them to one
of the 56 pre-determined categories, according to a classification scheme. This
technique can be time consuming, particularly when the amount of texts is quite
large. Given the high number of texts gathered (almost 500), and their wide
length (overall, they contained 3,599,740 words) hand-coding strategies would
not be suitable.
To cope with such concerns, scholars started to develop semi-automated and
automated techniques of text analysis. Among the others it is worth mentioning
Wordscores (Laver et al. 2003) and Wordfish (Slapin & Proksch 2008; Proksch
& Slapin 2009). They provide quicker (but still reliable) means to determine the
ideal points of political actors. Accordingly, I will analyze data through Word-
fish, for the reasons explained below.
Wordfish is an automated scaling technique that produces estimates of policy
positions comparing several textual documents. This method was immediately
applied to different text sources (manifestos, speeches, statements and pledges)
measuring policy positions of parties and interest groups in several countries
(e.g., Klüver 2009; Proksch & Slapin 2009, 2010) and it is now largely used by
political science scholars.
Wordfish assumes that the words’ usage is informative of the policy position
of each textual document relative to others.13 It further assumes that, for each word,
the probability of appearing in a document is independent of the presence of other
words, therefore it considers each text as a vector of randomly distributed words.14
Looking at the relative frequencies of words contained in each text, Wordfish
allows the differences between alternative documents to be determined, scaling
them on a common latent dimension thereby measuring their relative distance and
policy position. Following previous studies on text analysis, Wordfish assumes that
words follow a Poisson distribution, which is suitable to match the skewed distri-
bution of words in texts. The Poisson is defined by one parameter only as the mean
and the variance of the distribution coincide. This allows the analysis to be simpli-
fied, decreasing the computational effort. The distribution then could be summa-
rized as follows: yijt ~ Poisson(λijt), where yijt measures how often the word j appears
in the document related to party (faction) i at time t. In the Wordfish analysis four
parameters are estimated: the document position on the latent scale, at time t, along
with its fixed effect (which accounts for the different length of the texts), and the
discrimination parameter (β) of each word, as well as its fixed effect (which is the
logged mean count of each word across all the documents). This latter parameter
allows for the higher or lower usage of certain words across different texts to be
controlled. Common words not associated with any political meaning (i.e., articles,
prepositions, etc.) appear frequently but are uniformly spread across documents.
These words retain a strong fixed effect while their discriminating power will be
close to zero. As such, the fixed effect ensures that rare words are properly taken
into account. Words that appear only in one document should carry a theoretically
infinite discriminating power as they would be sufficient to identify that document;
to deal with this, Wordfish decreases the β values of very rare words.
The two main parameters of interests, however, are the policy positions of
textual documents (with mean 0 and standard deviation equal to 1) and the
words’ discrimination parameter.
Starting from textual documents of factions and subgroups, Wordfish will
then produce the estimates of factions’ policy positions. They are scaled on a
single latent dimension whose meaning has to be interpreted ex-post according
to the political connotation of the texts submitted to the analysis.
For instance, if documents are related to a single issue (e.g., environmental
policy) this topic will define the latent policy continuum. To the contrary, when
documents outlining the policy views of a political actor on a variety of issues
are considered (as for party manifestos or faction motions) the latent dimension
should be interpreted as a more general left–right scale.15
The other main parameter, β, is the word discrimination parameter. It measures
how much each word is helpful in classifying documents; it also corresponds to the
word’s “policy position” along the latent scale, and allows whether the word stands
on the left or on the right of the continuum to be evaluated. Words that appear with
a high frequency across all documents have a discriminating power close to zero
while words that appear only in a few documents are highly helpful to distinguish
between documents that are on opposite sides of the scale. These words will retain
a higher absolute value of β and will be located on the extremes of the latent scale.
β values are important also because they allow a first diagnostic of the results to be
performed, by comparing the estimated policy positions of words to evaluate
whether these match their actual usage in real-world political language.
Wordfish is particularly suitable for the present analysis. It allows with large
amounts of textual data to be dealt with and produces replicable estimates of
documents positions (Klüver 2009; Slapin & Proksch 2008).
It does not require a reference text to anchor the documents and does not
require the building of dictionaries that define the political meaning of words; in
fact, it analyzes all the words contained in each document (without the need to
remove some of them) assigning a substantial political meaning to each word
(i.e., the discrimination parameter) without any ex-ante judgment.
Wordfish is also suitable to analyze time series, assuming that word usage is
constant over time (things can be different if language changes over time, see
below). Documents related to the same political actor at different points in time
are considered to be independent with respect to each other so that the position
of one document at time t does not affect its placement at time t + 1. Its position
over time will be similar only if its word usage does not change. Conversely, if
the content varies over time, the estimated position of that political actor will
change too.
However, Wordfish can require some pre-processing steps in order to improve
the quality of the estimates. Such pre-processing strategies can involve the
choice to stem words, to remove very common or peculiar terms (“stopwords”)
as well as rare ones (reducing the sparsity of the documents–terms matrix) or to
consider only words that are used repeatedly over time. These quite arbitrary
decisions are ultimately left to the researchers (Greene et al. 2016), even though
such choices can somehow affect the estimates. This aspect will be largely dis-
cussed and addressed at the end of this chapter.
Policy positions of party factions in Italy

Wordfish requires the assumption that word usage has remained relatively con-
stant over time. Italian data though range from 1946 to 2010; textual documents
span from the end of the Second World War, through the Cold War, the Protests
of 1968, the fall of Berlin Wall and the collapse of communism, until Tangen-
topoli and the advent of the Italian Second Republic.
For this reason I want to control for any possible modification in word
usage. To make sure that the research fits the requirement of language
stability I split documents into two datasets, choosing 1989 as the break point.
Far from producing the “end of history” (Fukuyama 1992), the downfall of
communism certainly produced changes that affect several issues, such as
foreign policy (due to modification in the structure of international system) or
economic policy (the Soviet Marxist left extreme of the state–market policy
scale vanished).
These changes have reasonably altered the political meaning of some words;
this effect could be even more consistent in the Italian case, as the PCI was quite
strong in the electoral and parliamentary arena and the debate on communism
was a relevant issue of the political agenda.
Indeed, center-right parties (the DC, PLI and MSI) often emphasized their
anti-communist position. This strategy though became useless at the end of the
1980s; for instance, during the MSI congresses factions debated the need to
define the party as something else other than a mere barrier against communism.
The Fall of the Berlin Wall was the first factor that contributed to altering the
Italian party system at the beginning of the 1990s, in particular with respect to
the renewal of the PCI (Giannetti & Mulé 2006). Many scholars argued that the
Italian transition began the day after the fall of the Berlin Wall: as long as Italian
politics was dominated by the cleavage of communism/anticommunism, the end
of such international strife led to adjustments in party competition and redefined
the players of such competition (Bull & Rhodes 1997; Pasquino 2000).
It prompted the transformation of the PCI into the Democratic Party of the
Left (PDS), and favored the electoral success of the LN in 1992, as moderate
voters were no longer “forced” to support classic anti-communist parties such as
the DC and its allies (Pasquino 2000).
The Italian political language and the dynamics of political communication
changed indeed, leading to the rise of new styles of language and communica-
tion. Some relevant changes concerning the lexicon, tone and content of mes-
sages occurred (Ceron 2015c); luckily, until 2010, in intra-party debates
ideological concepts such as “left” and “right” were still relevant (becoming
even more frequent after 1989), and there was neither a strong decline in the use
of traditional ideological terms nor a wider usage of populist words (Ceron
2015c). The language, in sum, remained largely ideological and suitable to
perform Wordfish analyses. Nevertheless, to account for some stylistic changes,
I retain the year 1989 as the breaking point.
I performed two separate Wordfish analyses. In both analyses the documents
cover the whole spectrum of the left–right continuum. In the first period
(1946–1988) I analyze motions of factions belonging to the left-wing PSIUP as
well as documents related to congresses held by the far-right post-fascist MSI.
Similarly, for the second period (1989–2010) data include texts belonging to the
far-left PRC, born from a fusion between the left-wing heirs of the PSIUP and
the left-wing heirs of the PCI, and motions of the AN (heir of the MSI). Due to
the lower number of center-right texts available in the second period, I included
in the analysis few party manifestos of center-right parties.16
Overall, the textual documents included in the analysis contain 1,718,412
words, with a number of unique words equal to 33,629 for the first period and to
35,391 for the second.
Figure 2.1 plots the policy positions of all the Italian factions. The latent
dimension can be interpreted as a left–right scale. Different symbols for each
party family are used. Post-fascist party factions (MSI and AN; denoted by a
square) are located on the far-right. The liberal–democratic family (PLI; hollow
square) is on the center-right. Christian democrats (DC and UDC; plus) are
located in the center during the first period, and move to the center-right in the
second period. Social–democratic and non-Marxist left party factions (PSDI,
PDA, PRI, DS, PD, New Italian Socialist Party (NPSI) and the Greens; triangle)
appear on the center-left in the first period but shift toward the center in the
1980s. Finally, socialists and communists (PCI, PSI, PSIUP, PRC and PDCI;
hollow triangle) are located on the left-wing. On the whole, party factions
belonging to different party families are properly located and there are few
Figure 2.1 Factions’ position on the left–right scale (Italy).

Note
First and second time period displayed together; factions clustered by party family.
overlaps: for instance, socialist and social–democratic factions are almost never
to the right of Christian–democratic or liberal–democratic factions.
Figure 2.2 confirms this idea; it shows the weighted mean of all the factions’
positions in each party congress. For simplicity, only the labels of the main
parties are displayed; left-of-center parties are displayed in black, right-of-center
parties are in gray. The picture confirms that the DC holds centrist positions, the
liberals and the post-fascists are on the center-right, communists, socialists and
social–democrats stand mainly on the center-left (with the far-left PSIUP being
the most extreme-left party before 1989). In the second period, the communists
PRC and PDCI are on the extreme-left; the AN and UDC (partners of Berlusco-
ni’s center-right coalition) are on the right; the DS, the PD and the Greens lay
somehow in the middle between these two extremes. The PD seems very close
to center-right parties, but this seems coherent with the recent history of this
party, which formed government coalition with the PDL and with splinter groups
of center-right parties (additionally, in the recent legislatures there have been
many switches from PD to centrist/center-right parties and vice versa).
There are some changes in the position of parties over time. The DC is closer
to the center-left in the 1960s, when the center-left formula was launched and
the PSI formed coalition governments with the DC. Overall, after the 1970s
and more deeply during the 1980s, many parties (particularly the PSI, PSDI, PLI
Figure 2.2 Weighted positions of Italian party factions in each party congress.
Note
First and second time period displayed together; main parties only.
and PRI) are affected by a rightward drift. This shift, however, is consistent with
the changes in their economic position after the 1973 oil crisis, and with the
growing need for economic policies able to drop inflation and cut government
spending and public debt.
Interestingly, the moderate shift of the PSI over time is consistent with the
expectations. The PSI moves to the left after 1947, when moderate factions exit
from the party; later, it starts to converge toward the center after 1959, when the
moderate faction Autonomia won the congress – this drift is even strengthened in
the 1970s (due to the breakaway of left-wing factions) and after the 1978 and 1981
congresses, when the reformist faction headed by Bettino Craxi took control of the
party (Ceron & Negri 2017). Analogously, the moderation of the heirs of the com-
munist party (PCI–PDS–DS–PD) is in line with the scholarly literature and the
empirical evidence (Ceron & Negri 2017; Giannetti & Mulé 2006).
Italian case: validity and reliability of the estimates

Let us turn to discuss the validity and the reliability of these estimates. Both in
the first and in the second period the length of the texts and the number of unique
words considered was large enough to obtain accurate estimates. On average, the
length of each text is equal to 5,403 words. There is, however, a wide variation,
ranging from a minimum of 70 words (one motion in the 1946 PDA congress),
to a maximum of 32,849 (PRC mainstream in 2002). Roughly 50 percent of
documents contain more than 3,400 words while only 37 (14 percent, mostly
related to PRI congresses) are below the threshold of 1,000 words.
To evaluate whether the Wordfish analysis caught the actual differences in the
use of political language across parties, let us take a look at the values of
the discrimination parameter ( β ). Figure 2.3 displays the β parameters of each
word and highlights, as an example, the label of some words related to the first and
second period. In the first period words such as “god,” “motherland” and “family”
(dio, patria and famiglia), which are traditionally associated with conservative
values, appear on the right side of the scale. On the left there are words like “class”
(classe), “solidarity” (solidarietà), “nationalization” (nazionalizzazione).17 In the
second period “no-global,” “collective bargaining” (concertazione)18 and “peace”
(pace) stands on the left, while “meritocratic” (meritocratici), “soldiers” (soldati)
and “devolution” (devoluzione) are on the right.
On both sides of the scale there are words that refer to different issues (e.g.,
economy, civil rights and foreign policy). For instance, concerns about “unem-
ployment” (disoccupazione) and “redistribution” (redistribuzione) are attributed
to left-wing parties, while “disinflation” (disinﬂazione) and “privatization”
Figure 2.3 Discriminating power of words in Italy.

Note
Beta coefficients of words’ estimates; first and second period displayed together.
(privatizzazioni) stand on the right. Concerning civil rights, the left emphasizes
“laity” (laicità) and “homosexuality” (omosessualità), while the right criticizes
the “Islamization” (islamizzazione) and emphasizes “birth-rate” (natalità) and
crime (criminalità). The same applies to other issues, such as foreign policy:
“peace” and “disarmament” (disarmo) are on the left, “gulag” and “soldiers” are
on the right. These examples confirm that β values are coherent with the actual
meaning of words in the Italian political language, therefore they appear on the
proper side of the left–right scale.
Let us move to discuss the estimates of factions’ positions to see if left-wing
and right-wing factions are properly located inside each party. To do that, per
each party congress, I aligned factions from left to right according to the liter-
ature (e.g., Giannetti & Laver 2009; Mershon 2001) and newspaper reports.
These expected ordinal positions have been compared with the observed rank of
factions according to Wordfish: the polychoric correlation is strong and signi-
ficant (0.68).19 Furthermore, there are clear differences in the policy preferences
of rival factions: in the 84 percent of cases (70 congresses out of 83), at least two
factions have positions that are statistically different from each other at the 90
percent level of confidence.
Factions seem to be properly located on the whole left–right scale too. There
is a strong and positive correlation between the estimated position of the median
faction, in each party congress, and several external measures of party positions
drawn from expert surveys (r = 0.8) or CMP data (r = 0.7). In both time periods
the mean position of factions belonging to the same party is coherent with the
expected placement of that party.
Finally, to assess the reliability, I hand-coded a subset of motions to compare
them with the estimates of Wordfish analysis. I selected a subsample of 49
motions, 38 related to DC factions, three to the PSI, two to the DS and six to the
PD.20 These motions have been analyzed by means of hand-coding, following
the coding scheme provided by Ceron et al. (2019). Wordfish results are com-
pared with the estimates of a left–right ratio scale related to economic issues; the
correlation is strong and positive (0.59), suggesting that Wordfish estimates are
reliable. Figure 2.4 plots these estimates together with a fitted regression line
(with 95 percent confidence interval).
Policy positions of party factions in France

For the French case a single Wordfish analysis that includes all the textual docu-
ments has been performed. Words have been stemmed, even though this choice
does not affect the results (see below). Overall, the documents included in the
French analysis contain 1,331,186 words, with a number of 39,231 unique words
(22,033 after stemming).
The policy positions of French factions are displayed in Figure 2.5. On the
left, there are communists (mean: –0.34; standard deviation: 0.65) and socialist
party factions (mean: –0.51; standard deviation: 1.04) as well as the Greens,
whose position matches that of the PS both in the 1980s and after 2000, though
Figure 2.4 Reliability of Wordfish estimates compared to hand-coding.

Note
Comparison based on a sample of 49 DC, DS, PD and PSI motions; the black line indicates the linear
fit with 95% confidence interval.
they look overall moderate due to unbalanced data (mean: 0.53; standard deviation:
0.50). On the right, there are UMP factions (mean: 1.02; standard deviation: 0.35)
as well as documents related to the MoDem Party (mean: 0.80; standard deviation:
0.13). Overall, the position of the PS remains quite constant until 1990. Later, the
PS shifts toward the center-right, especially after 2000, when the party started to be
led by Hollande and its liberal–democratic faction. The PS looks rather cohesive in
the 1970s and in the 1980s, when it held some unitary congresses and even the most
relevant left-wing minority, CÉRÉS, was recurrently involved in the coalition of
factions in charge of ruling the party. This unity ends up in 1990 and it is not by
chance that party splits involving the PS have been quite frequent since 1990.
Remarkably, when looking at the data, one can observe a huge internal disagree-
ment concerning the PS in 1990, when a highly conflictual party congress was held,
exacerbating internal divisions. Analogously, the estimates display a wide disagree-
ment also in 2005, when the PS held another very divisive congress (Ceron &
Greene 2019). Interestingly, when excluding unitary congresses, the correlation
between such measure of polarization inside the PS and alternative sources of data
based on the PS congress speeches (Greene & Haber 2016) is positive, as expected,
though not markedly strong (0.32). Notwithstanding this, the alignment of factions
Figure 2.5 Factions’ position on the left–right scale (France).
seems in line with the expectations. Inside the PS the left-wing CÉRÉS always
stands to the left of the mainstream faction Mitterrandistes; the average position of
CÉRÉS (–1.42) is lower (more left-oriented) compared to the mainstream faction
(–1.29) and this gap is statistically significant at the 95 percent level of confidence.
Analogously, inside the PCF the position of the mainstream (0.08) is statistically
more moderate compared to left-wing minority factions (–0.38). It is also worth
noticing that one of the most right-wing factions inside the PS is La Gauche
Moderne. This faction, headed by Bockel, split (2005) to create a movement that
promoted liberal–democratic views and merged inside the Gaullist UMP; interest-
ingly, La Gauche Moderne (0.93) is closer to the position of the PS mainstream
faction in 2015 (0.97), which was mostly composed of members that left the PS to
support Emmanuel Macron and his centrist movement La République En Marche.
Indeed, the faction promoted by Manuel Valls, who represented the right-wing of
the PS in the 2017 primary elections and later joined LREM, was considered in
continuity with La Gauche Moderne. Looking inside the center-right camp, one can
notice that the position of the mainstream MoDem faction is to the left, compared
to members (Hervé Morin) who wanted to preserve the alliance with the Gaullists.
Concerning the UMP, one of the most right-wing positions is that of Debout la
République (2004), the faction headed by Nicolas Dupont-Aignan, which later
broke away to create a right-wing sovereignist party (Debout la France); similarly,
the right-wing faction La Droite Forte is also located on the right end of the scale.
Policy positions of party subgroups in Germany
For the German case, I split data and performed two different Wordfish analyses
considering 2007 as the breaking point. In the German case, the textual docu-
ments included in the analysis contain overall 550,142 words (with 27,296
unique words related to the first period and 26,177 related to the second). The
policy positions of German subgroups are displayed in Figure 2.6 in black (SPD)
or gray (CDU).
In the first period, the MIT is consistently the most right-wing group (mean:
1.12; standard deviation: 0.31) and this placement is in line with the position of
the association of entrepreneurs within the conservative CDU. Conversely, the
workers’ association (CDA) is located more in the middle of the policy space
(mean: –0.11; standard deviation: 0.49), and its documents look quite left-
oriented in some years, particularly in the early 1990s (for a similar result: Debus
& Bräuninger 2009). Finally, SPD factions tend to be more on the left. The
Frankfurter Kreis holds the most left-wing position and overall the left-wing
factions (FK and DL21) are indeed more on the left (mean: –0.49; standard devi-
ation: 0.96) compared to the moderate Seeheimer Kreis, which has been located
as centrist in three documents out of four (mean: –0.003; standard deviation:
0.73). In the second period the alignment of factions remains the same, though
Figure 2.6 Factions’ position on the left–right scale (Germany).

Note
First and second time period displayed together; SPD factions in black, CDU subgroups in gray.
the reciprocal distances shrink. The MIT (mean: 0.78; standard deviation: 0.06)
stands on the right compared to the CDA (mean: 0.44; standard deviation: 0.07),
then, in a centrist position, one can find the two moderate groups within the SPD,
i.e., Netzwerk Berlin and SK (mean: 0.31; standard deviation: 0.04); finally, the
leftist DL21 remains the most left-wing subgroup among those considered here
(mean: 0.09; standard deviation: 0.04). After 2007, according to the Wordfish
estimates, the most left-wing group is the SPD workers’ association (Arbeitsge-
meinschaft für Arbeitnehmerfragen (Workers Association, AfA)), which has
strong ties with the German trade unions (that in turn supported some demon-
strations against the Agenda 2010 platform of welfare and labor market reforms).
Surprisingly, in the first period the position of the AfA was instead more
moderate and this group was located, on average, halfway between the SK (the
right-wing of the SPD) and the MIT (the most right-wing CDU subgroup), close
to the group of self-employed inside the SPD (Arbeitsgemeinschaft Selbststän-
dige (Self-Employed Association, AGS)). However, as discussed above, I do not
consider the AfA or the AGS in the rest of the book.
Admittedly, while in the Italian and French cases the Wordfish analysis pro-
duced estimates that appeared immediately valid and not particularly affected by
temporal trends, pre-processing strategies or the inclusion/exclusion of some
documents, in the German case the interpretation of the output is not completely
straightforward. In particular, splitting the analysis in two groups, choosing 2007
as the dividing line, was crucial to obtain estimates that are not aligned on a tem-
poral basis.
One possible reason is connected with the heterogeneous nature of the intra-
party subgroups considered and the textual documents included in the analysis.
In fact, not all the documents were similar to each other. There are differences in
the structure of the text when comparing position papers or factional contribu-
tions to the party’s manifesto with records of conferences held by different intra-
party organizations (factions or associations) as well as with the practical
resolutions voted and adopted during these conferences. The substance of some
documents could also be problematic given that the content of generalist texts,
which address a variety of topics, is intrinsically different from that of texts
focusing on a single policy field (or subfield), which address only very specific
aspects of it.
In principle, one should try to consistently select documents based on their
style and content. For the German case, though, I had to relax such criteria in
order to expand the analysis, gathering more data. Notwithstanding this, the
number of documents collected is still quite low compared to Italy and France,
and this can be another reason to explain why this textual analysis was tricky.
Interestingly, an analysis performed by applying another automated technique
of text analysis (Wordscore) on a subsample of these textual documents also
reports similar lights and shades (Debus & Bräuninger 2009). That study prop-
erly locates the CDU subgroups (CDA and MIT), while the position of the SPD
subgroups was in part unusual, particularly with respect to the AfA, which
appeared to be unexpectedly more moderate than the core of the SPD, while the
right-wing Seeheimer Kreis was estimated to the left of the party, from 1987 to
1998 (Debus & Bräuninger 2009: 137). Nevertheless, this comes as good news
in terms of reliability. I compared the estimates of my Wordfish analysis (con-
sidering also documents that will not be used in the rest of the book, such as the
PDS manifestos or documents of the AfA and the FDP) with the estimates pro-
vided by Debus and Bräuninger (2009). The correlation, concerning 49 docu-
ments, is high and positive (0.80) and it gets even stronger (0.88) when leaving
aside the AfA, AGS, FDP and PDS, to focus only on the estimates of subgroups
considered in the course of the book. For all these reasons, I consider the data
robust enough to profitably investigate intra-party dynamics inside the SPD and
the CDU.
Text analysis: robustness checks, insights and best practices

In light of the ongoing debate in the field of text analysis about pre-processing
strategies and best practices (Greene et al. 2016), I conclude this chapter by
evaluating the robustness of the different Wordfish analyses presented above. By
doing that, I will provide some insights concerning the actual relevance of pre-
processing choices in different contexts and across different languages. This test
will also be useful to evaluate the robustness of the estimates that will be used in
the empirical analyses to be discussed in the next chapters.
It has been argued that Wordfish can be run on the entire set of words included
in textual documents, although Proksch and Slapin (2009) recommend to use a
subsample of words. This can be particularly crucial if the language does not
remain stable over time. Furthermore, words that are used very rarely could be
removed too (Proksch & Slapin 2009) to facilitate comparisons over time. This
strategy can be done by reducing the number of cells with zero-word frequency
in the document–term matrix. Finally, stemming is another pre-processing
strategy that is often used by researchers (Greene et al. 2016). It consists in redu-
cing complex forms of words to their simplest root, to better connect substan-
tively similar words; as such, words with similar meanings but different endings
are considered as a unique stem given that these words usually convey similar
information about the latent position of a textual document.
Based on this, I have replicated the analyses on Italy, France and Germany by
adopting alternative pre-processing strategies compared to those discussed
above. Table 2.2 summarizes the results of such comparison, reporting the
correlation between the estimates obtained by applying pre-processing strategies
and the estimates produced when no pre-processing has been adopted.
To sum up, the following strategies have been tested. First, I compared docu-
ments without pre-processing and documents in which stemming was performed.
Second, to account for any possible change of language over time, I replicated
the analyses by considering only words that consistently appear both before and
after a certain date (see below for details on each country). Third, I reduced spar-
sity, lowering the size of the document–term matrix, by removing terms that
occurred in fewer than 1 percent (low reduction) or 20 percent (strong reduction)
Table 2.2 Correlation of Wordfish estimates when using pre-processing strategies or not
Wordfish Analysis Italy 1st Italy 2nd France Germany 1st Germany 2nd
Stemming 0.96 0.99 0.99 0.99 0.97

No temporal break 0.93 0.81 0.88 0.28 0.60
Words consistent over time 0.96 0.97 0.95
Reduce sparsity 1% (low) 0.90 0.98 0.98 0.99
Reduce sparsity 20% (strong) 0.74 0.93 0.95 0.93
Note
The baseline for the comparison is the analysis ran without any pre-processing.
of the documents. Finally, in light of the discussion on temporal trends, which

pushed me to split the analyses on Italy and Germany, I want to evaluate what
happens when no temporal break is applied, running a unique analysis in these
two contexts. Conversely, in the French case, in which a single analysis on the
whole period was run, I will assess what happens when imposing a temporal
break to split data, analyzing a subsample of documents related to a shorter lapse
of time (i.e., using only post-1989 data and getting rid of pre-1989 motions that
are exclusively related to the PS).
In the French case, for brevity, the analysis has been replicated on a sub-
sample of 132 documents excluding the motions of the PCF that have been gath-
ered later; notice that comparing the estimates with or without stemming on the
whole set of motions (153) produces the same results. With respect to France,
the scenario in which I only considered the same words over time has been done
by using the year 1992 as a breaking point (given that there are data from 1971
to 1991 and then there is a gap until 1997). Accordingly, I run this analysis com-
paring word usage only for words that existed both before and after 1992.
For the Italian case, to control for the consistency of the language over time, I
replicated the analysis on pre-1988 data only. In fact, the time period related to
the first analysis was quite long and could have been affected by changes in lan-
guage that should only play a limited role in the subsequent shorter period (post-
1989). I evaluate the robustness in two different ways. First, by selecting words
that appear both before and after 1988; second, by selecting words that were
already in use before 1972. In both cases the results are similar.
Finally, with respect to Germany, the test about sparsity has been done, for sim-
plicity, on a single analysis that includes textual documents of both time periods.
Analogously, the effect of consistent word usage over time has been done running
a single analysis and considering only words that were used both before and after
2007 (which is the breaking point considered in the main analysis).
The results suggest that, overall, adopting alternative pre-processing solutions
do not really make the difference with respect to my data. The correlations dis-
played in Table 2.2 are almost always equal to 0.9 or above.
In more detail, stemming words or not, the results remain very similar with
correlations close to 1. Reducing the sparsity of the document–term matrix, in
turn, does not really make the difference and the size of the reduction does not
affect the results too. When reducing sparsity, in seven cases out of eight there is
a strong correlation (above 0.9) between these estimates and the original non
pre-processed data. The unique exception is represented by Italy, in the first
period (1946–1988), though only if a strong reduction of the document–term
matrix is applied. Here the correlation remains strong (0.74) but it falls below
the level of 0.9.
Turning to the consistency of word usage, I do not find strong differences
when considering only words that have been continuously used over time. Once
again, the estimates of these robustness checks are still almost perfectly correl-
ated (between 0.95 and 0.97) with the original results. Although from a theoret-
ical point of view it seems quite reasonable to perceive a change in language
over time, from an empirical point of view, the Wordfish estimates hold the same
and are not affected by removing words that are solely used in shorter time
periods.
Results are slightly different only when contrasting analyses in which data
are split based on a temporal threshold, instead of running a single time series
analysis on the whole amount of texts. This can be taken as a signal that lan-
guage is indeed changing over time (for additional details and some empirical
evidence: Ceron 2015c), and such change can partially affect the estimates.
However, for Italy (first period: 0.93) and France (data after 1989: 0.88) the
differences are quite minimal; the estimates are strongly correlated with those
obtained without any pre-processing. In the German case time seems to matter
more. Indeed, when running a single analysis, the results are still correlated,
but to a lower extent (0.60) with post-2007 data, and the correlation with
pre-2007 data is even lower (0.28). This empirical check highlights the
importance of splitting data in the German case, as discussed in the previous
section.
In the end, the estimates seem quite stable no matter the pre-processing strat-
egies adopted. This is strikingly evident with respect to Italian and French data,
i.e., those that will be more extensively used to perform large-N statistical ana-
lyses in the course of the book. Based on this, I can be confident about the
robustness of the empirical findings presented in the next chapters.
Notes
1 They held roughly 8 percent of posts in party body: around 13 seats out of 172 in the
board and three out of 36 in the executive branch. In the Senate 10 out of 146 MPs
were loyal to Fini while in the House their percentage was slightly higher: at least 34
MPs belonged to Finiani (around 12 percent). Such figures, compared to the share of
other former AN factions, are not far from the actual relative size of Finiani inside
the AN.
2 Finiani faction was organized through a net of associations and foundations, linked
with and somehow coordinated by Fini’s think-tank FareFuturo. Some of these sub-
groups strongly opposed the PDL leadership (i.e., Generazione Italia, led by Fini’s
lieutenant, Italo Bocchino, and Libertiamo, led by a former Radical party leader
Benedetto Della Vedova); others, such as Area Nazionale or Spazio Aperto (created
by Silvano Moffa) were more willing to negotiate with the PDL leadership (indeed
Moffa left the PDL in July 2010 but re-entered in December).
3 “Zur Sonne, zur Einheit,” Der Spiegel November 2010, p. 28.
4 These associations exist even inside the SPD, however, when analyzing this party I
consistently focus on factional subgroups only.
5 Due to parliamentary rules, in the Italian context the voting behavior includes the
absence as a fourth alternative. When strategic, in fact, the absence allows a policy
position to be expressed (Ceron 2015b).
6 It could be argued that talk (and language) is cheap. However, Giannetti and Laver
(2009) found a relationship between politician speeches during party congress and
their consequential political behavior in the parliamentary arena. Therefore, “if the
politician speaks and behaves in a consistent manner, […] may well become associ-
ated with a particular position on a particular dimension” (Laver & Shepsle
1996: 248).
7 Notice that different subgroups might decide, for instrumental reasons, to contest the
congress together, presenting a common list and a common motion. However, for
both theoretical and practical reasons, I do not investigate the process leading to such
inter-factional alliance and I measure their common position and their common
strength assuming that inside the party they will bargain as a single subgroup.
8 Parties’ newspapers are the followings: Avanti! (PSI); La Giustizia, L’Umanità,
Critica Sociale (PSDI); L’Unità (PCI and DS); La Voce Repubblicana (PRI); Il
Secolo d’Italia (MSI).
9 Parties’ websites were the followings: Greens (www.verdi.it), NPSI (www.socialisti.
net), PD (http://beta.partitodemocratico.it), PDCI (www.comunisti-italiani.it), PRC
(www.rifondazione.it), PSOC (www.partitosocialista.it), UDC (www.udc-italia.it);
for AN (www.alleanzanazionale.it) and DS (www.dsonline.it), I relied mostly on the
Internet Archive (www.archive.org).
10 In 1998 the PRC mainstream split and during a party body meeting four factions com-
peted to take control over the party in order to set the new line; each faction presented
a motion voted by members of the party National Political Committee. Accordingly, I
considered it as if it was a real congress.
11 http://62.210.214.184/cg-ps/ladocps.php.
12 For the SPD see: www.parlamentarische-linke.de; www.forum-dl21.de; www.
seeheimer-kreis.de; www.netzwerkberlin.de. For the CDU: www.cda-bund.de or
www.mit-bund.de/.
13 It could be argued that single words are uninformative about policy positions. On the
contrary, words matter. For example, during the 2009 PD congress Bersani’s oppon-
ents criticized him arguing that his victory would have transformed the PD into a
social–democratic party. The PD president Rosy Bindi (a Bersani’s ally) answered
claiming that “Bersani’s motion does not contain the word ‘social-democracy’ ” (La
Repubblica, August 15, 2009, translation mine).
14 Wordfish estimates are robust to its main assumptions (Slapin & Proksch 2008).
15 Multidimensional analyses can be performed by dividing these documents into single
pieces of texts related to different topics and running distinct analyses on each topic.
16 Excluding these documents leads to very similar results.
17 Interestingly, the position of words used by party members to refer to each other is
also coherent: “comrades” (compagni) used by socialists and communists, stands on
the left; “friends” (amici) used inside the DC, is located on the center; “companions”
(camerati), used by the MSI is on the right.
18 See Ceron & Negri 2018.
19 Factions location on the internal left–right scale usually tends to be steady, though
scholars report a few of exceptions related to the DC: Nuove Cronache (tied to Amin-
tore Fanfani) moved from left to right over time; Andreotti’s faction (Primavera)
temporarily shifted from right to left in 1980, and back to the right afterwards
(Bettcher 2005; Boucek 2010); the left-wing Forze Nuove moved to the right in the
early 1980s, when it firmly opposed the idea of a “Historic Compromise” (i.e., a coali-
tion government between the DC and PCI). The estimates successfully caught these
three shifts.
20 For the PD I also included more recent motions, see Chapter 6.
References
Bäck, H., Debus, M., & Müller, W.C. (2016). Intra-party Diversity and Ministerial Selec-
tion in Coalition Governments. Public Choice, 166(3–4), 355–378.
Benoit, K., Bräuninger, T., & Debus, M. (2009). Challenges for Estimating Policy Prefer-
ences: Announcing an Open Access Archive of Political Documents. German Politics,
18(3), 441–454.
Bettcher, K.E. (2005). Factions of Interest in Japan and Italy: The Organizational and
Motivational Dimensions of Factionalism. Party Politics, 11(3), 339–358.
Boucek, F. (2009). Rethinking Factionalism: Typologies, Intra-party Dynamics and Three
Faces of Factionalism. Party Politics, 15(4), 455–485.
Boucek, F. (2010). The Factional Politics of Dominant Parties: Evidence from Britain,
Italy and Japan. In Bogaards, M., & Boucek, F. (eds) (2010). Dominant Political
Parties and Democracy, 117–139, Routledge/ECPR Studies in European Political
Science, London.
Budge, I., Ezrow, L., & McDonald, M.D. (2010). Ideology, Party Factionalism and Policy
Change: An Integrated Dynamic Theory. British Journal of Political Science, 40(4),
781–804.
Budge, I., Klingemann, H.-D., Volkens, A., Bara, J., & Tanenbaum, E. (2001). Mapping
Policy Preferences: Estimates for Parties, Electors, and Governments 1945–1998.
Oxford University Press, Oxford.
Bull, M., & Rhodes, M. (1997). Crisis and Transition in Italian Politics. Frank Class,
London.
Calise, M. (2010). Il partito personale. I due corpi del leader. Laterza, Roma-Bari.
Carrubba, C., Gabel, M., Murrah, L., Clough, R., Montgomery, E., & Schambach, R.
(2006). Off the Record: Unrecorded Legislative Votes Selection Bias and Roll-Call
Vote Analysis. British Journal of Political Science, 36(4), 691–704.
Ceron, A. (2012). Bounded Oligarchy: How and When Factions Constrain Leaders in
Party Position-taking. Electoral Studies, 31(4), 689–701.
Ceron, A. (2015a). The Politics of Fission: Analysis of Faction Breakaways among Italian
Parties (1946–2011). British Journal of Political Science, 45(1), 121–139.
Ceron, A. (2015b). Brave Rebels Stay Home. Assessing the Effect of Intraparty Ideo-
logical Heterogeneity and Party Whip on Roll-call Votes. Party Politics, 21(2),
246–258.
Ceron, A. (2015c). Changing Politics, Changing Language. The Effect of Institutional
and Communicative Changes on Political Language Measured through Content Ana-
lysis of Italian Intra-party Debates. Journal of Language & Politics, 14(4), 528–551.
Ceron, A. (2017). Social Media and Political Accountability: Bridging the Gap between
Citizens and Politicians. Palgrave Macmillan, Cham.
Ceron, A., Curini, L., & Negri, F. (2019). Intra-party Politics and Interest Groups:
Missing Links in Explaining Government Effectiveness. Public Choice, doi: 10.1007/
s11127-019-00644-0.

Leaders, Factions and The Game of Intra-Party Politics PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Leaders, Factions and The Game of Intra-Party Politics PDF

Uploaded by

Copyright:

Available Formats

2 Factional preferences in Italy,

France and Germany

Factional politics: the Italian case

Beyond Italy: intra-party politics in France and Germany

Party congress, the arena where factional bargaining

Data collection on Italy, France and Germany

From words to fact(ion)s: estimating factions positions using

Policy positions of party factions in Italy

Figure 2.1 Factions’ position on the left–right scale (Italy).

Italian case: validity and reliability of the estimates

Figure 2.3 Discriminating power of words in Italy.

Policy positions of party factions in France

Figure 2.4 Reliability of Wordfish estimates compared to hand-coding.

Figure 2.5 Factions’ position on the left–right scale (France).

Figure 2.6 Factions’ position on the left–right scale (Germany).

Text analysis: robustness checks, insights and best practices

Stemming 0.96 0.99 0.99 0.99 0.97

of the documents. Finally, in light of the discussion on temporal trends, which

You might also like