Fieldwork

Organizing the fieldwork for a candidate survey requires considerable effort. This is particularly true for the Italian case, as the parliamentary elections for the Chamber of Deputies held in February 2013 saw the participation of about 11,000 candidates, grouped into 47 party lists. As a first step in our research, we reduced the universe of candidates to a sample of about 2900 potential respondents selected on the basis of two main criteria: (a) relevance of the party list and (b) competitiveness of the candidate. According to the first criterion, we excluded all party lists failing to pass the electoral threshold and thus having no representatives in parliament. The only exception is Civil Revolution (RIV), a left-wing list that did not reach the electoral threshold, but was included in our research owing to RIV’s salience on the national political stage. This reduced the number of sampled party lists to 11. Regarding our second criterion, we used a candidate’s position on the party list as an indicator of her competitiveness. As a rule of thumb, in each district we picked all candidates elected plus an equal number of unsuccessful runner-up candidates from the party lists. We then applied some corrections based on district magnitude by including a higher number of non-elected candidates for small parties. These sampling rules resulted in a sample of about 2600 candidates distributed across 11 party lists. Finally, we added a further 300 candidates belonging to the centre-right People of Freedom (PdL) because of a low response rate from candidates for this party.

Having defined our sample, we collected candidates’ personal contact details. This turned out to be a particularly problematic task: in Italy there is no nationwide public register of candidates’ addresses. We then gathered personal addresses from several sources including the electoral offices of the Courts of Appeal – where the candidacies to the electoral lists are collected – and the Italian Tax Agency National Register. The ICS survey was fielded in the summer of 2013. We sent all sampled candidates a mail questionnaire based on the one developed by the CCS international team (module 1), together with a cover letter outlining the purposes of our research. After some months we sent a first postal reminder offering the respondent the opportunity to complete an online version of the questionnaire. Later on we sent a second reminder, attaching again a paper questionnaire and all the information needed to complete the online version. Finally, a third reminder was sent to elected candidates and to those belonging to the PdL. We closed the survey at the end of September 2014.

Response rates

Politicians are usually busy people and have little time to fill in long surveys.Footnote ³ Moreover, distrust for research like this one may contribute to a generally lower response rate compared with mass opinion surveys (Hoffman-Lange, Reference Hoffman-Lange2008). Despite these limitations, we obtained an acceptable number of respondents, both in absolute and relative terms. In roughly 1 year of fieldwork, we collected 672 completed questionnaires out of 2878 surveys sent, yielding a response rate of about 23%. More precisely, 558 mail surveys and 114 online questionnaires were completed. Table 1 illustrates the response rate by each party list included in the ICS research. The results in this table show that the overall mean response rate is higher for centre-left parties (PD, SEL), the centrist SC list, and the M5S, whereas it is lower for centre-right parties (PdL and FdI). In particular, it should be noted that the PdL remains under-represented in our data, despite over-sampling and sending a greater number of reminders (see above). Almost 21% of the total responses were made by elected candidates. The distribution of parliamentarians’ responses among the party lists follows more or less the same trends shown in Table 1.

Even though our sample is acceptable in terms of absolute numbers, our response rates may be a problem if they are biased (Bailer, Reference Bailer2014). This can be assessed by comparing the distribution of respondents in our ICS sample with the distribution of the whole population of candidates in terms of potentially relevant variables such as gender, party group, and constituency. A measure useful for this purpose is the ‘Duncan Index of Dissimilarity’ (Duncan and Duncan, Reference Duncan and Duncan1955; Deschouwer et al., Reference Deschouwer, Depauw and André2014: 10–11), which measures the percentage differences between the sample and population distributions. This index ranges from 0 indicating no differences to 100, suggesting maximum dissimilarity. In our case, the overall deviation between the population (i.e. the list of competitive candidates in all sampled party lists) and sample distributions is 3 for gender, 13 for party groups, and 11 for constituencies. In all the cases, the Duncan index is <15, suggesting that our respondents closely represent the population of candidates in terms of key aspects. Altogether, this evidence supports the notion that analyses based on our data should not suffer from selection bias or distorted results.Footnote ⁴

The ICS in comparative perspective

Although mass surveys based on quota and probability samples are well established within the Italian research community, elite surveys involving politicians are rare. According to Bailer (Reference Bailer2014), we can count four prominent examples of elite surveys including samples of Italian politicians: the European Parliament Research Group (EPRG), PartiRep, the European Election Candidate Survey (EECS), and the CCS, to which ICS belongs. The first two projects focus on elected representatives: the EPRG (Scully et al., Reference Scully, Hix and Farrell2012) provides data on individual members of the European Parliament, whereas PartiRep surveys national and regional legislators in 15 advanced democracies (Deschouwer and Depauw, Reference Deschouwer and Depauw2014).Footnote ⁵ Rather than focussing only on elected representatives, the EECS, which is included in the PIREDEU project (Giebler and Wüst, Reference Giebler and Wüst2011), and the CCS have collected data about candidates running for European and national parliamentary elections, respectively. An overview of the most recent elite surveys covering Italian politicians is provided in Table 2, which shows that the response rate we achieved is higher or at least comparable with other projects of similar scope.

In comparison with PartiRep, the only other research targeted to national politicians, the ICS/CCS response rate is 10 points higher; 14 points higher if we consider only the universe of elected represented in the Chamber of Deputies (the figures in parentheses in the table). Although there are significant differences in the response rates, our survey shares with PartiRep several features, which can be therefore considered as consolidated trends in elite survey research in Italy. In particular, the PD and centre-left parties in general are slightly over-represented among the respondents in both surveys, whereas the PdL and centre-right parties are somewhat under-represented. Similarly, the balance between genders both in the candidate population and among the ICS respondents is slightly skewed in favour of males. However, in terms of party groups our sample more closely resembles the candidate population than the PartiRep sample (Duncan index: 13 vs. 16, see Deschouwer et al., Reference Deschouwer, Depauw and André2014: 10).

As noted above, the ICS is part of a cross-national survey programme, which combines an internationally agreed core questionnaire and a locally adapted set of questions that try to capture the specifics of the national political and electoral system. CCS currently has comparable data from 20 countries, including Italy, for 25 elections.Footnote ⁶ Table 3 reports the sampling procedure, sample size, and response rate for the countries participating in the CCS project for which such information is available. As we can see, Italy is among the few countries included in the project that adopted a sampling procedure. However, the ICS research team could hardly have done otherwise as the universe of Italian candidates (about 11,000) is more than double than the one recorded in Austria (4800) – the country with the highest number of candidates among the other CCS participants. Regarding response rates, the figure for Italy is significantly lower than that observed in Nordic or non-European countries. However, the ICS response rate is similar to that recorded by CCS in countries in Southern and Eastern Europe.

Modes of political representation

Political representation has multiple facets. If we look at its outcomes, representation can be described either through politicians’ backgrounds or in terms of their preferences on salient policy issues. Conversely, when the very nature of representativeness is at stake, political representation can be analysed by looking at the most important aspects of the relationship between voters and representatives (Andeweg and Thomassen, Reference Andeweg and Thomassen2005). In contemporary democracies, a common way to depict this relationship is describing it as a chain of delegation from voters to governing actors (Strøm, Reference Strøm2000, Reference Strøm2003). Acting as principals, voters can mitigate possible agency problems through a series of control mechanisms that operate ex ante: contract design, screening, and selection of potential candidates and parties. As an alternative, voters can count on ex post mechanisms in order to keep agents under control: monitoring, reporting requirements, and institutional checks. Therefore, elections can be understood either as a mechanism of selecting ‘good types’ for office or as a sanctioning device. Within the principal agent framework, elections are hypothesized to translate the popular will into government policy, that is, representation is ‘from below’. Alternatively, a more active role can be expected to be played by parties and representatives, who ‘enter the political process with their views and put these views to the citizens for their approval’ (Andeweg and Thomassen, Reference Andeweg and Thomassen2005: 511). In this case, representation is ‘from above’. An example of this mode of representation is the Responsible Party Government Model (Ranney, Reference Ranney1954).

Andeweg and Thomassen (Reference Andeweg and Thomassen2005) used the approaches described above to build a typology of modes of representation. By combining the bottom–up and top–down link between voters and representatives and the ex ante and ex post types of control mechanisms, four modes of political representation, which capture the most important aspects of the relationship between citizens and politicians, may be examined. First, representation from below and ex ante controls define the ideal-type model of delegation. In this model, which is best suited for describing parliamentary democracies, parties translate voters’ preferences and act as prior screening devices in order to select prospective parliamentarians as well as cabinet members. Second, representation from below and ex post control characterize the responsiveness model. Here the focus is on ex post sanctions, which force representatives to respond to changes in public policy preferences. Third, representation from above and ex ante controls describe the authorization model. Within this model, representation aims at giving a popular mandate to a party for implementing its manifesto in the subsequent period. Finally, representation from above and ex post controls illustrate the accountability model. This mode resembles the liberal vision of representative democracy where the only function of voting is to control officials.

Data from the ICS allow us to assess the attitudes of Italian candidates towards these four models of representation. Following Andeweg and Thomassen (Reference Andeweg and Thomassen2005), we operationalized the dimensions defining their typology by using responses to the following questions:

1. (1) In their relationship with their voters, politicians may emphasize different aspects. Which of these two aspects do you think is most important? (a) translating the political views of citizens into policy as accurately as possible; or (b) seeking support from the voters for the political views of their own party; and

2. (2) in our political system, elections have various functions. Which of these two functions do you think is most important? (a) in elections, politicians account to the voters for their actions in the past; or (b) in elections, politicians put their plans for the future to the voters.

The first question differentiates between the direction of the interaction between voters and their representatives (from below or from above); the second one distinguishes between the two types of control mechanism: ex post (retrospective voting) or ex ante (prospective voting).

Table 5 shows the responses by Italian candidates to these questions.Footnote ⁸ According to the results presented in the last row of this table, candidates in 2013 Italian elections assign much more importance to representation from below than from above (84 vs. 16%). On the other hand, when asked about the type of control mechanisms, the majority of ICS respondents emphasize the importance of ex ante rather than ex post controls (59 vs. 41%). Italy has been described as a country in which parties exert a strong control over each phase of electoral democracy, from candidate recruitment to electoral campaigns and the organization of parliamentary activity (Cotta and Verzichelli, Reference Cotta and Verzichelli2007). Consequently, it seems reasonable to believe that variations in the individual candidate’s attitudes towards representation can be accounted for by parties to which these candidates belong. However, Table 5 reveals that candidates from almost all the parties included in the analysis assign more importance to the delegation model of representation (from below and ex ante controls). Let us also underline that the M5S candidates give greater importance to representation from below and ex ante controls (58%) than respondents from other parties (49% registered in the entire sample). This does not come as a surprise, as the M5S emphasizes the virtues of the ‘ordinary people’ vis-à-vis the perceived self-referential and corrupt nature of political elites (Pedrazzani and Pinto, Reference Pedrazzani and Pinto2013, Reference Pedrazzani and Pinto2015). Moreover, M5S candidates cannot rely on any previous political experience as they belong to a brand new party (Pinto and Pedrazzani, Reference Pinto and Pedrazzani2015).

The only exception to this pattern of responses towards delegation comes from candidates from the two centre-right parties: the PdL and the UDC, who favour a bottom–up form of representation with ex post controls as the best model of political representation. This pattern in answering might be explained by the fact that the PdL had been part of governing coalitions for a number of years before the ICS fieldwork. Consequently, it seems reasonable to think that candidates belonging to this political group consider elections as a mean to validate the political decisions taken by the party when in office. It must be added that the PdL and the UDC are the parties in our sample with the lowest share of first-time candidates (67 and 61%, respectively, vs. 82% observed in the entire ICS sample). It appears then that more experienced politicians tend to emphasize what they have done in the past; and consider elections as an opportunity to have their previous political behaviour evaluated by the electorate.

The delegation model of representation, which was supported by a majority of the ICS respondents, relies mainly on ex ante screening and selection devices. In parliamentary democracies, ex ante controls are guaranteed by an internal selection of agents through centralized, socially cohesive, and policy-oriented political parties. Parties rely strongly on prior screening devices, such as experience, seniority, or ascriptive social criteria in order to recruit candidates who will best represent their principals (Strøm, Reference Strøm2003). Hence, voters are asked to choose between different packages of candidates sharing policy preferences, where citizens have enough information to assessing their potential representatives’ skills as agents (Strøm, Reference Strøm2000). Nowadays, this mode of representation is criticized for several reasons. First, ascriptive social characteristics, such as class background, provide less information about principals’ voting decision than before (Dalton, Reference Dalton1996). Second, increasing volatility in the policy agenda makes information based on candidates’ prior experience a less effective device for providing useful indications about their preferences and skills. Third, parties’ tendency to select candidates who have not followed a traditional political career and the growing use of primaries as candidate selection mechanisms may make politicians more responsive towards voters than parties leading to divided loyalties (Strøm, Reference Strøm2000).

What general conclusions can be drawn from our analysis of candidates’ attitudes towards representation? Given the problems highlighted above, the effectiveness of representation from below is, from the perspective of our ICS respondents, questionable. On the one hand, the emphasis on responsiveness gives politicians incentives to continuously monitor voters’ preferences and to behave in ways that best please their principals even in election campaigns that are not very competitive. This generates a climate of never-ending electoral campaigning, which can potentially harm the day-to-day functioning of the policy-making process. On the other, the reduced role of parties as a proper ex ante screening and selection device combined with the decline of class voting may broaden policy differences between agents and their principals, thereby challenging the very functioning of the chain of delegation. This may threaten the citizens–elite policy linkage that is assumed to characterize democratic political representation, as we will show in the next section dealing with preference congruence.

Party–voter congruence

Democratic representation implies that the actions of policymakers are responsive to the wishes of the electorate. This responsiveness relationship between politicians and voters is realized thanks to a set of institutional arrangements among which the most crucial is free and competitive elections. However, it is often presumed that democratic representation can only work well in the presence of a certain level of affinity or ‘congruence’ between the preferences of political elites and the opinions of voters resulting in a correspondence between citizens’ views and government policies (see, e.g., Dalton, Reference Dalton1985; Thomassen, Reference Thomassen1994, Reference Thomassen1999; Katz, Reference Katz1997). Assessing congruence requires estimates of politicians and voters policy preferences and a measure of the correspondence in policy positions of both groups. Fortunately, the ICS data allow us to map the policy congruence between candidates running for office in the 2013 Italian elections and their voters.

Research into political representation shows a persistently high level of policy congruence in democratic countries (Powell, Reference Powell2004; Thomassen, Reference Thomassen2014). Party–voter congruence has been defined and measured in different ways. ‘Relative’ congruence is based on the explained variance of bivariate correlations between the ideological scores of voters and those of their representatives (Miller and Stokes, Reference Miller and Stokes1963; Kitschelt et al., Reference Kitschelt, Mansfeldova, Markowski and Toka1999). ‘Absolute’ congruence is instead measured as the distance between the average position of parties’ electoral base and the average score of party members along the same scale (Achen, Reference Achen1978; Rohrschneider and Whitefield, Reference Rohrschneider and Whitefield2012). A second distinction in this literature concerns the domains in which the level of congruence is assessed. Although most studies analyse congruence on the general left-right scale, there can be economic, social, or political reasons why some policy issues yield more congruence than others. Some recent works have therefore measured congruence across different policy domains (Thomassen and Schmitt, Reference Thomassen and Schmitt1999; Costello et al., Reference Costello, Thomassen and Rosema2012).

In this section, we use the ICS data to evaluate ‘candidate–voter’ policy congruence at the time of the 2013 Italian general elections. We compare the preferences of political candidates with those of voters from the same party using an absolute definition of congruence. As the ICS data set enables to take into account the multidimensionality of the policy space of party competition rather than concentrating on a general left-right scale, we map congruence on a set of specific policy issues. In other words, are there some policy domains where a substantial gap exists between political elites and the Italian electorate? And is this divide greater for some parties? Assessing these differential policy gaps is important for understanding the dynamics of party competition, as the existence of a substantial mismatch on issues such as immigration and the EU can, for instance, help explain the success of populist parties (Costello et al., Reference Costello, Thomassen and Rosema2012). In addition, in this study we also assess the extent to which political candidates are aware of their policy distance from their electoral base: an aspect of political representation that rarely receives attention in studies of policy congruence. Do candidates of Italian parties acknowledge that they are far away from their voters on certain issues? Indeed, newcomer parties are likely to have a less precise idea of their voters’ policy preferences because there is no electoral history to act as a guide.

The ICS questionnaire asked each candidate to place herself/himself and the voters of her/his party on a set of 11-point scales (0–10) dealing with four key policy issues. The first policy scale measures support for greater spending in public services (0) vis-à-vis tax reduction (10), and it is the dimension most clearly related to economic left-right. The second domain deals with disapproval of (0) or support for (10) further European integration. EU-related issues are gaining increasing relevance in shaping political competition in EU member states (Thomassen and Schmitt, Reference Thomassen and Schmitt1999; Hooghe et al., Reference Hooghe, Marks and Wilson2004; Kriesi et al., Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2008; Schmitt and Thomassen, Reference Schmitt and Thomassen2009; Di Virgilio et al., Reference Di Virgilio, Giannetti, Pedrazzani and Pinto2015). The third dimension relates to family policy, dealing with support for (0) or opposition to (10) the legal recognition of new types of family such as same sex domestic partnerships. Finally, the fourth policy scale measures attitudes towards immigration with 0 indicating opposition to having more migrants in Italy and 10 showing support for accepting more non-nationals. These two last issues have been linked with the libertarian–authoritarian domain of GAL–TAN (Hooghe et al., Reference Hooghe, Marks and Wilson2004), and with the cultural dimension of contestation generated by the process of globalization and defined as a conflict between ‘integration’ and ‘demarcation’ (Kriesi et al., Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2008). The self-placement of voters on the four policy dimensions examined in this study are taken from a mass (post-election) survey fielded in February 2013.Footnote ⁹

Figures 1 to 4 contain estimates for voters and candidates for four policy positions. First, there is a mean estimate of the self-placement of voters indicated by solid black squares. The ‘whiskers’ on these squares represent the 95% confidence intervals around the mean. Second, the dark grey squares show the mean estimates of the party candidates’ self-placement on the four policy scales where again 95% confidence intervals are shown. Third, the light grey squares show mean estimates of where candidates locate their own party’s voters on a policy issue. The final mean estimates illustrate the extent to which candidates misperceive their own voters’ policy preferences. The first general lesson that may be drawn from Figures 1 to 4 is that preference congruence between Italian candidates and voters of the same party varies across issues and political formations. The same pattern is observed with candidates’ perceptions of voters’ policy positions.

Figure 1 Party–voter congruence: taxes vs. services.

Figure 2 Party–voter congruence: European integration.

Figure 3 Party–voter congruence: family policy.

Figure 4 Party–voter congruence: immigration.

If we compare the proximity between candidates and voters across the four policy dimensions, we observe that the overall policy congruence is greater for some issues and there are domains about which candidates have in general a more accurate idea of the opinions of their electoral base. Considering all parties together, the two policy issues that Italian candidates perceive themselves to be most distant from their party’s voters are EU integration and family policy. On European issues, the sum of absolute distances between the mean self-placement of candidates and the average position of voters indicated by candidates from the same party is 8 points. This means that on average candidates from each of the 10 parties studied perceive a policy distance of 0.80 points from their own party’s voters on EU matters. Similarly, in area of family policy the sum of absolute distances is equal to 7.9 points indicating a mean absolute distance is 0.79 points. The issue about which candidates perceive themselves to be closest to their electorates is immigration. Here, the sum of absolute distances is just 1.7 points.

Likewise, the dimensions concerning European integration and family policy are the ones where candidates are farthest away from their voters, as the sum of absolute distances between candidates’ self-placements and voters’ self-placements is 20.9 and 19.3 points, respectively. With regard to European integration, the mean differences between candidates’ and voters’ positions towards the EU are statistically significant for almost all parties (the only exceptions being the LN and the FdI). This is consistent with previous comparative studies on congruence, which show a stark contrast in views towards the EU between voters and national- and European-level political representatives (e.g. Thomassen and Schmitt, Reference Thomassen and Schmitt1999; Best et al., Reference Best, Lengyel and Verzichelli2012).

According to our ICS data, the issue about which candidates’ preferences resemble most the opinions declared by voters is tax cuts and public expenditure, where the sum of absolute distances is about 10 points. This difference represents a mean of 1 point on the 0–10 policy scale for each of the 10 parties studied. Despite what Italian candidates perceive, the ICS data for the 2013 election indicates that immigration is not the issue where there is greatest congruence between candidates and voters. In fact, this is the issue where candidates’ misperception of voters’ preferences is highest (the sum of absolute distances between voters’ self-locations and the scores attributed to them by candidates is 18.3). Unlike the EU integration issue where Italian candidates know their preferences are quite far from those of their voters, it seems that Italian candidates in 2013 really misunderstood the more extreme anti-immigration position of their constituency. Candidates from almost all Italian parties tend to place their voters at least 1.8 points more towards the pro-immigration pole than was actually the case. The only exceptions are LN, FdI, and PdL, that is, centre-right parties that, to different degrees, exploit or foment xenophobic sentiments in the population for electoral purposes. The same candidate–voter gap also appears, although to a smaller degree, in the case of family policy. Here candidates from almost all parties tend to be more liberal than their voters. With family policy, centre-left politicians tend to view their electoral base as being less conservative than they actually are, whereas centre-right candidates seem to have a more accurate idea of the preferences of their voters.

The dimension dealing with taxes and public services is the one where candidate–voter congruence is the greatest. These results matches with Costello et al. (Reference Costello, Thomassen and Rosema2012) findings that mass–elite congruence remains strong for issues associated with economic left-right, which still proves to be the main dimension of contestation in European politics and may be seen as an effective basis for political representation. In contrast, voters’ attitudes and parties’ policy positions in domains such as EU integration and immigration tend not to be constrained by the left-right dimension (Hooghe et al., Reference Hooghe, Marks and Wilson2004; Kriesi et al., Reference Kriesi, Grande, Lachat, Dolezal, Bornschier and Frey2006). This would help to explain the lower levels of policy congruence in these specific domains.

Let us now examine how candidate–voter congruence varies across parties. The 2013 Italian elections saw the participation of important new parties such as Beppe Grillo’s M5S and Mario Monti’s SC. We could reasonably expect a rather low knowledge of the views of voters for candidates in parties running in their first general election. In contrast, for parties with a longer electoral history (such as LN, PD, and PdL) candidates should have more accurate perceptions of the actual policy preferences of their electoral base. Data from the ICS does not support these expectations. Overall, it is PD candidates who have the largest policy preference gap with their electoral base. These candidates believe themselves to be distant from their party’s voters for European and family policy issues, where their perceived distance from voters is >1 point on the 0–10 policy scales used. This difference is statistically significant (p<0.01). Candidates who believe themselves to be the closest to their voters come from the SEL (although not on EU matters), SC, and FdI parties.

Comparison between the self-declared positions of voters and candidates tells a different story. This is because the actual preference gap between candidates and voters is higher for almost all of the left and centre-left parties (the CD, RIV, and the PD), and is lower for the right and centre-right parties (the LN, the PdL, and the FdI). The largest misperception by candidates of the policy preferences of their voters is for CD and RIV. Here the sum of absolute policy difference distances on the four dimensions is 10.4 and 7.8, respectively. In contrast, PdL candidates have the most accurate view of their party’s voters’ policy positions where the sum of absolute distances is 1.9. The level of misperception by candidates from new parties (M5S and SC) is not particularly high, that is, mean absolute policy differences are 6.0 and 5.1, respectively. Our data suggest that PD candidates know their policy preferences are different from those of their voters. This is especially true for the issues of EU integration and family policy where the policy differences are large and statistically significant; however, this is not the case for immigration preferences. In contrast, would-be legislators from CD (particularly in the domains of EU integration and immigration) and RIV (especially for family, immigration and tax-services policies) were not aware of the policy preference gaps between themselves and their respective electorates. In general, left and centre-left parties seem, during the 2013–14 ICS fieldwork period, to have been unaware of the anti-immigrant preferences of their voters.

Party cohesion

Political parties are central actors in democratic regimes (Schattschneider, Reference Schattschneider1942; Sartori, Reference Sartori1976). Beyond being crucial for democratic representation, parties guarantee the functioning of legislative assemblies as they build a link between the legislative process and policy outputs by grouping legislators under common labels (Aldrich, Reference Aldrich1995). These labels give voters information about how candidates are likely to vote across a large number of policy issues and represent the main vehicle through which politicians can enter public office and influence policymaking (Snyder and Ting, Reference Snyder and Ting2002). In order to play their role as a mechanism of political accountability in electoral democracies, parties need to be cohesive (Strøm and Müller, Reference Strøm and Müller2009).

Cohesion refers to the extent to which members of parliamentary party grouping act in concert because they share common values and preferences (Bowler et al., Reference Bowler, Farrell and Katz1999). Legislative party cohesion needs to be distinguished from related concepts such as ‘party unity’ and ‘party discipline’ (Hazan, Reference Hazan2003). Party unity refers to the degree to which party members, who are observed, act in a united manner (Sieberer, Reference Sieberer2006). Higher levels of unity are important because they give parties a stronger bargaining position with competitors in government formation and in policy-bargaining processes. Party unity is often the product of two mechanisms as follows: (1) shared preferences and (2) party discipline. Both mechanisms are the basis of a strategic game played within a party, whereby representatives respond to positive incentives and sanctions created by an internal party decision-making regime (Giannetti and Laver, Reference Giannetti and Laver2009).

To assess party unity, empirical research has most often relied on the behaviour of legislators in roll-call voting or other parliamentary activities, such as cosponsoring legislation, committee participation, or speech making (Carroll and Poole, Reference Carroll and Poole2014). However, measures of behaviour in legislative activities may be contaminated by partisan and institutional constraints, including party discipline, agenda setting, and logrolling. These ‘contaminants’ make it almost impossible to distinguish between preference- or strategic-induced unity (Saiegh, Reference Saiegh2009). This results in a high degree of conceptual overlap, which can confound inferences about the level and sources of party unity.

Unlike measures of behaviour, candidate or parliamentary survey responses are not contaminated by legislative or partisan institution effects. This allows scholars to assess exogenous, non-behavioural measures of party cohesion. Moreover, focussing on party cohesion facilitates study of internal groups (or factions) in political parties with programmatic viewpoints that diverge from a majority of fellow party members or the current party leadership. A growing body of literature on intra-party politics highlights the prevalence of both ‘internal dissent’ and party cohesion in Western countries (see Giannetti and Benoit, Reference Giannetti and Benoit2009).

In this section, we use the ICS data to evaluate party cohesion on the left-right dimension during the 2013 Italian general elections. Looking at the important role played by cohesion in providing information to voters regarding party policy preferences, assessing the coherence of party labels in the specific case of Italy may cast further light on the relationship among voters, candidates, and the parties to which they belong. In particular, we asked each candidate to place themselves on an 11-point scale (0–10) left-right self-placement scale. The results reported in Figure 5 provide kernel density estimates of the (1) mean party positions indicated by vertical lines and (2) the distributions of candidates’ left-right self-placements for the three main parties (i.e. PD, M5S, and PdL) studied in the ICS.

Figure 5 Distribution of candidates’ left-right self-placements by party.

Looking first at the kernel density distributions, Figure 5 shows that the leftist PD (light grey short-dashed curve) and rightist PdL (dark grey long-dashed curve) are mainly located towards opposite ends of the ideological spectrum as one would expect. There is not much overlap in the centre of the left-right dimension. In contrast, the M5S candidates’ left-right positions cover a much wider range (solid black curve): the distribution peaks at the very centre of the spectrum where the M5S distribution covers substantial parts of both the left and right parts of the dimension. The lines representing mean party positions show an almost symmetrical location for the PD and the PdL, whereas the M5S appears closer to the centre (point 5 on the 0–10 point scale). In light of the distribution of preferences described above, the centrist position of Grillo’s M5S cannot be considered as an expression of an ideologically cohesive group of moderate candidates, but as the product of an aggregation of a wide range of contrasting preferences. However, this evidence must be treated with caution, as a relatively high proportion of respondents from M5S (the 27% against the 3% for the PdL and the 1% for the PD) refused to answer this question. This is because M5S party members stress that the terms ‘left’ and ‘right’ are obsolete concepts that are not useful for understanding contemporary politics (Pedrazzani and Pinto, Reference Pedrazzani and Pinto2013, Reference Pedrazzani and Pinto2015). A final element deserving attention is the ‘shape’ of the three distributions. Although the PdL and the M5S show a unimodal distribution, the PD distribution has three local peaks: the highest, close to the mean party position, and two lower modes, one close to the extreme left and another close to the centre. This evidence may reflect the existence of different factions within this party.