Leadership, media, arenas of competition: an overview

One of the aims of this article is to understand the role of leadership in the individual decisions to vote either for a party or for a candidate to the party leadership. In the last years leadership has been widely studied (Manin, Reference Manin1997; King, Reference King2002; Poguntke and Webb, Reference Poguntke and Webb2005; Garzia, Reference Garzia2014) as a factor that could be able to modify established frameworks analysing voting behaviour such as the ones by Downs (Reference Downs1957); Stokes (Reference Stokes1963), and Lipset and Rokkan (Reference Lipset and Rokkan1967). Nonetheless, as underlined by Barisione (Reference Barisione2009), researchers studying the impact of leadership on voting behaviour should not limit to ask themselves ‘do leaders matter?’, rather, they should include some other questions in their analysis: ‘[t]he point to be considered is not only “how much”, but also “who” can make a difference, along with “where”, “when”, “to whom”, and “how”’ (Barisione, Reference Barisione2009: 474).

Some scholars have also taken into consideration the fact that voters obtain a noticeable amount of information about candidates and parties from candidates and parties themselves or from TV, newspapers and, more recently, the internet. This is also one of the strands of research we deal with in this study. The literature on this topic is quite wide, and different theories have been developed (Thompson, Reference Thompson2004). More reductionist approaches (Katz and Lazarsfeld, Reference Katz and Lazarsfeld1955; Klapper, Reference Klapper1960) contrast with theories according to which the role of media is more pervasive (Meyrowitz, Reference Meyrowitz1985), while some scholars take a more nuanced position (Neuman et al., Reference Neuman, Just and Crigler1992). On a different perspective, more recent contributions underline that different media can either create different ‘hybrid systems’ that influence political communication and voting behaviour (Chadwick, Reference Chadwick2013) or have quite different effects also thanks to the influence of audience fragmentation and political polarization (Bennett and Iyengar, Reference Bennett and Iyengar2008). Moreover, the features and consequences of the diffusion of Web 2.0 are more and more relevant for scholars. The bottom-up process through which information is spread into the web, the development of a collective knowledge and the process of the disintermediation of communication that characterize new media (Mosca and Vaccari, Reference Mosca and Vaccari2011) could be relevant discriminatory factors in individual voting decisions.

An interesting question is whether Italy is a relevant case to consider regarding the impact of leadership and of (possibly different) media diets on voting behaviour. Concerning media, many studies analyse the transformation of the political communication tools in the last Italian general elections and, more generally speaking, during the so-called ‘Italian Second Republic’ (the post-1992–94 period). Some scholars study the influence of ‘old media’ (mainly, but not exclusively, TV) in the formation of political preferences and in voting decisions (Campus, Reference Campus2000; Legnante, Reference Legnante and Baldassarri2002; Legnante and Baldassarri, Reference Legnante2010), without identifying a unique and generalizable causal mechanism underlying such relationship. Moreover, even the recent inception and development of new media is extensively analysed (Bentivegna and Ceccarini, Reference Bentivegna and Ceccarini2013), and the general conclusions of such studies underline only a marginal, albeit significant, role of new media in voting decisions, both in general and in primary elections (on this very last point see Mazzoni and Vicentini, Reference Mazzoni and Vicentini2014).

As far as leadership’s role in Italian voters’ decision-making is concerned, several studies underline the growing importance of leaders for voting behaviour. Specifically, in the last 20 years the image of party leaders ‘has turned into the most relevant driver of individuals’ party identification and (indirectly) vote choices’ (Garzia and Viotti, Reference Garzia and Viotti2012: 19). More nuanced conclusions, focussing on the 2006 Italian general election, are reached by Catellani and Alberici (Reference Catellani and Alberici2012), who underline that candidates’ features had an impact on voting behaviour also in connection with other variables such as ideology and political sophistication, and by Bellucci et al. (Reference Bellucci, Garzia and Lewis-Beck2015), who show that leaders’ images had a role in voting decisions, but an even stronger role was that of the perceived ideological distance between voters and coalitions.

Finally, this article also analyses the differences between voting behaviour in party leadership races and in national elections. Do these different competitions have an impact on voting behaviour? Are there relevant differences between the selectorate and the electorate of a party? According to Key’s study (Reference Key1956) on the American case, it is reasonable to expect that, under certain conditions, such differences exist: primary voters could not be considered as representative of party voters, and therefore selectors may hinder the chances of the party to win general elections. Other studies do not confirm Key’s conclusions: Ranney and Epstein (Reference Ranney1966) and Ranney (Reference Ranney and Epstein1968), working on Wisconsin primaries, posit that these differences between party selectors and party voters are hardly detectable, at least in that specific context. Another local study (DiNitto and Smithers, Reference DiNitto and Smithers1972) confirms that only as far as information levels are concerned there are differences between the selectorate and the electorate. Such ‘reductionist’ views are substantially confirmed by the US nationwide study by Norrander (Reference Norrander1989). On the contrary, Geer (Reference Geer1988), analysing Democratic and Republican primaries in several US states in 1976 and 1980, finds out that selectorate and electorate do show some important differences regarding, for instance, the level of education or the ideological self-positioning.

Even if the history of the diffusion of the election of the party leader (or of the coalition leader) by party members or voters in Italy is not as long as in the United States, some scholars consider a number of points which can be useful for our analysis. As we have already argued in the Introduction section, there are no studies dealing with differences in the causal mechanisms driving voting decisions of Italian voters and selectors. Nonetheless, some contributions focus on the characteristics of PD’s selectorate and electorate. For instance, Cavataio and Fasano (Reference Cavataio and Fasano2014), analysing the primaries for the selection of the centre-left Prime Minister candidate in 2005 and 2012 and the primaries for the selection of the PD leader, show that there are some differences regarding the propensity to vote for the PD – or for its ancestorsFootnote ³ – in the national elections for different selectors who took part in the centre-left coalition primaries between 2005 and 2012. Moreover, focussing on the PD selectorate, a study on the 2012 centre-left coalition primaries (Sandri and Seddone, Reference Sandri, Seddone and Venturino2015), underlines differences between different categories of selectors with regards to their ideological self-positioning, interest in politics, and voting intentions in the subsequent 2013 Italian general election.

With respect to issue saliency within primaries and general elections, the analysis by De Sio (Reference De Sio2011) shows that it is reasonable to expect that in party leadership races there is a stronger focus by the selectorate on leadership-related issues and a weaker focus on ideology-related (or positional) issues. That is, during primary elections candidates are particularly interested in keeping the party united on ideological issues and selectors are interested in finding the right person to represent the party in national elections according to her or his personal qualities. From a different perspective, a specific analysis of the party leadership races within the PD in 2007 and 2009 (Seddone and Venturino, Reference Seddone and Venturino2013) underlines that there are differences between the PD selectorate and the PD electorate and that one of the consequences of these differences may be a ‘primary penalty’ (Carey and Polga-Hecimovich, Reference Carey and Polga-Hecimovich2006) for the PD: being the selectorate more ideologically extreme than the electorate (i.e. being the median selector of the party not in the same position of the median voter of the party), the party leadership may find itself in a difficult situation when appealing to the broad electorate. It can be argued that this strand of the literature underlines the role of the ideological component on voting decisions by the selectorate and the electorate of a party.

What differences can be found in the analysis of a party’s selectorate and a party’s electorate? Are these groups influenced by the same voting determinants? In the next section, we underline why the Italian PD is a relevant case to answer these questions, while in the third section, we will put forward some hypotheses concerning the differences we expect to find in the analysis of the PD’s electorate and selectorate.

Case study, data selection, and hypotheses

The PD is an appropriate case for the research question we have posed in the introduction for many reasons. First of all, the PD is the first and the only main Italian party that allows its voters to select the national leader and, therefore, this pushes for the comparison between its selectorate’s electoral behaviour and its electorate’ electoral behaviour.

Second, since 2005 the PD and its ancestors have used primary elections to nominate the national leader and the Prime Minister candidate for a relevant number of times. Since 2005 five national elections have been organized: the PD elected three party leaders (Walter Veltroni in 2007, Pier Luigi Bersani in 2009, and Matteo Renzi in 2013) and nominated two candidates for the Presidency of Council of Ministers (Romano Prodi in 2005 and Pier Luigi Bersani in 2012). Thus, this selection pattern appears to be well institutionalized within the PD.

Third, since in 2013 the PD incurred both in the general election and in the party leader selection, it is possible to compare these two elections without discounting all those time-related biases that can affect behavioural analysis. Indeed, not only is the gap between the Italian general election and the PD primary election in 2013 not very wide, but we have also decided to consider in the empirical analysis of the PD primary election only those who have stated that they had voted for the PD in the precedent Italian general election. This decision has been taken in order to compare two populations of which one (the selectorate) is the subsample of the other one (the electorate), also following Duverger’s reasoning (Reference Duverger1951).

Fourth, there are similarities between the electoral law used in Italian general elections and the rules used in PD primary elections that further consolidate the comparison we perform. In the 2013 Italian general election voters used a Proportional Representation (PR) system with a majority bonus for the winning coalition (no preferential voting was allowed), while PD selectorate in 2013 voted, without preferential voting, with a PR system for a list representing a motion that in turn was connected to a candidate for leadership, who needed to obtain the absolute majority of delegates in order to avoid a runoff (Fiorini and Porcellato, Reference Fiorini and Porcellato2014).Footnote ⁴ The rules of the 2013 PD primary election granted the right to vote to all those citizens who declared to be PD electors, agreed with PD’s fundamental political values, agreed to be inserted into a ‘primary voters list’ and gave 2 Euro to the party (Venturino, Reference Venturino2014). Differently from the rules for national election, 16-year-old citizens and immigrants with the residence permit were entitled to vote (but we have excluded them from the empirical analysis in order to facilitate the comparison with the PD voters in 2013 Italian general election). The different requirements to be admitted to the vote, in addition to the different magnitude of the two populations (PD voters in the 2013 Italian general election were more than 8,600,000 while in 2013 PD primary election ~2,800,000 people cast their vote) can be also seen with respect to the socio-demographic characteristics of the two samples (which will be described later in this section). Indeed, there is a higher proportion of women in the PD electorate (47%) than in the PD selectorate (41%); moreover, PD voters are on average slightly older than PD selectors (53 vs. 52 years) – albeit PD selectors are older than the average Italian elector in 2013 (52 vs. 49 years) – and have a lower proportion of people with a university degree than PD selectors (15 vs. 35%). Finally, the 53% of PD voters is fairly or very interested in politics; this is a higher percentage than that of the average Italian voter in 2013 (37%) but lower than the average PD selector in the same year (82%).

Finally, 2013 represented a turning point in Italian politics. The noticeable result obtained in the general election by the Movimento Cinque Stelle (M5S) – the best electoral result of a new party in a post-Second World War election in Western Europe, excluding the democratic foundational ones (Chiaramonte and Emanuele, Reference Chiaramonte and Emanuele2013) – signalled an important change not simply for electoral and political reasons, but also for political-communication-related ones. In fact, several scholars underline the role of the internet in fostering the message of M5S (Mosca et al., Reference Mosca, Vaccari and Valeriani2015). For all these reasons, we believe a comparison between PD electors and PD selectors can be very fruitful.

As far as data selection is concerned, we use 2013 Candidate and Leader Selection survey data concerning the PD’s selectorate.Footnote ⁵ Regarding the electorate, we use 2013 Italian National Election Study (ITANES) survey data.Footnote ⁶

Moving to the presentation of our hypotheses, we first expect that, as already seen above, a candidate-related issue is more salient during a primary election than during a general election. Indeed, because of the higher homogeneity within the selectorate, and of the desire by the national leadership and cadres not to make the party appear divided on ideological positions (De Sio, Reference De Sio2011) in light of the election, we expect candidates for party leadership to compete mostly on non-policy-related issues and, consequently, selectors give more relevance to those issues. It means that candidates make their personality the main issue on which selectors make a decision. Thus, we argue the relevant differences between voters and selectors regard the voting determinants of the two groups. Starting with the role of leadership’s evaluation, we hypothesize that:

On the contrary, during electoral campaigns for general elections, parties have the interest to differentiate among themselves on policies (Merrill and Grofman, Reference Merrill and Grofman1999). Indeed, despite the fact that candidates have gained importance in electoral campaigns (Garzia, Reference Garzia2014; Franchino and Zucchini, Reference Franchino and Zucchini2015), we expect the left-right dimension to be still relevant (Bellucci and Segatti, Reference Bellucci and Segatti2011), unlike to what we expect to happen in primary elections. This is another difference we expect to find between a party’s selectorate and a party’s electorate. Therefore, a second hypothesis is tested:

Turning to the role of mass communication tools, old media (TV, newspapers, and radio) may have a role in influencing voting decisions both in primary elections and in general elections. Specifically, the Italian case is very interesting to analyse because of its specific media environment. As underlined by Hallin and Mancini (Reference Hallin and Mancini2004), in Italy there has been a high political parallelism between media and politics and a certain degree of control of media environment by politics (especially in, but not limiting to, the case of TV state-controlled and private service). Nonetheless, this is true for old media but not for new media, because in Italy there is not a relevant top-down relationship between politics and the internet (for instance, regarding funding systems). Therefore, we can expect that, all other things being equal, old media have a less watchdog role than new media, because of the political links of the former ones. This means that leaders are more likely to find a comfortable and not hostile environment on TV channels, unlike what could happen on the internet. Moreover, it must also be added that old media, and especially TV, increase the relevancy of personality traits (Caprara and Zimbardo, Reference Caprara and Zimbardo2004) and, in a situation of strong personalization of leadership, acquiring information from TV could have an impact of candidates’ evaluation by voters (Maricchiolo and Bonaiuto, Reference Maricchiolo and Bonaiuto2011). Finally, we can imagine that in a country like Italy, where there is a strong digital divide (ISTAT, 2015), new media could not have a noticeable role in voting decisions, unlike other means of communication. Thus, two hypotheses are considered:

We now turn to the empirical analysis in order to test the hypotheses we have just put forward.

Empirical analysis

In order to test the hypotheses set in the previous section, we perform two multivariate alternative-specific conditional logistic regressions (Greene, Reference Greene2012): one on ITANES’ data and one on Candidate and Leader Selection’s data. The alternative-specific conditional logistic regression is a particular case of conditional logistic regression (McFadden, Reference McFadden1974; see an application of this model in Mauerer et al., Reference Mauerer, Thurner and Debus2015) that allows researchers to consider both decisor-related and alternative-related variables. The difference between the two types of variables is straightforward: decisor-related variables only vary across individuals since they represent individuals’ characteristics (for instance, media-related variables and the level of interest towards politics); on the contrary, alternative-specific variables are related to the alternatives of the dependent variables and consequently vary across individuals and alternatives (for instance, leadership evaluation and distance from candidates or from parties). In order to facilitate the readability, in the text we simply refer to the models by calling them ‘regressions’.Footnote ⁷

Specifically, we test the impact of several independent variables on the vote for the PD in the 2013 Italian general electionFootnote ⁸ and on the vote for all the candidates in the 2013 PD primary election. These are our two dependent variables. The use of this specific statistical model allows us to compare the overall subset population of selectors with the more general and numerous set of PD voters. Moreover, this model allows us to compare the way in which electors and selectors take decisions and to evaluate the saliency of each factor in their voting behaviour for the first time, to our knowledge, in the Italian case.

We test the impact of several independent variables. In order to test Hypotheses 1 and 2 we have created Leadership and Distance. The first variable is a 0–10 scale where 0 represents the lowest possible vote and 10 the highest possible vote attributed to a certain leader by voters or by selectors. Distance, instead, is the squared distance between, on the one hand, the position of the candidate in the PD primary election or the position of the party in the Italian general election (computed as the mean of its voters’ or his selectors’ positions) and, on the other hand, the position of each voter or selector.Footnote ⁹

In order to account for the effect of media, we have created Old Media, Prevalent Old Media, Prevalent New Media, New Media, and Contact with a candidate (or Contact with a party). Old Media is a dichotomous variable in which 1 corresponds to the fact that the prevalent and the secondary sources of information of an individual are both old media.Footnote ¹⁰ Prevalent Old Media is also a dichotomous variable, where 1 corresponds to the fact that an individual’s prevalent source of information is an old media. Instead, Prevalent New Media is a dichotomous variable whose value 1 corresponds to the fact that an individual’s prevalent source of information is the internet. Furthermore, New Media is a dichotomous variable in which 1 corresponds to the fact that the prevalent and the secondary source of information of an individual are both the Internet. Finally, Contact with a candidate (Contact with a party) is a dichotomous variable in which 1 corresponds to the fact the individual has taken information directly from a candidate or from his supporting committee (for the PD primary election) or from a party (for the Italian general election).

Theoretically Hypothesis 3 can be tested by using the interaction between Old Media and Leadership, while Hypothesis 4 can be tested by using Prevalent New Media and New Media. Nonetheless, in order not to bias the results of the empirical analyses we have decided to consider all media-related variables in both regressions.

Moreover, we control for Interest in Politics (which is a 0–4 scale, where 0 represents no interest and 4 high interest, on which respondents are asked to position themselves), Age, Gender, Education (which measures the level of education of respondents) and for the Geographical Area of respondents (measuring where they resideFootnote ¹¹ ).

The analysis of 2013 Italian general election and of the 2013 PD primary election

Table 1 presents the results of the regression performed on data about the 2013 Italian general election. We have decided to include, for data reliability reasons, the five biggest political parties that ran in the 2013 Italian general election: the PD, M5S, Popolo delle Libertà (PDL), Scelta Civica, and Lega Nord. Respondents who declared to have voted for such parties correspond to the 85% of those who have declared they had cast a vote. We have decided to use the PDL as the base outcome for the regression. In this case we present two models: in Model I only the control variables are considered, while Model II includes all the variables used in the empirical analysis.Footnote ¹²

Table 1 Alternative-specific conditional logistic regression on 2013 Italian general election (PD-related variables)

PD=Partito Democratico; BIC=Bayesian information criterion.

Log odds coefficients (Coeff.) reported; SE in parentheses; variable ‘new media’ not reported for quasi-complete separation.

⁺ P<0.10; *P<0.05; **P<0.01; ***P<0.001.

First of all, Table 1 shows that only in Model I Age, Education, and Interest in Politics are significant predictors of the probability to vote for the PD compared with the base outcome. Indeed, by adding in Model II media-related variables and also Leadership and Distance all these control variables become non-significant predictors of the decision to vote for the PD.

Passing to the analysis of Model II, as far as media-related variables are concerned, Contact with a Party positively affects (P<0.05) the probability to vote for the PD compared with the base outcome, while Old Media, with respect to the base outcome (Prevalent Old Media), decreases the probability to vote for the PD (P<0.10) compared with PDL. Moreover, the interaction between Old Media and Leadership, which is necessary to test Hypothesis 3, is not significant. The other media-related variable reported does not have a significant impact on the decision to support the PD compared with the base outcome.Footnote ¹³ Further considerations on the relevance of these findings for Hypotheses 3 and 4 can be drawn by comparing these results with those of the analysis of the 2013 PD primary election, which we will be presented in the following paragraphs.

Moving to the political variables, we analyse the effect of Distance: as the ideological proximity between the voter and the party decreases, the probability to vote for the PD increases. In other words, being closer to the PD on the ideological continuum significantly increases (P<0.001) the probability to vote for such party. This finding confirms that the ideological dimension is relevant for PD voters in the Italian general election of 2013. Nevertheless, following the comparative nature of Hypothesis 2, we need to verify the effect of this variable on PD selectors before considering this hypothesis as confirmed. Moreover, Leadership has a positive and very significant (P<0.001) effect on the probability to vote for the PD. Even in this case, an analysis of the effect of this variable on PD selectors’ voting decisions is necessary before confirming or disconfirming Hypothesis 1.

Moving to the analysis of the 2013 PD primary election, Table 2 presents the results of the regression analysis performed on the PD primary election data. Even in this case, Model I deals with control variables, while Model II presents all the variables included in our analysis. We have decided to use Gianni Cuperlo as base outcome.Footnote ¹⁴

Table 2 Alternative-specific conditional logistic regression on 2013 Partito Democratico primary election

BIC=Bayesian information criterion.

Log odds coefficients (Coeff.) reported; SE in parentheses.

⁺ P<0.10; * P<0.05; ** P<0.01 *** P<0.001.

The first model presented in Table 2 shows a significant impact of many control variables on the decision to support one of the three candidates to PD party leadership. Once we add the explanatory media-related and political variables in Model II, only Gender, Age, and Education positively affect the decision to vote for Matteo Renzi compared with the base outcome, while South negatively affects the probability to vote for Giuseppe Civati compared with the base outcome. Turning to the media variables, only Contact with a Candidate significantly and negatively affects the probability to vote for either Giuseppe Civati (P<0.05) or Matteo Renzi (P<0.10) compared with the probability to support Gianni Cuperlo. Conversely, Old Media, Prevalent New Media, and New Media do not have any significant effect on the probability to support any candidate to PD leadership compared with the base outcome (Prevalent Old Media). If this is something expected for new media, it is a quite surprising result for Old Media. All in all, considering the analysis of Prevalent New Media for PD voters and PD selectors and of New Media for PD selectors, we can consider Hypothesis 4 as totally confirmed for PD selectors and only partially confirmed for PD voters.

Turning to Hypothesis 3, we have seen in Table 1 that the interaction between Old Media and Leadership, which is necessary to test this hypothesis, is not significant as far as 2013 Italian general election is concerned. What about the effect of such interaction in the 2013 PD primary election? Surprisingly, even in this case we can see, by looking at Table 2, that there is not a significant effect of Leadership on the decision to vote for a candidate when both the primary and the secondary sources of information of respondents are old media. We can conclude Hypothesis 3 is disconfirmed by the empirical analysis.

Passing to the analysis of the political variables, Distance is significant (P<0.01) for the decision to support Giuseppe Civati or Gianni Cuperlo (P<0.10), while it has not a significant impact on the decision to support Matteo Renzi. Finally, as we expected, Leadership has a very significant (P<0.001) and positive impact on the decision to support all the three candidates. In other words, the more Matteo Renzi, Gianni Cuperlo, or Giuseppe Civati are positively considered by a selector, the higher is the probability that this selector decides to vote for one of them.

PD’s electorate and selectorate in 2013

Up to now we have separately analysed the impact of our independent variables on the decision to vote for the PD in 2013 Italian general election and on the decision to vote for one of the three party leader candidates in the 2013 PD primary election. In order to test Hypotheses 1 and 2 we need a rigorous comparison of several variables’ relevancy in these two races.

Table 3 reports the percentage changes in the odds of Leadership and Distance in the analysis of the 2013 Italian general election and in the analysis of the 2013 PD primary election. Starting from Hypothesis 1, Table 3 tells us that Leadership has a higher impact on the decision to vote for one of the three candidates in the PD primary election than on the decision to vote for the PD in the 2013 Italian general election. Indeed Leadership’s percentage change in the odds for the PD is equal to 109.4 (P<0.001). It means that for a one-unit increase on the independent variable Leadership, the probability to vote for the PD with respect to the probability to vote for the other parties increases by 109.4%. Passing to PD primary election, Leadership’s percentage change in the odds for Matteo Renzi is equal to 488.3 (P<0.001), for Gianni Cuperlo is 471.7 (P<0.001) and for Giuseppe Civati 375.9 (P<0.001). In other words, while the statistical significance of the variable is equal for both PD electors and selectors, the magnitude of the percentage change in the odds is noticeably different: as Leadership increases, ceteris paribus, the increase in the probability to vote for one of the candidates for PD leadership is bigger than the probability to vote for the PD as the evaluation of its leadership increases by one unit. It follows that Hypothesis 1 is confirmed by the empirical analysis.

Table 3 Percentage changes in the odds of Leadership and Distance on 2013 Italian general election and on 2013 Partito Democratico (PD) primary election

Percentage changes in the odds and P-value reported.

We now turn to Hypothesis 2. The analysis of Table 3 says that the percentage change in the odds of Distance for PD electors is equal to −11.4 (P<0.001). Conversely, as we have also seen in the analysis of Table 2, Distance’s percentage change in the odds is not statistically significant for Matteo Renzi. Conversely, Distance has a significant effect for Giuseppe Civati and Gianni Cuperlo, albeit this effect is not at the maximum level of confidence (P<0.01 for Civati; P<0.10 for Cuperlo), differently from the case of PD. Therefore, the probability that these results have been caused by chance is higher than in the case of the PD, as the P-values of the three percentage changes show (P<0.001 for the PD vs. P<0.01 and P<0.10). This does not allow us to confirm that the effect of Distance is more statistically relevant for the selectors of Giuseppe Civati and Gianni Cuperlo than for the PD voters. Moreover, if we add the fact that in the case of Matteo Renzi’s selectors the effect of Distance is not statistically significant, we can consider Hypothesis 2 as confirmed by the empirical analysis. That is, we can safely posit that a decrease in the distance on the ideological continuum between one candidate for PD leadership and one selector has a lower impact on the vote for such candidate than the same decrease on the ideological continuum between the PD and one voter on the vote for such party.

A final consideration on the importance of Distance is related to the alleged differences between party voters and selectors underlined in the American and the Italian literature on the topic. The implicit assumption of many theories is that the ideological continuum is a very relevant voting determinant for both party voters and party selectors – otherwise the difference on the ideological continuum between party voters and party selectors would not make a lot of difference for a party. This assumption does not hold as far as 2013 PD voters and PD selectors are considered: it can indeed be noticed that many PD selectors do not consider the ideological positioning as the most relevant issue on which to evaluate whether to support a party leader candidate instead of another one.

PD voters and PD selectors: irreconcilable differences?

This article has tested, for the first time, whether there are differences in vote choice determinants between the electors and the selectors of an Italian party. Going into detail, the analysis of the 2013 Italian general election and of the 2013 PD primary election has shown that PD electors and PD selectors are different, because their voting decisions are based on different motivations and because there are diverse dimensions on which the electoral competition develops in party primaries and in general elections. Specifically, leadership evaluation and the distance between a voter (or a selector) and a party (or a candidate) have a different impact on the decision to vote taken by PD selectors and PD voters. Going into detail, the average PD selector in the 2013 party primary election gives a lot of importance to leadership evaluation, while such evaluation has a lower importance for the average PD voter in the 2013 Italian general election. Conversely, the distance on the ideological left-right continuum mostly affects PD voters. Indeed, concerning the 2013 Italian general election, the higher the distance between a voter and the PD, the lower the probability he or she will vote for this party. Turning to PD selectors, this relationship only holds for the two minority candidates, Giuseppe Civati and Gianni Cuperlo, while those selectors who voted for Matteo Renzi were not influenced by this factor. Nevertheless, the impact of the ideological distance on the vote for these two candidates is noticeably lower than that on the vote for the PD.

Are differences between PD voters and selectors also detectable as far as information sources are concerned? The answer to this question is substantially negative: both PD voters in the 2013 Italian general election and PD selectors in the 2013 party primary election are not influenced, in their voting decisions, by a number of media diets. The only exception is represented by the direct contact of voters and selectors with the party or with the candidates for party leadership and by the use of old media by PD voters.

This result is somewhat surprising. It was foreseeable, also according to the literature on the topic, to expect that old media could have had a role in influencing voting decisions especially when old media have an intervening role with respect to the impact of leadership evaluation. Nonetheless, the empirical analysis has disconfirmed this idea. Further pieces of research could deeply investigate this issue, which surely represents an interesting and novel point to deal with. Moreover, it must be stressed that while many commentators have highlighted the role of the internet in the mobilization and subsequent electoral inception of the M5S, new media seem not to have influenced neither the decision to support the PD in the 2013 Italian general election nor the decision to vote for a candidate to party leadership in the 2013 PD primary election.

To sum up, we have underlined some relevant differences between PD voters and PD selectors in 2013. Such differences confirm the relevancy of leadership over policy issues in the intra-party arena (De Sio, Reference De Sio2011) also in a crucial year for Italian politics, the 2013. This year was crucial not only for the PD, since it did not manage to obtain a viable majority in both houses of the Italian parliament in the general election and it underwent a leadership crisis that ended with the victory of Matteo Renzi in the subsequent party primary for leadership, but also for the Italian political system as a whole, since it saw the inception of M5S, that was able to obtain a quite noticeable electoral result. While some could argue the results of the PD primary election were influenced by the outcome of the general election held in the same year (partly following the argument by Bille, 1997), some others could point out at the role of the particular electoral campaign before 2013 Italian general election in shaping voters’ behaviour (Garzia, Reference Garzia2013). Even if at this stage it is not possible to take a clear position, this represents a starting point from which future analyses can easily depart. Indeed, a logical question is whether these differences between party voters and selectors will change in the future. Starting from 2013, the first year in which PD selectors were asked to evaluate candidates to party leadership, this question could be addressed in a diachronical way by analysing the next PD congress and the next Italian general election. But this will be another story.