1In Italy, referenda are relatively common. As Morel (2012) highlights, Italy is one of the countries that uses referenda the most frequently, along with Switzerland, Liechtenstein, Ecuador, and Micronesia. However, none of these other countries have a population as large as Italy’s, and this combination of a large population and frequent use of referenda is what makes it a unique case. In sharp contrast with the Brexit referendum, for which a burgeoning literature has analyzed different aspects of voting behavior (e.g., Becker et al., 2017; Clarke et al., 2017; Goodwin & Heath, 2016; Hobolt, 2016; and Los et al., 2017) the 2016 constitutional referendum in Italy has so far attracted much less attention. While Brexit had profound consequences for the United Kingdom, Europe, and beyond, the Italian referendum “only” led to the resignation of the prime minister (PM), Matteo Renzi. However, the long-term consequences, be they direct or indirect, also had an impact on the next general election in 2018, after which two anti-establishment parties—the Movimento 5 Stelle (M5S) and Lega (called Lega Nord before the election) formed a coalition, following lengthy negotiations.

2Unlike normal elections, referenda are direct democracy devices through which “the people are asked to vote directly on an issue or policy” (Morel, 2012, p. 501). The Italian Constitution allows for two types of referenda: popular referenda in which the electorate votes on whether to abrogate existing legislation with a 50%+1 quorum of participation; and referenda without a quorum, which can be either advisory or constitutional. Morel’s definition fits all the cases specified in the Constitution. Drawing on the literature on referenda across Europe and the US, this paper highlights an uncommon pattern in voting behavior: the relevance of the leader in shaping voters’ positions. While voters’ preferences on an issue may be influenced by party identification, this paper shows that beyond party identification, voters’ evaluations of party leaders played a crucial role in the 2016 referendum. Using pre- and post-referendum data drawn from ITANES (Italian National Election Studies) surveys, I test three different hypotheses in order to evaluate which aspects had the most impact on voting behavior.

3The paper proceeds as follows: the first part analyzes the literature on direct democracy and voting behavior. Rather than providing an in-depth analysis of referenda around the world (e.g., Cronin, 1999; Hollander, 2019; LeDuc, 2003; Qvortrup, 2014), this overview is intended to both frame the typology of the 2016 referendum and to provide a starting point for analyzing the main patterns that impact on voters’ choices. Based on these premises, I formulate a set of research questions related to the different interpretation of voters’ preferences. The second part contextualizes the case study by providing an overview of past referenda in Italy and of the content of the 2016 referendum. The third focuses on the political context before the 2016 referendum, along with the electoral campaign and the alleged personalization process that took place in the months running up to the vote. This part not only better contextualizes the referendum, but it also links back to the theoretical framework delineated in the first part. The fourth part presents the data, the methodology, and the hypotheses, while in the fifth and final part, I describe and discuss the results.

Voting for what? Direct democracy and voting behavior in Western countries’ referenda

4Direct democracy is on the rise around the world: in her comparative analysis, Susan Scarrow (2001) found that while referenda usage is still relatively low, an increasing number of countries have implemented institutional reforms to include direct democracy devices. This is unsurprising: despite disagreement over which sectors of society are more likely to accept the expansion of direct democracy (Dalton et al., 2001; Inglehart, 1997), several scholars have highlighted that worldwide public opinion is increasingly demanding more direct involvement in decision-making, through both conventional and non-conventional forms (Dalton, 2002). This growth has crucial implications for the institutional frameworks in which referenda take place, for the interactions between the actors involved—be they institutional (parties, elected elites) or non-institutional (social movements, interest groups, and so on)—, and for representative democracy in general (e.g., Setälä, 1999, 2009; Hollander 2019; Uleri 2011).

5Even though referenda can have a wide range of typologies (see below), they all have a specific feature in common: they allow citizens to exercise their right to vote on a political issue, which can range from a constitutional reform to change the political regime, to a local non-binding referendum on a specific policy. As the literature has highlighted (Scarrow, 2001), local referenda substantially outnumber national ones, which take place far less frequently but have a much greater overall impact. The main examples of these include the numerous referenda on European integration that have taken place across Europe (Hobolt, 2006a), those on Scottish and Welsh devolution (Denver, 2002), abortion referenda in Italy and Ireland (Sinnott, 2002, Uleri, 2002), and constitutional referenda in countries such as Italy, Ireland, and Canada.

6However, not all referenda are the same: Hollander’s recent typology (2019) divides referenda into five types according to two main criteria: who triggers the vote (representatives or citizens); and whether it takes place on the basis of aggregation and majority rule or integration and minority protection. The five types are: referenda triggered by a parliamentary majority (i.e., legislative majority referenda); presidential referenda; referenda triggered by a parliamentary minority (i.e., legislative minority referenda); citizen-initiated referenda; and mandatory referenda. The fifth type—mandatory referenda—are peculiar cases of referenda initiated formally by representatives, who are required to call for a referendum by the constitutions. The 2016 Italian referendum falls within this category, as it had a clear-cut entrepreneurship: the parliamentary majority and the PM at the time, Matteo Renzi (Pasquino & Valbruzzi, 2017). [1]

7Mandatory referenda are much less common than the other types; in Italy there have only been three, plus the one in 1946 that resulted in the monarchy being replaced with a republic. Abrogative referenda (triggered by citizens), on the other hand, are much more common (Uleri, 2002). Due to this, the literature analyzing voting behavior in constitutional referenda is lacking. However, findings on other referenda are an important starting point. Unsurprisingly, the literature has found that several variables impact voters’ choices, most of which are well-known aspects of voting behavior: party identification (Franklin et al., 1994 and 1995), the campaign issues (Svensson, 2002; Christin et al., 2002); left-right positioning; the information environment (De Vreese & Semetko, 2004); political awareness (Hobolt, 2006b); or a mixture of these different variables (LeDuc & Pammet, 1995). However, one of these variables has attracted comparatively more attention: party identification, one of the crucial explananda in the literature on classic voting behavior (Campbell et al., 1960; Converse, 1976). The literature has debated about whether referenda (namely those triggered by representatives) can be considered similar to second-order elections to test the government in power, rather than a vote on the policies at stake (Franklin et al., 1994, 1995). In a nutshell, in a similar way as in general elections, party identification in this perspective is regarded as the most relevant explanatory factor. If this is true, in referenda initiated by parliamentary majorities and minorities, these groups may have a specific interest in strengthening or weakening the government. This may also be true in the case of mandatory referenda: the parliamentary majority, as in this case study, may trigger the constitutional reform, despite knowing that it lacks the minimal requirements to have the reform approved in Parliament (see below), in order to strengthen its position vis-à-vis the main competitors in the opposition.

8When looking at the case of Switzerland, Treschel and Sciarini (1998) found that for mandatory referenda, consensus (or a lack of it), between the elites in the National Council increases (or decreases) the probability of the (de)alignment within the electorate. Svensson (1994, 2002) challenges this position, showing that the second-order hypothesis is at best overstated since voters’ beliefs were quite independent from party allegiances. My first research question is therefore: To what extent did party identification play a role in the 2016 constitutional referendum? And, on the contrary, did voters’ beliefs about the reform play a role in shaping their choice?

9Beyond party identification and voters’ beliefs, another variable is also crucial: the role of the elites, whose importance in structuring voters’ preferences has already been brought to light. This is particularly true for Italy (Barisione, 2006), a country with a lengthy track record of personalizing politics (Campus, 2010), and where leader-led parties have dominated the political scene for three decades. According to LeDuc (2002), party leadership was a key factor in two crucial referenda: that held in France on the Maastricht Treaty, and that held in Spain on NATO membership. The French president at the time, François Mitterrand, and the Spanish PM, Felipe Gonzáles, were at the forefront of their respective referendum campaigns. Personalization may also be a significant variable in second-order elections, however. In presidential or “presidentialized” systems (Poguntke & Webb, 2005), personalization processes, especially those put in motion by the incumbent president or PM, may lead to the referendum being framed as a further assessment of the country’s leader, rather than an electoral competition centered on policy issues. As the Italian referendum campaign was heavily influenced by a personalization component (see below), was this factor just as important as party identification, if not more?

Abrogative and constitutional referenda and the 2016 Italian constitutional referendum

10From the approval of the 1946 Constitution up until 2016, seventeen referendum days were held in Italy. The referenda held on these days dealt with a wide variety of issues, and in twelve out of seventeen cases, more than one referendum was voted on on the same day. During this time, there have also been three constitutional referenda (2001, 2006, and 2016), which were mandatory because the constitutional reforms were approved with a parliamentary majority of less than two-thirds (high and low chamber). Other amendments to the Constitution have been approved by Parliament with a large majority, thus without the need for a confirmatory referendum. [2] In only one of these three constitutional referenda (2001) was the reform was approved, and this was also the only constitutional referendum in which the turnout was lower than 50% (35.8%). In the other two cases, 2006 and 2016, the overall participation was higher than 50% (52.5% and 62.5% respectively), and in both cases the reforms were rejected by almost 60% of the voters (Table 1). According to the Italian Constitution (art. 75), “No referendum may be held on a law regulating taxes, the budget, amnesty or pardon, or a law ratifying an international treaty. […] The referendum shall be considered to have been carried if the majority of those eligible has voted and a majority of valid votes has been achieved” (Senato della Repubblica, 1947, p. 20.)

11Overall, the number of referendum days held in Italy has increased substantially over the last few decades: there were only two in the 1970s and three in the 1980s, but six in both the 1990s and the 2000s. More importantly, what has been dubbed the “era of referenda”—the nineties, in which thirty-two referenda took place—was followed by two equally referenda-heavy decades: between 2000 and 2016, there were twenty-one referenda, plus three constitutional referenda. Although the number of referendum days has not decreased since the nineties, what has changed is the overall turnout. Before 2011, the last referenda in which the turnout threshold was reached was in 1995. Between 1995 and 2011, twenty-four referenda (across seven referendum days) failed to reach the threshold requirement. Only in the 2006 constitutional referendum, in which a minimum threshold was not required—more than a half of the electorate (52.46%) participated. The last abrogative referendum, which failed to reach the minimum threshold, took place a few months before the constitutional referendum, in April 2016.

12The 2016 constitutional reform was approved by Parliament in April 2016. Members of both the majority and of the opposition supported the call for a referendum, and despite being mandatory (since it failed to reach the required majority in Parliament), it had the clear-cut sponsorship of the government, which promoted the reform in the first place. The reform comprised an unprecedented number of articles (47) and titles (5). Yet, some of them were only marginally amended as a consequence of the amendment of other articles (Gazzetta Ufficiale, n.88, 2016). According to Ceccarini and Bordignon (2017), the reform had three main themes: the reorganization of the territorial structure of the state; the abolition of symmetric bicameralism (the limitation of the prerogatives held by the Senate); and the relationship between the executive and the legislative power. The Senate would have been transformed into a chamber representing the regions, and the number of senators would have been reduced to 100. The reform also aimed to abolish the provinces and the Consiglio nazionale dell’economia e del lavoro (National Council for Economics and Labour, CNEL), which was targeted by YES supporters as being symbolic of how public funding was wasted on non-productive research centers. Other important reforms concerned the requirement for popular initiative laws to be presented in Parliament, the requirement for signatures to be collected in the case of both advisory and abrogative referenda, and the electoral laws for the Chamber of Deputies and the Senate.

The political background of the 2016 Italian constitutional referendum (2014-2016)

13The aftermath of the Great Recession (November 2011) saw the resignation of the PM at the time, Silvio Berlusconi. This paved the way for a grand coalition between the two mainstream parties (Partito Democratico, PD, and Forza Italia, FI), which supported a technocratic government led by the economist Mario Monti (2011-2013). The outcome of the 2013 election was a multiparty government led by Enrico Letta (PD) (2013-2014). When Letta stepped back and the new PD secretary, Matteo Renzi, became the new PM, the grand coalition split due to the withdrawal of part of FI. However, FI’s split, Nuovo Centro Destra (New Center-Right), continued to support the Renzi government.

14Once elected as the PD’s secretary in December 2013, Renzi started his fight for the party’s ideological renovation. He embodied a post-ideological and outsider-styled leadership, whose core was represented by the concepts of “efficiency, rapidity, and merit” (Bordignon, 2014, p. 7). Between December 2013 and February 2014, Renzi decided to start cooperating with Berlusconi, as a way to isolate Movimento 5 Stelle (M5S), the successful anti-establishment party led by the well-known comedian Beppe Grillo. In a famous meeting at the PD’s headquarters in January 2014, the two leaders agreed on a reform of electoral law and of the Constitution. This agreement lasted until the election of the new President of the Republic, Sergio Mattarella, a jurist and former Christian-Democratic MP. Berlusconi criticized the unilateral method adopted by Renzi in this choice and, consequently, decided to dismantle the agreement. The 2014 European elections were a personal success for Renzi: the PD achieved its best electoral result (40.8%) (Figure 1), while Berlusconi’s and Grillo’s parties suffered relative setbacks. Before the constitutional referendum in 2016, Renzi’s government passed controversial reforms, the most of significant of which was the labor market reform, renamed the Jobs Act, which aimed to reduce protection from dismissal for workers on new open-ended contracts by abolishing Article 18 of the Workers’ Statute, [3] while at the same time reforming the public employment agencies and improving unemployment benefits. Renzi’s government also had to deal with the 2015 migrant crisis: in the space of a year, almost one million people attempted to cross the Mediterranean Sea to reach border countries such as Italy, Greece, and Malta. The migrant crisis immediately became a salient issue in the Italian public debate: the European-led Operation Triton, which replaced the Italian-led Operation Mare Nostrum, proved to be ineffective both in saving lives and stopping the migrant crisis. Another issue of utmost importance during Renzi’s mandate was Italy’s relationship with the European Union (EU): while presenting a marked pro-European attitude, Renzi denounced Brussels’ hawkish attitude toward Italian budgetary law.

Figure 1

Elections in Italy (legislative and European) – 2009-2018

Elections in Italy (legislative and European) – 2009-2018

The electoral campaign and the referendum result

15In April 2016, after several modifications, the final text of the constitutional reform was approved by the parties of the majority, while all opposition parties left Parliament during the voting operation. The elites soon polarized, for the most part following party allegiances.

16The Yes camp, called Basta un Sì (Just a Yes), was represented by the PD and its leader, Renzi. The main governing parties endorsed the Yes vote, including FI’s split Nuovo Centro Destra, and the centrist Scelta Civica (Civic Choice, SC), whose former leader, Mario Monti (Italian PM from 2011 to 2013) in fact voted No. A tiny yet relevant parliamentary party, Alleanza Liberalpopolare (Liberal Popular Alliance, ALA), headed by former FI MP Dennis Verdini, supported the Yes camp, as did other minor liberal parties. In any case, the burden of the campaign was mainly on the PD and, particularly, on Renzi and Maria Elena Boschi, the promoter of the reform along with Renzi. Both politicians invested all their political credibility into the referendum: in their view, a Yes vote was a vote for changing and modernizing Italy. The Yes camp tried to impose its agenda, which was focused on the widespread anti-politics sentiment that had grown exponentially after the unexpected success of M5S in the 2013 elections (Baldini, 2013). In this sense, the question in the paper ballot exemplified what the government considered to be the main cornerstone of the reform: reducing the operating costs of government institutions. [4] According to Pasquino and Valbruzzi (2017, p.153), Renzi had two goals: first, a greater legitimization for the government he was leading, and, second, to show that, despite the criticism of the opposition, he had “the support of a large majority of the Italian people.” Ceccarini and Bordignon (2017, p. 289) share the same view: “It was Renzi himself who personalised the vote, identifying a ‘yes’ to the reform with support for his premiership.” According to the two scholars, Renzi transformed the constitutional reform into a vote of confidence for his government.

17On the other hand, the positions in the No camp were much more varied. The No vote was endorsed by all parties in opposition to the Renzi government: despite expressing different motivations behind their choice, M5S, FI, Lega, Fratelli d’Italia (Brothers of Italy, Fd’I), Unione di Centro (Union of the Center, UDC), radical-left parties, and the Federazione dei Verdi (Federation of the Greens, FdV) all opposed the constitutional reform. A minority faction within the PD also supported the No vote. Two committees, Committee for No and I Vote No, were created to rally for the No vote. However, both were headed by professors and jurists, while the parties campaigned independently. In particular, M5S organized a tour around Italy to explain its No position. M5S’s position on the referendum was confrontational: it accused Renzi of attempting a coup d’état, because of the alleged excessive power that the new Constitution and the new electoral law would grant him. Another controversial point that M5S raised during the electoral campaign was that the reform would give indirectly elected senators parliamentary immunity, and thus enable them to escape trials.

18Personalization was not only present in the Yes camp’s electoral campaign, but also in that of the No camp, according to Ceccarini and Bordignon (2017). Renzi used the heterogeneity of the No camp to highlight its internal divisions and its hidden anti-Renzi platform.

19According to the polls, the personalization strategy did not work properly: although before the approval of the reform—and, thus, before, its politicization—a clear-cut majority (66%) endorsed it, as soon as it became a matter of political contention and parties took their positions, public opinion realigned in favor of the No camp. In particular, while the majority of M5S, FI, and Lega supported the content of the reform before its approval, before the referendum, only 15% of Lega voters, 14% of M5S voters, and 34% of FI voters endorsed the Yes vote (Ceccarini & Bordignon, 2017). In October, two months before the referendum, the Yes camp was minoritarian in the electorate according to the polls (Figure 2). It never recovered from then on.

Figure 2

Pre-referendum polls

Pre-referendum polls

20The result of the referendum was (partially) disappointing for Renzi: 60% of the electorate rejected the reform (with turnout at 68.5%). Soon after the results, he resigned as PM, while staying in charge as PD secretary. The No vote won in all regions apart from Emilia-Romagna, Tuscany, Trentino-Alto Adige, and among Italians voting from abroad, thus suggesting a clear-cut national trend with the partial exception of part of the so-called Red Belt, those regions in which the Partito Comunista Italiano (Italian Communist Party, PCI) and its heirs usually prevailed in elections, sometimes with a landslide (Emilia-Romagna, Tuscany, Umbria, Marche) (Figure 3).

Figure 3

Results of the 2016 constitutional referendum: Italy, Italian regions, and Italians voting from abroad

Results of the 2016 constitutional referendum: Italy, Italian regions, and Italians voting from abroad

21The referendum had long-lasting consequences. Once Renzi resigned, Paolo Gentiloni, a PD MP and former minister, became PM. The majority supporting the new government remained the same. The 2018 general election saw success for M5S (32.7%), along with unexpectedly good results for Lega (17.4%). On the other hand, the PD (18.8%) and FI (14%) were the main losers.

Data, hypothesis, and methodology

22The data on the Italian referendum are taken from ITANES’s pre- and post-referendum surveys. The questions both of these surveys asked were the same, the only exceptions being the perspective voting question in the pre-referendum survey and the retrospective voting question in the post-referendum survey. The questions were on sociography, policy, leadership, and the referendum, as well as parties, the quality of democracy, and prospective votes. The post-referendum data approximate fairly well with the actual result of the referendum, when excluding the respondents who claim to have not voted. This category is underrepresented—9.3% of respondents claim they did not vote, although the actual abstention rate was 34.5%—while the other three categories—Yes Vote, No Vote, and Blank Vote [5]—are fairly representative of the referendum results: the No:Yes ratio is 61:39 while in the referendum it was 59.1:40.9. The pre-referendum survey, on the other hand, overestimates the number of Yes voters. For this reason, the best way to approximate the results is to focus on the post-referendum surveys, taking into account the respondents who had already decided which way to vote in the referendum.

23Drawing on the theoretical framework detailed in the first part of this article, and taking into account the contextualization on the specific nature of the Italian case explained above, I will now test three hypotheses to determine what were the determinants of the vote in the referendum and, which of them is best fitted to explain the result.

24The first two contrast the “Svensson” hypothesis on the importance of beliefs, and the “Franklin” hypothesis on government performance. As for the first, it has been noted that the referendum concerned several amendments to the Constitution, some of which involved very sensitive issues that were raised both by the Yes and the No camps about the reduction of the number of senators, the abolition of Italy’s symmetric bicameralism, and the state’s acquisition of new competences, which were currently co-managed by the state and the regions. If the content of the reform and voters’ positions on those issues were the main determinants of the vote, then it could be expected that

• a) approval of all the content of the reform explains the Yes vote, and opposition to the content of the reform explains the No vote.

25On the other hand, as the Yes camp and the No camp were divided along party lines (with the PD and other parties who supported the government in the Yes camp, and the opposition parties in the No camp, with a few exceptions on both sides), it may be expected that the polarization of the elite would bring party identification to the forefront, to the point that, following the Franklin hypothesis, this variable was the main determinant of voters’ choices. Thus,

• b) party identification was the main explanatory factor determining both the Yes and the No choices.

26However, not only was the government the main trigger of the referendum (although strictly speaking it was a mandatory referendum), but in the years preceding the referendum, the Renzi government passed several controversial bills, including the labor market reform (the Jobs Act) and the school act (La Buona Scuola [The Good School]), while promising to reverse the power balance within the EU on several issues, such as the EU budget and immigration. Thus, a corollary of the second hypotheses is that, beyond party identification, this referendum was a second-order election to test government approval. In particular, the main policy issues on which the government had acted in previous years (the EU, immigration, and the labor market) were a good predictor of the vote. The corollary of the second hypotheses is that

• b.1) approval of the most relevant issues that Renzi’s government dealt with before the referendum is a predictor of the Yes vote.

27Finally, because the referendum campaign was characterized by strong personalization and because this variable can be used to explain the outcome of other referenda, I hypothesize that a negative/positive judgment of the main leaders in government can explain the outcome of referenda votes. This third hypothesis is not in contrast with the “Franklin hypothesis” on the relevance of government performance: admittedly, personalization—particularly in a case in which the government initiated the reform process—may be just another way for the incumbent elite to test its approval. However, the data at our disposal does not enable us to determine the nature of the personalization process, i.e., whether it is a heuristic shortcut that voters use to disentangle the position of the government and of the opposition or, on the contrary, whether it is indicative of the real trust that voters have in party leaders. Be as it may, if personalization does occur, I expect that it has a higher explanatory power for the party leader who is in the spotlight the most during the electoral campaign; in this case, for Renzi, as he was the main protagonist of the constitutional reform bill. On the other hand, I hypothesize that, while important, the overall evaluation of other leaders—Silvio Berlusconi (FI), Matteo Salvini (Lega) and Beppe Grillo (M5S) [6]—is less relevant than that of Renzi, since according to the literature, the focal point of the referendum campaign was the PM, rather than the opposition leaders. Thus, my personalization hypothesis is the following:

• c) positive judgment of the Yes leader—PM Matteo Renzi—explains the Yes vote. Positive judgment of No leaders, while linked to the No vote, has a lower explanatory power.

28I ran seven separate logistic regression models, one for each area on its own (policy issues, reform content, and leader evaluation) and four that combined these three areas, in order to check the robustness of the evidence and to test which combination has a greater explanatory power. In each model, I controlled for the following sociographic variables: age, gender, education, income, and region of residence. For this last variable, I opted for a different operationalization that distinguished Red Belt regions from northern, central, and southern ones, while other variables’ re-scaling operations were marginal. Since the so-called Red Belt is the PD’s core in terms of electorate, I expected that living in these regions, which were the ones in which Yes had better results, increased the likelihood of voting Yes.

Results

29The first model examined the impact of the electorate’s policy preferences in the referendum vote (Table 2). As was the case for other models (see below), party identification had a significant impact on the electorate’s choices. What emerged was that the divisiveness created by the labor market reform—and Renzi’s personalization of it—had a significant impact on the electorate’s choices. Among the policy-related variables, a positive evaluation of the Jobs Act was strongly associated with voting Yes in the referendum, while views on immigration and on the EU and the Euro had a lower degree of significance. In line with my expectations, a negative evaluation of the economy favored the No vote in the referendum. Control variables were not significant. Thus, both party identification and the evaluation of government performance proved to be relevant in the first model, providing a first support to the Franklin hypothesis.

Table 2

Regression models – single issue (policy issue, reform content, or leader evaluation)

Policy issue Reform content Leader evaluation (Intercept) 0.518 (0.758) -2.491*** (0.650) -1.372* (0.612) Gender: Female -0.352* (0.178) -0.331 (0.176) -0.307 (0.181) Age: Adult 0.306 (0.253) 0.109 (0.247) 0.299 (0.246) Age: Elderly 0.435 (0.276) 0.058 (0.269) 0.432 (0.269) Region: North 0.278 (0.233) 0.465* (0.236) 0.160 (0.243) Center -0.218 (0.317) 0.137 (0.315) -0.206 (0.326) South -0.426 (0.260) -0.543* (0.258) -0.508 (0.273) Income: Middle class 0.175 (0.237) 0.451 (0.238) 0.180 (0.237) High income 0.054 (0.236) 0.294 (0.238) -0.085 (0.240) Education: College 0.016 (0.321) 0.095 (0.309) 0.141 (0.327) University degree -0.091 (0.338) -0.088 (0.323) -0.004 (0.345) Employed: No 0.037 (0.242) 0.146 (0.231) 0.066 (0.233) Party_ID: Movimento 5 Stelle -2.306*** (0.280) -2.856*** (0.278) -0.996** (0.306) Center-right parties -2.350*** (0.275) -2.847*** (0.264) -0.708* (0.326) Centrist parties -1.215 (0.630) -0.988 (0.564) -0.158 (0.660) Radical-left parties -1.385*** (0.367) -1.729*** (0.367) -1.566*** (0.362) Others -2.330*** (0.627) -2.538*** (0.624) -2.190*** (0.626) None -1.119*** (0.247) -1.383*** (0.243) -0.584* (0.256) Profession: Blue-collar/Low-skilled -0.042 (0.449) -0.367 (0.429) 0.141 (0.435) Self-employed 0.313 (0.369) 0.116 (0.366) 0.067 (0.376) Salaried 0.218 (0.333) -0.030 (0.330) -0.003 (0.336) Immigration -0.100* (0.040) Economy -0.451*** (0.129) Jobs Act 0.479*** (0.042) EU: Bad -0.598* (0.247) Neither good nor bad -0.528* (0.208) Senate -0.028 (0.047) Bicameralism 0.438*** (0.044) Centralization 0.179*** (0.035) Renzi 0.605*** (0.040) Berlusconi -0.136** (0.044) Salvini -0.063 (0.043) Grillo -0.216*** (0.038) Log-likelihood -465.455 -468.854 -450.547 AIC 982.9 969.6 954.3 BIC 1116.604 1093.57 1084.09 McFadden 0.7435 954.3 0.7461 0.7509 N 1264 1312 1326

Regression models – single issue (policy issue, reform content, or leader evaluation)

30The second was the reform-content-related model, which tested the Svensson hypothesis. As may have been expected, among the control variables, party identification still played a relevant role, with being a PD supporter a predictor of voting Yes, compared to being a supporter of the opposition parties. Unsurprisingly, the difference with the supporters of centrist parties was not significant, since most of them supported the constitutional reform. However, beyond party identification, the second model shows that it matters what voters think about the content of the reform: a positive attitude toward a) the abolition of symmetric bicameralism and b) the centralization of competences was strongly associated with voting Yes in the referendum. On the other hand, voters’ attitude toward the reduction in the number of senators (var. Senate) was not significant, as this variable’s explanatory power may have been absorbed by the other content-related variables. An alternative explanation may be that this issue was evaluated positively by No voters too, since it was associated with a shared feeling among the electorate, i.e., dissatisfaction with the political class.

31The third model included voters’ evaluation of the leader. Of the three models, this is the one with the highest explanatory power. Even when controlling for party identification, how voters evaluated leaders mattered. Yet, a positive/negative evaluation of Renzi was not the only significant variable. That of the other leaders—Grillo and Berlusconi—was equally important. Personalization therefore works both ways: on the one hand, a positive evaluation of Renzi is associated with the Yes vote, while on the other, a positive evaluation of Grillo and Berlusconi is associated with the No vote.

32Nonetheless, the impact of the evaluation of Renzi is the highest among the leaders, meaning that Renzi supporters were highly mobilized in voting Yes. Since one of the aims of Renzi’s campaign was to intercept the moderate vote of the center-right parties, particularly Berlusconi voters, it seems that Berlusconi’s endorsement of the No vote did not help Renzi.

33Alongside these three separate models, which initially tested the three hypotheses separately, I created four other models that tested the strength of the variables when combined. Among these four different logistic models that combined the policy issues, the content of the reform, and the evaluation of the leader, those with the highest fits were the two that included the content of the reform (policy issue and reform content; reform content and leader evaluation). However, it is worth noting that at least one of the two reform-content-related variables—centralizing competences by shifting them from the regional to the national level, and reducing the number of senators—either decreased or maintained the absence of significance in the model combining all of the three previous models. On the other hand, not only did voters’ evaluations of the leaders—namely Renzi and Grillo—retain their significance in the last model, but also their inclusion in the model, as may be expected, reduced the relevance of party identification.

34Among the different leaders under analysis in the fourth model, Renzi’s evaluation is the most important driver for voting Yes. However, a content-related variable (bicameralism), is a powerful explanatory factor. Equally important, a policy-issue-related variable (the evaluation of the labor market reform [the Jobs Act]) is highly significant, although the strong correlation between the evaluation of the Jobs Act and the evaluation of Renzi (.78) suggests that caution should be taken before proposing a purely policy-issue-related explanation of the vote. The weak correlation between the Jobs Act and the evaluation of both Grillo (-.27) and Berlusconi (-.005) and the considerable explanatory power that the latter have indicates that while the Yes vote can be explained by a positive evaluation of Renzi’s policy-making (and its personalization), the No vote has a much deeper explanation. In particular— and this may appear striking for a constitutional referendum—, immigration is a significant predictor of the No vote in all three models in which the variable is included, even though only in the model combining policy issues and leader evaluation does it have a significance of 0.01.

35In terms of explanatory power, the last model, which combined policy issue, reform content, and leader evaluation, added few clues to the results produced by the model combining reform content and leader evaluation (Table 3). Although it increased the fit, the overall pattern was not significantly modified: the Yes vote can be explained through a combination of a positive evaluation of Renzi and the policies his government implemented and of the content of constitutional reform, while the No vote is associated with a positive evaluation of the No leaders and, in contrast with the previous pattern, with a negative evaluation of immigration (Figure 4). More importantly, in the last model, party identification loses part of its explanatory power.

Table 3

Regression models – Combination of policy issue, reform content, and leader evaluation

Policy issue and reform content Policy issue and leader evaluation Reform content and leader evaluation Policy issue, reform content, and leader evaluation (Intercept) -1.898* (0.951) -0.751 (0.879) -3.003*** (0.757) -2.238* (1.044) Gender: Female -0.344 (0.195) -0.395* (0.199) -0.284 (0.201) -0.354 (0.216) Region: North 0.420 (0.260) 0.099 (0.266) 0.184 (0.273) 0.138 (0.287) Center 0.117 (0.356) -0.332 (0.370) 0.127 (0.365) -0.063 (0.399) South -0.517 (0.287) -0.431 (0.294) -0.667* (0.304) -0.612 (0.318) Age: Adult 0.256 (0.276) 0.370 (0.276) 0.121 (0.274) 0.214 (0.301) Elderly 0.168 (0.302) 0.545 (0.303) 0.093 (0.303) 0.166 (0.333) Income: Middle class 0.302 (0.269) 0.065 (0.261) 0.403 (0.273) 0.204 (0.296) Low income -0.004 (0.268) -0.199 (0.263) -0.121 (0.273) -0.350 (0.295) Education: College 0.085 (0.351) 0.029 (0.358) -0.026 (0.362) -0.033 (0.380) University degree -0.207 (0.370) -0.133 (0.377) -0.377 (0.382) -0.390 (0.401) Employed: No 0.187 (0.263) 0.048 (0.263) 0.118 (0.259) 0.162 (0.284) Party_ID: Movimento 5 Stelle -2.247*** (0.312) -0.888** (0.341) -1.019** (0.350) -0.918* (0.381) Center-right parties -2.599*** (0.303) -0.867* (0.358) -0.828* (0.363) -0.972* (0.390) Centrist parties -1.433* (0.660) -0.940 (0.735) -0.743 (0.669) -1.282 (0.760) Radical-left parties -0.991* (0.424) -0.853* (0.407) -1.111** (0.422) -0.516 (0.465) Others -2.102** (0.725) -2.069** (0.732) -1.973** (0.757) -1.815* (0.830) None -1.048*** (0.276) -0.572* (0.283) -0.553 (0.292) -0.540 (0.314) Profession: Blue-collar/Low-skilled -0.328 (0.497) 0.196 (0.487) -0.180 (0.496) -0.084 (0.527) Self-employed 0.375 (0.416) 0.161 (0.417) 0.163 (0.436) 0.304 (0.459) Salaried 0.154 (0.373) 0.178 (0.372) -0.061 (0.387) 0.104 (0.406) Immigration -0.104* (0.045) -0.181*** (0.047) -0.154** (0.051) Economy -0.358* (0.146) -0.245 (0.144) -0.218 (0.160) Jobs Act 0.340*** (0.047) 0.356*** (0.050) 0.273*** (0.055) EU: Bad -0.378 (0.270) -0.113 (0.281) 0.074 (0.305) EU: Neither good nor bad -0.473* (0.232) -0.253 (0.233) -0.201 (0.256) Senate -0.016 (0.053) -0.035 (0.054) -0.046 (0.058) Bicameralism 0.378*** (0.049) 0.422*** (0.053) 0.401*** (0.056) Centralization 0.104* (0.042) 0.108* (0.043) 0.087 (0.047) Renzi 0.469*** (0.053) 0.436*** (0.044) 0.368*** (0.056) Berlusconi -0.153** (0.048) -0.146** (0.047) -0.168*** (0.051) Salvini -0.049 (0.048) -0.080 (0.047) -0.069 (0.051) Grillo -0.209*** (0.041) -0.198*** (0.042) -0.196*** (0.045) Log-likelihood -385.155 -383.261 -364.375 -323.982 AIC 824.4 833.3 781.6 718.8 BIC 972.53 987.45 925.85 886.92 McFadden 0.7889 0.7869 0.8001 0.8202 N 1235 1243 1289 1217

Regression models – Combination of policy issue, reform content, and leader evaluation

Figure 4

Main positive and negative effects on voting behavior in the policy issue, reform content, and leader evaluation model

Main positive and negative effects on voting behavior in the policy issue, reform content, and leader evaluation model

Discussion

36Returning to my three different hypotheses, the results are mixed. When it comes to party identification, the first three models prove that this variable is of crucial importance: in each of them, its impact is the highest among the variables under analysis. Thus, the Franklin hypothesis holds its relevance when policy issues, reform content, and leader evaluation are analyzed separately. However, the reform content model has a greater explanatory variable compared to the policy issue model, although the leader evaluation model has the highest explanatory power of the three, thus providing support for hypothesis three. The support for hypothesis three is confirmed when analyzing the last four models; a positive evaluation of Renzi is a significant predictor of voting Yes, even when controlling for party identification. Similarly, positive evaluations of Berlusconi and Grillo are associated with a higher likelihood of voting No in the referendum. On the other hand, party identification, contrary to previous findings, loses its importance when the first three models are combined. Rather than party identification in itself, the positive/negative evaluation of the Renzi-sponsored policies (the Jobs Act and his immigration policies) increase the likelihood of a Yes/No vote in the referendum. The corollary of the first hypothesis (b.1) proves to be more accurate than hypothesis b on party identification. Despite this, the policy issues model has the lowest explanatory power among the first three models, and the model combining policy issues and leader evaluation has the lowest explanatory power among the last four models. And what about the reform content? Even though the reform content model alone is less powerful than the leader evaluation model, the picture is reversed when looking at the three combinations: in fact, the ones that included variables related to the reform content have the greatest explanatory power. In particular, the most relevant reform, i.e., the one aimed at abolishing symmetric bicameralism, is significant in all models, even when combined with leader evaluation and policy issues (Figure 4). Yes voters did therefore approve of the reform content: beyond the personalization effect, they took into consideration the proposed changes to the Constitution. Finally, the combination of reform content and personalization provide an almost equally well-fitted model compared to the model with all three areas combined, and a much better model compared to the leader evaluation model.

37To sum up, and in order to assess the four hypotheses I put forward, hypothesis b (Franklin thesis) is only partially confirmed: party identification is important when paired with policy issues, reform content, and leader evaluation on their own. However, it loses its significance when combined with more than one single aspect. Accordingly, while significant, policy issues are less relevant in explaining voters’ choices, be they isolated or in combination with reform content and leader evaluation. All in all, the electorate was less prone to using this as a heuristic shortcut to vote Yes in the referendum. In partial contrast with this trend, the importance of the Jobs Act—a highly personalized reform promoted by Renzi—proves to be significant. However, the extent to which it is possible to isolate the content of the labor market reform to its staunch promoter is difficult to assess at this stage.

38On the other hand, hypothesis c is confirmed: approval of Renzi is the most powerful predictor of the Yes vote. Accordingly, the No vote is explained by two political leaders who endorsed the No vote: Grillo and Berlusconi. Personalization was therefore not unidirectional, although approval of the No leaders does have a lower explanatory power. Personalization on its own mattered more than reform content and policy issues on their own.

39At this stage, it was not possible to evaluate whether the personalization of the referendum can also be considered a clear-cut (negative) evaluation of the government’s performance. The literature on the Italian referendum seems to show that this is the case, thus complementing the Franklin thesis. The findings suggest that personalization played a role for Yes voters, who approved of both the policies promoted by the Renzi government and the content of the constitutional reform. However, beyond personalization, the content of the reform was also crucial for the Yes camp, thus providing partial support for hypothesis a as well. Reform content was not as powerful a predictor as leader evaluation, but it did matter, at least more than policy issues. Voters knew what they were voting for and the significance of one of the crucial aspects of the reform, bicameralism, testifies to this fact.

40Was it all about personalization, then? No, but personalization, all other variables being equal, played the most important role in this referendum. Equally importantly, the answer to this question has contributed to the ongoing debate on voting behavior in referenda: voters’ evaluation of leaders matters, but when crucial aspects of the political institutions are at stake, the content of the reform is just as important.