Negativity in the news and electoral behavior between first- and second-order elections

Abstract The effects of news coverage on political attitudes in election campaigns have been widely studied in academic research. In particular, a fertile branch of the literature investigated the impact of news media negativity on turnout. To date, however, findings are mixed, precluding to state a clear relationship. This paper aims to shed a light on this topic by testing whether negative coverage may affect voters' turnout and to what extent. It approaches this research question by accounting for two different dimensions, controlling whether the interplay of media negativity (press and TV coverage) with the type of election (first- vs. second-order elections) has an impact on citizens' propensity to turnout. We test our hypotheses by taking Italy as a case study because it offers a combination of systemic and media characteristics that allows addressing properly the topic. We rely on four datasets covering the 2018 Italian general elections and the 2019 European Parliament Elections in Italy, respectively, with opinion data (2018 ITANES survey and 2019 ITANES-University of Milan survey) and two datasets measuring media negativity by means of a human content analysis carried out on media coverage during the 7 weeks before the election days (2018 and 2019 ITEM data). Our findings show that individual exposure to negative coverage leads to an increase in turnout mainly to the detriment of indecision. This pattern holds both in first- and second-order elections.


Introduction
The impact that mass media messages have on political attitudes and behavior in election campaigns is a widely debated topic in academic research. Starting from the late 1980s, the literature has investigated the impact of news media negativity on citizens' beliefs and behaviorssuch as, for instance, turnout (e.g. Ansolabehere and Iyengar, 1995), voting behavior (Kahn and Kenney, 1999), knowledge (De Vreese and Boomgaarden, 2006), perceived salience of certain issues (Roberts et al., 2002) producing mixed evidence. In this paper, we aim at assessing whether and to what extent negative coverage may harm or boost citizens' propensity to cast their ballot. Also, we are interested in understanding whether the impact of negativity of media coverage on voters' turnout differs in first-or second-order elections.
The literature in this field is extensive, but far from being settled. Several scholars argue that negativity has a de-mobilizing effect on citizens' propensity to vote: negative campaignsnamely, a political debate characterized by mutual attacks, conflict, and criticism among candidatesare most likely to alienate citizens (Ansolabehere et al., 1994;Ansolabehere and Iyengar, 1995). By increasing dissatisfaction with politics and cynicism, negativity keeps people far from the polling stations. However, other studies contend the opposite, suggesting that negative messages contribute to stimulating citizens' attention and political interest, turning to be a boost for political participation (Finkel and Geer, 1998;Wattenberg and Brians, 1999;Goldstein and Freedman, 2002).
When the attention moves specifically on media and negative coverage, the picture is even more blurred. Indeed, scholars agree that negativity is becoming a common standard and routinized trait of journalism (West, 2009), to the extent that media coverage is even more negative than party communication (Geer, 2006(Geer, , 2012Pedersen, 2014). Contemporary media outlets are keen to offer higher visibility to negative campaigning: not just by providing visibility to negative contents and messages derived from political actors, but also by adopting a negative tone within the coverage (Jamieson, 1992). This pattern mirrors precisely the media logic, which puts special emphasis on sensationalism, conflict, and a negative frame. Politicians are aware of this dynamic: given that candidates seek higher visibility within media coverage, they tend to engage in negative campaigning because they know that then media will be more likely to cover them (Hansen and Pedersen, 2008;Walter and Vliegenthart, 2010;Haselmayer et al., 2019).
However, Min (2004) points out that, for a better understanding of media coverage impact, additional individual and contextual features have to be taken into account, suggesting that the type of connection between negativity and political attitudes is less direct and obvious than expected. Media coverage could be received differently by the audience, depending on their traits and on the type of political context in which they are embedded. This latter argument leads us to consider the impact of media negativity on turnout from a wider perspective, considering the political supply and, in particular, the electoral context in which the individual decision process is carried out.
In this respect, the literature raises a very clear point: the elections are not all equally relevant from the voters' perspective. In particular, the paradigm of the second-order elections (Reif and Schmitt, 1980) argues that European Parliament Elections are less prominent if compared to the national elections, where voters understand more clearly how their vote will affect the future government of the country. Instead, European Parliamentary Elections are intended as elections where there is 'less at stake'. This is mainly due to the institutional framework and the functioning mechanisms of the decisional power of the European Parliament. The entire electoral process appears ambiguous and citizens feel to not have a say on the EU Commission formation (Decker, 2002), perceived as a distant institution. However, it should be noticed that, although the recent recession has brought the EU at the center of the domestic political agenda, EU issues still appear as domesticated, namely depicted (and addressed) according to a national perspective (Kriesi, 2016). Similarly, although studies on the Europeanization of national public debates (Koopmans and Erbe, 2004;Koopmans and Statham, 2010) showed that media tend to offer an increasing and larger visibility to EU and related issues (Pfetsch and Eschner, 2008), the EP elections still maintain specific traits in terms of voting behavior and lower turnout rates (Franklin and Hobolt, 2015). In short, European Parliamentary elections fail to mobilize citizens' interest and do not incentivize their participation in the vote.
In light of these arguments, in this article, we test the effects of negative media coverage on voters' propensity to turnout in different settings. On the one hand, we consider whether negative media coverage per se affects individuals, and in which way. On the other hand, we test whether negative media coverage may impact differently in different electoral contexts, namely distinguishing between national and EP elections. The aim of the paper is therefore twofold: (i) assessing the effect that negative political coverage might have on the propensity to turnout in a context, the Italian one, characterized by a strong presence of negative campaigning; (ii) testing whether the electoral setting (first-order elections vis-à-vis second-order elections) might alter the magnitude and substantive significance of these effects.
To test our research questions, this study relies on four different datasets. Two of them are based on individual data measuring our dependent variable (turnout) as well as individual control variables and media exposure. Individual data were collected via CAWI mode in the election campaigns that preceded the 2018 Italian General Elections (2018 ITANES survey) and the 2019 European Parliament Elections in Italy (2019 ITANES-University of Milan Survey). The other two datasets provide information about the media coverage of the 2018 and 2019 election campaigns carried out through a human content analysis of the political contents published or broadcasted by the main Italian daily newspapers and evening newscasts during the 7 weeks before the vote (2018 and 2019 ITEM project -ITalian Election on the Media). We derived here our main independent variable, namely the negativity of coverage tone. Since each individual-level dataset reported information concerning the media exposure, we have thus been able to build a merged datasetproviding, for each election year, both individual and media coverage information.
Our findings suggest a positive impact of negative coverage on turnout on the whole audience, especially at the detriment of pre-electoral uncertainty. Negativity exposure, on the other side, does not affect people who are decided not to vote in the upcoming election. For what concerns the effect driven by the electoral context (firstvis-à-vis second-order election), we do not find any impactnamely, the effects that we see in the first-order election are substantively equal to those seen in the second-order election. Indeed, the effects of media negativity on undecided voters are similar in both first-and second-order elections.
The paper is organized as follows: the next section provides literature background and illustrates the hypotheses tested within the study, the following section presents methodological details concerning datasets, variables, and models employed, while the fourth section describes results. The final section discusses the findings, outline the limits, and refine the theoretical contribution that the work brings to existing literature.

Background
Over the past decades, political communication research has been committed to understanding and testing the presence of the so-called 'media effects', namely, the way in which mass-media-conveyed information is able to change people behavior and attitudes (Kahn and Kenney, 1999;Slater, 2004;De Vreese and Boomgaarden, 2006). After a long phase driven by the Klapper's (1960) claim that the effects of mass communication were not relevant, in the last two decades scholars argued the opposite, by providing some evidence that media effects were 'far from minimal' (Iyengar and Simon, 2000). From a psychological point of view, exposure to mass media might influence people's political ideas through processes of learning (the acquisition of new information), which, in turn, are potentially able to (partly or completely) change one's system of beliefs (Lee, 2015). An extensive political communication literature has demonstrated that media effects may affect a variety of attitudes and behaviors. For instance, De Vreese and Boomgaarden (2006) show that, consistently with learning mechanisms, exposure to news outlets presenting a larger amount of political content contributes to increasing political knowledge. Also, the ways in which politicians and parties are (positively or negatively) covered in the mass media might shape the audience's attitudes and behaviors (namely, voting) toward various political objects (see Ansolabehere and Iyengar, 1995;Hillygus and Jackman, 2003;Hopmann et al., 2010).

Media effects and the relationship between negativity and turnout
The academic debate aiming at connecting media effects and participation is particularly extensive Ansolabehere et al., 1994;Ansolabehere and Iyengar, 1995;Germond and Witcover, 1996;Finkel and Geer, 1998;Kahn and Kenney, 1999;Min, 2004), and it has been mainly committed to assessing whether people participation might be increased/decreased by negative political advertising, a topic that is particularly relevant in the US literature. Likewise, the relationship between negativity aired by mass media (through advertising or news coverage) and turnout has represented a leitmotif of the academic debate since the beginning of the 1990s (Ansolabehere and Iyengar, 1995;Finkel and Geer, 1998). With this respect, several scholars argue that candidates and parties engaging in mudslinging may alienate citizens from the electoral process, with the consequence of depressing turnout (see Ansolabehere and Iyengar, 1995;Germond and Witcover, 1996). In particular, these studies contend that being exposed to advertising that negatively addresses or attacks political competitors suppresses turnout. Additional literature, moreover, clarifies that negative political debate, incivility, and conflict contribute to depress citizen's engagement in politics and contribute to increasing distrust (Mutz and Reeves, 2005). People dislike negative campaigning, they declare to prefer positive discussion of the issues, rather than attacks and mudslinging strategies. Therefore, it seems reasonable to assume that negativity will hurt citizens' turnout, contributing to consolidating their negative feelings concerning politicians and politics.
However, research proves that 'people think is often at odds with how they act' (Brooks, 2006: 684). As concerns negativity in election campaigning, for example, we know that negative information tends to be fixed in memory longer than positive information and tends to offer more substantive information than positive ones Bradley et al., 2007). This may contribute to developing a more sound knowledge about the candidates and the political agenda, favoring in the end awareness and mobilization (Lau, 1982;Garramone et al., 1990;Geer, 2006). Furthermore, subsequent studies contend that negative campaigning may even increase turnout (Finkel and Geer, 1998;Wattenberg and Brians, 1999). This strand of literature argues that negative campaigning in the mass media exposes citizens to alternative opinions and, thus, is more effective in increasing citizens' knowledge. Citizens, in other words, may count on a larger variety of tools and information for evaluating candidates and parties and, eventually, to cast a more thoughtful vote. Similar arguments come from further research showing that negativity observed in political campaigns contributes to raising voters' attention and incentivizing their participation (Lau andPomper, 2001, 2004;Clinton and Lapinski, 2004;Ladini et al., 2020). Martin (2008), starting from Zaller (1992) seminal research, claims that negative coverage stimulates citizens' awareness about the country's problems and, in the end, may boost turnout. Other studies (Mutz and Reeves, 2005), instead, while recognizing the media's negativity effect on political interest and knowledge, also point out that negative coverage could harm citizens' political trustwhich is usually correlated with propensity to political participation.
All things considered, the literature connecting negative campaigning and turnout has come to a stalemate presenting contradictory results. According to Min (2004), this heterogeneity of findings hugely depends on the research setting (experimental vis-à-vis observational), the type of data (individual vis-à-vis aggregate), and the type of campaign (national or local). In this respect, this article aims at assessing whether negative coverage may affect people's turnout, and, in particular, at clarifying in which way. We test our hypotheses in Italy, which represents a very appropriate case study for our purposes. Italian politics, indeed, is characterized by a fierce contrast between parties (Sani and Sartori, 1982;Bellucci and Segatti, 2011), it presents a high level of polarization at the mass-level (Mancini, 2013), and it also shows high levels of negative coverage of politics and election campaigning (Bobba et al., 2013). Given that previous literature on this topic provides mixed results, we state two alternative hypotheses, which read as follows: H1a: People exposed to high levels of negative political coverage in the outlet they read/watch have higher propensities to turnout.
H1b: People exposed to high levels of negative political coverage in the outlet they read/watch have lower propensities to turnout.
First-and second-order elections: does the electoral context affect media effects?
Countless moderating or mediating factors, not directly related to the individual characteristics of the audience, can be identified in the literature that attempts to link media negativity and turnout. Although previous studies have investigated separately the role of media effects in different electoral contexts, by analyzing those effects in the election campaigns before national (see Hopmann et al., 2010) or European Elections (see van Spanje and de Vreese, 2014), little research has committed to systematically comparing different types of elections. More precisely, investigations on whether media effects are equally effective, irrespective of the electoral context in which those effects take place, are virtually absent.
Drawing upon the seminal work of Reif and Schmitt (1980), the literature addressed the study of parties' strategies in the various European Election campaigns, as well as outcomes from both the supply-side (namely, party and leaders campaigning, communication, etc.; for instance, see De Vreese, 2009;Schuck et al., 2011) and the demand-side viewpoint (namely, interest in the campaign by voter, turnout rates, and voting behavior; see, for instance, Franklin, 2001;Mattila, 2003). This scholarship agrees in underlining that in European Elections there is much less at stake with respect to a national election. While in National Elections the stake is the government of the country, in European Elections the stake is sending amore or less limitednumber of MEPs in the new (and distant) parliament. The most relevant consequence of this asymmetry is that, compared to national elections, in European elections parties fail to mobilize voters, and thus the electorate shows little interest in the campaign (Lodge, 1996;Norris and Reif, 1997). 1 Literature underlines also that, being more focused on European issues, second-order elections are usually less interesting for the electorate which is usually more attentive to domestic issues (see Hobolt and Wittrock, 2011;van Spanje and de Vreese, 2014). As a result, because of the less-at-stake argument discussed above, parties are also less prone to invest at the monetary level in election campaigns (Petithomme, 2012), strengthening the citizens' estrangement from second-order elections.
All those elements lead to arguing that the European Election campaign context might entail a low-profile campaign. This implies that also media effects influencing individuals' propensity to turnout are less relevant with respect to the National elections. The second hypothesis thus reads as follows: H2: The effect of negative political coverage on turnout is weaker in the context of the European Elections campaign.

Data
The data employed to test our hypotheses are derived from four datasets. Two of them refer to individual data (measuring our dependent variable, turnout, as well as individual moderator and control variables) gathered from the 2018 ITANES survey and 2019 ITANES-University of Milan survey. Both the surveys were collected by means of CAWI mode before the 2018 National Elections (held on 4 March 2018) and the 2019 European Parliament elections (held on 26 May 2019). However, the two surveys present several differences: first, the companies collecting the data in the two elections are different (IPSOS for 2018 and SWG for 2019 elections); second, the 2018 ITANES survey is collected following a rolling cross-section design (Johnston and Brady, 2002), which started about 2 months before the election days (every day about 200 cases were randomly selected from an opt-in basket). The 2019 ITANES survey is collected in five waves before the election day. Therefore, in order to maintain the comparability of the two datasets, we had to reach the 'lowest common denominator' of the two surveys. Accordingly, we had to harmonize the two datasets so that the cases selected were contacted in the same n days before 1 This is mainly why Elections present lower levels of participation compared to national elections (Reif and Schmitt, 1980). the respective Election Day. We thus subdivided both the datasets into three waves: the first one covering from 46 to 32 days before the Election Day; the second one covering from 31 to 18 days; the third one covering from 17 to 11 days.
Both the datasets provide information concerning media exposure since respondents were asked to declare: (a) their main source of information and (b) the TV/newspaper outlet they were exposed to in the previous few days. This information has been matched with data from the ITEM project, which studied the media coverage of the 2018 and 2019 electoral campaigns throughout human content analysis of the political contents published or broadcast by the main Italian daily newspapers and TV-news programs. For each election, the analysis has been performed covering the 7 weeks before the election day. The set of outlets analyzed was defined according to two main criteria: the readership/audience and the outlet's political leaning. As the press is concerned, we considered the following outlets: Corriere Della Sera, Il Giornale, Il Fatto Quotidiano, la Repubblica, and La Stampa. Within this selection of newspapers, all political news items published on the first-page and recalled in internal pages were coded. As regards the TV, both private and public companies have been covered: more specifically, we included the following newscasts in their evening editions: TG1, TG2, Tg3, Tg5, and La7 News. In this case, due to the differences between printed and TV media, all political news items broadcasted were coded.
Once harmonized with the days available for the individual datasets, the final media datasets include 2776 news items in 2018 and 3227 items in 2019. Both the media datasets provide a measure of negative tone (Lengauer et al., 2012)  The data employed for the analysis are thus a pooled dataset containing respondents interviewed before both the 2018 and 2019 elections in Italy, with, as an additional independent variable, the level of negativity of the media outlet to which they are exposed.

Measures
The literature investigating the relationship between turnout and negativity generally operationalizes turnout as a dichotomous choice (namely, one might be more or less prone to vote). However, more recent literature addressing the effect of media/external factors on voting behavior stresses the importance of voting uncertainty in the processes of opinion formation during the election campaigns (see Redlawsk, 2002;Ohme et al., 2018). From both a theoretical and an empirical point of view, uncertainty (namely, the impossibility for the respondent of declaring whether one will go to the polls or not) represents a relevant factor that might impinge the way in which results are interpreted. We argue that there is a strong difference in interpreting a result assessing that, for instance, negativity increases turnout at the expense of non-voters, or, rather, at the expense of the group of the uncertain voters, or both. In order to deepen the relationship between electoral behaviors and media negativity, this study employs an operationalization of the turnout variable that distinguishes between three diverse/possible outputs. For both 2018 and 2019 election campaign datasets, we construct a variable that scores 0 when the respondent declares to be undecided, 1 when the respondent declares that he/she will not vote, and 2 when the respondent declares a valid vote choice in the upcoming election.
The main independent variable is the negativity of the media outlet the respondent is exposed to. Empirically, it is derived from the 'general tone of the news' variable within the ITEM dataset.
2 Negativity was operationalized through a categorical variable accounting for the overall tone of the news item. 'Indications of negative tonality are the framing of the story as a political failure, fiasco, disaster, crisis, frustration, collapse, flop, denial, rejection, neglect, default, deterioration, resignation, skepticism, threats, cynicism, defeatism, or disappointment. Indications of positive tonality are depictions of political success, problem solutions, achievement, improvement, advance, prosperity, accomplishment, enthusiasm, hope, benefit, gain, sustainability, gratification, or accomplishment. If a report does not reflect indications of negative tonality or positive tonality, then it has to be coded as 'neutral'. In case of the presence of both negative and positive frames, the tone was coded as 'balanced/ambivalent' (Lengauer et al., 2012: 195).
We coded a dummy variable that scores 1 when a news item presents a generally negative tone and 0 otherwise; after that, we calculated the average of the variable by wave (namely, over the days of each harmonized wave exposed above) and outlet, and multiplied it by 100. We, therefore, merged the variable to the individual data, in a way that every respondent exposed to a nonmissing outlet is coupled with a variable indicating the average negativity that he/she has been exposed to in every wave (the procedure is performed in a way that, to every respondent, only the single main outlet is coupled).
In addition to the other moderator of the effect, we included a further variable identifying the electoral context. It is the case of a dummy variable indicating whether the respondent has been interviewed in the first-or second-order election. We have added a set of additional controls that represent possible alternative explanations for predicting our dependent variable: gender, age, educational level (subdivided in primary, secondary, and tertiary), left-right self-placement (subdivided in right, center-right, center, center-left, left, not located, do not know), political involvement, 3 and a dummy variable indicating whether the respondent is primarily exposed to a newscast or newspaper.

Models
The two hypotheses are tested through a set of multinomial logistic regression models, having, as the dependent variable, the trichotomous turnout variable. Multinomial logistic regressions allow predicting the probabilities of the three different outcomes at different levels of the same independent variables. Model 1 tests the alternative hypotheses H1a and H1b by simply assessing the effects of the negativity variable, while Model 2 (testing H2) is fitted by performing a two-way interaction between outlet negativity and electoral context. However, interpreting multinomial logistic regression might be particularly complicated, in particular for what concerns the significance and magnitude of the effects of the different outcomes. Also, the interpretation becomes even more difficult when evaluating the levels and significance of interaction terms (Long, 1997). We thus rely on graphical summaries of probabilities and discrete changes in probabilities (average marginal effects) in order to measure the substantive effects of our independent scores. Table 1 reports the coefficients for the two models. As regards control variables, it shows that, for what concerns the gender variable, women present lower likelihoods with respect to men to vote compared to uncertain respondents. The same applies to more centrist/non-located people. This trend is consistent in both the models and does not change according to their different parametrizations (namely, the presence/absence of the interaction term).

Results
As pertains to our first hypothesis, Model 1 shows that being exposed to more negative news outlets significantly increases turnout in our sample, at the expense of indecision. Figure 1 illustrates the predicted probabilities for the differential effect of negativity on each of the three outcomes.
The increase of decided voters at increasing levels of negativity corresponds to lower levels of indecision. In other words, negativity leads people to be more sure to vote (corroborating 3 In this case, since the dataset does not provide variables measuring precisely the concept of political sophistication, we constructed the variable relying on a method already employed by Solvak and Vassil (2015) and, similarly, by Vegetti et al. (2014). They used the highest score on the 'Propensity to vote' (PTV) for five parties (a variable measured on a 0-10 scale, where 0 is 'not at all likely' and 10 is very likely (Franklin et al. 1996). In order to solve the issues of skewness, the variable has been successively recoded into a dummy which is 0 when the highest PTV stops at 5 and is 1 when the highest PTV is 6 or more. A 0 for this variable indicates that the respondent does not consider any of the relevant parties as a good voting option. According to Solvak and Vassil (2015) 'this low score […] would mean dissatisfaction with the party choice on the ballot'. Standard errors in parentheses. ***P < 0.01, **P < 0.05, *P < 0.1. hypothesis H1b) at the expense of those who are undecided, while it does not affect people who are sure not to go to the polls. H2, the hypothesis arguing that the electoral context moderates the effect of negativity, is tested in Model 2. In this case, the effect of the interaction between the main independent variable, negativity, and the moderator is not significant. Figure 2 indicates the average marginal effects of the negativity variable on the three outcomes of the dependent variable, according to whether the electoral context is a first-order (FOE) or a second-order (SOE) one. Our analyses reveal that the soleslightdifference between the two figures is the effect of negativity on undecided voters, which is non-significant to the 5% threshold in the FOE (the figure is, however, significant to the 10%) and stronger and significant in the SOE. However, a Wald test on the difference between the two average marginal effects resulted as non-significant (P = 0.40). We can thus say that the effects of negativity are not substantively different in the two electoral contexts.

Discussion and conclusion
In this article, we were interested in clarifying the linkage between media negativity and voters' turnout. Despite the extensive literature on this field, we have shown that research results are mixed and struggle to provide a clear answer about the impact of negative coverage on participation. Literature suggests that negativity depresses citizens' turnout (Ansolabehere et al., 1994;Ansolabehere and Iyengar, 1995). Negative campaigning provides the image of conflictual politics, and deriving cynicism turns to alienate from politics, disincentivizing citizens to go to the polls. However, other scholarship states thaton the contrarynegativity could bring in virtuous dynamics. In this respect, for instance, Finkel and Geer (1998) argue that negative messages when reporting important and relevant information are able to catch the citizens' attention contributing to stimulate their mobilization. Goldstein and Freedman (2002) come to similar conclusions. Also, Wattenberg and Brians (1999) posit that the intent of negative campaigning is not to demobilize the people, but rather to change their minds. Negativity, in other words, does not keep away from the polling station the supporters of a candidate, but rather it results to change and shifts their support.
Although findings are controversial, they all agree that negativity does not play alone, as several factors may moderate the effect of media negativity on citizens' political behavior. In this respect, Min (2004) points out that both individual and contextual factors may shape how individuals receive media inputs, and, in turn, the media impact could be less direct than how scholars are used to positing.
Accordingly, in this article, we focused on the contextual dimension, by considering the type of electoral context (whether first-or second-order election).
First, we test the effects of negative coverage on turnout; our analyses reveal that increasing levels of negativity exposure appear to boost turnout, in particular at the expenses of undecided voters, and do not affect non-voters. Put in other terms, negativity increases the certainty to vote, while it does not impinge those who are sure to desert the vote. As regards the type of election and, more specifically, whether the salience of the election makes citizens more sensitive to negativity, we did not find clear differences between first-and second-order elections.
This study has some limitations. They pertain mainly to the research design, and the different methodological setting applied in 2018 and 2019 for the individual data collections. Due to the organization per waves of the CAWI survey in 2019, and for harmonization necessities, we were forced to re-design also the 2018 dataset. This entailed a loss in data detail since we had to avoid employing the daily data that the RCS design of 2018 provided. An additional element that represents a limitation of the paper is the fact that it focuses on the influence that TV newscast and newspaper have on voters, by neglecting other external elements, such as, for instance, interpersonal communication (such as agreement/disagreement in one's political network) or strictly online news outlets. Unfortunately, comparing first-and second-order elections led us to choose a relatively small subset of variables that were present in both datasets. For this reason, it was for us impossible to test possible alternative explanations related to these issues.
Summarizing, our study found that negativity in Italy, a largely polarized environment, is able to boost turnout, and in particular leads people to be more certain to go to the polls. On the other side, negativity does not affect the likelihood of abstaining. This difference is particularly relevant because it suggests that exposure to negative information affects voters who have some chances to be mobilized (namely, undecided voters). On the other side, our results show that media negativity does not spur people to go to the polls if they are already sure to abstain. These results are in line with the literature arguing that political media coverage has a positiverather than negativeeffect on turnout (i.e. De Vreese, 2005;De Vreese and Boomgaarden, 2006;Schuck et al., 2013;Schuck et al., 2016), disconfirming the idea of political cynicism induced by media negativity. While we can rely on the extensive literature on the effects of media negative frames on leader evaluation (i.e. Druckman and Perkin, 2005;Hopmann et al., 2010;Ha, 2011), few studies so far have addressed the turnout dimension (i.e. Min, 2004). Previous research focused in particular on voting behavior, rather than participation in elections. For example, Krupnikov (2011) argues that media bias has a demobilizing effect when negativity focuses on the voter's preferred candidate. According to our results, indeed, a negative frame in media coverage is effective in mobilizing those who are not sure to go to the polls, while among those who are certain about their voting strategies (even to desert the election) no effects are found. Our research, thus, contributes to shed a light on undecided voters, and the mechanisms of media effects on their shift toward participation or abstain.
Furthermore, our analyses show that this mechanism is not dependent on the electoral context (namely, whether the election is a first-or second-order election). This latter result can be the combination of several factors: it could be due to the high level of polarization that Italy has experienced in the last years (which, in turn, tends to make an event of every election), by a contingent, strong, and violent style of communication that emerged in the last years, mainly engaged by populist entrepreneurs (see, for instance, Cremonesi et al., 2019), or by the very fact that, also for what concerns the communication styles and their effects, the differences between first-and second-order elections are becoming increasingly thinner. Further results (both employing a comparative and a longitudinal analysis) will be able to assess whether the invariance between electoral contexts shown here is an exceptional result or signals that some aspects of the second-order framework are becoming less crucial.
Funding. The research received no grants from public, commercial, or non-profit funding agency.