Attack Politics from Albania to Zimbabwe PDF

946410 research-article2020 IPS0010.1177/0192512120946410International Political Science ReviewValli and Nai Article International...

946410 research-article2020 IPS0010.1177/0192512120946410International Political Science ReviewValli and Nai Article International Political Science Review Attack politics from Albania 2022, Vol. 43(5) 680­–696 © The Author(s) 2020 to Zimbabwe: A large-scale Article reuse guidelines: comparative study on the drivers sagepub.com/journals-permissions https://doi.org/10.1177/0192512120946410 DOI: 10.1177/0192512120946410 of negative campaigning journals.sagepub.com/home/ips Chiara Valli University of Bern, Switzerland Alessandro Nai University of Amsterdam, Netherlands Abstract There is little comparative research on what causes candidates in elections across the world to ‘go negative’ on their rivals – mainly because of the scarcity of large-scale datasets. In this article, we present new evidence covering over 80 recent national elections across the world (2016–2018), in which more than 400 candidates competed. For the first time in a large-scale comparative setting, we show that, ceteris paribus, negativity is more likely for challengers, extreme candidates, and right-wing candidates. Women are not more (or less) likely to go negative on their rivals than their male counterparts, but we find that higher numbers of female MPs in the country reduces negativity overall. Furthermore, women tend to go less negative in proportional systems and more negative in majoritarian systems. Finally, negativity is especially low for candidates on the left in countries with high female representation, and higher for candidates on the right in countries with proportional representation (PR). Keywords Negative campaigning, comparative political communication, expert survey, elite behaviour Introduction This article investigates the drivers of negative campaigning worldwide using a dataset of more than 80 national elections. The phenomenon of negative campaigning – that is, the use of political attacks against the program, record, policies, or persona of opponents during election campaigns – has Corresponding author: Alessandro Nai, Amsterdam School of Communication Research (ASCoR), University of Amsterdam, Postbus 15793, Amsterdam, 1001 NG, The Netherlands. Email: [email protected] Valli and Nai 681 received increased attention in recent decades. Yet, the question of whether negativity is a detrimen- tal force for contemporary democracy is highly contested. Several studies have shown that negative messages tend to demobilize voters, negatively influence public trust (Ansolabehere and Iyengar, 1995), and increase political cynicism (Cappella and Jamieson, 1997). At the same time, other scholars suggest instead that negative campaigning has the potential to mobilize voters (Goldstein and Freedman, 2002) because it stimulates interest and participation (Geer, 2006). Regardless of its effects, few would argue that negative campaigning is inconsequential, which is why the literature increasingly focuses also on its drivers. Why, and under what conditions, are parties and candidates more likely to go negative on their opponents? Negative campaigning is unlikely to always work as intended, and often ends up harming the sponsors more than their des- ignated target (Roese and Sande, 1993). Also, because of this, the decision to make use of attack messages often becomes a strategic one. Little is, however, known about how the individual pro- files of candidates affect this decision, especially in a comparative perspective. Most of the research on negativity in politics focuses on the American case. Outside this case, existing evidence is either on specific countries – such as Brazil (da Silveira and de Mello, 2011) or Germany (Maier and Jansen, 2015) – or, if comparative, often limited to only a handful of countries (e.g. Walter et al., 2014). A recent study (Nai, 2018) compared the use of negative campaigning by candidates who competed in 35 national elections worldwide, but mostly focused on differences among the spon- sors and the targets of the attacks, and only addressed the issue of cross-country comparison marginally. All things considered, we still know surprisingly little about (i) whether negative campaigning in elections across the world – from Albania to Zimbabwe, so to speak – follows the same logics identified in the US literature, and (ii) to what extent differences across countries drive the use of negativity in campaigns differently. Is negative campaigning a global phenomenon, driven by uni- versal ‘rules’? To what extent do the characteristics of the context shape the use of negative cam- paigning strategies? This article contributes to the emerging comparative political communication literature by look- ing at the use of negativity in elections worldwide. We do so by comparing the content of election campaigns of 404 ‘top’ candidates (that is, party leaders and leading candidates for office) who competed in 84 elections in 71 countries worldwide between June 2016 and December 2018, rated by selected samples of scholars (1,321 experts in total). We aim to shed light on the effects of can- didate profiles and two specific contextual aspects, namely the electoral system and the proportion of female representatives in parliament. As will be discussed in the following sections, we claim that the increased need for consensus in proportional electoral systems reduces the incentives to go negative and that more female representatives lead to a convergence towards ‘kinder’ communica- tive norms and hence, less negativity. All data and materials, including the full Appendix, are available for replication at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/ Why and under which conditions candidates go negative Candidate profile. The literature claims that incumbents are less likely to go negative than challeng- ers (Nai, 2018; Walter and Vliegenthart, 2010). Incumbents have a larger political record which they can use to promote themselves; because challengers are less likely to have this option, they need to provide good reasons as to why voters should turn against the incumbents (Kahn and Ken- ney, 2004). Challengers tend to receive weaker media coverage (Hopmann et al., 2011), which encourages them to find ways to increase their visibility, by, for example, attracting attention through negative rhetoric (Nai, 2018). Moreover, challengers have no office to lose, which makes 682 International Political Science Review 43(5) it electorally less risky for them to use negativity. The first hypothesis is thus formulated: H1: Challengers are more likely to use negativity than incumbents. Evidence from the US further suggests that right-wing candidates have a higher tendency to use negativity than left-wing candidates (Lau and Pomper, 2001). The right-wing inclination to view ‘the world as a ruthlessly competitive jungle in which the strong win and the weak lose’ (Duckitt, 2006, p. 685) may increase the acceptance of attack messages among their electorate. This assump- tion is at least partially supported by studies that show that Democrats, in comparison with Republicans and Independents, exhibit less sympathy for negativity (Ansolabehere and Iyengar, 1995). Accordingly, hypothesis two goes as follows: H2: Right-wing candidates are more likely to use negativity than left-wing candidates. It further seems likely that parties and candidates on the extremes of the political spectrum have a higher tendency to use negative rhetoric. The more extreme a party is, the more it disagrees with other parties on political issues (Elmelund-Præstekær, 2010, p. 142), making it less likely for them to form coalitions or policy agreements. Assuming that political disagreements may lead to rhetori- cal attacks, we expect that the ideologically extreme are more negative in their campaigns than mainstream candidates. Results of previous studies do indeed suggest that parties far from the ideo- logical centre are more inclined to go negative (Elmelund-Præstekær, 2010; Walter et al., 2014). This logic leads to the following hypothesis: H3: Extreme candidates are more likely to use negativity than mainstream candidates. It remains much debated how gender influences the choice of going negative (Grossmann, 2012). Scholars, including Fox (1997) and Kahn and Kenney (2004), find support that males are more likely to go negative. Fewer studies claim that females increasingly engage in attack behaviour (e.g. Kahn, 1993), and the preponderance of research does not find any significant gender differences in the use of negativity (Grossmann, 2012; Lau and Pomper, 2001; Panagopoulos, 2004). One rationale behind the assumption that females refrain from going negative is as simple as follows: The usage of negative rhetoric violates the female stereotype. The different attributes assigned to males and females trigger concrete expectations of ‘appro- priate’ social behaviours (Ennser-Jedenastik et al., 2017) that also translate into stereotypical expectations of campaign behaviour (Dinzes et al., 1994). Women thus have to choose between dispelling or exploiting gender stereotypes while acknowledging that the disruption of the expected behaviour might have damaging electoral consequences (Trent and Friedenberg, 2008). Although plausible in theory, current research challenges this logic. The lack of consen- sus will be addressed in this large-scale study by testing the controversial hypothesis four on a cross-national sample: H4: Female candidates are less likely to use negativity than male candidates. The role of the context. Candidates do not campaign in a vacuum, and the nature of the context is likely to drive campaigning dynamics. More specifically, we could expect that the context both drives a differential use of negative campaigning and alters the incentives of competing candidates to go negative. Because of the lack of large-scale comparative data, these dynamics have not yet received the attention they deserve. In this article, we explore the direct and indirect effects of the Valli and Nai 683 election context by testing the assumptions that negative campaigning has a lower incidence in proportional representation (PR) systems and democracies with higher shares of female MPs. First, as discussed in Nai (2018) we expect that elections held under PR should have lower levels of negativity. PR tends to create multiparty systems, which are likely to reduce negativity for at least two reasons. First, when multiple parties (or candidates) compete, the positive return on attacks are very uncertain. Negative campaigns might decrease support for the target, but because of the non-zero-sum nature of the political competition, this does not necessarily result in increased support for the sponsor (Nai, 2018; but see Elmelund-Præstekær and Svensson, 2014). Second, in multiparty PR systems, competing parties and candidates have enhanced incentives for cooperative behaviour, due to the possibility of coalition bargaining (e.g. Elmelund-Præstekær, 2010; Walter et al., 2014). This assumption has been addressed by Lijphart (1999), who claimed that PR systems tend to have a stronger community orientation and social consciousness and, thus, stand in contrast to self-interest and power politics. In line with this idea, PR and multiparty systems should reduce political conflicts, as the risks of alienating potential coalition allies is at odds with the systemic incentives for successful governance (Lijphart, 1999). Hence, we test the following hypothesis: H5: Negativity is less likely in countries with proportional representation (PR). Whereas in majoritarian elections citizens cast their ballot for specific candidates, in propor- tional systems voters often opt for a party list, which reduces the importance of the individual candidates (Thames and Williams, 2010, p. 1579). The diverging foci of these systems are expected to affect the relative importance of the candidate profile in so far as politicians’ personal differ- ences become less relevant in PR systems. In other words, we expect that the influence of ideology, extremism, and gender on negativity decreases in PR systems. The expected interactions can be summarized as follows: H6: The effect of the left–right positioning of candidates on negativity is weaker in countries with proportional than majoritarian systems. H7: The effect of extremism on negativity is weaker in countries with proportional than majori- tarian systems. H8: The effect of gender on negativity is weaker in countries with proportional than majoritar- ian systems. There is also reason to believe that negativity declines with an increase in female representation. In general, women are said to practice a kinder politics that is ‘characterized by cooperation rather than conflict, collaboration rather than hierarchy’ (Norris, 1996, p. 93). This ‘feminized’ style of politics is expected to radically transform not only institutions, and public policy, but also the political behaviour once a ‘critical mass’ of elected women is reached (Studlar and McAllister, 2002). The cited literature makes it plausible that more female representatives lead to a greater convergence of these ‘kind’ norms and political styles. Yet, one should not disregard the mixed results regarding the general attack pattern of women discussed in the preceding section. The fol- lowing hypothesis, therefore, serves the purpose of investigating the gender difference on the insti- tutional level and posits that: H9: Negativity is less likely in countries with a higher proportion of female MPs. 684 International Political Science Review 43(5) In their work, Ennser-Jedenastik et al. (2017) further suggest that the candidates’ attack behav- iour varies with intraparty gender balance. Although the authors do not uncover evidence for this claim, the theoretical argument that highlights the role of the social context deserves some atten- tion. Ennser-Jedenastik et al. (2017) argue that a more equal gender distribution leaves females less ‘“conspicuous” since female politicians are naturally a more common occurrence in an environ- ment with higher overall female representation’ (p. 88). This, in turn, may decrease women’s responsiveness to gender stereotypes and thus increase their willingness to go negative. Testing this hypothesis aims to shed light on the ongoing debate about gender effects and follows Eagly et al.’s (2004, p. 280) suggestion that it might be the social context that explains the variation in gender behaviour. The hypothesis to be tested can be formulated as follows: H10: Female candidates are more likely to use negativity in democracies with a higher propor- tion of female MPs. Data and methods Dataset Due to the complexity of measuring discourse comparatively, very little data exists that compares the content of election campaigns worldwide, across different cultures, languages, and political systems. In this article, we rely on the Negative Campaigning Comparative Expert Survey Dataset (Nai, 2018; Nai and Maier, 2018; Nai et al., 2019), covering all national elections held worldwide between June 2016 and December 2018. The dataset is based on a systematic survey distributed to election-specific samples of national and international scholars1 in the weeks following each elec- tion. Experts were asked a series of questions about the campaign in general, as well as actor-spe- cific questions. The average response rate across all elections in the dataset is just below 20%. After excluding the missing values on all relevant variables and considering only elections for which at least five different scholars rated the campaign, our models are run on 404 candidates who competed in 84 elections worldwide. The information is based on answers provided by 1,321 experts. Appendix A lists all elections and candidates in our dataset; the number of responses for each election is signalled in Table A1 (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/). On average, scholars in the dataset lean to the left (M = 4.33/10, SD = 1.80), 74% are domestic (that is, have a professional appointment at a university in the country for which they were asked to evaluate the election), and 33% are female. Overall, experts declared themselves very familiar with the elections (M = 8.01/10, SD = 1.77) and estimated that the questions in the survey were relatively easy to answer (M = 6.51/10, SD = 2.38). Table C1 in Appendix C (available at the fol- lowing Open Science Foundation (OSF) repository: https://osf.io/4vhea/) presents the composition of the expert samples according to these characteristics for each election in our database. Using experts to measure the tone of the campaign might seem unorthodox, as the literature usually focuses on systematic content analysis of communication messages (e.g. Russmann, 2017). Yet, there are several advantages to using experts. There is a long tradition of using ‘expert’ knowl- edge to assess party dynamics and positions (e.g. Hooghe et al., 2010), and using experts is a par- ticularly cost-effective way to obtain systematic and comparable data across a wide variety of contexts. In the case of discursive dynamics, as we study here, we could even argue that expert ratings are likely to be the most indicated approach for very large-scale comparative studies because they are able to ‘circumvent’ fundamental issues such as coding in different languages and the availability of comparable communication outputs to code in the first place. Furthermore, Valli and Nai 685 experts can be used to assess the campaign on the whole, independently of specific communication channels (e.g. manifestos or TV ads, which are both idiosyncratic and strongly contextual). We return to this important feature in the following when discussing the measurement. To be sure, using experts does not come without downsides, especially in terms of sample composition and profile effects (e.g. Curini, 2010; Walter and van der Eijk, 2019). Furthermore, a frequent issue with expert surveys is the complexity in assessing which information experts rely upon when pro- viding their ratings (e.g. Budge, 2000). We address some of these issues in the following when discussing the measure of campaign tone and provide some robustness checks to exclude any major profile effects (see the following). Case selection The dataset includes information for all national elections (legislative and presidential) held across the globe between June 2016 and December 2018, with the exclusion of elections in microstates (e.g. the election in Palau in November 2016) or in case of flagrant violations of basic rules of free and fair elections and/or absence of competition (e.g. Somaliland, or the Egyptian election of March 2018). The elections in our database are thus not a random selection of elections and not statistically representative of all elections worldwide per se. Yet, because of their sheer number and geographical spread (Figure 1), they can be seen as providing a comprehensive image of contem- porary electoral competition across the globe. All analyses are run on a dataset that includes measures of campaign negativity for ‘top candi- dates’: party leaders during legislative elections or main candidates in presidential elections. Especially during legislative elections, the focus on ‘top’ candidates is likely to yield a conserva- tive estimate of campaign negativity. Evidence from the Austrian case shows, for instance, that a ‘division of labour’ takes place within parties, where the ‘dirty work’ of attacking the opponents is delegated to lower-level party leaders (Dolezal et al., 2015). Robustness checks will assess whether the difference between these two types of elections matters in our case (it mostly does not; see in the following). Measuring negativity. Experts were asked to assess the ‘tone’ of the campaign (Lau and Pomper, 2001; Nai and Walter, 2015) used by competing actors, that is, to what extent they ‘talked about the opponents in the race by criticizing their programs, attacking their ideas and accomplish- ments, questioning their qualifications, and so on’ instead of ‘talking about one’s own accom- plishments, qualifications, programs and ideas by praising them’ (quoted directly from the questionnaire). Concretely, experts had to evaluate the tone on a scale from -10 very negative to +10 very positive. Table C2 in Appendix C (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/) reports, for each candidate, the number of experts which provided ratings – whose answers are aggregated into the measure of campaign tone – and the standard deviation resulting from this aggregation. The standard deviation can be seen as an indicator of how ‘consen- sual’ the experts were about the campaign tone of any given candidate; the average standard devia- tion across all candidates is approximately 3.7, which is relatively modest on a 21-point scale (from -10 to +10); there is no significant relationship between the standard deviation and number of experts, r(402) = 0.01, p =.846 nor between the standard deviation and the overall tone score, r(402) = -0.06, p =.234, as also shown in Figure C1 (Appendix C) (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/). In other terms, experts were not significantly more or less precise in assessing the campaign of candidates that went particularly negative or positive. 686 International Political Science Review 43(5) Figure 1. Geographical coverage. Valli and Nai 687 Due to the complex nature of the concept measured (Sigelman and Kugler, 2003) and because the concept itself could suffer from cross-cultural comparability issues, the questionnaire included six ‘vignettes’ – examples of campaign messages2 that experts also had to rate using the scale for the campaign ‘tone’. We used those vignettes to ‘anchor’ the experts’ ratings, starting with the assumption that answers to these vignettes provide a useful benchmark across experts. More spe- cifically, we ran a series of parametric adjustments (King et al., 2004) through ordered probit models (gllamm models). The models yielded an adjusted measure of campaign negativity that simultaneously takes into account the experts’ vignettes ratings plus a series of cross-sectional set parameters (an election identifier, plus the expert’s gender, domestic/international status, self- reported familiarity with the election, and left–right positioning – this latter is of particular interest in the light of research suggesting that expert ratings might be influenced by their individual orien- tations; e.g. Curini, 2010). The obtained variable is a continuous measure of negative tone that ranges between 1 ‘very positive’ and 7 ‘very negative’. The original and adjusted variables are very strongly correlated, r(402) = -0.97, p <.001. Nonetheless, and to err on the side of caution, we replicated all analyses also using the original expert rating (see Appendix B, Tables B7 to B9) (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/). It is important to note that, because they were cued to assess the campaign ‘on the whole’ – that is, beyond idiosyncratic differences across channels or specific communication events (e.g. rallies) – experts can be expected to provide an assessment of the general negativity of the campaign. We can expect experts to pick up not only the tone of messages used by the competing candidates but also the volume of the negative messages they use (e.g. Stevens, 2009). Experts are also likely to include in their assessment the overall ‘harshness’ or incivility of the exchanges between the can- didates; indeed, they tend to rate the harsher vignettes as higher on the negativity scale (Nai, 2018). Furthermore, campaign negativity, as measured by experts, strongly correlates with the use by candidates of harsher character attacks and fearmongering (Gerstlé and Nai, 2019). In this sense, using such a ‘holistic’ measure of campaign negativity could help reduce the disconnect between the ‘narrow’ scholarly definition and measurement of channel-specific negativity and the more broad, general impression of the public (Lipsitz and Geer, 2017; see also Haselmayer, 2019; Sigelman and Kugler, 2003). Table A3 (Appendix A) (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/) illustrates the differences in the overall cam- paign tone in the 71 countries in our dataset, ranked from the most ‘positive’ to the most ‘negative’ campaign (overall assessment). Due to the large-scale scope of the dataset, and the discursive nature of the phenomenon we are dealing with, the external validity of our measures cannot be assessed by comparing them with other existing ones – quite simply, there is no other dataset that measures the tone of candi- dates’ campaigns across the globe. Nonetheless, as we report elsewhere (Maier and Nai, 2020), we were able to set up a triangulation check for our measures within a different context – an expert survey we ran for the 2018 US Senate election Midterms using an identical protocol and measure of campaign tone as the ones discussed here. For all candidates having competed in the 2018 US Senate Midterms we compared the ratings of our experts with two independent data sources: the tone of the candidates’ campaigns on Twitter, and the percentage of negative TV ads of competing candidates in the Midterms elections. Even controlling for covariates at the candi- date and US state levels (gender, party affiliation, age, state turnout, state leaning), the expert measure is significantly and positively correlated with the negativity in Twitter and in TV ads. This triangulation suggests that experts are able to pick up the content (at least in terms of tone) of candidates’ campaigns, beyond the idiosyncrasies of different campaign channels (Walter and Vliegenthart, 2010); we have no reason to believe that this should not be the case outside of the US Midterms as well. 688 International Political Science Review 43(5) Candidate and country characteristics. Gender, age, and incumbency status of candidates are infor- mation easy to find, and their measure is straightforward. Less so is their left–right position; to compensate for the lack of cross-sectional datasets including information about the ideology of candidates worldwide, we use the information provided by Wikipedia. For all parties, Wikipedia includes an ‘information box’ in which demographic and biographic information of the party are presented, as well as the party’s ‘political position’. The latter is classified using a systematic series of labels, ranging from ‘far left’ to ‘far right’; sometimes, intermediate positions are presented (e.g. UKIP is currently classified as ‘right’ to ‘far right’). All of this information can be reduced to a 13-point scale (‘far left’, ‘far left/left’, ‘left’, and so forth), which we simplified into a 7-point scale ranging from 1 ‘far left’ to 7 ‘far right’. The dataset available in the OFS repository (available at https://osf.io/4vhea/) includes both versions of the measure. Although not ideal, Wikipedia can often provide quality factual information (Brown, 2011). Indeed, an external validity check dis- cussed in Nai (2018) shows that comparing the information in Wikipedia with left–right measures in the Chapel Hill Expert Survey (CHES; Polk et al., 2017) and the data in Benoit and Laver (2007) yields very high correlations. It is also important to note that the left–right scale is an ‘amorphous vessel’ (Huber and Inglehart, 1995, p. 90) unlikely to capture ideological cleavages in exactly the same way in every context (e.g. Rohrschneider and Whitefield, 2009), and some scholars argue instead for the use of party ‘families’ (e.g. ‘social democrat’, ‘communist’, ‘liberal’, and so forth; Franzmann and Kaiser, 2006). Nonetheless, the left–right distinction still represents a clear and simple heuristic to classify parties. The fact that the information is easily available on Wikipedia also suggests, in our opinion, that the left–right distinction still has a relevant appeal for the public at large. The left–right variable is then folded on itself to create the ‘extremism’ variable, which takes the value 0 for ‘low extremism’ (centre left to centre right), 1 for ‘moderate extremism’ (left, right) and 2 for ‘high extremism’ (far left, far right). We also used a more restrictive binary measure, differ- entiating only between ‘extreme parties’ (far left and far right) and all the other ones, and obtain very similar results (see robustness checks discussed in the following). To quantify the ‘level’ of female representation, we rely on the percentage of female politicians in the national parliament, using data from the Institute for Democracy and Electoral Assistance (IDEA).3 We use a binary variable that sorts countries with a PR electoral system from countries with a plurality/majority system. Our models are controlled, at the contextual level, by the com- petitiveness of the election, effective number of competing candidates, and country’s level of democracy. We measure the competitiveness of the election via a question in the expert survey that asked experts to evaluate how much they agree that ‘the race was not competitive, the winner was clearly known beforehand’. We use the formula of Laakso and Taagepera (1979) to measure the total effective number of candidates (ENC), which yields a number reflecting the number of com- peting candidates with a similar strength. Controlling for the ENC is important, as recent research suggests that party system fragmentation has a role to play in the overall level of negativity of the campaign (Papp and Patkós, 2019). Descriptive statistics for all variables are in Table A4 (Appendix A) (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/). Results Our main results are presented in Tables 1–3. Table 1 estimates firstly the use of negative cam- paigns as a function of individual and contextual determinants. Comparing the two models in Table 1, our results show that, even controlling for the nature of the context, individual differences across candidates drive the use of negativity quite substantially. As expected (H1), incumbents are significantly less likely to go negative on their opponents. More Valli and Nai 689 Table 1. Drivers of negative campaigning, direct effects. Model 1 Model 2 Coef Se Sig Coef Se Sig Female –0.03 (0.15) –0.00 (0.15) Incumbent –0.40 (0.13) ** –0.40 (0.13) ** Extremism 0.52 (0.08) *** 0.53 (0.08) *** Left–right 0.17 (0.03) *** 0.16 (0.03) *** Year born –0.01 (0.00) –0.01 (0.00) Female representation –1.64 (0.75) * Proportional representation (PR) 0.22 (0.17) Competitiveness 0.11 (0.08) Effective N of candidates 0.03 (0.04) Democracy index –0.08 (0.05) † Intercept 17.33 (8.86) † 15.98 (8.90) † N (candidates) 404 404 N (elections) 84 84 R2 0.171 0.216 Model χ2 98.53 110.2 Note: All models are random-effect hierarchical linear regressions (HLM) where candidates are nested within elections. The dependent variable (negative campaigning) varies between 1 ‘very positive’ and 7 ‘very negative’. ***p < 0.001, **p < 0.01, *p < 0.05, †p < 0.1. extreme candidates are significantly and substantially more likely to go negative and use character attacks in their campaigns (H3), and so are right-wing candidates (H2). Against our expectations (H4), but in line with the idea that gender plays a more complex role than expected, we do not find any support for the notion that female candidates are less likely to use negative elements in their campaigns. Turning to the characteristics of the context, we find that candidates in countries with more female MPs tend to campaign more positively (H9); for each increase of 10% of female MPs in the national parliament, campaigns tend to be approximately 0.16 points (on a 1–7 scale) more posi- tive. This being said, the difference between a PR and a majoritarian electoral system does not seem to drive a different use of campaign negativity; we thus reject H5. This lack of effect is some- what of a surprise, especially in light of case-specific evidence. For instance, Ridout and Walter (2015) show that election campaigns in New Zealand have become more positive since the country moved away from its first-past-the-post electoral system in the mid-1990s. Our results suggest that, broadly speaking, the importance of the electoral system should not be overestimated – or, at least, that its effects are subordinate to the effects of candidates’ profile and electoral dynamics. This being said, contextual factors can alter the incentives to go negative for candidates with different profiles. Tables 2 and 3 test the moderating effect of the percentage of female MPs and the electoral system. Results suggest that some moderation dynamics are at play for the effects of candidates’ ideologi- cal placement. Substantively, the effect of candidates’ left–right position is much stronger in coun- tries with higher female representation or a PR electoral system, all things considered. Figure 2 substantiates the interactions in Tables 2 and 3. As Figure 2 shows, negativity is especially low for candidates on the left when female representation is high (left-hand panel), and high for candidates on the right in PR countries (right-hand panel, which rejects H6). In both cases, the effect of 690 International Political Science Review 43(5) Table 2. Drivers of negative campaigning, moderated by female representation. Model 1 Model 2 Model 3 Coef Se Sig Coef Se Sig Coef Se Sig Female 0.19 (0.42) –0.02 (0.15) –0.01 (0.14) Incumbent –0.40 (0.13) ** –0.40 (0.13) ** –0.40 (0.13) ** Extremism 0.53 (0.08) *** 0.31 (0.21) 0.53 (0.08) *** Left–right 0.16 (0.03) *** 0.16 (0.03) *** –0.12 (0.09) Year born –0.01 (0.00) –0.01 (0.00) –0.01 (0.00) † Female representation –1.57 (0.76) * –1.94 (0.79) * –6.12 (1.46) *** Proportional representation (PR) 0.23 (0.17) 0.22 (0.17) 0.24 (0.17) Competitiveness 0.11 (0.08) 0.11 (0.08) 0.11 (0.08) Effective N of candidates 0.03 (0.04) 0.03 (0.04) 0.03 (0.04) Democracy index –0.08 (0.05) † –0.08 (0.05) † –0.09 (0.05) † Female repr * Female –0.66 (1.36) Female repr * Extremism 0.81 (0.73) Female repr * Left–right 1.08 (0.30) *** Intercept 15.61 (8.94) † 16.20 (8.91) † 20.17 (8.86) * N (candidates) 404 404 404 N (elections) 84 84 84 R2 0.218 0.221 0.239 Model χ2 110.2 111.4 126 Note: All models are random-effect hierarchical linear regressions (HLM) where candidates are nested within elections. The dependent variable (negative campaigning) varies between 1 ‘very positive’ and 7 ‘very negative’. ***p < 0.001, **p < 0.01, *p < 0.05, †p < 0.1. left–right – which was quite substantial when only direct effects are accounted for – is virtually nonexistent in countries with low female representation or a majoritarian system. We do not find any moderation effects for candidates’ extremism, rejecting H7. Second, we find some evidence that male and female candidates react differently to the nature of the context. As shown in Table 3, female candidates tend to go less negative in PR systems and attack more than their male counterparts in majoritarian systems. This suggests again that the over- arching expectation that female candidates adopt a gentler style of campaigning is unfounded – not only do they not do so overall, but in more competitive settings (majoritarian elections), they go even more negative than men, partially confirming H8. We, however, find no indication whatso- ever that female candidates are more or, for that matter, less likely to go negative on their oppo- nents in countries with a higher or lower percentage of female MPs. We thus reject H10. Robustness checks Appendix B (available at the following Open Science Foundation (OSF) repository: https://osf. io/4vhea/) includes a series of robustness checks. First, Tables B1 to B3 replicate the main analyses but use a more restrictive sample of candidates and elections, for which at least 10 experts provided ratings. Results are slightly weaker on the effects of contextual variables, but broadly in line with the main results. Second, Tables B4 to B6 are run for democracies only – that is, countries that score a minimum of 4 points out of 10 in the Economist ‘democracy index’. Results are in line with the main effects discussed in the text. Third, Tables B7 to B9 replicate the main analyses, but using Valli and Nai 691 Table 3. Drivers of negative campaigning, moderated by electoral system. Model 1 Model 2 Model 3 Coef Se Sig Coef Se Sig Coef Se Sig Female 0.55 (0.24) * –0.00 (0.15) 0.03 (0.15) Incumbent –0.40 (0.13) ** –0.40 (0.13) ** –0.40 (0.13) ** Extremism 0.52 (0.08) *** 0.57 (0.13) *** 0.52 (0.08) *** Left–right 0.15 (0.03) *** 0.16 (0.03) *** 0.06 (0.05) Year born –0.01 (0.00) –0.01 (0.00) –0.01 (0.00) Female representation –1.62 (0.73) * –1.63 (0.75) * –1.65 (0.75) * Proportional representation (PR) 0.33 (0.17) † 0.25 (0.19) –0.45 (0.33) Competitiveness 0.11 (0.08) 0.11 (0.08) 0.11 (0.08) Effective N of candidates 0.02 (0.04) 0.03 (0.04) 0.03 (0.04) Democracy index –0.08 (0.05) † –0.08 (0.05) † –0.08 (0.05) † PR * Female –0.87 (0.30) ** PR * Extremism –0.07 (0.16) PR * Left–right 0.16 (0.07) * Intercept 13.74 (8.85) 15.52 (8.98) † 17.98 (8.88) * N (candidates) 404 404 404 N (elections) 84 84 84 R2 0.237 0.216 0.225 Model χ2 120.8 110.2 117.4 Note: All models are random-effect hierarchical linear regressions (HLM) where candidates are nested within elections. The dependent variable (negative campaigning) varies between 1 ‘very positive’ and 7 ‘very negative’. ***p < 0.001, **p < 0.01, *p < 0.05, †p < 0.1. Figure 2. Negative tone, by left–right * female representation, electoral system. Marginal effects with 95% confidence intervals, based on coefficients in Table 2, M3 (panel a) and Table 3, M3 (panel b). All other variables fixed at their mean. In panel (a) the two groups represent two critical values for female representa- tion, respectively at one standard deviation below the mean value (low representation), and one standard deviation above the mean value (high representation). 692 International Political Science Review 43(5) the original unadjusted dependent variable; results are virtually identical in direction and magni- tude. Fourth, a case is often made that expert ratings are potentially affected by the profile of the experts and the composition of expert samples (e.g. Budge, 2000; Curini, 2010; Walter and van der Eijk, 2019). With this in mind, we replicated all analyses controlling for each election’s average expert profile in terms of gender, percent of domestic experts, average familiarity with elections in the country, average simplicity in answering the questionnaire, and average position on the left– right scale. Tables B10 to B12 present the results for these additional models and show that (i) the effect of the average experts’ profile is absolutely negligible, and (ii) the main results resist such controls. We can thus confidently exclude any major profile effects. Fifth, we replicated all models using bootstrapped standard errors (300 replacements; Tables B13 to B15) and find robust results. Sixth, Table B16 shows then that the difference between legislative and presidential elections seems to influence some effects (for instance, the effects for left–right are stronger during presiden- tial elections); to avoid any spurious effects there, we replicated all models controlling for the dif- ference between legislative and presidential elections (Tables B17 to B19); results are not only robust, but in many cases even stronger. Similarly, seventh, the difference between Western and Non-Western countries does not seem to alter the dynamics studied here (Tables B20 to B23), beyond the fact that female candidates seem less likely to go negative in Western countries, ceteris paribus. Notably, the effect of political ideology, which has been shown to follow different dynam- ics in Non-Western countries (e.g. Rohrschneider and Whitefield, 2009), does not seem to differ depending on this broad geographical classification. Finally, eight, we replicated all models using a more restrictive binary measure of political extremism (Tables B24 to B26); results are robust. Discussion and conclusion In a nutshell, our results show that challengers, right-wing, and ideologically extreme candidates have a higher tendency to use attack messages. Females are not more likely to attack their rivals in general, but they are more negative than their male counterparts in majoritarian countries. Higher numbers of female representatives reduce negativity overall, but right-wing candidates tend to particularly go negative on their opponents in countries with higher female representation and PR electoral systems. Taken together, these results indicate that (a) the candidates’ profile matters, as expected, but also that (b) the context matters, but mostly to moderate the effects of candidates’ profile. In turn, this suggests that comparative analyses looking at contextual drivers of campaign- ing strategies should be careful in only assuming that the context alters the incentives to campaign in a certain fashion – it certainly does so, but not across the board. This article is not without caveats: first, these results are likely to be conservative, especially for legislative elections. Evidence suggests that in this type of contest, the dangerous and ‘dirty work’ of attacking competitors is left to lower-level candidates (e.g. Dolezal et al., 2015), whereas our analyses are run on top candidates only. In this sense, it is likely that the dynamics at play here are even more intense when looking beyond the party leadership. This being said, as shown in Appendix D (available at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/), experts tended to rate rather consistently the negativity of top candidates and the negativity of their party overall, even if incumbents seem slightly less likely to go negative than their own party, cet- eris paribus. Second, the study solely relied on ratings provided by experts, which cannot provide all the nuances of other more traditional approaches to the study of campaign content (e.g. struc- tured content analysis). Nonetheless, relying on experts is particularly handy in large-scale com- parative settings. Expert ratings provide comparability across diverse contextual settings where it is unrealistic to implement traditional comparative methodologies, due to language barriers, for example. Third, the analysis only provides a snapshot of the elections and looks at the respective Valli and Nai 693 campaigns as a whole. It, thereby, largely ignores temporal dynamics such as, for example, the proximity of the election day (Damore, 2002). Fourth, we did not discriminate between the differ- ent communication channels, which may influence not only the amount of negativity but also the nature of the attacks (Walter and Vliegenthart, 2010). Fifth, of course, structural differences in political systems across countries go well beyond the rather simple dichotomy between countries with PR and majoritarian systems. Recent research shows indeed that additional elements, like electoral system disproportionality or the polarization of the electorate, can have an important role to play as well (Papp and Patkós, 2019). Despite these shortcomings, this study contributes to the literature in multiple ways: by provid- ing evidence from a large-scale comparative dataset, we contribute to the comparative research on elections and political campaigning. The analysis of 84 elections worldwide enabled us to test prominent communication trends on a broader sample and assess to what extent these can be con- firmed in cases that, up to this day, have received little to no attention from the literature. Moreover, while most articles focus on the isolated effects of either candidate characteristics or contextual factors, this article is one of the first to integrate both levels and look at them simultaneously. This provides a more realistic and comprehensive picture of why and under which conditions candidates choose to use negative rhetoric. The possibilities of extending this approach are manifold: future research could, for example, include other aspects of the political culture such as the citizens’ atti- tudes toward oppositional forces, or the degree of polarization. From a broader perspective, this paper deepens our knowledge of an international phenomenon, whose practical and normative implications for modern democracies are not to be underestimated. Acknowledgements We are very grateful to the anonymous reviewers and IPSR editors for their support, critical assessment, and constructive suggestions during the publication process. We take, of course, full responsibility for any remain- ing mistakes. A previous version of this article was presented at the 2019 annual meeting of the Canadian Political Science Association (CPSA) in Vancouver. Many thanks to all participants to the panel, and in par- ticular to Scott Matthews, for precious comments and feedback. A sincere thank you to all experts that partici- pated over the years in the NEGex study for their time and insights. Funding The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publica- tion of this article: Alex Nai acknowledges financial support from the Swiss National Science Foundation (Grant P300P1_161163), and logistic support from the Electoral Integrity Project at the University of Sydney and Harvard; thank you in particular to Pippa Norris. We also acknowledge the generous institutional and financial support from the Amsterdam School of Communication Research (ASCoR) at the University of Amsterdam. ORCID iD Alessandro Nai https://orcid.org/0000-0001-7303-2693 Supplemental material Supplemental material for this article is available online. All data and materials, including the full Appendix, are available for replication at the following Open Science Foundation (OSF) repository: https://osf.io/4vhea/ Notes 1. As discussed in Nai (2018) we define an expert as a scholar who has worked and/or published on the country’s electoral politics, political communication (including political journalism) and/or electoral 694 International Political Science Review 43(5) behaviour, or related disciplines. Expertise is established by existing relevant academic publications (including conference papers), membership of a relevant research group, professional network, or organ- ized section of such a group, and/or explicit self-assessed expertise in professional webpage. 2. For example, ‘I care about people’, ‘You cannot trust my opponent’, ‘My opponent is dishonest and corrupt’. See Nai (2018) for a discussion about how experts rated these vignettes in a comparative perspective. 3. https://www.idea.int/data-tools/data/gender-quotas References Ansolabehere, Stephen and Shanto Iyengar (1995) Going Negative: How Attack Ads Shrink and Polarize the Electorate. New York: Free Press. Benoit, Kenneth and Michael Laver (2007) Estimating Party Policy Positions: Comparing expert surveys and hand-coded content analysis. Electoral Studies, 26(1): 90–107. Brown, Adam R (2011) Wikipedia as a Data Source for Political Scientists: Accuracy and completeness of coverage. Political Science and Politics, 44(2): 339–343. Budge, Ian (2000) Expert Judgements of Party Policy Positions: Uses and limitations in political research. European Journal of Political Research, 37(1): 103–113. Cappella, Joseph N and Kathleen Hall Jamieson (1997) Spiral of Cynicism: The Press and the Public Good. New York: Oxford University Press. Curini, Luigi (2010) Experts’ Political Preferences and Their Impact on Ideological Bias. Party Politics, 16(3): 299–321. Damore, David F (2002) Candidate Strategy and the Decision to Go Negative. Political Research Quarterly, 55(3): 669–685. Da Silveira, Bernardo S and João MP De Mello (2011) Campaign Advertising and Election Outcomes: Quasi- natural experiment evidence from gubernatorial elections in Brazil. The Review of Economic Studies, 78(2): 590–612. Dinzes, Deborah, Michael D Cozzens and George G Manross (1994) The Role of Gender in ‘Attack Ads.’ Revisiting negative political advertising. Communication Research Reports, 11(1): 67–75. Dolezal, Martin, Laurenz Ennser-Jedenastik and Wolfgang C Müller (2015) Who will Attack the Competitors? How political parties resolve strategic and collective action dilemmas in negative campaigning. Party Politics, 23(6): 666–679. Duckitt, John (2006) Differential Effects of Right Wing Authoritarianism and Social Dominance Orientation on Outgroup Attitudes and Their Mediation by Threat From and Competitiveness to Outgroups. Personality and Social Psychology Bulletin, 32(5): 684–696. Eagly, Alice H, Wendy Wood and Mary C Johannesen-Schmidt (2004) Social Role Theory of Sex Differences and Similarities: Implications for the partner preferences of women and men. In Alice H Eagly, Anne E Beall and Robert J Sternberg (eds) The Psychology of Gender. New York: Guilford Press, 269–295. Elmelund-Præstekær, Christian (2010) Beyond American Negativity: Toward a general understanding of the determinants of negative campaigning. European Political Science Review, 2(1): 137–156. Elmelund-Præstekær, Christian and Helle Mølgaard Svensson (2014) Ebbs and Flows of Negative Campaigning: A longitudinal study of the influence of contextual factors on Danish campaign rhetoric. European Journal of Communication, 29(2): 230–239. Ennser-Jedenastik, Laurenz, Martin Dolezal and Wolfgang C Müller (2017) Gender Differences in Negative Campaigning: The impact of party environments. Politics and Gender, 13(1): 81–106. Fox, Richard Logan (1997) Gender Dynamics in Congressional Elections. Thousand Oaks, CA: Sage. Franzmann, Simon and André Kaiser (2006) Locating Political Parties in Policy Space. Party Politics, 12(2): 163–188. Geer, John G (2006) In Defense of Negativity: Attack Ads in Presidential Campaigns. Chicago: University of Chicago Press. Gerstlé, Jacques and Alessandro Nai (2019) Negativity, emotionality and populist rhetoric in election cam- paigns worldwide, and their effects on media attention and electoral success. European Journal of Communication, 34(4): 410–444. Valli and Nai 695 Goldstein, Ken and Paul Freedman (2002) Campaign Advertising and Voter Turnout: New evidence for a stimulation effect. The Journal of Politics, 64(3): 721–740. Grossmann, Matt (2012) What (or Who) Makes Campaigns Negative? American Review of Politics, 33: 1–22. Haselmayer, Martin (2019) Negative Campaigning and Its Consequences: A review and a look ahead. French Politics, 17(3): 355–372. Hooghe, Liesbet, Ryan Bakker, Anna Brigevich, Catherine De Vries, Erica Edwards, Gary Marks, Jan Rovny, Marco Steenbergen and Milada Vachudova (2010) Reliability and Validity of the 2002 and 2006 Chapel Hill Expert Surveys on Party Positioning. European Journal of Political Research, 49(5): 687–703. Hopmann, David Nicolas, Claes H de Vreese and Albæk Erik (2011) Incumbency Bonus in Election News Coverage Explained: The logics of political power and the media market. Journal of Communication, 61(2): 264–282. Huber, John and Ronald Inglehart (1995) Expert Interpretations of Party Space and Party Locations in 42 Societies. Party Politics, 1(1): 73–111. Kahn, Kim Fridkin (1993) Gender Differences in Campaign Messages: The political advertisements of men and women candidates for US Senate. Political Research Quarterly, 46(3): 481–502. Kahn, Kim Fridkin and Patrick J Kenney (2004) No Holds Barred: Negativity in U.S. Senate Campaigns. Upper Saddle River, NJ: Pearson Prentice Hall. King, Gary, Christopher JL Murray, Joshua A Salomon and Ajay Tandon (2004) Enhancing the Validity and Cross-Cultural Comparability of Measurement in Survey Research. American Political Science Review, 98(1): 191–207. Laakso, Markku and Rein Taagepera (1979) ‘Effective’ Number of Parties. Comparative Political Studies, 12(1): 3–27. Lau, Richard R and Gerald M Pomper (2001) Negative Campaigning by US Senate Candidates. Party Politics, 7(1): 69–87. Lijphart, Arend (1999) Patterns of Democracy: Government Forms and Performance in Thirty-Six Countries, New Haven: Yale University Press. Lipsitz, Keena and John G Geer (2017) Rethinking the Concept of Negativity: An empirical approach. Political Research Quarterly, 70(3): 577–589. Maier, Jürgen and Carolin Jansen (2015) When Do Candidates Attack in Election Campaigns? Exploring the determinants of negative candidate messages in German televised debates. Party Politics, 23(5): 549–559. Maier, Jürgen and Alessandro Nai (2020) Roaring Candidates in the Spotlight. Campaign negativity, emo- tions, and media coverage in 107 national elections. The International Journal of Press/Politics. Epub ahead of print 2020. doi: 10.1177/1940161220919093 Nai, Alessandro (2018) Going Negative, Worldwide: Towards a general understanding of determinants and targets of negative campaigning. Government and Opposition, 55(3): 430–455. Nai, Alessandro and Jürgen Maier (2018) Perceived Personality and Campaign Style of Hillary Clinton and Donald Trump. Personality and Individual Differences, 121: 80–83. Nai, Alessandro, Ferran Martínez i Coma and Jürgen Maier (2019) Donald Trump, Populism, and the Age of Extremes: Comparing the personality traits and campaigning style of Trump and other leaders world- wide. Presidential Studies Quarterly, 49(3): 609–643. Nai Alessandro and Annemarie Walter (eds) (2015) New Perspectives on Negative Campaigning. Why Attack Politics Matters. Colchester: ECPR Press. Norris, Pippa (1996) Women Politicians: Transforming Westminster? Parliamentary Affairs, 49(1): 89–102. Panagopoulos, Costas (2004) Boy Talk/Girl Talk. Women and Politics, 26(3–4): 131–155. Papp, Zsófia and Veronika Patkós (2019) The Macro-Level Driving Factors of Negative Campaigning in Europe. The International Journal of Press/Politics, 24(1): 27–48. Polk, Jonathan, Jan Rovny, Ryan Bakker, Erica Edwards, Liesbet Hooghe, Seth Jolly, Jelle Koedam, Filip Kostelka, Gary Marks, Gijs Schumacher, Marco Steenbergen, Milada Vachudova and Marko Zilovic (2017) Explaining the Salience of Anti-elitism and Reducing Political Corruption for Political Parties in Europe with the 2014 Chapel Hill Expert Survey Data. Research and Politics, 4(1): 1–9. Ridout, Travis N and Annemarie S Walter (2015) Party System Change and Negative Campaigning in New Zealand. Party Politics, 21(6): 982–992. 696 International Political Science Review 43(5) Roese, Neal J and Gerald N Sande (1993) Backlash Effects in Attack Politics. Journal of Applied Social Psychology, 23(8): 632–653. Rohrschneider, Robert and Stephen Whitefield (2009) Understanding Cleavages in Party Systems. Comparative Political Studies, 42(2): 280–313. Russmann, Uta (2017) Negative Campaigning in Party-Controlled Communication Channels: Party com- munication strategies in campaign posters, newspaper advertisement, and press releases during the 2008 Austrian National Election Campaign. Journal of Political Marketing, 16(2): 95–117. Sigelman, Lee and Mark Kugler (2003) Why Is Research on the Effects of Negative Campaigning So Inconclusive? Understanding citizens’ perceptions of negativity. The Journal of Politics, 65(1): 142–160. Stevens, Daniel (2009) Elements of Negativity: Volume and proportion in exposure to negative advertising. Political Behavior, 31(3): 429–454. Studlar, Donley T and Ian McAllister (2002) Does a Critical Mass Exist? A comparative analysis of women’s legislative representation since 1950. European Journal of Political Research, 41(2): 233–253. Thames, Frank C and Margaret S Williams (2010) Incentives for Personal Votes and Women’s Representation in Legislatures. Comparative Political Studies, 43(12): 1575–1600. Trent, Judith S and Robert V Friedenberg (2008) Political Campaign Communication: Principles and Practices. Lanham, MD: Rowman & Littlefield. Walter, Annemarie S, Wouter Van der Brug and Philip van Praag (2014) When the Stakes Are High. Comparative Political Studies, 47(4): 550–573. Walter, Annemarie S and Cees Van der Eijk (2019) Measures of Campaign Negativity: Comparing approaches and eliminating partisan bias. The International Journal of Press/Politics, 24(3): 363–382. Walter, Annemarie S and Rens Vliegenthart (2010) Negative Campaigning across Different Communication Channels: Different ball games? The International Journal of Press/Politics, 15(4): 441–461. Author biographies Chiara Valli completed a Research Masters at the Department of Communication Science at the University of Amsterdam where she specialized in political communication. Her research interests lie in the area of persua- sive communication, campaigning, political psychology, and electoral behaviour. From September 2020, she works as a research assistant and PhD student at the Institute of Communication and Media Studies, University of Bern. Alessandro Nai is an assistant professor of Political Communication and Journalism at the Department of Communication Science, University of Amsterdam. His research broadly focuses on political communica- tion, voting behaviour, political psychology, and campaigning effects. His recent work appeared in, among others, Political Psychology, European Journal of Political Research, West European Politics, International Journal of Press/Politics, European Journal of Communication, Government & Opposition, Personality and Individual Differences, Journal of Elections, Public Opinion & Parties, Political Studies, and Electoral Studies. He recently co-edited the volumes New Perspectives on Negative Campaigning: Why Attack Politics Matters (ECPR Press, 2015, with Annemarie S. Walter) and Election Watchdogs (Oxford University Press, 2017, with Pippa Norris). He is currently Associate Editor of the Journal of Social and Political Psychology.