When “Following” the Leader Inspires Action: Individuals’ Receptivity to Discursive Frame Elements on Social Media

ABSTRACT

How political actors convey information–that is, the discursive frames they use–can alter individuals’ attitudes, preferences, and behaviors, especially during campaigns. Although scholars have shown that discursive frames using populist rhetoric evoke particularly strong reactions, we do not yet know how the individual elements that make up the populist frame, like anti-elitism or pro-people, fare relative to other ways of seeing the political world or what kinds of messages engage individuals beyond populist ones. In this paper, I evaluate the effectiveness of thirteen frame elements in stimulating online engagement. I derive frame elements not only from populism, but from competing discursive frames, including technocracy, pluralism, and neutral rhetoric. I find support for my argument that frame elements using populist rhetoric, are less cognitively demanding, and evoke emotions produce observable framing effects. To test my argument, I evaluate campaign Tweets from 18 actors in Brazil, Mexico, Colombia, Italy, and Spain (N = 1,577). My findings affirm the existence of framing effects in campaigns while identifying the generalizable content of the messages that produce these framing effects, as well as identifying the type of message content that most effectively competes with populist frame elements in this sample.

KEYWORDS:

• Populism
• online engagement
• participation
• frames
• social media
• campaigns

Acknowledgments

I am grateful for the enormously helpful comments on earlier drafts of this article by Larry Bartels, Kirk Hawkins, Jon Hiskey, Mitch Seligson, and Liz Zechmeister; the Political Communication editor and two anonymous journal reviewers; and the Liz Zechmeister Lab participants at Vanderbilt University. This work was supported by the exceptional research assistance of Bianca Herlory, Stacy Horton, Miriam Mars, and Joy Stewart, and funding from Vanderbilt University’s Russell G. Hamilton Graduate Leadership Development Institute.

Disclosure Statement

No potential conflict of interest was reported by the author.

Data Availability Statement

The data described in this article are openly available in the Open Science Framework at https://doi.org/10.17605/OSF.IO/TPA6U.

Open Scholarship

This article has earned the Center for Open Science badge for Open Data. The data are openly accessible at https://doi.org/10.17605/OSF.IO/TPA6U.

Supplementary Material

Supplemental data for this article can be accessed on the publisher’s website at https://doi.org/10.1080/10584609.2020.1829761.

Notes

1. These discursive frames reflect politicians’ understanding of the relationship between the people and the elites. Other conceptions are possible, such as issue positions, but extend beyond the scope of this study.

2. Existing studies do not utilize technocracy and elitism as separate categories. For example, Akkerman et al. (2014) measure elitism in surveys not only as a moralistic distinction between “the people” and the elite (Mudde & Kaltwasser, 2013, p. 152), a conception in line with elitism, but also as important business leaders or independent experts, which is very much in line with technocracy.

3. I refer to Manichean discourse as dispositional blame attribution, but the underlying sentiment behind these two concepts, as they are used here, is similar. In both of these frames, one group is blaming another group, implicitly creating an “us versus them” dichotomy.

4. Existing studies focus on anywhere from four to six frame elements, but in earlier iterations of this project, I found that other frames either were biased toward right-wing populism, occurred extremely infrequently (less than 1% of the sample), or were not distinct from the three final frame elements used in the study. For example, Casero-Ripollés et al. proposes a single frame for “the people,” Engesser et al. (2017) two frames (“sovereignty of the people” and “advocacy for the people”), and Cranmer (2011) three frames (“advocacy for,” “accountability to,” and “the legitimacy of the people”). All three articles include a frame for “attacking the elite” and “exclusion of outgroups.” Engesser et al. proposes a frame that invokes “the heartland;” Casero-Ripollés et al. a frame for “narrative of a crisis,” and Cranmer a frame for “homogeneity or threat.”

5. I also find that the actors that regularly using populist communication on Twitter significantly overlap with the actors that experts identify as “populist,” including the four datasets outlined in footnote 10; see Appendix B.1 for additional information.

6. Social media users in general and Twitter users specifically tend to be whiter, more educated, younger, and male (Lupu et al., 2019). In particular, scholars have started pointing out the differences between social media users who actively post/receive political content and those that use social media for other purposes, finding that the former group is more interested in politics, has higher political knowledge, and is more likely to vote than the overall population (Bode & Dalrymple, 2016; see also Wojcik, 2019). However, representativeness is not necessarily a concern unless one tries to generalize beyond the population of interest. A potentially greater threat to inference is if Twitter users are more likely to engage with populist messages than other kinds of messages, thus biasing the results. While more research is needed, previous research has shown that populist supporters tend to be less educated and more economically insecure (Elchardus & Spruyt, 2016; Inglehart & Norris, 2016; Spruyt et al., 2016), in stark contrast to the traits that characterize Twitter users.

7. Framing effects can be attenuated when individuals are exposed to multiple competing messages (Chong & Druckman, 2007a).

8. Several scholars have demonstrated that individual predispositions are strongly associated with framing effects (see, e.g., Chong & Druckman, 2007a; Kam & Simas, 2010).

9. Why framing effects occur is widely debated. The accessibility perspective (Chong & Druckman, 2007b; Druckman, 2007, 2011; Zaller, 1992) argues that frames are more likely to produce framing effects when they are available, accessible, and applicable to individuals. An alternative perspective is that framing effects occur because they are more important than other considerations (see, e.g., Nelson et al., 1997). Intuitively, this approach suggests that individuals weigh more important considerations differently than less important ones, where framing effects occur when the frame aligns with an important consideration. A related perspective suggests that framing effects occur when frames resonate with individuals (McDonnell et al., 2017; Snow & Benford, 1988), arguing that resonance (which I consider a proxy for framing effects) occur when a narrative structure which diagnoses a problem, prescribes a solution, and contains a call to action. Outside of an experimental design, it is difficult to ascertain which of these theoretical mechanisms is at play. For the purposes of my analysis, I focus predominantly on what makes for a strong frame among the four discursive strategies (and subsequent thirteen frames) I investigate rather than why strong frames produce framing effects. To make my argument, I draw most heavily on the accessibility perspective as that has the most associated information on frame strength.

10. The language variable is trichotomous: positive, neutral, or negative and is based on the RAs’ interpretation of the overall tone of the message. See Appendix A.3 for additional information.

11. The Tweets span the period of late December 2017 through April 2019 (16 months).

12. Costa Rica satisfied the populist criteria, but I opted not to include this case due to the particular combination of populism and evangelism that the populist actor (Fabricio Alvarado) displayed, which strongly limited the generalizability of this case. El Salvador had a strongly anti-elite candidate (Nayib Bukele), but existing accounts did not support this candidate as being populist.

13. Italy: 5.46% as of March 2018; Mexico: 19.45% in August 2018 (this number dropped precipitously post-election, and is at 7.47% as of August 2019); Brazil: 5.48% in October 2018; Colombia: 6.8% in June 2018; Spain: 6.2% in April 2019. Data from the country pages at https://gs.statcounter.com/social-media-stats/.

14. Two parties did not meet the minimum number of Tweets. FI, and MS5. FI was sampled at 80 Tweets (the non-populist amount) as the existing classification information available at the time the study was conducted indicated that FI was not-populist. Since then, later datasets indicated that FI was considered a populist by a majority of indicators, thus they are coded as populist here. For FI, I included Tweets where the party retweeted the party leader’s (Silvio Berlusconi) Tweets. Though this was not done for other cases, it is consistent with other parties who, instead of re-Tweeting leader’s Tweets (as FI did), simply use the same Tweet between candidate. MS5 is sampled at 77 Tweets total, representing their entire universe of Tweets during the campaign. I also collected separate Tweets from the party leader for a robustness check, which is why I did not combine the MS5 with Luigi Di Maio’s Tweets.

15. Official campaign periods are hard to pin down in many countries. I selected campaign dates that reflected the official kickoff of the campaign marked by the first major campaign event, and ended either the day before the election, or a few days before in certain cases that observe a few days of non-campaigning (aka “reflection periods”). The campaign periods covered in this analysis are: 1) Italy: 12/27/2017 (when Parliament was dissolved) – 3/3/2018; 2) Colombia: 3/11/2018 (when primaries were held) – 6/16/2018 (excluding the 1^st round election day, 5/27/2018); 3). Mexico: 3/30/2018 – 6/27/2018; 4) Brazil: 7/20/2018 (registration for parties’ candidates opened) – 10/27/208 (excluding the 1^st round election day, 10/7/2018); 5) Spain: 2/15/2019 (snap elections were called) – 4/26/2019. Two candidates, Ciro Gomez of Brazil and Sergio Fajardo of Colombia did not make it to the 2^nd round; thus, their campaign period ended the day before the 1^st round election in these countries.

16. I classify who is and is not a populist according to four existing datasets: three expert surveys (the Chapel Hill Expert Survey – CHES, the Negative Campaigning Comparative Expert Survey – NEGex, and the Global Party Survey – GPS) and one based on speech analysis (the Global Populism Database – GPD). I used four datasets to ensure external validity as well as adequate coverage of the actors in this sample. I classify candidates as “populist” if the majority of these datasets considered the candidates to be somewhat or very populist and “non-populist” otherwise. Full details are available in Appendix A.3. I go against the existing data in only one instance: FI of Italy. I do so because I evaluate FI as a party, not the party leader (Silvio Berlusconi) or as a coalition. While existing accounts generally view Berlusconi as populist, FI is not necessarily a populist party. Bobba and Roncarolo (2018), for example, classify only 8.1% of FI’s Tweets as populist (making the “not populist” designation more appropriate). I also include Cs of Spain as a populist party – this was the only actor in the sample that had an even split of populist/non-populist in the existing data sets. However, my data indicate that Cs falls on the lower end of populism, thus I opt to include them as populist.

17. I log-transformed both likes and re-Tweets due to a positive skew toward lower values – 50% of “likes” are below 800 with an average of 4,055 and a high value of 91,000, while the average number of re-Tweets in the sample is approximately 1,500 despite a high value of 21,000.

18. See Appendix D.3 for an explanation of the subsample case selection, and Appendices B.5 and B.6 for the results.

19. About 1/3 of the Tweets in this sample contained relevant media that may have (though did not necessarily) revealed the speaker’s identity.

20. See Appendix F.1 for a detailed discussion of the coding procedures.

21. The model tests whether the average number of likes/re-Tweets for each frame is zero. Given that the average number of likes is 4,055 and the number of re-Tweets 1,544, it is unsurprising that every frame is statistically differentiable from 0. In contrast, Model 2 tests whether the average number of likes/re-Tweets for each frame is different than that of the information frame.

22. AMLO is chosen as the reference candidate because he attracts the most re-Tweets and second most likes.

23. Coupled with the low prevalence with which actors used these frame elements, there seems to be agreement between actors opting not to use pluralistic messages and, when they do, individuals not engaging with these elements as often. One possible explanation for this finding is that pluralist elements are more conducive to forming a government (in the case of parliamentary systems) and governing more broadly, with less utility during a campaign when each actor is attempting to maximize their individual support at the polls. This explanation is especially applicable to the “emphasis on compromise” element, which specifically refers to compromise in the political sphere. However, the elements of “inclusivity,” “legalistic view of democracy,” and “situational blame attribution” are less obviously temporally bound. An alternate possibility is one that depends on context: in Appendix D, which breaks down the findings by region, pluralist messages actually perform slightly better in Europe than they do in Latin America. Thus, it may not be a temporal story but one of institutional differences, with parliamentary systems more favorable to pluralism than presidential ones.

24. Relative magnitudes are calculated using the formula 100[e^β – 1] to interpret the logged dependent variable as a percentage difference compared to the base category, populism.

25. This finding also holds when I test frames capitalizing on positive emotions or negative emotions separately (not pooled) against the low-fit category. It also holds if, instead of focusing on particular frame elements, I focus on the predominant tone of the individual Tweets themselves (whether they are negative, neutral, or positive). The group means of Tweets with emotive content are significantly higher than the group means for neutral Tweets for both likes and re-Tweets, significant at p<.01. I present the frames results at it is more consistent with the objective of my hypotheses – to identify the particular aspects of frames that are associated with higher engagement.

Additional information

Funding

This work was supported by the Vanderbilt University Russell G. Hamilton Graduate Leadership Development Institute under two grants awarded in 2019.

Notes on contributors

Kaitlen J. Cassell

Kaitlen J.Cassell received her Ph.D. in Political Science from Vanderbilt University in 2020. Her research centers on issues of political behavior and communication in Latin America and Europe, with a specific emphasis on elites’ online communication strategies and how individuals engage with these strategies. Her methodological specializations include text analysis, for which she has developed a method to measure different kinds of discourse in social media posts, and survey research methods. More information on her work can be found at www.kaitlencassell.com.