Populism and Resistance to Science: Assessing Antivaccine Sentiments in Chile During the COVID-19 Pandemic

Abstract

Introduction

Inquiries into antivaccine sentiments are as old (Kaufman, 1967; Porter & Porter, 1988) as they are extensive (Larson, 2018; Larson, Jarrett, Eckersberger, Smith & Paterson, 2014; Salmon, Dudley, Glanz, & Omer, 2015), but only recently they have been closely tied to populist attitudes in the political arena. Since populism entails a profound distrust of elites and experts, as well as a general anti-establishment attitude that targets public officials and scientific authorities, it has been viewed as a proxy for scientific skepticism in general (Brown, 2014; Huber, Greussing, & Eberl, 2022; Mede & Schäfer, 2020; Merkley, 2020; Staerklé, Cavallaro, Cortijos‐Bernabeu, & Bonny, 2022; Zapp, 2022) and for vaccine hesitancy in particular (Kennedy, 2019; Speed & Mannion, 2020; Tomasi, 2020; Żuk & Żuk, 2020). Other studies report an increase in antivaccine beliefs with the rise of social media, as they are reinforced by interactions with similarly minded people in echo chambers and by the proliferation of fake news (English, 2017; Kata, 2010; Maci, 2019). While the bulk of this literature is pre-coronavirus disease 2019 (COVID-19), the latest pandemic has provided a vast number of additions to this multidisciplinary research, at the intersection of antivaccine sentiments, the global phenomenon of contemporary populism, and the study of online interactions (Germani, & Biller-Andorno, 2021; Hameleers, & Van der Meer, 2021; Kohler & Koinig, 2023; Rietdijk, 2021; Sorell, & Butler, 2022; Wang & Catalano, 2022).

This paper aims to contribute to the literature that connects populism and anti-science attitudes by investigating online antivaccine conversations in Chile, thus helping decentralize a research field that has been mostly centered in Europe and the US, by focusing on a Latin American country.

Chile provides a particularly interesting case because of two unique features. On one hand, it boasts one of the most successful COVID-19 vaccination campaigns worldwide, due to a combination of factors, including the timely availability of vaccines, the country’s long history of mandatory immunization programs (which helped create a favorable vaccine culture), efficient coordination between the central health system and local authorities (Aguilera, Mundt, Araos, & Weitzel, 2021; Castillo, Dintrans, & Maddaleno, 2021; Luna et al., 2023), and a well-established capability of the Chilean State to deliver on health outcomes in general (Brieba, 2018). On the other hand, Chile has been usually singled out as fairly resistant to the lure of the otherwise ubiquitous Latin American populism, mainly due to its institutional strengths, sane economic policies, and the stability of its party system (Mainwaring & Scully, 1995; Navia, 2003; Walker, 2008). Note, however, that recent scholarship points to the rise of a “populist radical right” (PRR) in Chile, led by José Antonio Kast and his Partido Republicano (Campos, 2021; Díaz, Rovira Kaltwasser, & Zanotti, 2023; Kestler, 2022; Zanotti & Roberts, 2021), borrowing from Mudde’s characterization of a PRR in Europe (Mudde, 2007). Although he did not downplay the risks of COVID-19 as some populists (such as Jair Bolsonaro) in other countries, Kast did entertain populist sentiments during the pandemic to criticize government policies.

Against this contextual background, in this work we disaggregated and categorized distinctive grounds for vaccine skepticism and the resistance to the Chilean COVID-19 official vaccination scheme by analyzing online posts and conversations related to the vaccine on a massive social media platform such as Twitter, during a six-month period in 2021 (given that Twitter changed its name to X only in mid-2023, we will continue to refer to it as Twitter to match the term used during our analysis period). This dataset provided an excellent sample of the different types of antivaccine arguments voiced in public debates. To be sure, Internet penetration and social media usage in Chile is very high, similar to the US or Singapore, and grew from 82% to 92% during the pandemic. Regarding Twitter, in 2021 there were 2.9 million users in the country, corresponding to 18% of the total adult Chilean population (Datareportal, 2022). Although it is known not to represent proportionally the various population segments, Twitter (nowadays X) remains the only massive open network where opinions are publicly voiced and can be systematically collected along with information on their origin and reach, thus allowing their characterization. Twitter interactions in Chile related to vaccination were also significantly higher during the vaccine rollout in 2021: using the same keywords as in the current study, the number of tweets collected during the same period the year after, in 2022, was reduced by a factor of 30 and the number of accounts that produced them, by a factor of 10 (Social Listening Lab, Sol-UC, 2023). These observations, combined with the significant restrictions on mobility and social gatherings that were imposed in the country during the pandemic, allow us to consider the captured online arguments as a reasonable proxy for the types of antivaccine objections raised by the Chilean population.

Although resistance to vaccines has been related to a variety of phenomena, such as religious beliefs (Grabenstein, 2013), a preference for alternative medicine (Attwell, Ward, Meyer, Rokkas, & Leask, 2018), or plain misinformation (Garett & Young, 2021; Orsini, Bianucci, Galassi, Lippi, & Martini, 2022), here we evaluate it only against the kind of science-skepticism that is grounded in populist discourses and attitudes. More specifically, we analyze antivaccine arguments in a framework that disaggregates the populist resistance to science into three main categories: moral objections, democratic objections, and epistemic objections (Bellolio, 2024). After testing this theoretical framework by means of empirical observation and data analysis of the antivaccine discourse in Chilean social media, we found that these three objections are not only identifiable but are also represented in similar proportions: some tweets denounce the vaccination scheme as a conspiracy led by global selfish elites; others complain that the authorities, following scientific advice but with no democratic warrant, are taking away the citizen’s freedom to decide whether to be inoculated; and yet others point to a broad distrust of the scientific procedure, arguing either that it produced the vaccine in a timeframe that was too short or that its results contradict personal evidence.

Theory

Populism is widely believed to be anti-expertise, as it praises commonsense and folk wisdom (Hawkins, 2010; Mudde & Rovira, 2017; Müller, 2016; Taggart, 2020). Given that science is a paradigmatic form of expertise and technical knowledge, it is commonly inferred that populism must also be anti-science. In broad strokes, the literature stresses the anti-elitist and people-centric core of the populist mentality, which includes: a “popular suspicion of organized power,” such as the “power of science” (Brown, 2014, p. 129); an anti-intellectual stance of “resistance to expert consensus” (Merkley, 2020, p. 24); a general distrust of expert knowledge “in scientific and health-related issues” (Tomasi, 2020, p. 223); a politically motivated “opposition against the system and the elite,” in which “scientists are inevitably part of a highly educated cultural elite” (Rekker, 2021, p. 8); a typically anti-scientific rhetoric in the “post-truth era” prompted by populist politics (Zapp, 2022); the growing role of “politicized common sense in motivating…opposition to measures and policies based on scientific expertise” in the face of global crises (Staerklé, Cavallaro, Cortijos‐Bernabeu, & Bonny, 2022, p. 913); and the conspiratorial belief that selfish elites are “up to something” when making scientific claims (Castanho Silva, Vegetti, & Littvay, 2017); among others. Although some authors still argue that hostility to science is not an “intrinsic part of the toolkit of political populism in general” (Szabados, 2019, p. 229), or that populism is only secondarily related to anti-scientific views (Müller, 2020), a large part of the literature shows that “populism is a consistent predictor of declining support for science across all models, regardless of the political system” (Zapp, 2022). This relationship has thus led to conceptualizing a phenomenon described as “science-related populism,” which corresponds to:

“a set of ideas suggesting an antagonism between an (allegedly) virtuous ordinary people and an (allegedly) unvirtuous academic elite—an antagonism that is due to the elite illegitimately claiming and the people legitimately demanding science-related decision-making sovereignty and truth-speaking sovereignty” (Mede & Schäfer, 2020; p. 484; for an empirical testing of the concept, also see Mede, Schäfer, & Füchslin, 2021).

In this context, recent research has consistently shown that populist actors and voters worldwide display a rather skeptical or plainly denialist position toward the scientific consensus on the climate crisis (Brown, 2014; Fiorino, 2022; Huber, 2020; Huber, Greussing, & Eberl, 2022; Lockwood, 2018; Machin, Ruser, & Andrian, 2017; Norton, 2016; Staerklé, Cavallaro, Cortijos‐Bernabeu, & Bonny, 2022), although the COVID-19 experience has complicated this analysis. Indeed, in the early stages of the pandemic, many commentators rushed to decree the end of the populist era: as populism misprizes scientific expertise, and the pandemic would make us all revalue scientific expertise, it was argued that populism was on its way out (Betz, 2020; English, 2020; Mead, 2020; Patman, 2020; Rachman, 2020; Wright & Campbell, 2020). But the fact is that populists in different governments and oppositions reacted to the COVID-19 health crisis in very different ways: while some initially downplayed the health threat, others applied highly restrictive measures from the outset; while some lost elections following the pandemic, others came out strengthened; while some found the opportunity to deploy their nationalist sentiments with “rally around the flag” discourses amidst what they portrayed as an imported crisis, others reinforced a libertarian discourse and demanded economic openness (see, for example, Müller, 2020; Ringe, & Rennó 2022; Rovira & Taggart, 2022; Stavrakakis & Katsampekis, 2020; Urbinati, 2020). In either case, it seems safe to infer that populism did not die with the pandemic.

Recently, some theoretical work has been advanced to frame the relationship between populism and science in more systematical terms. According to Bellolio (2024), populist actors can typically wield three distinctive objections to science:

• (a) a moral objection that is aimed at scientists who have been allegedly corrupted by alien interests or malicious intents. It is moral to the extent that it singles out an enemy of the common and decent people, portrayed as a powerful villain, who is often viewed as participating in secret plots and machinations to carry out an egoistic agenda, thus connecting to the conspiratorial facet of populism (see also Bergmann & Butter, 2020; Castanho Silva, Vegetti, & Littvay, 2017; Pirro & Taggart, 2022).

• (b) a democratic objection that is raised against scientific experts who purportedly aim to rule by shortcutting the popular will. It follows from the conceptualization of populism as a democratic reaction to an increasingly undemocratic liberalism that has transferred decision-making from the people to non-elected bodies governed by technocratic rationality (for a conceptualization of populism as illiberal democracy, see Eatwell & Goodwin, 2018; Mudde, 2015; Mounk, 2018; Pappas, 2019; for a description of the “essential tension” between scientific facts and democratic politics, see Arendt, 2006; Meyer, 2023; Swyngedouw, 2022; Guston, 1993; for a defense of the populist challenge to elitist expertise, see ).

• (c) an epistemic objection that targets scientific rationality, which is said to be inferior to commonsensical reasoning when describing the reality of the world. It is epistemic in the sense that it contests the way in which established mainstream science justifies its knowledge-claims, which are perceived to be artificially sophisticated and divorced from real people’s everyday experiences, emotions, and personal evidence (see also Harsin, 2018; Mede, 2023; Mede & Schäfer, 2020; Saurette & Gunster, 2011; Van Zoonen, 2012).

These objections are all described as populists insofar “they all speak to the core feature of populism… which is the people versus elites cleavage” (Bellolio, 2024, p. 497). Although it is almost a cliché to say that populism is a disputed concept, most scholars agree that its core feature is the adversarial divide between “the people” and “the elite” (Moffitt, 2020; Rooduijn, 2014). As summarized by Bellolio,

“The moral objection targets scientists as members of an elite in cahoots with alien powers; the democratic objection targets an unelected elite that seeks to undermine the people’s rule; and the epistemic objection questions that the standard to validate knowledge-claims reflects a complex and detached-from-ordinary- experience rationality instead of folk rationality” (2022, p. 12).

These populist grounds to distrust mainstream science are thought as general categories: they can describe skepticism or plain hostility toward the scientific consensus over anthropogenic climate change, genetically modified food, nuclear power, or any other area of public controversy, even if there is no real controversy within the relevant scientific communities.

In what follows below, we will be specifically interested in the way in which this threefold framework of populist objections to science may be able to describe and organize the antivaccine sentiment during the COVID-19 pandemic. Even more specifically, the question we aim to address is whether these types of objections can adequately reflect the range of arguments aired on social media against the vaccine rollout in Chile.

What do we expect the populist antivaccine arguments to look like? The moral objection can be assumed to target scientific elites and public officials who support their advice as members of a global conspiracy, maybe run by an international organization such as the UN, or by a recurrent villain such as George Soros, perhaps fueled by big pharmaceutical companies seeking profit. We can also expect references to a “plandemic”—the idea that the whole COVID-19 crisis was intentionally created with fraudulent motives- (Eberl, Huber, & Greussing, 2021; Kearney, Chiang, & Massey, 2020), as well as paranoid theories about 5G and the implantation of chips for social control (Ahmed, Vidal-Alaball, Downing, & Seguí, 2020). All these narratives are moral in nature since they are deeply suspicious about the intentions behind the vaccination, which are presumed corrupt or deviant.

In turn, the democratic objection should appear in discourses that challenge the scientists’ influence outside their laboratories and hospitals, that is, opposing the extension of their epistemic authority into political authority. Although COVID-19 vaccination was not mandatory in Chile, vaccine passports or “green passes” were required to access to most venues and events. In this context, the democratic objection might look like a libertarian claim, one asserting that we are not children in a paternalistic state and should instead be treated as adult citizens who can bear the risks of our freedom (see Butler & Sorell, 2022).

Last but not least, the epistemic objection of a populist mindset can be found in discourses that call into question the proficiency of the COVID-19 vaccine, that is, its safety and effectiveness, placing instead more confidence in personal experiences and anecdotal evidence (such as the case of a family member who was vaccinated and got sick or had terrible side effects), as well as in alternative—nonofficial—sources of information (like a Facebook group linking vaccines to autism or a charismatic doctor speaking on TV). This objection is not directed against science per se, but instead against a way of doing science that is considered flawed or just too risky due to its high uncertainty. As Müller (2020) notes, the indeterminacy and changeability of scientific pronouncements due to the developing knowledge during the COVID-19 pandemic were often used by populist politicians as a license for contestation. Thus, populist objections to the vaccine that we call epistemic can be expected to reflect a broad mistrust of the procedure that established the science employed to react to the virus.

As a final theoretical caveat, we point out that, although the goal of this work is to empirically relate populist attitudes with skeptical or rejectionist views of science, one may wonder whether these populist grounds could also activate pro-scientific feelings, and, in this case, pro-vaccine attitudes. We find this hard to conceive, however, since the vaccination process was, in most cases worldwide and certainly in Chile, promoted, financed, implemented, and monitored by the established powers, that is, by the political and scientific elites. If populism is an essentially “reactive” phenomenon (Taggart, 2000), akin to a “plebeian” resistance (Vergara, 2020), it seems difficult to think of a populist motivation to favor a process led by national and/or global elites. Here, we thus operate under the theoretical premise that populist attitudes mainly relate to vaccine skepticism, for three different but related anti-elite reasons. Note that, although prima facie the Cuban experience could stand as a counterexample (given that the Communist-ruled island had a successful vaccination process promoting a homegrown vaccine named SOBERANA), at closer look, the discursive framing of the Cuban authorities was more chauvinistic than “plebeian”: they celebrated the national character of the vaccine, whilst the common enemy was always the virus and not big corporations (Soler Mas & Villota Oyarvide, 2020; Yaffe, 2021).

Method

To assess the distribution of online opinions regarding COVID-19 vaccines, we began by collecting all tweets produced in Chile from March 1st until August 31st, 2021, that used any of the following keywords or hashtags: vacunas, vacuna, vacúnense, vacunado, vacunada, vacunación, dosis, Pfizer, Sinovac, CoronaVac, CanSino, AstraZeneca, Johnson & Johnson, efectos secundarios, coágulos, OMS, #YoMeVacuno, and #YoNoMevacuno (in English: vaccines, vaccine, get vaccinated, vaccinated, vaccination, dosage, Pfizer, Sinovac, CoronaVac, CanSino, AstraZeneca, Johnson & Johnson, side effects, blood clots, WHO, #IGetVaccinated, and #IDontGetVaccinated). We also included variations of these terms, such as common misspellings and different forms of capitalization, accentuation, abbreviation, etc. This specific period in 2021 was chosen because it contained a particularly rich conversation about the vaccine. During these six months, the fraction of eligible Chilean population that received at least one dose went from under 20% to over 75% while the number of confirmed COVID infections and deaths almost doubled (MINSAL, 2021). A fast and highly successful vaccination process was therefore accompanied by large increases in cases and deaths, which animated a significant level of debate and sparked strong criticisms by the antivaccine community. The collected dataset thus consisted of 351,573 tweets generated by 59,252 different accounts, which received over three million interactions (the sum of likes, retweets, and replies).¹

An artificial intelligence algorithm was then used to associate each account with a score between 0 and 1 that reflected its expressed views, as detailed by Villegas et al. (2022), where scores close to 0 corresponded to the most pro-vaccine accounts and scores close to 1, to the most antivaccine ones. For the current work, we then selected only the accounts and tweets with scores higher than 0.45, which opposed or resisted vaccination according to the AI model. This resulted in 72,441 tweets generated by 20,293 different accounts, which received 689,125 interactions. In order to further distinguish between mildly antivaccine attitudes and the most extreme ones, we followed Villegas et al. and subdivided this dataset into two groups. Accordingly, we categorized all accounts with scores between 0.45 and 0.71 (more moderate positions) as inhibitors and all accounts with scores between 0.71 and 1 (more radical attitudes) as antivaxxers. This resulted in 66,146 tweets generated by 18,589 inhibitor accounts, which received 641,276 interactions, and to 6,295 tweets generated by 1,704 antivaxxer accounts, which received 47,849 interactions. Overall, this implies that about 34% of the accounts in our dataset opposed or resisted vaccination (i.e., they were classified as inhibitors or antivaxxers), whereas only close to 3% expressed radical antivaccine sentiments (i.e., they were classified as antivaxxers).

We evaluated all the inhibitor and antivaxxer accounts according to the objections described in the Methods section (moral, democratic, epistemic) by carrying out the following procedure. First, we selected at random almost 1000 tweets and classified them by hand, identifying 119 tweets with a moral objection, 121 with democratic, and 145 with epistemic, as well as 595 tweets expressing unclear, neutral, or even moderately pro-vaccine views that had been misidentified by the AI algorithm detailed above.² This analysis allowed us to gain insights into the different types of arguments that are prototypical of each type of objection, which we illustrate in the Results section below. We then used these manually classified tweets as training sets for three new machine learning models (Allaire & Chollet, 2022), each trained to recognize one type of objection, by feeding the tweets with that objection as examples and the rest as counterexamples.³

Finally, the resulting AI algorithms were used to assess all 72,441 tweets in our dataset, associating three scores to each tweet: a moral objection score, a democratic objection score, and an epistemic objection score. We then also computed three scores for each account, given by the mean score per objection category of all its tweets.

To determine the score thresholds above which an account will be characterized as expressing a type of objection, we analyzed the scores given by the trained model to the training dataset. Although most tweets manually identified as presenting each type of objection obtained a higher objection score than the rest for that type of objection, at the lower end of their score values we also found tweets that were not associated with that objection, due to expected imperfections in the AI scoring algorithm. We thus defined the three objection score thresholds at values that yielded less than 15% false positive identifications of tweets that had not been manually associated with the corresponding category in the training set. Using these threshold values, we then classified all the accounts as expressing or not each of the three types of objections (moral, democratic, epistemic) by determining if their mean score in that category was above its corresponding threshold. Note that this determination was based on the objection scores of the accounts, and not of the tweets, since the average over their multiple tweets provided a more reliable measure of the type of arguments they expressed, and because characterizing the accounts allowed us to also investigate the online behavior associated to each type of argument.

Results

Using the methods described above, we identified the accounts in our dataset that opposed or resisted vaccination while expressing moral, democratic, or epistemic objections in their tweets. Overall, we found that almost 76% of all those opposing or resisting vaccination expressed at least one of these three types of objections (i.e., 15,413 accounts of the 20,293 that had been classified as inhibitors or antivaxxers). A direct examination of the remaining 24% of the accounts showed that they often did not articulate a clear objection⁴ or presented objections that were not grounded in populist attitudes, as discussed in the Introduction.

Figure 1 presents the statistics of all the characterized accounts. The series of left-hand panels show the results for our complete dataset of accounts classified as opposing vaccination, including both inhibitors and antivaxxers (which respectively display more moderate or more radical positions, as detailed in Methods). The right-hand panels only include the latter.

Figure 1.

Statistics of Twitter accounts that use moral, democratic, or epistemic objections to vaccination. The left-hand column includes the statistics of all accounts resisting vaccination, including milder (inhibitors) and stronger (antivaxxers) resistance, whereas the right-hand column includes only antivaxxer accounts. A: Total number of accounts that expressed arguments in each category. B: Mean number of tweets resisting vaccination generated by each type of account during the period of study. C: Mean number of interactions (likes, retweets, and replies) received by tweets from each type of account during the period of study.

Open in new tabDownload slide

Panel A displays the number of accounts that used objections of each type. We note in the left-hand panel that all objections are represented in somewhat similar amounts. In the right-hand panel, the fraction that argues with moral objections increases while the fraction that wields epistemic objections decreases, which is consistent with the higher emotional component shared by the moral objection and radical antivaxxers, and with the fact that epistemic objections tend to be less passionate. Note, however, that these numbers will always have a certain level of uncertainty, since determining the type of objection in a short tweet can often be subject to interpretation, even for humans, and AI algorithms tend to misinterpret tweets more than humans. Despite these limitations, we are confident that the rough proportions displayed in Panel A are valid, since they match our direct observations during the manual training set analysis and are similar to the proportions that we obtained in tests using different training sets, AI algorithms, and thresholds.

To make the automatic scoring as objective as possible, we took a deliberately parsimonious approach in the manual classification procedure of our training set by not considering the context of the statements or reading into the intentions of the user behind the account. A good example of this is the use of the term “experimental” to refer to the vaccine in a critical sense. We assumed in our manual classification that experimental, in a literal sense, means that it has not been tested enough. Under our categories, we thus classified its use as an epistemic objection (as it targets the scientific method) unless otherwise implied by the rest of the tweet. Alternatively, if we had considered the context, we could have argued that experimental was also commonly used to imply that someone is using us as “guinea pigs” without any regard for our well-being, probably with hidden purposes, which would imply a rather moral accusation (see Discussion for examples and further details). This latter interpretation would have added another layer of uncertainty in the classification, however, since we cannot access the thoughts of the users behind the tweets.

Panel B presents the corresponding mean number of tweets per account of each type. Here we find that the accounts that express moral objections are more active in promoting their views, since they generated more tweets per user in the analysis period. On the other hand, the accounts that formulate epistemic arguments appear to promote them less than other accounts. When focusing on the tweets by more extreme antivaxx accounts (right-hand panel), we find that both moral and democratic accounts rise to a similar higher level of activity, while epistemic accounts remain at the same level.

Panel C shows the mean number of interactions (the sum of likes, retweets, and replies) per tweet of the accounts in each category, which reflects the rate of response to posts from accounts expressing each type of objection. We find again that accounts that express moral objections elicit the highest levels of interaction and that those expressing epistemic objections produce the lowest. The right-hand panel shows that the situation remains the same when focusing on more extreme antivaxx accounts, although the reaction level differences between categories increase.

Figure 2 presents a Venn diagram showing all accounts classified in one or more objection categories for the inhibitors and antivaxxers (left-hand panel) and for only the antivaxxers (right-hand panel) when considering only the 2,402 users that tweeted at least four times in our database. The left-hand panel shows that a significant fraction of the accounts formulated both moral and democratic arguments in their tweets. When focusing exclusively on antivaxxers, the right-hand panel shows that this fraction grows even larger. By contrast, the figure also shows that the fraction that presented democratic and epistemic, or moral and epistemic, arguments is much lower and that it remains relatively constant when considering only radical antivaxxers.

Figure 2.

Venn diagrams of accounts that produced four or more tweets and expressed one or more populist objections to vaccination. The left-hand panel includes all accounts resisting vaccination; the right-hand panel considers only the more extreme antivaxxer accounts. We consistently observe that many users that tweeted multiple times expressed both moral and democratic objections, while other combinations of opinions were much rarer.

Open in new tabDownload slide

We note that the use of different types of objections by the same users shown in this figure is observed despite the fact that, in our manual analysis of individual tweets, we found almost no posts that simultaneously presented two different types of objections. This lack of tweets that include combinations of objections under the parsimonious classification scheme described in the Methods section can be explained by the limited maximum character count in tweets, which does not easily lend itself to presenting two types of arguments simultaneously. On the other hand, when considering users with more tweets, as in the figure, we find that they often do express different types of arguments throughout their posts. Although the exact amounts in the figure are somewhat subject to the choice of thresholds, as in the results in Figure 1A, we consistently observe that a significant fraction of accounts with multiple tweets presents moral and democratic arguments, in both the inhibitors and antivaxx communities.

Discussion

Before discussing the results, we shall explain the specific arguments that we encountered while categorizing the tweets that were used as training sets for the machine learning models. To do this, we will present below a few examples (translated from Spanish) of paradigmatic tweets for each category (moral, democratic, epistemic), which represent the main narratives.

The following three tweet texts (translated from Spanish) squarely represent examples of the moral objection to the vaccination scheme:

• - “[we] do not lend our bodies to the experiment of the pharmaceutical transnationals of the millionaires of the Davos forum—Bill Gates included—and for the Chinese communist party with its vaccine…”

• - “In fact, Bill Gates’ vaccine was ready before the pandemic...!”

• - “…with this plandemic and the massive vaccination they [the communists] are making a fortune, everything is part of a great plan of the great elites, it is a well-organized scheme!”

In the first tweet, note that the experimental nature of the vaccine is not meant as a critique of uncertainty, which would reflect an inadequate (epistemic) understanding of the way that science advances, but as a moral critique of pharmaceutical transnationals, which is accused of using us as “guinea pigs” for probably sinister purposes. This is evidenced by the fact that “powerful villains” are all over the tweet: not only big pharma, but also Bill Gates, Davos, and the Chinese Communist Party. There are no distinctions among them, thus confirming that the populists’ suspicions do not have a specific ideological target, insofar as they can be portrayed as elites. The text in the second tweet implies that vaccines and the pandemic in itself was a planned conspiracy by the elites, represented again by Bill Gates. The third tweet directly associates “the communists” with the elites, which shows the presence, in this case, of a right-wing populist discourse that mainly targets progressive elites.

We now turn to three paradigmatic examples of democratically grounded antivaccine objections, illustrated by the following tweets:

• - “The green pass is a violation of everything we know as freedom, it generates discrimination against people who have not been vaccinated, and that is totally unconstitutional, because the vaccine is not mandatory. #Saynotothegreenpassport…”

• - “How sad it is to see how people get vaccinated just for a pass. Where is the right to freedom?”

• - “[Although] the vaccine is voluntary… you can’t do ANYTHING if you don’t put it on □□♀▯”

In all three cases, the objection is primarily democratic in the sense that it does not claim moral corruption or scientific floppiness but instead shows discontent with a political decision that lacks democratic legitimacy. As it consistently invokes the liberty argument, which in this case implies the freedom to not get vaccinated without consequences, we argue that the democratic objection takes a libertarian form: it is a political claim to minimize state control. It does not dispute the effectiveness of the vaccine, demanding instead that the people themselves should decide whether to get it and to bear the corresponding risks. This was indeed José Antonio Kast’s claim, as the leader of the so-called Chilean “populist radical right” tweeted on May 9, 2021: “[the government] tells us that vaccination is voluntary, but those who do not get vaccinated will not be able to move freely. We have supported their policies… but transforming Chile into a permanent Dictatorship is unacceptable.”

Finally, we present three examples of antivaccine tweets containing epistemic objections.

• - “Alert: Testimony of a #Chilean teacher and her mother about the localized magnetism in the area where the inoculation of the #pfizer and #sinovac “vaccines” were applied…

• - “From the beginning and without being a doctor, I knew that Sinovac was less efficient than Pfizer. There is a reason why, in Europe, NOBODY KNOWS the very vaccine that most of us received”

• - “Robert Malone, virologist: What the vaccine is doing, in addition to killing and damaging people’s health, is actually developing the virus”

These tweets fit clearly into the epistemic category, as they doubt the science behind the vaccine. In the first case, vaccine-skepticism is based on anecdotal testimonial evidence that may or may not be real. The user nonetheless believes this alternative information source, rather than relying on established science. The second tweet adds a crucial feature of this type of objection: the self-assertion of epistemic autonomy and proficiency, akin to what Ruth Wodak has called “the arrogance of ignorance” (2015, p. 6). The user knew that one vaccine was less efficient than the other, despite citing no medical studies that could resolve the issue. Instead, a selective piece of information (the fact that Sinovac was not inoculated in European countries) is employed to support the claim. The third tweet opposes vaccination by bringing alternative experts to the fore, as proof that mainstream science is wrong.

In addition to the examples above where a single type of objection is presented in a post, we found a fraction of cases in which more than one is voiced in a given tweet. We present below two specific examples that illustrate the type of narratives that can contain more than one type of objection.

• - “The green pass is a hoax... many people got vaccinated to be able to move freely, but once again we are locked up… and maybe they put some shit in our bodies… another hoax from politicians”

• - “Antivaxxer? What a simplistic and stupid label! No one is antivaxxer. We just reject an experimental vaccine, imposed in a hypocritical and illegal way, in exchange for being able to exercise basic rights, that is a very different and altogether natural thing!”

In the first case, we can identify a moral objection entangled with a democratic one. On one hand, it suggests that politicians deceived on purpose by promising freedom to the vaccinated, while never intending to deliver. On the other hand, it wonders if the vaccine was an excuse to inoculate toxic substances. The second tweet seems to combine all three objections: the moral one, to the extent that authorities are branded as hypocrites, a type of ethical failure; the democratic one, to the extent that the rollout of the COVID-19 vaccine is deemed illegal for sidestepping the established political procedures; and the epistemic one, as it reasserts the rationality of a skepticism towards a vaccine that has not been sufficiently tested (and is thus referred to as “experimental,” in a proper epistemic sense).

Now that we have described the typical narratives found in tweets presenting each of the three types of objections, we can better characterize the results of our quantitative analyses. Our main finding is that our complete set of tweets resisting vaccination in Chile during 2021, including more moderate (inhibitors) and more extreme (antivaxxers) positions, fit well into the three theoretical anti-scientific categories indicated by the literature (moral, democratic, epistemic), in roughly similar proportions, thus confirming that antivaxxer sentiments do not respond to a single cause but can be traced over a range of allegedly populist grounds: science as corrupted by the elite’s selfish interests; science as the techno-elitist’s attempt to rule ignoring the people’s will; and science as an elitist way of thinking that spoils the innate good sense of ordinary citizens, and should therefore be challenged by popular counter-knowledge. The fact that all three objections thus described appear well represented in the data shows that there is empirical support for these theoretical distinctions in the context of antivaxxer sentiments, at least in Chile. In addition, our data also shows that, if we limit the sample to the hard anti-vaccine users, the moral objection clearly prevails whilst the epistemic objection decreases. Our analysis thus unveiled a connection between the degree of radicalism of the antivaccine sentiment and a higher representation of moral grounds for resisting vaccination. These results are intuitively sound: moral objections, in the populist sense, are directed against a common villain, which reinforces the us versus them divide. This way of tribal thinking sparks more emotional reactions than unaddressed claims, in which the enemy remains rather diffuse (see Greene, 2014), and emotional reactions have been shown to be more likely to go viral (Nelson-Field, 2013).

In further findings, we also observe that the users tweeting epistemic objections to the COVID-19 vaccine were the least active, that is, they generated the least tweets per user, compared to users tweeting moral or democratic objections. Similarly, general antivaccine tweets of an epistemic character generated less reactions, that is, fewer likes, retweets, and replies than tweets that contain moral or democratic criticisms of vaccination, which generated a similar number of reactions. This may be due to the fact that the users’ own experience regarding vaccination that contests scientific recommendations is not enough to produce intense, militant, or activist antivaccine commitment, which is more common in the case of moral and democratic objections. Consistently, users who were categorized as only mildly resisting vaccination tend to tweet more epistemic objections than radical antivaxxers, which mainly tweet moral and democratic objections. Hence, to rephrase these findings, when we broadly consider interactions by mild and strong antivaccine accounts together, the three types of populist objections appear in a similar proportion. But when we take only hard antivaxxers, tweets denouncing moral and democratic problems regarding vaccination increase considerably.

Finally, it appears that only a few tweets combine objections. This can be explained by the limited availability of characters on Twitter, which likely presses users to narrow their arguments, but also by the fact that we interpreted the tweets in a parsimonious sense, that is, we did not delve into the user’s intentions or his/her overall thought. At the same time, we find that many users can raise different types of objections in different tweets and that the two types of objections that are by far mostly combined by users are moral and democratic objections. On the other hand, we have observed that tweets that classify as epistemic, for example, may come from users that show their true conspiratorial colors (thus moral, under our nomenclature) in their Bio. We thus confirm that people who resist vaccines are willing to raise one objection or the other depending on the context, the interlocutor, or simply the need to win an argument. This is what we would expect if the relationship between populism and science is rather opportunistic, as part of the literature suggests (Bellolio, 2024; Szabados, 2019).

Conclusion

This study confronts recent theoretical literature that disaggregates the populist resistance to science with empirical data of the antivaccine sentiments aired on Twitter (now called X) throughout the Chilean vaccination process in 2021. We found that the three types of objections indicated by the chosen theoretical framework (moral, democratic, epistemic) are present in online interactions in similar proportions: a first set of antivaccine arguments denounce a global conspiracy fueled by dark agendas and profit-seeking; a second set blames political authorities for enacting restrictions on all kinds of liberties with no democratic legitimation; and a third set sows doubts regarding the safety and efficiency of a vaccine produced in record time, based on alternative information or anecdotal evidence. In further, more fine-grained results, we found that mild antivaccine accounts mostly appeal to the epistemic criticism, while hard antivaxxers mostly invoke the moral and the democratic objections. All in all, this paper provided empirical support to the work done in political theory and political science at large to unravel the understudied relation between populism and (hostility to) science. Our work also resulted in methods that could be applied to other vaccine-related datasets (e.g., from other countries) or extended to categorize other types of arguments, as long as they are clearly differentiable and represented in large enough datasets.

Funding

This work is part of the project The Place of Science in Liberal Democracy, FONDECYT Iniciación 11200056, funded by Chile’s national Association for Research and Development (ANID), and received support from the Millennium Institute on Immunology and Immunotherapy (ICN09_016/ICN 2021_045, former P09/016-F); FONDECYT # 1190830; SoL-UC; and CHuepe Labs Inc.

References

Biographical Notes

Cristóbal Bellolio is Associate Professor at the School of Government, Universidad Adolfo Ibáñez. He has a PhD in Political Philosophy and an MA in Legal & Political Theory from University College London, and a BA in Political Science and a BA in Law from Pontificia Universidad Católica de Chile. His research area is political theory, mostly within the intersection of liberalism, science and populism.

Abril Ortiz is a Social Anthropologist from Universidad Academia de Humanismo Cristiano and a researcher in Social listening lab SoL-UC at Pontificia Universidad Católica de Chile.

José Baboun is a Master of Science in Engineering student at Pontificia Universidad Católica de Chile and a researcher at SoL-UC. His research area is combinatorial optimization and social complex networks.

Eduardo Arriagada is a Research Professor at the School of Commuinications and codirector of the Social listening lab SoL-UC at Pontificia Universidad Católica de Chile, and Associate Investigator of the Millennium Institute on Immunology and Immunotherapy at Pontificia Universidad Católica de Chile. He has an MBA in Business from IE Business School and a Bachelor degree in Journalism from Pontificia Universidad Católica de Chile. His research areas are in mass media and social media.

Cristián Huepe is a Research Professor at ESAM and the Northwestern Institute on Complex Systems at Northwestern University, and the codirector of the Social listening lab SoL-UC at Pontificia Universidad Católica de Chile. He has a PhD in Theoretical Physics from the University of Paris and a Bachelor's degree in Physics from Universidad de Chile. His research areas are in complex physical, biological, technological, and social systems.

Footnotes