Master’s Thesis Pieter Attema Zalis
11824891
Graduate School of Communication Master’s programme Communication Science
2018-06-26
Supervisor Name Damian Trilling Email d.c.trilling@uva.nl
Abstract
This study investigates the role of fake news in the second turn of the 2018’s Brazilian
presidential election. Using a data set of tweets streamed between October 8, 2018, and October 27, 2018, we performed an automated content analysis of news articles. Two main comparisons are explored: right-wing vs. left-wing fake news and fake news vs. traditional news. Importantly, fake news here is defined as news which follows different intentions and processes of traditional news (Lazer et al., 2018). First, it is demonstrated that (H1) right-wing fake news is shared more frequently than left-wing fake news. Second, we show that fake news articles (H2) have more emotional content and also that it is (H3) retweeted faster and by more individuals than traditional news. We end by demonstrating that despite these advantages on the article level, when we consider the domain level, it is demonstrated that (H4) traditional news domains still have a higher reach of unique users than fake news domains.
Keywords: Brazil, Twitter, Bolsonaro, Fake News, Traditional News, Automated Content Analysis
Fake News in the 2018’s Brazilian Presidential Election: An analysis of its diffusion on Twitter
Introduction
Since the year of 2016, with Donald Trump’s election in the United States and the Leave
movement victory in the Brexit referendum, conventional politics have been turned upside down. As a consequence, a new set of studies and theories to describe and to understand the phenomena emerged. If before 2016 social media was mainly a platform for democratic revolutions, after 2016, its predominant role in political communication raised concerns about how network platforms, such as Facebook, might destroy democracy (Persily, 2017; Sunstein, 2017). As lying became part of communication strategies for attacking and destabilizing opponents (Bennett & Livingston, 2018), the discipline of verifying information, which has been one of the basic elements of any journalism (Kovach & Rosenstiel, 2001) for decades, became, under the umbrella of fact-checking, a separate field in itself. Reports trying to understand the "fake news" phenomena also flourished (Benkler et al., 2017; Jack, 2017; Tucker et al., 2018; Wardle & Derakhshan, 2017).
This paper focus on the latter concern: fake news. By doing so, it tries to extend the current knowledge about fake news in the democratic debate. The theoretical context of this thesis is the idea, as suggested by Lazer et al. (2018), that fake news is about different intentions and processes of traditional news. The contextual setup is the 2018’s Brazilian presidential election. Finally, the data setup is a collection of messages streamed from Twitter during the second turn of the presidential election. Brazil is an interesting country to analyze due to its similarities with the 2016’s American election. Against all initial expectations, the right-wing candidate, Jair Bolsonaro, got elected by defeating a center-left candidate, Fernando Haddad. Due to Bolsonaro’s anti-establishment populist style and social media presence (BBC, 2018), international media outlets got to the point of nicknaming him as the "Tropical Trump" (Phillips, 2018; Rathbone & Schipani, 2019). Moreover, as claimed in the US, fake news poisoned Brazilian political campaign (Tardaguila, Benevenuto, & Ortellado, 2018).
This study goes as follows. We start justifying the use of fake news, despite recognizing many weaknesses of the definition. The key element is to put fake news into perspective with traditional news. Public interest guides the intentions of traditional news (Kovach & Rosenstiel,
2001). In the case of fake news, it does not. If balance and objectivity dictate the processes of traditional news, previous studies suggest that the use of more emotional, partisan content in fake news could explain why these stories successfully generate a lot of interactions on social media (Bakir & Mcstay, 2018; Brady, Wills, Jost, Tucker, & Van Bavel, 2017).
With the definition established, the first step is to follow previous American attempts to analyze how ideology might moderate the differences between right-wing and left-wing fake news in social media. Next, this paper looks deeper into the processes followed by fake news publishers and its effects on patterns of information diffusion. As more shares of the content of specific news domain do not mean that it will reach more people, we will measure, in the final analysis, if the Brazilian case follows patterns found in other countries, where traditional news reach more unique users than fake news. To sum up, the research question is to what extent, during the second
round of the 2018’s Brazilian elections, is fake news from one ideology more present than the other, and do they, together, have different patterns of diffusion and reach compared to traditional news on Twitter?
This study would like to have contributed to the debate in the following ways.
Theoretically, this study tries to show why the concept of fake news should not be discarded from the academic debate in the way defined by Lazer et al. (2018). Secondly, scholars have
hypothesized that fake news is associated mainly with right-wing populist movements (Bennett & Livingston, 2018). Finding a predominance of fake news favoring Bolsonaro over Haddad would provide further evidence of this hypothesis and suggest that we might be talking about a globally right-wing strategy of political campaign. Thirdly, the acceptance of the suggested hypotheses would confirm the validity of the main theories on fake news in a less westernized, but democratic country where empirical evidence is still needed.
Theoretical background Defining a problematic term: Fake News
The more popular the term fake news becomes, the less the academia seems to like its usage. There are good reasons why to avoid using fake news. First, politicians - and Bolsonaro is an example - started to appropriate the term to discredit news organization that brought negative
news about them (Jankowski, 2018; Wardle & Derakhshan, 2017). Moreover, the problem might lie on how we define fake news. Tandoc Jr., Lim, and Ling (2018) showed, in a review of the 34 studies, that the term was used to describe six different ideas with little in common: news parody, news satire, manipulation, fabrication, advertising, and propaganda. These two concerns
prompted the usage of other concepts: disinformation and misinformation. Generally speaking, disinformation is information that is deliberately false or misleading (Jack, 2017; Wardle & Derakhshan, 2017). Misinformation is also false, but without the intention of causing harm (Jack, 2017; Wardle & Derakhshan, 2017).
The main problem of restricting the discussion into disinformation or misinformation is that it does not solve the third critic implied in fake news. False information might be a small piece of a bigger problem. Bennett and Livingston (2018) extend their definition of disinformation to encompass the idea of news formats that advance political goals. Wardle and Derakhshan (2017) talk about the existence of an information disorder environment that adds to misinformation and disinformation the idea of mal-information, i.e., information that is based on reality, but created to inflict harm on others intentionally. In a different direction, Jack (2017) defines the idea of problematic information, which includes not only misinformation and disinformation, but also the concept of agitprop, i.e., a propaganda campaign designed to provoke the audience to take a particular action.
Even if all raised weaknesses are relevant, this thesis keeps using fake news. Mainly because, I believe, the best description of the phenomena goes as follows: "we define ‘fake news’ to be fabricated information that mimics news media content in form but not in organizational process or intent." (Lazer et al., 2018, pg. 1094). In other words, the defining element of fake news is that they have different intentions and processes from traditional news. Thus, the “fakenes” is not at the level of the story but at the publisher (Grinberg, Joseph, Friedland, Swire-Thompson, & Lazer, 2019). The reasonable question to be asked, then, is how to define fake news intentions and processes. First, intentions. Public interest is the primary guide of traditional news. To set an agenda that favors political groups or ideology is what motivates fake news. In the process of traditional news, facts are sacred. In the process of fake news, facts can be interpreted or forked. To put it differently with a common metaphor in journalism, the
separation between church (opinions) and state (facts) has no blurred lines in fake news. Additionally, focusing on a conceptualization that has news (e.g., fake news) instead of information (e.g., disinformation) serves as a reminder of how those producers of fabricated information depend on the past strength of traditional journalism and its current existential crisis. First, fake news only exists today because, during the 20th century, the establishment of objectivity and balance as fundamental norms allowed news to get high levels of public trust (Lazer et al., 2018). To put it differently, for decades news has been a synonym for veracity. Second, we can not dissociate the escalation of the fake news problem from the shift of an offline to an online dominant news environment. First, this transformation, especially with the
predominance of social media communications, decreased the cost of entrance for new
competitors, which do not necessarily follow the norms of objectivity and balance (Lazer et al., 2018). Second, the abundance of information sources in the online environment makes individuals rely on heuristics and social cues to outline their beliefs and determine the credibility of
information (Lazer et al., 2017).
To sum up, the definition of fake news encompasses the following characteristics. It can be false stories, but not necessarily. What matters is the intentions and processes followed by the publishers. Fake news has no blurred lines between news articles and opinions. It also advocates for a political agenda. The rules of objectivity and balance in the text are not respected. In many cases, it might overlap with ideas embedded by others in concepts such as disinformation
(Bennett & Livingston, 2018), hyper-partisan media (Moretto Ribeiro & Ortellado, 2018), or online ideological news (Garrett, Weeks, & Neo, 2016).
Fake News as Intention
The literature split the intention of fake news in two main domains: financial and ideological. The former is a result of individuals willing to profit from advertising revenues generated by clickbait content (Allcott & Gentzkow, 2017). In some cases, publishers can own different websites that support two conflicting agendas, such as one website that spreads fake news in favor of left-wing candidates, while the second promotes the agenda of right-wing candidates (Subramanian, 2017; Victor, 2017). When the intention is ideological, publishers will try to boost the agenda of
like-minded candidates and attack politicians from a different political spectrum without
necessarily thinking about financial returns. Benkler et al. (2017) argue, for instance, that highly partisan media was the leading incubator of disinformation and propaganda in social media during the 2016 American campaign. They found that, in the top 100 most in linked media sources on social media, seven outlets from both the partisan right and the partisan left -received more attention than the other media outlets. Previous research claims a similar portrait of the political debate in the Brazil (Moretto Ribeiro & Ortellado, 2018). In other words, when we think about the intentions, the concept of fake news here suggested might overlap with the idea of hyper-partisan media, i.e., media outlets which report a biased portrait of the political debate Moretto Ribeiro and Ortellado (2018).
As we focus on the ideological intentions of fake news, it is vital to define ideology in the context of this study. Mainly because, from the perspective of a foreigner, Brazilian politics can look “inchoate”(Lucas & Samuels, 2010). Liberals and conservatives, different from the US, are both on the right side of the national political spectrum. Longitudinal surveys also reveal high degrees of instability in ideological self-identification among voters (Ames & Smith, 2010). A way suggested by Moretto Ribeiro and Ortellado (2018) to make sense of this inconsistency is to group individuals into antagonist groups. Thus, we define right-wing as being in favor of Bolsonaro or against Haddad. Left-wing is simply the other way around, i.e., being in favor of Haddad or against Bolsonaro.
It is reasonable, so, to expect that many right-wing voters might enjoy reading stories that Haddad would turn Brazil into Venezuela or that left-wing individuals would easily share online articles that classify Bolsonaro as a Nazi. Readers do not question the credibility of such
misleading information because it supports their preconceptions (Flynn, Nyhan, & Reifler, 2017; Kahne & Bowyer, 2017; Lazer et al., 2018) and make them feel connected to the piece of
information. In other words, their system of beliefs influences how the information is proceeded and understood (Kunda, 1990). That is the core of the motivated reasoning theory. There is enough evidence to support the theory. Allcott and Gentzkow (2017) suggest that Democrats and Republicans were both about 15 percent more likely to believe ideologically aligned headlines during the 2016 American election. Republicans trusted more Donald Trump statements,
independently if they were true or false than Democrats (Swire, Berinsky, Lewandowsky, & Ecker, n.d.). Among youths, social media posts aligned to prior political views are considered to be more accurate than posts that did not align (Kahne & Bowyer, 2017).
On the other hand, how the cognitive role performed by motivated reasoning acts differently on the right-left scale is a debatable subject. There are claims that ideologically motivated
cognition is an information processing mechanism that works as an indicator of an individual’s loyalty to social group (Kahan, 2013). This mechanism would mean that individuals from left to right are equally susceptible. Others, however, defend that there would be a prevalence among right-wing individuals (Jost, Glaser, Kruglanski, & Sulloway, 2003). The main reason for the second argument is that right-wing individuals have specific personality traits, such as intolerance towards ambiguity, and cognitive styles, such as the need for cognitive closure, which make them more likely to engage in motivated information processing. This second argument, in the context of fake news during political campaigns, is confirmed by the disproportionately higher
consumption of fake news observed among Trump’s supporters in the 2016’s election compared to Clinton’s supporters (Allcott & Gentzkow, 2017) and the significantly higher relevance of partisan right media compared to the left (Benkler et al., 2017). This final evidence leads to our first hypothesis: (H1) Right-wing fake news will be shared more than left-wing fake news on Twitter.
Fake News as Process
In the previous section, it was clarified how the intentions of fake news work. Now, we dive deeper into the idea of the processes, the second defining element of fake news. Traditional journalism follows specific methods and routines. The norms of objectivity and balance guided much of its development during the 20th century (Lazer et al., 2018). Newsmakers should distinguish facts from values and report only the facts (Schudson, 2001). They also need to maintain independence from those they cover (Kovach & Rosenstiel, 2001) and have a stable compromise with the truth (Kovach & Rosenstiel, 2001).
The processes of fake news go in a different direction. Bakir and Mcstay (2018) claim that the economics of emotion is at the heart of the fake news problem. The authors argue that these news domains leverage emotions and preconceptions of readers to generate attention to their
content. When analyzing data of the 2016 American election, Benkler et al. (2017) mentions, for instance, the predominant framing of anti-immigrate stories that focused on disease, crime and terrorism. Coming back to Brazil, Bolsonaro, for instance, was framed as a possible racist neo-fascist dictator despite running a democratic election (Forum, 2018), while Haddad was accused of throwing a Bible in the trash without any fact-checking (Rovervan, 2018).
A possible explanation for the efficacy of such stories might lie in the consequences of more emotional content. Weeks (2015) indicates, following the argument of the Affective Intelligence theory, that anger can enhance partisan influence. Moreover, Vosoughi, Roy, and Aral (2018) suggest also that the degree of novelty and the emotional reactions may be responsible for diffusing more false stories than true ones. In other words, emotion drives the process of fake news, in opposition to the balance and objectivity of traditional news.
The consequence of more emotional content is that fake news might have a better
performance in social media than traditional news. As fake news, online social networks, where individuals consume today significant amounts of information (Amy, Katerina Eva, & Hannah, 2018), tend to follow the same rules of the economics of emotions. Their architectures are engineered such that "every time a user posts content - and it is liked, commented up or shared further - their brain releases a tiny hit of dopamine" (Wardle & Derakhshan, 2017, pg. 13). Previous empirical evidence show, for instance, that emotional Twitter messages tend to be retweeted more often (number of times the tweet has been retweeted) and quicker (time lag between the tweet and the retweet) compared to tweets with low levels of emotions (Stieglitz & Dang-Xuan, 2013). In a similar direction, Brady et al. (2017) demonstrate that the presence of emotional words increases the chances of a message going viral (i.e., retweeted by many people) by a factor of 20% for each emotional world present in a tweet.
If fake news is expected to have more emotional content, we can expect it to be shared faster (it takes less time to retweet a message) and generate more engagement (more people will share the message on Twitter). Vosoughi et al. (2018) show that fake news articles (in the sense of false stories) spread indeed faster and deeper (in the sense of reaching more people) than true stories, especially for political content. Moreover, a journalistic analysis by Buzzfeed showed that top fake election news stories generated more engagement on Facebook than top election stories
from 19 major news outlets combined (Silverman, 2016). This reasoning suggests that fake news articles from both left and right in Brazil will have higher levels of emotions, will spread faster and have more people retweeting it than traditional news. Thus, the second and third hypotheses go as follows:
(H2) Fake news articles have more emotional content than traditional news articles. (H3a) Fake news articles will be retweeted faster and (H3b) by more individuals than traditional news.
Fake News vs. Traditional News
Until now, we formulated the hypotheses on the level of articles. What if we analyze the units of news domains (e.g., bbc.uk, nu.nl or globo.com.br)? Higher rates of story diffusion or engagement do not necessarily translate into higher levels of absolute audience per domain. One media outlet can have, for example, a group of active users that retweet more often, but a second media outlet can have more people retweeting less. In this case, we can claim that the reach of the second domain is higher than the reach of the first. It seems to be the case when comparing the audience of traditional and fake news. In the US, Nelson and Taneja (2018) concludes that the fake news audience comprises a small number of heavy Internet users, a finding that was corroborated by Grinberg et al. (2019). Additionally, Fletcher, Cornia, Graves, and Nielsen (2018) concluded that websites that publish false stories had an average monthly reach of over 3,5% in Italy and France. This number is significantly smaller to the most visited traditional news websites, Le Monde, France, and La Repubblica, Italy. Thus, our final hypothesis is:
(h4) The reach of traditional news domains is bigger than the reach of fake news domains
Methodology Research Design and Data
The hypotheses are tested in a data set of tweets collected during the second turn of the Brazilian presidential election. Twitter users are not representative of the underlying populations
(Jungherr, 2016), but it is the only major social media network that is both used for political discussion in a variety of countries and makes a large amount of data available for research
(Bright, 2018; Jungherr, 2016). Between October 8, 2018, (the day after the first round of the election) and October 27, 2018, (the day of the second round of the election), 19,649,008 messages were streamed by the author through Twitter streaming Application Programming Interface (API). The scale of the data set impedes the usage of traditional forms of content analysis. Thus, this study relies mainly on automated content analysis, a collection of techniques used to
automatically analyze media content (Trilling & Jonkman, 2018). Manual content analysis will be used, however, in two moments: when we label domains as fake news or not and when we classify the ideology of fake news. Two coders performed the task independently.
Some pre-processing steps were taken to fit the data according to the necessities of the hypotheses. As the interest of this research lies on only understanding patterns of fake and traditional news, we started by filtering messages which contain URLs of news domains that fall into the concepts of fake news and traditional news. Additionally, following previous studies with similar designs, we established a threshold of URLs that were shared at least five times (Machado et al., 2018), which allows us to discard 40,791 news stories that correspond to only 4% of the total amount of URL shared. As a consequence, the final data set is composed of 18,838 unique articles from 72 different news domains that appeared 1,609,738 times on Twitter. Regarding the research units, in most cases, the analysis will be performed at the level of each article (which sometimes are called URLs). Only for the last hypothesis, tweets are aggregated into the level of news domains.
Observed Variables
Fake News. As Lazer et al. (2018) suggest, we focused the “fakeness” not at the level of the story but of the publisher. As previously explained, the focus of this approach is how fake news processes and intentions are different from the ones in traditional news. Thus, we developed a codebook (Appendix A) that highlights these differences. Two coders manually labeled (fake news/not fake news) each domain based on the content of 10 random articles per domain. In the end, we arrived to a list of 44 fake news domains.(α = 0.93).
Ideology of Fake News. In the case of coders classifying an outlet as fake news, a second step was to classify the ideology of the news domain. Previous studies show that the
online political debate in Brazil has been mainly focused on contrasts between individuals being anti something or in favor of something (Moretto Ribeiro & Ortellado, 2018). Thus, a domain is considered right-wing fake news if its news coverage consistently favors Bolsonaro or consistently attacks Haddad. When the opposite occurs - i.e., the news coverage consistently favors Haddad or consistently attacks Bolsonaro - the outlet is considered left-wing fake news. From the 44 domains labeled as fake news, 25 were classified as left fake news, and 19 right fake news (α = 0.91).
Traditional News. News domains were classified as traditional news if they fall into one of the following criteria: national newspaper, a regional newspaper, TV channel, news magazine, radio station, established news portal or international news agencies with content in Portuguese. Importantly, we excluded any foreign news outlets (e.g., Guardian, New York Times or CNN) and "Digital-Born News Media" domains, such as the Brazilian versions of Buzzfeed and Vice, which follow different business models and goals of traditional media (Nicholls, Shabbir, & Nielsen, 2017). In the end, we formed a list of 29 traditional news domains.
Emotions on Articles. It is a continuous variable (M =3.63, SD=1.73) based on Language Inquiry Word Count (LIWC), a text analysis software that calculates the degree of usage of different psychologically meaningful categories in text files. Previous studies already used it to label emotions on political tweets (Brady et al., 2017). Additionally, its Portuguese version was already validated as a reliable measurement for tasks such as sentiment analysis (Filho, Pardo, & Aluísio, 2013). Due to the computational costs of extracting the text from all URLs from the data, we selected a random sample of one thousand fake and traditional news articles each to test the hypothesis about emotions.
Article Velocity. It is a continuous measure of how many seconds it takes on average for a new user to retweet an article (M =2401.22, SD=3113.19). In other words, velocity is the shelf-life of the article divided by the number of unique users who shared the article. We define shelf-life as the time passed between the time the first tweet containing the article was published and the time at which the article received a fraction of 90% of the visits (Castillo, El-Haddad, Pfeffer, & Stempeck, 2014). The usage of a 90th percentile avoids the effect of possible outliers, such as a user who retweets weeks later.
Reach of News Domains. The number of unique users that tweeted articles from each news domain (M =22051.20, SD=46048.43).
Results Right-Wing vs. Left-Wing Fake News
The data shows that right-wing fake news had more retweets than left-wing fake news on Twitter. While URLs classified as right-wing fake news had, in total, 475,815 URLs shares during the analyzed period, left-wing fake news URL added up to 306,384. In other words, for every left-wing fake news article tweeted, there was 1.55 right-wing article also shared. Moreover, from the top hundred most shared fake news articles, 81 articles were labeled as right-wing, while only 19 came from left-wing media domains. Additionally, as a way to check if the difference is statistically significant, we compared the number of retweets per each article of right-wing fake news to left-wing fake news. A natural way would be to perform a t-test, but as both distributions violate the assumption of normality, we conducted the Mann Whitney U test, a non-parametric version of the t-test. The results confirm, with strong evidence, that the shareness of right-wing fake news (Mdn = 19) is greater than the shareness of left-wing fake news, Mdn =19, U = 1,0259,911.5, p < .01, r = 105,930.01. Thus, we confirmed, with strong evidence, the first hypothesis.
Figure 1 . The graph on the left shows the total amount of URL shared per news type. In the center, the comparison of means with 95% confidence interval. In the right, the number of shares from the top 100 most popular fake news articles
Emotions and Patterns of Diffusion
Next, we tested the hypothesis that fake news has more emotional content than traditional news. We took the measure of affect, a variable in LIWC which combines the sum of positive and negative emotions in a text file. Figure 2 demonstrates the distribution of affect scores between the two groups. The average affect score in the Fake News (M =3.93, SD = 1.97) lies with 95% between 3.79 and 4.07. This differs significantly, but with weak effect size, from the average of Traditional News articles (3.38), t(1406.8)=6.4491, p < .001, 95% [3.28, 3.47], d= 0.32.
Figure 2 . The two histograms show the distribution of the affect variables for Fake News and Traditional News
An aspect of LIWC is that the affect dimension is a combination of two sub-dimensions: positive and negative emotions. Regarding the later, it can be even split further in the other three sub-sub-dimensions: sadness, anger and anxiety. Figure 3 shows the mean values with a 95% confidence interval for each one of the five sub-dimension, plus affect. Only in the case of sadness, there is an overlap of values. This result indicates that, except for sadness, the Fake News group scores significantly higher than Traditional News. Negative emotions mainly due to anger -present a higher discrepancy than positive emotions.
Hypothesis three stated that fake news articles would spread faster (H3a) on social media and also have more people retweeting them (H3b). The statistical analysis confirmed the rationale in both cases. Kolmogorov-Smirnov test, a non-parametric test that compares the
Figure 3 . The mean comparison, with a 95% CI, of Fake News and Traditional News in all six dimensions of LIWC.
distribution of the articles velocity in the two groups, is conducted. The results are significant: D(2000) = 0.234, p < .01. The same test was performed for the number of people retweeting articles. Similar results were achieved: D(12589) = 0.11, p < .01. A good way of interpreting the results that fake news articles travel faster and spread to more people than traditional news is to compare the same quantiles between cumulative distributions shown in figure 4. While the 90th percentile of the velocity of fake news articles is 19, which means that 90% of the data is below 19 seconds, in the case of traditional news, the same quantile is 48.63, i.e., 90% of traditional articles are faster than 48.63 seconds. If we analyze the 90th percentile of the number of user retweeting, we see that 90% of the fake news articles are below 258, while in the traditional news case, 90% of the data is below 115.
Figure 4 . The cumulative distributions of articles and news domains. Fake News is the orange line and traditional news is the green.
Traditional News vs. Fake News
The last hypothesis stated that the reach of traditional media domains would be bigger than the reach of fake news domains. We checked if the difference is statistically significant by performing again a Kolmogorov-Smirnov test. As predicted in my hypothesis, the difference between traditional news and fake news is significant, D(73) = 0.359, p = .01. Thus, we confirm the hypothesis. As Figure 4 shows, while all fake news domains were at their maximum tweeted by 48,286 unique users, 17,25% of the traditional outlets have a higher reach than that. The result is corroborated by the sum of unique users for both types of domains. In our data set, 295,525 members make the community of traditional news users, while the community of fake news has 124,399. Thus, we confirm the hypothesis. The results have even higher relevancy if we compare the number of URLs shared with the number of unique users per domain shown in figure 5. If for every user tweeting fake news, there are 2.37 users sharing traditional news, the sum of URL shares, not considering the uniqueness of users, is barely equal. Additionally, if a right-wing fake news domain, oantagonista, leads the rank of the number of URLs shared per domain, the same outlet is beat by four traditional outlets in the rank of the number of unique users tweeting articles per domain.
Figure 5 . Distributions of unique users and amount of shares. The graphs at the top show the data aggregated by news type. The graphs at the bottom show the data aggregated by domain, with green indicating traditional news and orange fake news.
Conclusion
This study started with the following research question: to what extent, during the second round of the 2018’s Brazilian elections, is fake news from one ideology more present than the other, and do they, together, have different patterns of diffusion and reach compared to traditional news on Twitter? The answer to the question is divided as follows: right-wing fake news was more common than left-wing fake news (H1). When comparing fake to traditional news, we showed that fake news articles had more emotional content than traditional news articles (H2). Fake news also traveled faster and had more retweets than traditional news articles (H3). Finally, when we considered news domains, traditional news outlets had more unique users retweeting its content than fake news outlets (H4).
As studies on fake news and information pollution are only at the earliest stage (Wardle & Derakhshan, 2017), this paper brings a valuable contribution to the field as it extends to Brazil, the fourth largest democracy in the world, some of the findings already identified in the US and Europe. By confirming all hypotheses, we provided, for instance, further evidence that theories, such as motivated reasoning and the significant role of emotions in information diffusion, can partially explain the virality of fake news on social media. Additionally, also similar to what was discovered in other countries (Fletcher et al., 2018; Grinberg et al., 2019; Nelson & Taneja, 2018), we showed that fake news still is not capable of reaching the same amount of unique users as traditional news.
This research brings some implications to the theory and practice of fake news studies. From all findings here, the one with higher effect size was the difference between right-wing and left-wing fake news. Even if it is early to claim that this trend of right-wing predominance is an establish global pattern, it is fundamental to investigate this further in other democracies. As suggested by Bennett and Livingston (2018), there appears to be a correlation between fake news and the emergence of radical right parties. The authors’ argument is straightforward: declining citizens confidence in institutions, such as center parties and the mainstream press, undermines the credibility of traditional news (Bennett & Livingston, 2018). Radical right parties, with their anti-establishment discourse, seem to profit more than other political forces and use alternative facts to boost their electoral performance.
If checking the rationale that fake news sharing is a consequence of institutional distrust goes beyond the scope of this study, this is a hypothesis that should be further investigated. A possible way of doing so is to adjust the method suggested earlier in this study. If, here, we focus on the analysis of the content of fake news articles, future work could use LIWC dictionary to make an automated content analysis of the emotions of users who share fake news articles. If we find that individuals who share right-wing fake news show higher signs of anger in their tweets, it would be possible to investigate if this angry reaction is directed to conventional democratic institutions and democracy in itself. Additionally, another important contribution to better understanding the global predominance of right-wing fake news is to extend quantitative studies to other countries beyond Brazil, U.S. and Europe.
Regarding the differences between traditional and fake news, we demonstrated that, in Brazil, fake news domains still need twice as many unique users to have the same amount of shares than traditional news domains on Twitter. In this sense, as it was already discussed, this paper is another study which demonstrated that society, in general, might be overestimating the real reach of fake news when compared to the reach of traditional news. This, however, does not mean fake is not a problem at all. Another way of framing the same information is that fake news and their amateur newsgathering process is still capable of reaching half the audience of established
newspapers, magazines and TV channels. To give another example, more than 20 fake news outlets had a higher reach of unique users than respected outlets as the BBC Brazilian version.
Thus, the main open question left on this research concerns how big or small the fake news problem actually is. Many directions can be imagined to investigate this. However, if we agree that the first obligation of journalism is to the truth (Kovach & Rosenstiel, 2001), an initial path to follow is to check if fake news can undermine the most pressing element of journalism: make people believe on the truth when they see it. In other words, even if people still read more traditional news than fake news, the later can have a significant impact on the former if it makes people believe in less credible sources. In this regard, future work could add surveys which measure the capacity of fake news consumers to recognize true information. The next question, if the hypothesis is confirmed, is how representative those fake news consumers who disregard the truth are on the entire population. If they represent just a niche, its impact is probably
minimum. If, however, the respondents are relevant societal groups, the concerns about how information flows in a democracy might increase.
Finally, readers should consider two limitations of this study. The most important one concerns the whole idea of fake news used here. Based on the idea of Lazer et al. (2018), we defined the fakeness on the domain level. However, what the content analysis showed is that there is a certain level of nuance in specific domains classified as fake news that was not captured in the analysis. There are certain domains that only publish false stories. Others do not publish false stories, but only highly partisan content. In others, we can find both. In this sense, future work could label the fakeness on the article level and change the dichotomy between fake news vs. traditional news to something more nuanced such as fake news vs. partisan news vs. traditional news. The second limitation is about Twitter. As in other countries, Twitter has a limited audience in Brazil compared to other social media platforms, such as Facebook and WhatsApp. Thus, the finding suggested here should not be extended to other networks and future work on these platforms could be conducted to bring a full picture of how fake news travels through social media in Brazil.
References
Allcott, H., & Gentzkow, M. (2017). Social media and fake news in the 2016 election. Journal of Economic Perspectives, 31 (2), 211–236. doi: 10.1257/jep
Ames, B., & Smith, A. E. (2010). Knowing left from right: Ideological identification in Brazil, 2002-2006. Journal of Politics in Latin America, 2 (3), 3-38.
Amy, M., Katerina Eva, M., & Hannah, K. (2018). News use across social media platforms 2018. Washington, DC: Pew Research Center. Retrieved from https://www.journalism.org/ Bakir, V., & Mcstay, A. (2018). Fake news and the economy of emotions: Problems, causes,
solutions. Digital Journalism, 6 (2), 154–175. doi: 10.1080/21670811.2017.1345645
BBC. (2018, December 31). Jair Bolsonaro: Brazil’s firebrand leader dubbed the Trump of the tropics. Retrieved from https://www.bbc.com/news/
Benkler, Y., Roberts, H., Faris, R. M., Etling, B., Zuckerman, E., & Bourassa, N. (2017).
Partisanship, propaganda, and disinformation: Online media and the 2016 U.S. presidential election. Cambridge, MA: Berkman Klein Center for Internet & Society. Retrieved from https://cyber.harvard.edu/node/99981
Bennett, W. L., & Livingston, S. (2018). The disinformation order: Disruptive communication and the decline of democratic institutions. European Journal of Communication, 33 (2), 122-139. doi: 10.1177/0267323118760317
Brady, W. J., Wills, J. A., Jost, J. T., Tucker, J. A., & Van Bavel, J. J. (2017). Emotion shapes the diffusion of moralized content in social networks. Proceedings of the National Academy of Sciences of the United States of America, 114 (28), 7313-7318. doi:
10.1073/pnas.1618923114
Bright, J. (2018). Explaining the emergence of political fragmentation on social media: The role of ideology and extremism. Journal of Computer-Mediated Communication, 23 (1), 17–33. doi: 10.1093/jcmc/zmx002
Castillo, C., El-Haddad, M., Pfeffer, J., & Stempeck, M. (2014). Characterizing the life cycle of online news stories using social media reactions. In Proceedings of the 17th acm conference on computer supported cooperative work & social computing (pp. 211–223). New York, NY, USA: ACM. doi: 10.1145/2531602.2531623
Filho, P. P. B., Pardo, T. A. S., & Aluísio, S. M. (2013). An evaluation of the Brazilian Portuguese LIWC dictionary for sentiment analysis. In Proceedings of the 9th Brazilian symposium in information and human language technology. Retrieved from
https://www.aclweb.org/anthology/W13-4829
Fletcher, R., Cornia, A., Graves, L., & Nielsen, R. K. (2018). Measuring the reach of fake news and online disinformation in europe. Oxford, UK: Reuters Institute for the Study of Journalism.
Flynn, D. J., Nyhan, B., & Reifler, J. (2017). The nature and origins of misperceptions: Understanding false and unsupported beliefs about politics. Political Psychology, 38 (1), 127–150. doi: 10.1111/pops.12394
Forum, R. (2018, October 14). Postagens que associam bolsonaro ao nazismo dominam o twitter. Retrieved from https://revistaforum.com.br
Garrett, R. K., Weeks, B. E., & Neo, R. L. (2016). Driving a wedge between evidence and beliefs: How online ideological news exposure promotes political misperceptions. Journal of
Computer Mediated Communication, 21 (5), 331–348. doi: 10.1111/jcc4.12164
Grinberg, N., Joseph, K., Friedland, L., Swire-Thompson, B., & Lazer, D. (2019). Fake news on twitter during the 2016 u.s. presidential election. Science, 363 (6425), 374-378. doi:
0.1126/science.aau2706
Jack, C. (2017, September 08). Lexicon of lies: Terms for problematic information. New York, NY: Data & Society Research Institute. Retrieved from
https://datasociety.net/output/lexicon-of-lies/
Jankowski, N. W. (2018). Researching fake news: A selective examination of empirical studies. Javnost-The Public, 25 (1-2), 248–255. doi: 10.1080/13183222.2018.1418964
Jost, J. T., Glaser, J., Kruglanski, A. W., & Sulloway, F. J. (2003). Political conservatism as motivated social cognition. Psychological Bulletin, 129 (3), 339—375. doi:
10.1037/0033-2909.129.3.339
Jungherr, A. (2016). Twitter use in election campaigns: A systematic literature review. Journal of Information Technology & Politics, 13 (1), 72–91. doi: 10.1080/19331681.2015.1132401 Kahan, D. M. (2013). Ideology, motivated reasoning, and cognitive reflection. Judgment and
Decision Making, 8 (4), 407–424.
Kahne, J., & Bowyer, B. (2017). Educating for democracy in a partisan age: Confronting the challenges of motivated reasoning and misinformation. American Educational Research Journal, 54 (1), 3–34. doi: 10.3102/0002831216679817
Kovach, B., & Rosenstiel, T. (2001). The elements of journalism : what newspeople should know and the public should expect. New York, NY: Crown Publishers.
Kunda, Z. (1990). Psychological bulletin the case for motivated reasoning. Psychological Bulletin, 208 (3), 480–498. doi: 10.1037/0033-2909.108.3.480
Lazer, D., Baum, M., Grinberg, N., Friedland, L., Joseph, K., Hobbs, W., & Mattsson, C. (2017). Combating fake news: An agenda for research and action. Cambridge, MA: Shorenstein Center On Media, Politics and Public Policy. Retrieved from
https://shorensteincenter.org
Lazer, D., Baum, M. A., Benkler, Y., Berinsky, A. J., Greenhill, K. M., Menczer, F., . . . Zittrain, J. L. (2018). The science of fake news. Science, 359 (6380), 36–44. doi:
10.1126/science.aao2998
Lucas, K., & Samuels, D. (2010). The ideological “coherence” of the brazilian party system, 1990-2009. Journal of Politics in Latin America, 2 (3), 39-69.
Machado, C., Kira, B., Hirsch, G., Marchal, N., Howard, P. N., Graphika, T. L., & Barash, V. (2018, October 5). News and political information consumption in Brazil: Mapping the first round of the 2018 brazilian presidential election on twitter. Oxford, UK: Project on
Computational Propaganda. Retrieved from comprop.oii.ox.ac.uk
Moretto Ribeiro, M., & Ortellado, P. (2018). From fake news to hyper-partisan media. SUR 27 , 15 (27), 69–81.
Nelson, J. L., & Taneja, H. (2018). The small, disloyal fake news audience: The role of audience availability in fake news consumption. New Media & Society, 20 (10), 3720-3737. doi: 10.1177/1461444818758715
Nicholls, T., Shabbir, N., & Nielsen, R. K. (2017). The global expansion of digital-born news media. Oxford, UK: Reuters Institute for the Study of Journalism. Retrieved from https://reutersinstitute.politics.ox.ac.uk/
Persily, N. (2017). Can democracy survive the internet? Journal of Democracy, 28 (2), 63–76. doi: 10.1353/jod.2017.0025
Phillips, T. (2018, April 19). Trump of the tropics: the ’dangerous’ candidate leading brazil’s presidential race. The Guardian. Retrieved from https://www.theguardian.com/ Rathbone, J. P., & Schipani, A. (2019, March 17). Brazil’s ‘tropical trump’ seeks to reset ties
with white house visit. Financial Times. Retrieved from https://www.ft.com/
Rovervan, R. (2018, October 22). Haddad joga biblia no lixo e humilha eleitor nordestino. Terça Livre. Retrieved from https://www.tercalivre.com.br/
Schudson, M. (2001). The objectivity norm in american journalism. Journalism, 2 (2), 149–170. doi: 10.1177/146488490100200201
Silverman, C. (2016, November 16). This analysis shows how viral fake election news stories outperformed real news on facebook. Buzzfeed. Retrieved from
https://www.buzzfeednews.com/
Stieglitz, S., & Dang-Xuan, L. (2013). Emotions and information diffusion in social media sentiment of microblogs and sharing behavior. Journal of Management Information Systems, 29 (4), 217-248. doi: 10.2753/MIS0742-1222290408
Subramanian, S. (2017, February 15). Inside the macedonian fake-news complex. Wired. Retrieved from https://www.wired.com/
Sunstein, C. R. (2017). #republic: divided democracy in the age of social media. Princeton, NJ: Princeton University Press.
Swire, B., Berinsky, A. J., Lewandowsky, S., & Ecker, U. K. H. (n.d.). Processing political misinformation: comprehending the trump phenomenon. Royal Society Open Science, 4 (3), 160802. doi: 10.1098/rsos.160802
Tandoc Jr., E. C., Lim, Z. W., & Ling, R. (2018). Defining “fake news”. Digital Journalism, 6 (2), 137-153. doi: 10.1080/21670811.2017.1360143
Tardaguila, C., Benevenuto, F., & Ortellado, P. (2018, October 17). Fake news is poisoning brazilian politics. Whatsapp can stop it. New York Times. Retrieved from
https://www.nytimes.com/
Measures, 12 (3), 158–174. doi: 10.1080/19312458.2018.1447655
Tucker, J. A., Guess, A., Barbera, P., Vaccari, C., Siegel, A., Sanovich, S., . . . Nyhan, B. (2018, March 19). Social media, political polarization, and political disinformation: A review of the scientific literature. Menlo Park, CA: Hewlett Foundation. doi:
http://dx.doi.org/10.2139/ssrn.3144139
Victor, F. (2017, February 19). Como funciona a engrenagem das notícias falsas no brasil. Folha de S. Paulo. Retrieved from https://www1.folha.uol.com.br/
Vosoughi, S., Roy, D., & Aral, S. (2018). The spread of true and false news online. Science, 359 (6380), 1146-–1151. doi: 10.1126/science.aao2998
Wardle, C., & Derakhshan, H. (2017, September 27). Information disorder: Toward an interdisciplinary framework for research and policy making. Strasbourg, FR: Council of Europe. Retrieved from
https://www.coe.int/en/web/freedom-expression/information-disorder Weeks, B. E. (2015). Emotions, partisanship, and misperceptions: How anger and anxiety
moderate the effect of partisan bias on susceptibility to political misinformation. Journal of Communication, 65 (4), 699–719. doi: 10.1111/jcom.12164
Appendix A
This document brings the instructions for coding 48 news domains/twitter accounts for my thesis about fake news in Brazilian Election.
Basically, by looking in the links, you will have to perform two tasks. First, check if it’s possible to classify each domain as fake news or not. Second, check if it’s possible to classify the ideology (left/right) of the accounts that you classified as ideological.
Fake news definition
In my thesis, We define fake news as online media domains which show a coverage consistently favoring one political perspective over the others and this can
Let’s start with the process of fake news.
Objectivity is a primary guide of traditional news. In other words, journalists should avoid the use of adjectives and split opinions from facts. Journalists should use a sober style while writing too.
Fake news, on the other hand, does not respect the above rules. Opinions are mixed with facts and writers/journalists are free to make value judgments about the information.
Now, the intentions of fake news.
Public interest is the primary guide of traditional news. To clearly set an agenda that favours political groups is what motivates fake news publishers.
Instructions
I’ll send to you a Google sheet with all necessary information. After you complete the coding, you can send it back to me via my email: pieterzalis@gmail.com.
The sheet is divided in four columns: fake, ideology, domain, url.
How should you proceed? For every different domain, open the links from the url column. Look for the content of the links and ask yourself the following questions:
• Has the headline and body of the text a very unbalanced (biased and subjective) style of writing?
• Do you see a lot of moralized adjectives in the texts?
• It’s possible to clearly define if the news domain or news account favours its coverage for a specific candidate or systematic attacks the other candidate?
• Do you see the domain reinterpreting original information from the original source
If at least one of the answers are yes, you can classify as fake news. Importantly, you
don’t need to classify each story, but only the domains. In some cases, there will be
twitter accounts that were suspended. This means that will not be able to access the information directly. A way to work around this is to research for the link in Google. From there, you can see the content of the message.
Which labels should you use?
• Fake news: 1
• Not fake news: -1
• It’s not clear: 0
After you classified domains as fake news or not, the next step is to check the ideology of the domains that you labeled as fake news. The main point here is find the answer of the following questions:
• Does the fake news domain systematically favour one candidate over the other?
• Does the fake news domain systematically attack one candidate over the other?
Which labels should you use in the ideology columns?
• Fake news that favours Bolsonaro and/or attacks Haddad: 1
• Fake news that favours Haddad and/or attacks Bolsonaro: -1
