Roles Conceptualization, Data curation, Formal analysis, Funding acquisition, Investigation, Methodology, Project administration, Resources, Software, Supervision, Validation, Visualization, Writing – original draft, Writing – review & editing * E-mail: garrett.258@osu.edu Affiliation School of Communication, Ohio State University, Columbus, Ohio, United States of America
There is considerable concern about the role that social media, such as Facebook and Twitter, play in promoting misperceptions during political campaigns. These technologies are widely used, and inaccurate information flowing across them has a high profile. This research uses three-wave panel surveys conducted with representative samples of Americans during both the 2012 and 2016 U.S. Presidential elections to assess whether use of social media for political information promoted endorsement of falsehoods about major party candidates or important campaign issues. Fixed effects regression helps ensure that observed effects are not due to individual differences. Results indicate that social media use had a small but significant influence on misperceptions about President Obama in the 2012 election, and that this effect was most pronounced among strong partisans. Social media had no effect on belief accuracy about the Republican candidate in that election. The 2016 survey focused on campaign issues. There is no evidence that social media use influenced belief accuracy about these topics in aggregate, but Facebook users were unique. Social media use by this group reduced issue misperceptions relative to those who only used other social media. These results demonstrate that social media can alter citizens’ willingness to endorse falsehoods during an election, but that the effects are often small.
Citation: Garrett RK (2019) Social media’s contribution to political misperceptions in U.S. Presidential elections. PLoS ONE 14(3): e0213500. https://doi.org/10.1371/journal.pone.0213500
Editor: I-Ching Lee, National Taiwan University, TAIWAN
Received: May 19, 2018; Accepted: February 23, 2019; Published: March 27, 2019
Copyright: © 2019 R. Kelly Garrett. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Replication data are published in Harvard Dataverse database (DOI: 10.7910/DVN/1JBL6U).
Funding: The author received funding from the National Science Foundation (IIS-1149599) and from the School of Communication at the Ohio State University for this work. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The author has declared that no competing interests exist.
On November 19, 2016, the New York Times’ editorial board published a scathing critique of Facebook’s failure to stop the spread of falsehoods in the lead up to the 2016 U.S. Presidential Election [1]. The opinion piece is emblematic of concerns about the threat that social media pose to democracy by corrupting citizens’ perceptions of political reality. Implicit in this claim is an assertion that the effects of social media on citizens’ belief accuracy are large, in contrast to the more limited effects associated with older media systems [2].
Reliance on social media for political news has increased rapidly. In 2012, about two in five Americans reported using social media for political purposes, and about one in three said they had encountered messages on social media promoting one of the candidates in the month leading up to the election [3, 4]. Four years later more Americans named Facebook as the site they most often used for political information in the month leading up to Election Day 2016 than named any other site, including those of high-profile news organizations such as Fox News, CNN, and major national newspapers (see Fig 1). This is troubling as online social networks have frequently been used to share political falsehoods, both about candidates and about important campaign issues [5–7]. Even more concerning, there is growing evidence that many of the falsehoods circulating during the 2016 U.S. Presidential Election were part of a Russian propaganda effort [8]. The idea that a foreign power would use social media in an effort to sow disinformation intended to sway an election is deeply troubling.
PowerPoint slide larger image original image Fig 1. Site most frequently used for political information in past month.Data collected from a representative sample of Americans, n = 618, by the GfK Group in the weeks immediately following the 2016 election, using weights.
Exposure to deceptive messages is not tantamount to belief in them. Individuals often exhibit credulity, and act in ways intended to prevent themselves from being misled [9, 10]. Nonetheless, in a complex information environment, individuals’ cognitive limits and biases do make them susceptible to (political) misinformation. People find messages to be more believable the more familiar those messages are [11], suggesting that repeated contact with falsehoods shared online will encourage their acceptance. More importantly, people are prone to believe messages that affirm their political viewpoint or identity [12–14] regardless of the strength of the evidence, which suggests that partisan falsehoods are particularly likely to take root.
Inaccurate beliefs threaten the foundations of democracy [15]. Citizens shape the social and political environment through their engagement with politics, and especially through their participation in elections. Falsehoods undermine democratic processes by distorting decision making. Support for a candidate or policy depends fundamentally on what one believes, and falsehoods can color individuals’ judgments, potentially leading them to support positions that run counter to their self-interests [16]. Of course, promoting inaccurate claims about candidates and issues is only one of many communication strategies used to manipulate voters. Advancing conspiracy theories, which weave together truth and fiction in ways that appear to justify attributing social phenomena to a small but powerful group of self-interested individuals, are a related approach [17]. Accurate information can also be used strategically, as when confidential communication is leaked (often selectively) in order to cast a political opponent in a negative light [18]. This article, however, is focused on endorsements of candidate and issue falsehoods.
Attributes of the technological environment could exacerbate the effects of deceptive political messaging. Empirical evidence suggests that email, which like social media tends to follow the contours of existing interpersonal relationships, promoted rumor acceptance in a prior election [19]. Exposure to partisan online news, which is frequently shared via social media, has also been shown to contribute to misperceptions [20]. The very “socialness” of social media can make individuals less likely to verify the information they encounter online [21]. Patterns of content engagement are also potentially damaging. Social media users can be divided into highly segregated communities based on what they choose to like, comment on, or share [22]. This does not mean that users in these communities never see information challenging their viewpoint [23], but the behavior pattern is harmful nonetheless. For example, the resulting segregation could undermine efforts to correct falsehoods circulating on social media because individuals are more responsive to corrections from those with whom they regularly interact than from strangers [24].
There are, however, reasons to expect social media’s effects to be more limited. Survey research conducted immediately after the 2016 U.S. Presidential election estimated that the average American likely saw only a handful of verifiably false news stories during the campaign season, and that many who saw these messages were at least somewhat selective about which they believed [25]. Observational data collected during the campaign season suggest that ‘fake news’ was shared relatively rarely, and was concentrated among small subpopulations [26]. Although there were times when falsehoods were shared more often than real news [6], it is important to recognize that political information typically has a much smaller reach than other types of content [27], which could lessen the risk of harm.
Limits on media systems’ ability to alter citizens’ attitudes and beliefs are well known. Concern about citizens’ susceptibility to war-time propaganda and rumoring motivated an extensive body of social science research beginning in the 1940s [28] and continuing to the present day. This literature suggests that citizens use a host of strategies to protect and reinforce their predispositions, including being selective about the information to which they expose themselves [29], and engaging in ideologically motivated interpretations of evidence encountered [14]. Indeed, how individuals respond to messages to which they are exposed is often contingent on their political predispositions. Politically palatable claims are more readily accepted than claims that are less compatible with their political worldview. Thus, media effects often vary by users’ political affiliation. Media are still influential, but their effects tend to be small and contingent on host of other factors. The conclusion that media effects are limited still holds today [2], which is not to say that they are unimportant. Even small changes in belief accuracy can have consequential downstream effects on political behavior, including vote choice [30].
This project brings empirical evidence to bear on the question of what influence social media use has on Americans’ political belief accuracy and whether these effects are contingent on users’ partisan identity or which social media platforms they use. Panel data collected during the 2012 and 2016 U.S. Presidential elections from representative samples of Americans suggest that the limited media effects paradigm persists in the face of these new technologies. Fixed effects regression conducted with longitudinal data are used to estimate the effect of social media use on political belief accuracy, controlling for all stable individual differences. These regression models are extended through the addition of interactions terms to test several likely moderators.
Data come from a pair of three-wave panel surveys conducted during the 2012 and 2016 U.S. Presidential Elections. In each election, a large, representative, general population sample of Americans responded to the same set of survey questions at three points during the election cycle. Surveys in both elections repeat measures of social media use and belief accuracy in each wave, but they differ in terms of the misperceptions assessed. The first focused on candidate misperceptions, and the second on campaign issue misperceptions. The goal was to better understand whether and how social media effects would differ across these two important types of falsehoods.
The GfK Group collected the data, drawing respondents from its KnowledgePanel, a general population panel recruited using address-based sampling methods. The Ohio State University Behavioral and Social Sciences Institutional Review Board approved the study (Study Number 2012B0247). Consent was given digitally, via online consent forms.
The 2012 baseline data were collected July 13-August 6, immediately after the major party conventions (N = 1,004). Wave 2 was conducted August 31-October 3 (N = 783; 79.2% completion rate). The final wave was fielded November 2-19, immediately after the election (N = 652; 84.3% completion rate). The 2016 baseline data were collected between July 29 and August 11, and included 965 valid respondents (62% completion rate). 763 (81% completion rate) completed wave 2, which was fielded September 14-22. The third wave included 625 valid respondents (83% completion rate) and was collected November 9-14. Participant demographics for both surveys were comparable to the U.S. population (see Tables A and B in S1 File for more information). Neither social media use nor belief accuracy significantly influenced attrition.
Data are analyzed using fixed effects regression, a statistical technique that greatly improves researchers’ ability to make causal inferences using panel data despite a lack of random assignment [31, 32]. Fixed effects models are widely recognized as an invaluable tool for assessing panel data in the social sciences [33]. By analyzing within-respondent changes over time, these models treat unobserved variables as stable parameters, controlling for the influence of unmeasured factors in the resulting estimates. Model coefficients can be understood as describing the change in an outcome variable corresponding to a one-unit change in the predictor holding constant all stable characteristics of each respondent. Although it is not possible to estimate the effects of time-invariant variables when using this statistical technique, we can test whether the effects of time-varying variables are conditioned on stable parameters through the use of interaction terms.
Significance testing is conducted using 90% confidence intervals for predictions about candidate beliefs given prior evidence suggesting that social media promote candidate misperceptions. Tests use 95% confidence intervals for campaign beliefs based on the greater uncertainty about the direction of the relationship. All confidence intervals are estimated using 10,000 bootstrapped samples.
The 2012 survey focused on Americans’ endorsement of false political statements about candidates following other studies in the literature on misperceptions [5, 19]. Misleading claims about candidates are widespread, and the personal nature of the attacks may make them uniquely effective campaign tools. The falsehood endorsement measure assessed respondents’ beliefs in eight falsehoods, four critical of each of the two major-party candidates, Barack Obama (D) and Mitt Romney (R) (see Table 1). Several dozen candidate statements were collected with the assistance of Amazon Mechanical Turk workers, and the survey design team selected a subset that were unambiguously false based on the assessments of credible fact checkers such as Politifact, FactCheck, and major national news organizations. The final set of statements represents a range of issues and includes items that vary in terms of the media coverage they received. The belief battery also includes one true statement about each candidate.