Insidiously Trivial: Meme Format Reduces Perceived Influence and Intent to Debate Partisan Claims

If citizens systematically respond differently to claims conveyed by memes, their effects on the broader information ecosystem may be underestimated. This US‐based study (N = 598) uses a 2 (partisan news/meme format) × 2 (congenial/ uncongenial message) design to examine perceptions of partisan memes’ influence on self and others, and the format’s effect on willingness to share disagreement in the context of partisan claims about corruption surrounding biofuels oper‐ ations. Results indicate that meme format enhances individuals’ tendency to see messages as less influential on oneself than on others and individuals less intent to share disagreement with claims presented in meme format. This decrease is mediated by the decrease in perceived influence over self. These findings call attention to the role format differences may play in the psychological processes underlying political discussion as it becomes increasingly mediated and visual.


Introduction
Concern about the quality of online information environment has inspired debate about the prevalence of misinformation and hyper-partisan news, public susceptibility, and potential mitigation approaches (Albright, 2017). However, this debate: Has mostly referred to one thing: the spread of inaccurate, misleading, or otherwise invented articles passed as real news. The fake news conversation has taken place in the realm of words, but that's missing a big part of the story. Much of the content that circulates on Facebook are images, often memes. (Renner, 2017, para. 1) In fact, as Renner (2017) points out, images were massively more popular than hyperlinks shared on Breitbart's Facebook page in 2016, for example. Therefore, understanding how the public perceives and responds to claims conveyed in the partisan meme format-"the perfect vessel for the spread of false information" (Renner, 2017, para. 15) is critical. This has implications for the composition of online information ecosystems in terms of the amount of both outright false information as well low-credibility, hyper-partisan content that circulates unchecked.
With high novelty value, political memes are popular subjects in mass media reporting (DeLuca et al., 2012;Freelon & Karpf, 2015;Huntington, 2013), where they are billed as curios of internet participatory culture (e.g., Miranda, 2016). Scholarship has tried to catch up to this moving target with descriptive and conceptual work undertaken by Milner (2012Milner ( , 2013, Shifman (2013Shifman ( , 2014, and others, often taking a qualitative approach to exploring media forms' cultural meanings (e.g., Rodley, 2016;Wetherbee, 2015). Some quantitative work has focused on their diffusion (Huntington, 2013) and usage (Chagas, et al., 2019;Martínez-Rolán, & Piñeiro-Otero, 2016;Moody-Ramirez & Church, 2019), but researchers seem to have neglected political memes from basic media effects and political psychology perspectives. This may be due in part to their deceptive appearance of triviality. But if citizens systematically respond differently to these messages than similar information in differing formats (e.g., more traditional forms of partisan media), then their effects on the broader opinion climate may be underestimated.
While political memes are not likely to cause vote switching in presidential elections, they may reinforce partisan attitudes or shape issue stances on low-information issues. And if they are a vehicle of misinformation that is less likely to receive interpersonal attention or correction, then they may be a weak spot in the information ecosystem. This study manipulated whether participants were exposed to political content in a common meme format-the image macro-or as a partisan news article and varied whether this content was oppositional or congenial to participants' party affiliation in the US.
Results show that participants saw partisan memes as less likely to influence both others' and their own views and were less likely in turn to say they would share disagreement. These findings suggest prominent but trivialized elements of socially mediated political communication may decrease the chances of deliberative or corrective exchanges.

Partisan Memes and the "Image Macro"
Like other new media formats before them, studying memes has become critical to "understanding the fabric of opinion formation," as it changes with technology and social trends (Banning & Sweetser, 2007, p. 453). Internet memes are "multimodal symbolic artifacts created, circulated, and transferred by countless mediated cultural participants" (Milner, 2013(Milner, , p. 2359. Of these media, those arguably most commonly referred to as political memes are partisan "image macros": template-based single images superimposed with two lines of bold text (Börzsei, 2013;Lyons, 2017;Rintel, 2013;Vickery, 2014). This easily recognizable format may influence reactions regardless of a meme's content (e.g., Schmierbach & Oeldorf-Hirsch, 2012;Veenstra et al., 2015). However, it is worth noting that political memes encompass a sprawling set of media objects beyond the image macro, though these are not directly examined here.
Political memes combine a number of qualities of older political media, and in other ways transcend these (Lyons, 2017). Several of these qualities may matter in terms of perception and response. Political memes are created by anonymous amateurs who generally cite no sources, and remediation further obscures their origin (Rodley, 2016). Like more traditional political satire (Becker et al., 2010), they attempt both humor and persuasion. They also tend to inject politics into casual social spaces (da Silva & Garcia, 2012;Lyons, 2017). For these reasons, citizens may be wary of political memes and be motivated to reject their claims (Banning & Sweetser, 2007;Gunther & Thorson, 1992;Paradise & Sullivan, 2012). Like political advertising, they tend to malign or ridicule political figures or parties (Chagas et al., 2019;da Silva & Garcia, 2012;Moody-Ramirez & Church, 2019), and so may motivate greater backlash among partisans who feel attacked (Becker et al., 2010;Veenstra et al., 2015).
At the same time, political memes are deeply rooted in internet subculture (Milner, 2013), and are often absurd (Chagas et al., 2019;Jurgenson, 2012;Katz & Shifman, 2017), objectively wrong, or obnoxious even to in-groups. For these reasons, they are trivialized in media coverage (Huntington, 2013), and citizens may likewise look down their noses at them. These perceptions matter because of the behaviors they encourage. If viewers see them as less consequential, they may not bother to correct them. The broader opinion climate may then suffer as biased information and misinformation goes unchallenged (Neubaum & Krämer, 2016).

Presumed Influence
A few related literatures explore how the presumed influence of media messages motivates behavior. Most prominently, the third-person effect hypothesis posits that third-person perception-the belief that others will be more influenced by a message than oneself-often spurs action, such as censorship or correction (Davison, 1983). Individuals see themselves as less susceptible to persuasion, particularly from what they perceive as low-quality sources, and particularly when hypothetical consequences of a message are socially undesirable (Gunther & Mundy, 1993). Researchers have found third-person perception across a wide range of media forms, including political advertising, satire, and social media (e.g., Banning & Sweetser, 2007;Becker et al., 2010;Gunther & Thorson, 1992;Paradise & Sullivan, 2012).
The evidence regarding the behavioral component is less clear, though, particularly when behaviors beyond censorship are considered (Xu & Gonzenbach, 2008). Based on the third-person effect literature, other scholars have forwarded a related theory of the "influence of presumed media influence" (Gunther & Storey, 2003). Instead of the gap in perceived effects on self and others driving behavior, this theory focuses on a more general belief in a message's influence (Cohen & Tsfati, 2009;Cohen et al., 2008;. Both third-person perception and overall presumed influence have been linked with behavioral adjustments, including corrective actions (Barnidge & Rojas, 2014;Rojas, 2010;Sun et al., 2008). While third-person effects and the influence of presumed influence are thought of as complementary rather than competing hypotheses (Gunther & Storey, 2003), some studies show the two variables do not necessarily translate into equivalent behavioral or attitudinal outcomes (Sherrick, 2016).

Corrective Action
In recent years, scholars examining the behavioral component of presumed influence have focused on corrective action (Barnidge & Rojas, 2014). The corrective action hypothesis holds that individuals respond to presumed media influence by expressing their own counteropinions or otherwise "correcting" views and claims they see as wrong in the public sphere (Rojas, 2010;Sun et al., 2008). Scholars have forwarded models of corrective action stemming from both third-person perception (Lim & Golan, 2011;Sun et al., 2008) and overall presumed influence (Barnidge & Rojas, 2014;Rojas, 2010). These studies provide evidence that both are linked to behaviors that for the participant rehabilitate or improve public debate.
This could take the form of "social media activism" in which individuals seek to counter political messages' influence by posting refutations (Lim & Golan, 2011). Those who perceive memes as misguided and influential on others may seek to correct them. However, if individuals see memes as irrelevant, they may choose to refrain. Willingness to correct one's peers on social media platforms is important because professional outlets are unable to do so (Shelly, 2017), because such unchallenged claims can distort perceptions of the opinion climate (Neubaum & Krämer, 2016), and because peer corrections are effective Hannak et al., 2014;Serrano, 2017;.

Hypotheses and Research Questions
Based on the prior review, I formulate the following hypotheses: The first group assesses the effects of the meme format-that is, the effects of presenting messages in the classic "image macro" format that superimposes two lines of bold white text over an image. This study first seeks to confirm the existence of thirdperson perception regarding such formatting, hypothesizing that this perceptual gap will be greater for claims presented in meme format than in traditional partisan media format, and that meme format will reduce corrective intent relative to partisan media format. H1: Partisan meme format induces greater third person perception (TPP) than traditional partisan media format.
H2: Partisan meme format induces lower corrective intent than traditional partisan media format.
Next, this study asks if the effects of partisan meme format on perceptions of influence and willingness to correct are conditional on the content's slant. RQ: Does the partisan congeniality of the message moderate the effects of format on beliefs or corrective action?
The final set of hypotheses address the mechanisms of format effects (though note that these tests are nonetheless correlational). The literature provides little guidance regarding media content that may decrease presumed influence and corrective action. For the sake of reconciling previous work (Lim & Golan, 2011), this study tests competing hypotheses regarding presumed influence (TPP and total presumed influence [TPI]) and behavioral outcomes (Sherrick, 2016). Perceived influence is then posited as the mediation path between message format and behavioral intention.
H3a: TPP is associated with greater corrective intent.
H3b: TPI is associated with greater corrective intent.
H4: Partisan meme format's reduction of corrective intent is mediated through reduced presumed influence.

Sample
Hypotheses were tested using a between-subjects online experiment in March of 2016. Five hundred and ninety-eight participants were recruited via Amazon Mechanical Turk in March 2016 and compensated with $0.75. Demographically, Turkers are marginally more diverse than the typical Internet sample, significantly more diverse than an undergraduate sample (Buhrmester et al., 2011), and "exhibit the classic heuristics and biases," (Paolacci et al., 2010, p. 417). Mullinix et al. (2015) conducted a series of parallel experiments, comparing effects across a nationally representative survey sample, a Mechanical Turk sample, a student sample, and other convenient samples. They found that not only were Mechanical Turk samples' effects in the same direction as those of a national sample but of the same significance threshold and similar magnitude for each topic examined.
Participants were 53% female and 75.4% white, with a mean age of 39.22 (SD = 13.5), median education of a bachelor's or associate's degree, and median income of $20-40K. They were 44.1% democrat, 19.9% republican, and 36% independent. Accounting for independents who leaned toward one party or another, the participants were 57.5% democrat, 27.3% republican, and 15.2% independent.

Design and Procedure
The experiment employed a fully crossed 2 (meme/ partisan news article) × 2 (congenial/uncongenial content) design. A partisan dispute over advanced biofuels operations was chosen as a controversial topic for the stimulus (Fung et al., 2014); this low-salience issue was chosen to reduce pre-treatment exposure effects (Druckman et al., 2010). Participants were exposed to either an image macro (Börzsei, 2013) style political meme or a partisan news article about biofuels funding. The photograph in each was held constant. Partisan news articles were depicted as being posted by Breitbart or Huffington Post Facebook pages. The anti-Democratic Party meme included the text "Big Biofuels shut down? Where will Dems get handouts now?" while the anti-Democratic article (from Breitbart) included the headline "Democrats lose source of handouts with biofuels shutdown in NC [North Carolina]." The anti-Republican meme included the text "Repubs in big oil's pocket? Better shut down biofuels," while the anti-Republican article (from Huffington Post) included the headline "Bending to big oil, Republicans shut down biofuels operation in NC." The number of likes and shares was redacted. Stimuli materials can be seen in Figure 1 of the Supplementary Material.
Participants were told the meme or article was popular on social media following the controversy and answered questions about their perceptions of the content's potential influence over themselves and others, as well as the likelihood of sharing disagreement via social media. The survey experiment took place in the context of a larger survey, following an experiment analyzed as a separate study (Lyons, 2018). Specifically, participants had previously engaged in brief writing exercises before selecting discussion partners and news stories, and then viewing news video to test hypotheses relating to the mitigation of partisan bias. Using Transue et al.'s (2009) procedures, I find no spillover effects of the writing task on the current study's outcomes (see Table A1 of the Supplementary Material).

Independent Variables
Based on the manipulations, a dummy variable for meme format and a three-level oppositional content variable constructed using valence of the content and respondent party affiliation (45.2% opposed, 39.6% supported, 15.2% neither opposed nor supported) were employed.

Dependent Variables
Third-person effects researchers favor competing analytical approaches to perception of influence. One set of authors operationalizes third-person perception as perceived influence on self subtracted from influence on others, and TPI as the sum of self and other items (McLeod et al., 1997;Neuwirth & Frederick, 2002;Schmierbach et al., 2011). Another set of authors include both self and other items as conjoint predictors of behav-iors, thus controlling for presumed influence on the self (others) when examining the effects of presumed influence on others (self; Cohen & Tsfati, 2009;. As recommended by Schmierbach et al. (2008) and Sherrick (2016), both methods are employed and reported below.
Presumed influence on others (M = 4.80, SD = 1.39) and presumed influence on self (M = 3.62, SD = 1.71) were measured on seven-point scales. Third-person perception was computed by subtracting presumed influence on self from presumed influence on others (M = 1.18, SD = 1.63). TPI was computed by summing other and self-measures (M = 8.42, SD = 2.65).
Corrective action was likewise measured on a sevenpoint scale (M = 3.14, SD = 1.70) based on the statement "I would be likely to share my disagreement with it on social media." Importantly, Rojas (2010) and subsequent work on "corrective action" define such behavior as any actions that contest the influence of media messages-including expressing opinions, communicating one's views, voting, persuading others to vote a given way, or other attempts to sway public opinion. This is an important point, given that the present study touches upon how users may respond to misinformation or misleading content; the corrective action hypothesis does not refer specifically to fact-checking endeavors, but discussion more broadly. In studies of social media activism in response to presumed media influence, this concept has been measured with items such as "how likely would you be to leave a negative comment…?" (Lim & Golan, 2011). We take a similar approach and ask about intent to share disagreement.

Random Assignment Check
In addition to the sample's demographics described above, two further variables were included in the random assignment check based on their potential as confounds. Strength of party affiliation was measured as strong (52%) or not strong (48%). ." Random assignment was checked using analysis of variance, which showed no significant differences in age, gender, race, education, income, party affiliation, strength of affiliation, or network homogeneity across either the meme format factor of the oppositional content factor.

Results
H1 and H2 were tested using ordinary least-squares (OLS) regression models, with a meme format dummy variable and a three-level oppositional content variable, as well as their interaction term. Results of the first model support H1: Messages induced greater third-person perception when conveyed in a meme format than when presented as a partisan news article ( = 0.16, p < 0.001). Likewise, the second model showed support for H2: Meme format decreased corrective intent ( = −0.11, p = 0.011). While oppositional content increased third-person perception and correction intention (which is in line with prior work), slant did not interact with format. That is, addressing RQ, the effects of meme formatting were not moderated by messages' alignment with participants' party affiliation. The full results of these models are reported in Table 1, in the first two columns.
In anticipation of the two-pronged approach to assessing the relationship between presumed influence and behavior (Schmierbach et al., 2008;Sherrick, 2016), two further OLS analyses were conducted to model the individual components of third-person perception and TPI-perceived influence on self and on others-as outcomes of message format. Reported in Table 1, in the third and fourth columns, meme format reduced perceived influence on self ( = −0.24, p < 0.001) and others ( = −0.11, p < 0.001).
As indicated previously, two hierarchical linear regression models were employed to analyze the effects presumed influence on corrective intent (H3). The first included third-person perception (others − self) and TPI (others + self) simultaneously, while the second included self and other influence simultaneously. In both cases, format and consonance were included in the first block, with presumed influence variables included in the second block.
Results of the first model showed that the perceptual gap variable was not a significant predictor of correction ( = −0.05, p = 0.205). TPI was, however ( = 0.15, p < 0.001). It is noteworthy that the meme format became nonsignificant with the addition of the second block to the model, suggesting its effect may be mediated by TPI.
Results of the second model showed that presumed influence on the self ( = 0.16, p < 0.001), but not on others ( = 0.04, p = 0.429) predicted correction. Again, meme format became nonsignificant with the addition of the block of perceptual variables, suggesting presumed influence on self, in particular, mediates the format effect. The full results of these models are reported in Table 2.
Together, these tests suggest that a perceptual gap is not the mechanism of the meme format's diminishing of corrective action. Unlike third-person effects demonstrated elsewhere in the literature, where concern about the effect of a media message's influence over others appears to motivate corrective behavior (e.g., Lim & Golan, 2011), this outcome suggests that partisan memes instead discourage correction by reducing overall presumed influence, and particularly perceived influence on the self.
Before discussing the formal mediation analysis, it is important to note shifts in understanding of mediation models in recent years. In particular, mediation tests assume no confounding bias as applied to the X to M and M to Y paths (i.e., the sequential ignorability assumption; Imai et al., 2010). Because this study, like most others and those dedicated to third-person effects in particular, only randomizes levels of X, not M, these assumptions are not fully upheld. Still, the tests below are included as they might speak to prior work regarding presumed media influence, though they should be viewed in light of this limitation.
To formally test the proposed mediation pathway (H4), Hayes' (2013) PROCESS Macro (Model 4) was employed. Meme format was entered as the independent variable, influence on self as the mediator, and corrective intent as the dependent variable. Message consonance and presumed influence on others were entered as covariates. The 5,000 bootstrap sample procedure generated a 95% bias-corrected confidence interval that did not include zero (−0.205, −0.029) for the indirect effect of meme format on correction through self-influence. After accounting for self-influence, the direct relationship between format and behavior became insignificant Table 1. Effects of media format and consonance on perceptions of influence and correction.

Discussion
By focusing on perceptions of and responses to partisan memes, this study contributes novel insight into a popular communicative form that has nevertheless flown under the radar of media effects research. I find that partisan claims engender less corrective intent when conveyed by partisan memes than by partisan news articles. Mediation tests show this is due to a decrease in the message's presumed influence over the self. In other words, people see partisan memes as trivial, and not worth corrective efforts. For this reason, however, memes may present a highly effective vehicle for the spread of misleading claims or outright misinformation. Not only are they often more likely to be shared than traditional news links (Renner, 2017), they are less likely to attract corrective efforts of professionals or peers. Image-based memes may therefore serve as a loophole for those wishing to intentionally mislead others online (Marwick & Lewis, 2017;Renner, 2017). Importantly, this study shows how presumed influence can explain instances where individuals choose not to act. Rather than see partisan memes as low-quality information sources from which they must protect others, individuals instead see less reason to engage in persuasive or informative efforts in the face of messages to which they feel impervious. As with Sherrick (2016), it is important to note that a third-person perceptual gap was present, but not associated with behavioral intention. This lends further support to the notion that the belief that others are more vulnerable to media than oneself and the belief that media has generally derogative effects are not always equivalent predictors of attitudinal or behavioral response (for a more in-depth discussion of why discrepancies may appear in studies examining the downstream behavioral effects of third-person perception, see Lyons, 2022).
This study also complements prior qualitative, rhetorical, and descriptive approaches to memes by providing an initial understanding of the psychological processes involved when individuals encounter partisan memes on social media platforms. Similarly, the findings add texture to recent technical reports calling attention to memes' potential roles in disinformation campaigns, though the stimuli tested here do not represent disinformation per se (Gorwa, 2017;Marwick & Lewis, 2017). Likewise, on the practical front, these findings might inform future expansions of efforts to enhance media literacy (Mullin, 2017).
However, this study is also limited in a number of ways that may be supplemented by future work. First, the study employed a measure of behavioral intention rather than observed behavior. While behavioral intentions are typically antecedents of behaviors (Ajzen, 1985), this approach could be extended with computational efforts using large social platform datasets, to observe how citizens actually systematically respond to different media formats. As an additional threat to external validity, the effects found here may not necessarily generalize to other types of memes beyond the image macro and to political topics beyond biofuels. While it is wise to be circumspect about the generalizability of the findings, it seems less likely that these effects are due to the biofuels-centric message but rather the partisan framing of any low-salience political issue. In other words, to assume that the effects of the meme format (lesser presumed influence, lesser corrective intent) relative to the partisan headline conveying the same claims about biofuels are due to the biofuels content, rather than the meme format, is to assume that there is a specific interaction effect whereby meme format drives down presumed influence relative to traditional media formatting, but only for biofuel-related partisan content. I do not see this as particularly theoretically plausible. Overall, single-message design is a common limitation for experimental research in communication (Pingree et al., 2014), and can be aided through replication or designs employing multiple message versions. Admittedly, this is the first step in what should become a line of research dedicated to examining visual communication effects in online political discourse. It is my hope that I and others will replicate and extend upon this finding.
It may also be the case that the stimuli construction influenced respondents' intent to share disagreement, as the image macro versions of the claims employed rhetorical questions (e.g., "Big Biofuels shut down? Where will Dems get handouts now?"). Some may view such a question as hard to share "disagreement" with, but in my view, partisans likely understand such questions to be equivalent to a claim that Democrats are receiving handouts (as is made in the headline). It should also be mentioned that although the results should speak to perceptions of misinformation conveyed across formats, it may be argued that the stimuli do not represent misinformation per se, but rather partisan slant. It is my view that the messages represent a form of unsupported claim-that the opponent party is unethically accepting handouts in return for policymaking. In any event, whether we are concerned with falsified information or hyper-partisan content more broadly, it is worth asking whether mere format differences can distort public response.
It is also an inevitable fact that (reported) engagement with media will differ between artificial exposure and when embedded in real-world social networks. Notably, though, this study is not attempting to explain engagement decisions per se, but rather whether the formatting of messages can influence this in the abstract. That is, all else equal, I ask whether meme format itself exerts any influence on engagement intent. It is possible that the influence of formatting effects and social connections interact such that individuals may be especially more or less likely to engage with a meme-relative to other political content-when posted by a friend rather than a stranger.
Most importantly, more dimensional work, in general, is needed to understand the nascent formatting of memes. For example, as conversations surrounding memes take place on social platforms, future research should consider the interplay of social and political cues (Messing & Westwood, 2014), the implications of context collapse (Davis & Jurgenson, 2014;Shmargad & Watts, 2016), and the role of perceived network heterogeneity (Veenstra et al., 2017) in individuals' corrective decisions. Likewise, the effects of intramedium interaction (Lyons & Veenstra, 2016;Veenstra et al., 2015), whereby comments on shared posts might reframe and alter responses, should be examined. In other words, more work is needed to understand how social media users perceive their need to act based on the presence of various overlapping social signals. With an eye toward making those corrections more effective, experiments can suggest which might be the best forms and sources of evidence for specifically debunking a meme vs. other media . Lastly, researchers may determine whether the threshold for inducing familiarity effects, wherein information comes to be seen as truer through repetitious exposure (Weaver et al., 2007), varies across media formats.
Regardless, this study calls attention to the role of format differences in the psychological processes underlying deliberation and political discussion as it becomes increasingly mediated and visual (Hendriks et al., 2017;Lyons, 2017). In doing so, it provides initial empirical evidence about a commonly overlooked form of contemporary political discussion. Political memes warrant further attention, even if it is their very triviality that poses consequences for the public.

Conflict of Interests
The author declares no conflict of interests.

Supplementary Material
Materials, data, and analysis script can be found at https://osf.io/pu2sb.