Abstract
Researchers condemn the effects of news but have only recently turned their attention to determining the extent to which individuals engage with news. Within the context of online uncivil news, the current project investigates whether negativity always increases engagement with news. The results of two experiments demonstrate that civility in the news increased news engagement, especially compared to news with the most incivility. News articles that included multiple types of incivility and news articles that prompted individuals to perceive that an out-group political party was behaving uncivilly discouraged people from engaging with online news. The studies contribute theoretically to negativity bias and incivility research and signal that negativity does not always attract clicks.
Political media and campaigns have been studied, and criticized, for the prevalence of negative messages (Ansolabehere & Iyengar, 1995; Patterson, 1994). People pay attention to negative tweets about news (Kätsyri, Kinnunen, Kusumoto, Oittinen, & Ravaja, 2016) and weigh negative information about candidates more heavily than positive information (Lau, 1985). Contact with impolite exchanges leads to decreases in political trust, perceptions that arguments are illegitimate, and less openmindedness (Borah, 2014; Mutz, 2015). Yet, despite decades of research investigating political negativity, the question of whether negativity provokes individuals to engage with news or whether it pushes people away has not been answered.
Media channels have multiplied since the rise of the Internet, giving people more control over the news they follow (Prior, 2007). By testing effects of forced media exposure, researchers may be over-estimating the influence news has in the world outside of a laboratory (Arceneaux & Johnson, 2013), including the influence of negative news. Online news represents an ideal setting in which to test news engagement due to the opportunities users have to select news that appeals to them (Knobloch-Westerwick & Meng, 2009) and digitally interact with articles once they have chosen to read the articles (Stroud, Scacco, & Curry, 2016). This project uses the context of online news to investigate the ways in which people do or do not engage with negativity.
On one hand, negativity may lead individuals to select news articles and to engage with news articles once they have decided to read the news. Negativity bias is the well-documented phenomenon that negative information weighs more heavily on individuals’ minds than positive information (Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; Ito, Larsen, Smith, & Cacioppo, 1998; Rozin & Royzman, 2001). News selection scholars have argued that, because negativity attracts attention, it should increase the likelihood that news users will interact with negative news (Jang & Oh, 2016; Meffert, Chung, Joiner, Waks, & Garst, 2006; Mutz, 2015).
On the other hand, aversion to negative news is a distinct possibility. When individuals are asked to make decisions about protecting communities from deadly disease, they choose options described as gains (that is, lives saved) rather than losses (that is, lives lost), even when the options are otherwise identical (Kahneman & Tversky, 1984). In online settings, positive information often is shared more than negative information (Berger & Milkman, 2012; Eckler & Bolls, 2011; Guadagno, Rempala, Murphy, & Okdie, 2013). Thus, even though negativity might gain attention, it may not always prompt more interaction.
Within the context of online news coverage of incivility in Congress, the current project investigates whether negativity always increases engagement with news. The context of incivility among politicians is important because journalists “have a professional responsibility to attend to the quality of public discussion” (Gutmann & Thompson, 1996, p. 125). Whether news users gravitate to coverage of quality public discussion, however, is not clear. Investigating coverage of politicians behaving uncivilly also extends the theoretical understanding of negativity bias in a few ways. First, by studying types of incivility, including personal-level incivility (impoliteness; see Mutz, 2015), public-level incivility (lack of reciprocity; see Uslaner, 1996), or both, we test whether the amount of negativity affects engagement with news. Second, by contrasting incivility with civility among politicians, we test whether negative news can produce a backlash and promote engagement with positive news. Third, by examining the indirect effects of incivility on engagement, we investigate the mechanisms through which negativity leads to behavioral effects. The results of two experiments demonstrate that civility increased news engagement, especially compared to news with the most incivility. News that included multiple types of incivility and news that prompted individuals to perceive that an out-group political party was behaving uncivilly discouraged people from engaging with online news.
Uncivil and Civil News
What counts as “negative” in studies of negativity bias varies widely. Extreme negativity—for example, an image of a gun pointed at a viewer, a drowning child, a bloody face, or close up video of politicians yelling (Ito et al., 1998; Knobloch, Hastall, Zillmann, & Callison, 2003; Mutz, 2015)—has been tested in visual settings. In online news contexts, however, the majority of people read news texts rather than watch videos (Levy, Newman, & Nielsen, 2015). Although intense negativity is quite common in text-based newspaper columns and blogs (Sobieraj & Berry, 2011), the consequences of this strong textual negativity are unknown. Three studies to date have examined whether people choose negative political information in textual settings. Two of the studies investigated whether individuals selected headlines with relatively mild negative words like “accuse,” “oppose,” and “criticize” compared with relatively mild positive words like “praise,” “hail,” and “support” (Meffert et al., 2006, p. 34; see also, Jang & Oh, 2016). Both studies found that people gravitated toward negative headlines. Zillmann, Chen, Knobloch, and Callison (2004) also found that individuals were drawn toward negative news content, particularly conflict and agony, but in contrast to factual news rather than positive news content. Yet stronger negativity might prompt aversion to information and decrease interactions with the news (MacKuen, Wolak, Keele, & Marcus, 2010), especially in situations when individuals have a contrasting positive news option.
To amplify the contrast between negative and positive news, we test whether news coverage of politicians behaving with different levels of incivility influences the way individuals interact with online news. Incivility is broadly understood as a negative violation of political norms (e.g., Ben-Porath, 2010; Sinopoli, 1995). Although incivility has been notoriously difficult to conceptualize, two general approaches to political incivility have emerged: a personal level and a public level of incivility. The first approach, which appears often in studies examining media effects, defines incivility as any action “that violates the norms of politeness” (Mutz, 2015, p. 6). Founded on interpersonal politeness theory (Brown & Levinson, 1987) and the argument that television makes politics personal (Hart, 1999; Mutz, 2015), this personal level of incivility includes interactions with politicians who threaten the face (or self-esteem) of political opponents, for instance, by using name-calling, ad hominem attacks, and rudeness in their interactions with others (e.g., Brooks & Geer, 2007; Mutz, 2015). Mutz (2015) examined the effects of personal-level incivility in newspapers, but, unlike the current project, she did not test whether people selected to read impolite news articles more often than polite news articles.
A second approach, which more often appears in theoretical overviews of incivility than in media effects research, conceptualizes incivility as more public than personal. Public-level incivility is closely related to democratic governing processes (Orwin, 1992), like disrespecting the rights of other citizens (Papacharissi, 2004) or refusing to make arguments based on the common good (Rawls, 1993). This conceptualization is related to what Uslaner (1996) described as comity: the political norm that views other than one’s own are “legitimate” and that “fair exchange” among politicians with varied political views is necessary (p. 8). Politicians who violate the comity norm refuse to compromise with members of the opposing political party and use political tactics in legislative bodies to produce gridlock (Davetian, 2009; Uslaner, 1996). Some elements of a public-level conceptualization of incivility have made their way into content analyses of online news comments (e.g., Papacharissi, 2004), but effects research has not examined the consequences of news that emphasizes lack of comity among political figures.
The personal and public types of incivility contrast with political civility, which is rooted in the idea of interpersonal respect (Mutz, 2015) and reciprocity for others in the political system (Uslaner, 1996). In studies of personal-level incivility, civility often is operationalized as an absence of incivility. The civil sentence “Democrats often criticize Bush for renouncing the Kyoto Protocol” becomes uncivil by adding the word “whiners” after “Democrats” (Thorson, Vraga, & Ekdale, 2010, p. 299) or an interview becomes civil by removing yelling and eye rolling from a video (Mutz, 2015). Taking the public-level approach to incivility into account allows researchers to further distinguish between civility and incivility. Civility, from a public perspective, also includes politicians who attempt to work together to solve problems, even if they disagree on policy positions, an approach related to public deliberation (Fishkin & Luskin, 2005; Gutmann & Thompson, 1996). Thus, civility can be conceptualized as both a sense of politeness and a sense of reciprocity toward other politicians who disagree.
Only one study has examined the draw of incivility in politics. Mutz (2015) found that impoliteness on political television attracts viewers. The current project differs in two substantial ways. First, we examine online, text-based incivility rather than televised incivility. Second, Mutz (2015) conceptualized incivility as personal-level impoliteness whereas we also study public-level incivility, focused on the abject refusal to work toward a common goal, in our news texts. In sum, the current project examines whether people are more likely to interact with online news that emphasizes political incivility compared with civility.
Theoretical Foundations of Negativity Bias
Beyond studying incivility for the democratically normative implications of the topic, determining whether civility or incivility drives interactions with news also uncovers information about the theoretical mechanisms undergirding negativity bias. Shoemaker (1996) posited that bad news is prevalent due to biological and cultural processes that lead people to attend to deviance—in this case, deviance from norms of politeness and reciprocity. More recently, two specific theoretical mechanisms have been posited in negative news research to explain effects: expectancy violations theory (Meffert et al., 2006; Soroka, 2014) and automatic evaluative-categorization processing (Ito et al., 1998; Mutz, 2015). The first explanation is based on expectancy violations (Meffert et al., 2006; Soroka, 2014), a theory developed to predict the consequences of interpersonal interactions that either meet or disrupt individuals’ expectations (Burgoon, 1993). As Soroka (2014) and Rozin and Royzman (2001) contend, since most information in people’s day-to-day lives is positive, negative information violates expectations and stands out in comparison with the expected positive information.
A second explanation is based on automatic evaluative-categorization processing of negativity. Proponents of this mechanism explain that negativity draws individuals’ attention not because it is unexpected but because people attend to negativity early in the cognitive process as they form initial, automatic evaluations of a message (Carretié, Albert, López-Martín, & Tapia, 2009; Ito et al., 1998). As described by psychologists Carretié et al. (2009), negativity is processed quickly by the brain so as to help people recognize and avoid dangerous situations. In Mutz’s (2007) words, individuals have “gut-level” reactions to negative content (p. 634).
Although both mechanisms have led researchers to equate negativity bias with engagement with news, the two explanations of negativity bias could predict a behavioral backlash to negativity. Automatic processing of negativity suggests that negativity bias is, at first, attitudinal and evaluative, not necessarily behavioral (Ito et al., 1998). The decision to respond to a stimulus comes after evaluating the stimulus (Carretié et al., 2009), leaving room for people to either interact with negative news or, once they recognize the negative information, to avoid it. Similarly, expectations violation theory could theoretically predict interaction with positive news. Negativity might be less prevalent in everyday life than positivity (Soroka, 2014), but negativity (Patterson, 1994) and incivility (Sobieraj & Berry, 2011) are prevalent in news. If individuals typically expect to encounter negative news about politics, then the civil news would be unexpected and increase the likelihood of interaction.
Thus, although there is scholarly consensus that negativity weighs strongly on individuals (Ito et al., 1998; Meffert et al., 2006; Rozin & Royzman, 2001), it is unclear whether this means that individuals will always engage with negative information more than positive information. Furthermore, because previous news selection research in online settings has focused largely on negative, rather than uncivil, information choices (e.g., Jang & Oh, 2016; Meffert et al., 2006), the negativity may not have been strong enough to induce a backlash. Given the conflicting predictions, we raise the following competing hypotheses:
Both explanatory mechanisms predict that the most negative information will have the greatest effects, as the negativity will either be the most unexpected (Soroka, 2014) or evaluated as the most dangerous (Carretié et al., 2009). Incivility, in its multiple forms, allows this prediction to be tested. News can cover positive behaviors of politicians (civility), each type of incivility (personal or public) on its own, or both types of incivility together within the same article. This last type of coverage, because it contains the most negativity, should have the strongest effects compared with positive, civil coverage. News that includes both personal-level and public-level incivility should encourage either the most engagement or the least engagement with news compared with news coverage of civility, leading to the following competing hypotheses:
The current project also compares the two theoretical mechanisms to determine whether either explains the effects of news negativity. Of the previous examinations of news negativity, two studies suggested that violations of news expectations drove the results (Meffert et al., 2006; Soroka, 2014) and one concluded that an automatic reaction to incivility drove the results (Mutz, 2015). Yet none directly tested whether their proposed mechanisms led to selection of news nor did they compare theoretical explanations within their studies. As O’Keefe (2003) has argued, communication researchers are in an ideal position to test whether messages affect outcomes through psychological states of individuals. Specifically, O’Keefe (2003) contended that researchers should offer hypotheses that connect manipulated messages to effects indirectly through the psychological states generated by the messages. We follow O’Keefe’s advice in the current project by measuring violations of news expectations, as well as automatic evaluation-categorization, to test whether either, or both, of these proposed mechanisms indirectly connect incivility and civility in the news to interactions with the news.
We turn first to the expectancy violations mechanism. If Soroka (2014) and Meffert et al. (2006) are correct, unexpected news content should influence the amount of interaction with news. This process could occur for either the uncivil or civil news. On one hand, the majority of news tends to be tonally neutral (Lengauer, Esser, & Berganza, 2012), suggesting that uncivil news will violate expectations compared with the news individuals typically consume. On the other hand, though incivility and negativity do not make up the majority of content, they are prevalent in news (e.g., Lengauer et al., 2012; Sobieraj & Berry, 2011), suggesting that civil news will violate expectations compared with the news individuals typically consume. Given the importance of news choice in the current fragmented media environment (Prior, 2007), we do not investigate expectations about news content in general. Instead, we expect that participants will compare the news we provide to news they experience in their day-to-day lives. If the content we provide aligns with what participants typically consume, their expectations are met, and interactions with the news should be low. If, the content we provide is not in line with the news they typically consume, their expectations are violated, and interactions with the news should be higher.
Alternatively, if automatic evaluative-categorization processing drives negativity bias, which is in line with Mutz’s (2015) finding that incivility on television increases physiological arousal, then we expect that a basic recognition of incivility in the news will lead to news interactions. Automatic evaluative-categorization processing only requires recognition that negativity, or in this case, incivility, is present in an article for incivility to have an effect on news interactions. In addition, if automatic evaluation of negativity leads to interaction effects, the effect should be amplified when partisan group status is taken into account. In line with self-categorization theory (Tajfel & Turner, 1979), partisan politics prompts individuals to categorize themselves as members of in-group and out-groups. People who identify with the Democratic Party apply in-group thinking to news about Democrats and out-group thinking to news about Republicans (and vice versa; for example, Hartmann & Tanis, 2013). Individuals react more strongly to negative behaviors enacted by their out-groups than their in-groups (Howard & Rothbart, 1980). Thus, the automatic evaluative-categorization processing explanation of negativity’s effects will receive support if participants recognize incivility when viewing news coverage of politicians from the out-group political party and that recognition predicts interactions with news. To summarize, the current project extends negative news research by testing the mechanism through which uncivil or civil news content leads to interaction with the news:
Study 1
Study 1 Method
For the first study, we created a news website and unobtrusively observed individuals’ news selection behaviors on the site. The quasi-experiment prompted participants to select from civil and uncivil news headlines. We also manipulated the stories to feature like-minded, counter-attitudinal, and balanced partisan news. Partisan lean of the headlines did not significantly affect selection of the news stories, so we do not discuss the results of this content characteristic. The design mirrors previous news selection research (e.g., Knobloch-Westerwick & Meng, 2009).
Participants
Participants were recruited between August 28 and September 8, 2014, using the online sampling firm Survey Sampling International (SSI; see also, Stroud, Muddiman, & Scacco, 2016). In total, 603 participants completed the study. Only the participants who clicked on at least one article (n = 387) were examined. Of these participants, 63% were female, 74% identified as White/Caucasian, 14% as Black/African American, 7% as Asian, 2% as American Indian, and 3% as another race. Sixteen percent of the participants identified as Hispanic/Latino. Participants had an average of 14 years of education (SD = 2.35). There was a range of household incomes, with 34% reporting less than US$30,000 per year, 19% reporting US$30,000 to US$50,000, 17% reporting US$50,001 to US$75,000, and 30% reporting more than US$75,000. Participants varied politically, though they leaned liberal—M = 2.79, SD = 1.22; 14% very liberal (1), 23% somewhat liberal (2), 42% moderate (3), 12% somewhat conservative (4), 9% very conservative (5)—and Democratic—M = 2.22, SD = 1.05; 31% strong Democrat (1), 31% lean Democrat (2), 23% lean Republican (3), 15% strong Republican (4). Although not selected randomly, the participants came from diverse backgrounds. The national sample allows us to extend research based on news selection studies that have used student samples (Knobloch-Westerwick & Meng, 2009; Meffert et al., 2006).
Procedure
Participants answered pretest questions about political party identification and media use. To avoid the threat of priming partisanship, participants answered distractor questions about their media use and attempted to identify non-partisan news reporters before browsing the site. Participants were then asked to browse a political news website. They had 4 min to browse the website because previous news selection research has limited participants’ time on page to 4 or 5 min (Knobloch-Westerwick & Johnson, 2014) and because individuals spend approximately 3.6 min on news sites per visit (Katz, 2013). Participants were told that they would spend 4 min on the site, after which the study would automatically move on. After browsing the site, the participants responded to distractor questions about the website quality, provided demographic information, and completed the study.
Stimuli
We created a website with links to news articles for the study. The site listed 12 headlines and lede sentences on which participants could click to read the article. 1 Only the text of the headlines and ledes was displayed. This text-only website design made the webpage less like a real-life news site but allowed us to strictly control the differences among the articles and focus on the effects of textual rather than visual negativity. It also followed similar designs used by previous experimental research regarding selection of political information (e.g., Knobloch-Westerwick & Meng, 2009; Meffert et al., 2006). The order in which the headlines and ledes appeared was randomized for each person to ensure that the order did not influence the results.
The articles were created specifically for this project, but content was drawn from the New York Times, NPR, CNN, and similar news sources. The articles touched on political conflicts among members of Congress, including debate over a farm bill, a contentious Internal Revenue Service scandal hearing, and general gridlock in Washington, D.C. The specific conflicts were not mentioned in the headlines or ledes on the landing page of the website. Instead, the headlines and ledes focused on congressional conflict generally. Congress may seem a dry choice for news headlines. However, in 2014, when the Study 1 data were collected, Gallup’s polling indicated that people in the United States considered Congress and governmental dysfunction to be the top most important problem facing the nation, suggesting that news about congressional conflict would be of interest to participants (Saad, 2015). This choice also allowed us to control for any influence specific issue interest may have had on individuals’ news choices. The headlines and ledes were all of the same length (ledes: 22 words, 1 paragraph, 1 sentence; headlines: 7 words), and the ledes were of the same reading difficulty (Flesch–Kincaid of 12). Because the headlines were not sentences, their reading difficulty levels could not be computed.
The 12 headlines and ledes varied according to the incivility present. Three headlines and ledes depicted civility and nine depicted incivility. This unbalance allowed us to vary the type of incivility present in the articles and, thus, test H2a and H2b. Three headlines and ledes were civil and emphasized respect and reciprocity, three included personal-level incivility by describing name-calling, three included public-level incivility by describing refusal to compromise, and three headlines and ledes included both types of incivility.
We pretested the headlines and ledes to ensure that they differed on incivility. We asked participants to rate each headline and lede using 10 semantic differential items (e.g., civil–uncivil, friendly–hostile, polite–rude). The responses were averaged for each headline and lede and resulted in a score that ranged from 1 = extremely civil to 5 = extremely uncivil. All of the civil headlines and ledes were significantly civil with scores less than the midpoint of three (means range from 1.69 to 1.72; t range from −20.47 to −23.86, p < .001). All of the uncivil headlines and ledes were significantly uncivil with scores greater than the midpoint of three (means range from 3.95 to 4.52; t range from 14.88 to 29.07, p < .001). 2
Clicks on news
The website unobtrusively tracked the behavior of participants. Individuals’ clicks on the news articles provided the dependent variables of interest (Knobloch-Westerwick & Meng, 2009). We measured the number of articles on which participants clicked that covered politicians using civility, personal-level incivility, public-level incivility, or a mix of both personal- and public-level incivility. Participants could select as few as zero articles of each type of content and as many as three articles of each type of content.
Study 1 Results
We tested the competing hypotheses that individuals would be more likely to select uncivil (H1a) or civil (H1b) news by conducting a one-way repeated-measures ANOVA using the number of clicks on articles with each type of content (civility, personal-level incivility, public-level incivility, mixed incivility) as the repeated measure. The ANOVA was significant, F(3,1158) = 3.19, p < .05,
Study 1 Discussion
Study 1 indicated that people are attracted to civil rather than uncivil news, especially compared with uncivil headlines that emphasize both impoliteness and an unwillingness to compromise. There are some limitations to this study. First, participants were told the amount of time they would spend browsing the news site. Time pressure could have increased participants’ need for closure, which previous research has suggested can prompt people to react less strongly to negativity (Van Kleef, De Dreu, & Manstead, 2004). Second, the headline content may have affected selection patterns. The headlines simply labeled politicians as behaving uncivilly, there were more uncivil than civil headlines available, and the headlines mentioned congressional conflict rather than specific issues. Details about the uncivil acts, a balance between civility and incivility in the news, and presence of a political issue may, however, change the way individuals react to incivility. Finally, the quasi-experimental design of the selection study prevented a test of the mechanisms through which negativity may lead to effects. Any measure of violations of news expectations or perceptions of incivility could have been caused by exposure to the archive webpage that listed all article choices or by the specific articles on which individuals clicked.
Study 2
The fully experimental design of Study 2 addressed the limitations from Study 1. Study 2 did not disclose the amount of time participants would spend with the articles, asked participants to interact with a full article describing immigration reform and uncivil behaviors among members of Congress rather than headlines about congressional conflict, and measured mechanisms through which negativity could affect news interactions. Furthermore, Study 2 only provided participants with one news article, so there was no imbalance between civility and incivility as there was in Study 1.
Study 2 Method
Participants in Study 2 were exposed to one of nine articles that varied according to the incivility present: in-group incivility, out-group incivility, or civility. We again varied the partisan content of the article (like-minded position on immigration reform, counter-attitudinal position, or both positions). The partisan position of the article did not influence interactions with news, so we do not discuss results related to partisan news.
Participants
Participants (n = 1,203) were recruited through SSI between May 1 and May 11, 2015. They were, on average, 44.15 years old (SD = 16.29). Fifty-four percent of the participants were female, 80% identified as White/Caucasian, 12% as Black/African American, 4% as Asian, 1% as American Indian, and 3% as another race. Twelve percent of the participants identified as Hispanic/Latino. Participants had an average of 14 years of education (SD = 2.17). There was a range of household incomes, with 34% reporting less than US$30,000 per year, 19% reporting US$30,000 to US$50,000, 16% reporting US$50,001 to US$75,000, and 30% reporting more than US$75,000. Participants varied politically, though they leaned liberal—M = 2.90, SD = 1.13; 12% very liberal (1), 22% somewhat liberal (2), 39% moderate (3), 17% somewhat conservative (4), 10% very conservative (5)—and Democratic—M = 2.26, SD = 0.98; 25% strong Democrat (1), 37% lean Democrat (2), 25% lean Republican (3), 13% strong Republican (4). An SSI representative explained to the authors that, at the time of the studies, the average length of time any individual was part of SSI’s panel was 6 months. Since there were 9 months between Study 1 and Study 2, it is unlikely that the same participants from Study 1 would have been part of SSI’s panel at the time of Study 2.
Procedures
Participants were first asked questions about their demographics and their political partisanship. To alleviate the potential of priming partisanship, we used the same distractor measures described for Study 1. Participants were then randomly assigned to an article discussing immigration reform posted to a webpage that included interactive buttons. Prior to viewing the article, participants were encouraged to read the article and interact with the digital buttons on the webpage as much or as little as they liked. The “next” button on the study was deactivated for 60 seconds, keeping participants on the article for at least a minute, after which participants could leave the page at anytime. Participants were not told in advance the amount of time they needed to spend on the webpage. After moving on from the article, participants answered questions about their perceptions of the article before finishing the study.
Stimuli
Each participant was randomly assigned to one news story about immigration reform negotiations taking place in the U.S. Congress. Study 2 participants could click on interactive buttons (like, share, and/or comment) rather than selecting a news article to read as in Study 1. Immigration reform was chosen because immigration is an issue that has remained in the news for a number of years, making it unlikely that the issue would be resolved while data were being collected. According to Gallup, immigration was the third most important problem to people in the U.S. in 2015 when the Study 2 data were collected (Riffkin, 2016).
The incivility in the article varied. First, in some articles, politicians were described as behaving civilly rather than uncivilly. All politicians were portrayed as respectfully disagreeing but being willing to work together to get immigration reform passed (e.g., “Senate Republican leaders announced their full support of policies that restrict immigration, but say they are open to discussing the policy with Democrats”). Second, in some articles, the Democratic politicians were acting uncivilly. Finally, the remaining articles included Republican politicians behaving uncivilly. Examples of personal- and public-level incivility were used in Study 2’s uncivil stimuli because the most substantial differences in Study 1 appeared between civil news headlines and headlines that included both types of incivility. We continued to vary the amount of negativity in Study 2 instead by varying whether an in-group political party (less negativity) or an out-group political party (more negativity) was behaving uncivilly. When either party was portrayed as behaving uncivilly, the article included partisans from that party name-calling and yelling at opponents from the other party (e.g., “Republicans are delusional and reckless”) and refusing to work with the opposing party on immigration reform (e.g., “We will stand our ground and refuse to pass any Republican-backed legislation”). The examples of incivility remained the same in all articles; only the names and parties of the politicians in the articles changed.
Two between-groups ANOVAs, one predicting perceptions that Democrats were behaving uncivilly in the articles and a second predicting perceptions that Republicans were behaving uncivilly in the articles, were run as manipulation checks. The tests were significant for both perceptions of Democrats [F(2, 834) = 12.11, p < .001] and of Republicans [F(2, 834) = 12.11, p < .001] behaving uncivilly, demonstrating that the articles manipulated to show Democrats (Republicans) behaving uncivilly prompted participants to perceive that the Democratic (Republican) politicians were behaving more uncivilly in those articles than they were in the civil articles or in the articles that showed the other party behaving uncivilly.
The articles were used to create a partisan incivility variable with three conditions: first, a condition where Democrats (Republicans) read about Democrats (Republicans) behaving uncivilly (in-group incivility, n = 345), second, a condition where Democrats (Republicans) read about Republicans (Democrats) behaving uncivilly (out-group incivility, n = 366), and, finally, a condition that portrayed the politicians as respecting their opponents and being open to working with them despite disagreements (civility, n = 363).
The independent variable required participants to report, at minimum, a lean toward either the Democratic or the Republican Party. One hundred twenty-nine participants did not report their partisanship, so they were dropped from the analyses in this study. The news articles were written to be as similar as possible on variables other than the immigration position and incivility. The news articles created for Study 2 were the same length (360 words, 10 paragraphs, 20 sentences) and of similar reading difficulty (Flesch–Kincaid range from 11.6 to 11.9).
Dependent variable
Using a functional website, we unobtrusively tracked participants’ use of interactive buttons. Participants could interact using any combination of three options: clicking a “Like” button (19%), clicking a “Share” button (10%), and clicking a “Comment” button or directly commenting in the comment section (18%). Overall, 29% of the participants interacted with at least one of the options on the website. We summed individuals’ clicks on the webpage to create a number of button clicks variable, ranging from 0 (clicked on no buttons) to 3 (clicked on all three buttons) (M = 0.47, SD = 0.83).
Indirect Effects Variables
Violate news expectations
News users have substantial choice in the news they consume (Prior, 2007). Because we randomly assigned one of the experimental articles to participants, the article a participant read may have aligned with content that participant typically viewed (i.e., met expectations of news content) or may have contrasted with the content that participant typically viewed (i.e., violated expectations of news content). After reading and interacting with the news article, participants were asked to report how likely (1 = very likely to 7 = very unlikely) it was that, in everyday life, they would select the news article they read (M = 3.07; SD = 1.51). A higher score indicated that a participant perceived the experimental news article as less in line with the news they expect to read outside of the experimental setting.
Automatic evaluation-categorization
Previous research has measured automatic evaluation-categorization by assessing brain activity (Ito et al., 1998). Due to the online context of Study 2, we, instead, measured perception of incivility in the news articles. We asked participants to respond to 10 5-point semantic differential items rating the Democrats in the article as friendly/hostile, unemotional/emotional, calm/agitated, agreeable/quarrelsome, polite/rude, good for democracy/bad for democracy, acceptable/unacceptable, cooperative/uncooperative, compromising/uncompromising, and working together/not working together (see, for similar measure, Mutz & Reeves, 2005). Participants then responded to the same items but were prompted to think about the Republicans in the article. We averaged these scales into Democratic incivility (M = 2.97; SD = 0.83; Cronbach’s α = .89) and Republican incivility (M = 3.06; SD = 0.85; Cronbach’s α = .89). We created two variables from the measures: perceptions of in-group incivility (M = 2.76; SD = 0.74) and perceptions of out-group incivility (M = 3.27; SD = 0.86). The variables were based on participants’ reported partisan leans, with Democrats’ (Republicans’) perceptions of in-group incivility based on the ratings of Democrats (Republicans) in the news article, and Democrats’ (Republicans’) perceptions of out-group incivility based on their ratings of Republicans (Democrats) in the experimental news article they saw. The measures were only minimally correlated (Pearson’s r = .18; p < .05).
Study 2 Results
Study 2 was designed to replicate Study 1 and investigate whether violations of news expectations (H3a) or automatic evaluation-categorization of negativity (H3b) indirectly lead to interaction with news. We ran analysis using MEDIATE (Hayes & Preacher, 2014) because the predictor variable of partisan incivility in the news was multicategorical. MEDIATE does not provide p values for indirect effects. Instead, indirect effects are significant when a confidence interval (CI) for an indirect effect does not include 0. For the indirect effects tests, we used bias-corrected 95% CIs computed from 1,000 bootstrapped samples.
The predictor variable in the indirect effects model was the partisan incivility condition. The in-group incivility condition was used as the reference group so any differences between the in-group and the out-group incivility conditions could be tested. The outcome variable was the number of button clicks on an article’s webpage. Three parallel mediating variables were included: violate news expectations, perceptions of in-group incivility, and perceptions of out-group incivility. The model is presented in Figure 1. We first overview the significant and nonsignificant effects in the model then discuss how the results support and challenge the hypotheses.
Indirect effects model predicting number of interactions with news website.
We begin by discussing the effects of the civil condition on number of button clicks compared with the in-group incivility condition. There was no direct effect of civil condition on number of button clicks (effect = −0.02, SE = 0.10, p = .82). The civil condition also did not indirectly affect number of button clicks through violate news expectations [effect = −.001, SE = .01, CI = (−.0283, .0193)], nor did it indirectly affect number of button clicks through perceptions of in-group incivility [effect = .01, SE = .01, CI = (−.0047, .0451)]. The indirect effect of the civil condition on the number of button clicks through perceptions of out-group incivility was significant, [effect = .03, SE = .01, CI = (.0025, .0628)], with the coefficient for this indirect effect indicating that exposure to civil news increases interaction by reducing perceptions of the out-group as uncivil.
We find a similar pattern of effects for the out-group incivility condition compared with the in-group incivility condition. There was no direct effect of out-group incivility condition on number of button clicks (effect = 0.13, SE = 0.09, p = .16). The out-group incivility condition did not indirectly affect number of button clicks through violate news expectations [effect = −.01, SE = .01, CI = (−.0343, .0131)], nor did it indirectly affect number of button clicks through perceptions of in-group incivility [effect = −.002, SE =.01, CI = (−.0283, .0046)]. The indirect effect of the out-group incivility condition on the number of button clicks through perceptions of out-group incivility was significant (effect = −.02, SE =.01, CI = (−.0563, −.0004)], with the coefficient for this indirect effect suggesting that exposure to out-group incivility decreases interaction by increasing perceptions of the out-group as uncivil.4,5
The results replicate those found in Study 1. As the civil news indirectly increased and the out-group incivility news indirectly decreased number of interactions with the news, H1b was supported over H1a. H2b also was marginally supported. The out-group incivility article prompted marginally more perceptions of out-group incivility than the in-group incivility article (see Figure 1), and, thus, decreased interactions with news more than the in-group incivility article. Once again, a condition with stronger negativity had stronger effects on interactions with the news than an article with weaker negativity. The results also provide support for H3b over H3a. There were no significant indirect effects through violations of news expectations (H3a). Instead, all significant indirect effects were through perceptions of out-group incivility (H3b).
Study 2 Discussion
Study 2 served two purposes. First, it replicated the results from Study 1: Participants were more likely to interact with news that portrayed politicians as civil and respectful than news that portrayed politicians as uncivil, rude, and uncompromising—at least indirectly. Second, Study 2 tested the mechanism through which negative news affected interactions with the news. The experiment showed that the effects on interactions with a news page only occurred through perceptions of out-group incivility. Perceptions of in-group incivility and violations of news expectations did not significantly influence interaction. This finding provides evidence that automatic reactions to negativity, rather than violations of expectations, explain the effects.
The indirect effects test was strong in that it contrasted two perspectives about why negativity produces effects, a move that previous studies of negative news and the interactions it encourages have not taken (Meffert et al., 2006; Mutz, 2015). Future research can continue this line of research by using additional indirect effects variables. Automatic evaluation-categorization could be tested by measuring physiological arousal. Violations of expectations could be tested in future research by asking participants to (a) respond to more than one item when measuring news expectation violations and (b) report their expectations of politician behavior in addition to their expected news use. The evidence from Study 2, however, is the first direct comparison between these two theoretical mechanisms that underlie negativity’s effects and indicates that recognizing out-group negativity decreased interactions with news.
General Discussion
This project demonstrated that incivility in the news matters in a different way than previous research has suggested (Meffert et al., 2006; Mutz, 2015). Rather than being drawn toward incivility, individuals selected (Study 1) and interacted with (Study 2) news depicting politicians as respectful and willing to compromise. Study 1 indicated that the largest difference in news choice was found between civil news and news that portrayed politicians as both impolite and promoting gridlock in Congress. Study 2 demonstrated that perceiving the political out-group behaving uncivilly after reading a news story decreased interactions with the news.
Negativity in the form of incivility, rather than prompting engagement (Baumeister et al., 2001; Jang & Oh, 2016; Meffert et al., 2006), discouraged interaction with news. We were able to replicate this effect across different types of online behaviors. The effect held both for headlines that described politician incivility vaguely and focused on general congressional conflict (Study 1) and for full articles that provided detailed examples of politicians behaving uncivilly within the context of immigration reform (Study 2). The effect also held when participants had more uncivil than civil news options (Study 1) and when participants received one article and, thus, could not compare the number of civil and uncivil articles (Study 2).
The finding that civility prompts engagement with online news aligns with research indicating that individuals share positive information online (Berger & Milkman, 2012; Eckler & Bolls, 2011; Guadagno et al., 2013), but it contradicts studies of negativity in news that found individuals willing to select negative over positive information (Jang & Oh, 2016; Meffert et al., 2006; Mutz, 2015). Why might the literature be split as to whether positive or negative information engages individuals? We posit two reasons for the varied results across studies.
First, negativity and positivity are conceptualized differently in nearly all of these studies. Humorous (Guadagno et al., 2013) and awe-inspiring (Berger & Milkman, 2012) positive content is more likely to be shared online compared with disgusting and sad negative content, respectively. Alternatively, conflict and agony (Zillmann et al., 2004), mildly negative headlines about candidates (Meffert et al., 2006), and aggressive impoliteness between candidates (Mutz, 2015) draw more news users than factual information, mildly positive headlines, and polite exchanges, respectively. These brief examples suggest that positivity becomes more attractive when there is a greater contrast between the positivity and negativity. The current project provides evidence to support this suggestion, as news articles with the most incivility were selected least compared with news that covered civil, deliberative politicians. Future research should test this pattern directly by varying the levels of positivity and negativity in the same study to determine whether the greater contrast pushes people more toward positive news.
Second, the context of the news content may influence engagement with positive and negative information. In the current project, for instance, the two contexts—Congress and immigration reform—were among the most important problems that U.S. residents said the country faced during the time of data collection (Riffkin, 2016; Saad, 2015). Within these contexts, politicians getting along and working together—that is, civility—may have appeared as a positive solution to the problems that face the nation. Indeed there is preliminary evidence that online news that presents solutions attracts clicks (Curry & Stroud, n.d.). Other news selection studies, which have examined campaigns (Meffert et al., 2006; Mutz, 2015) and international news (Knobloch et al., 2003), did not highlight potential solutions to top problems facing the U.S. Future researchers should directly compare various political contexts to determine whether civility draws attention when news offers solutions to negative political problems.
The finding that people gravitate toward news that highlights respect and deliberation is, on one hand, encouraging. Mediated incivility tends to decrease trust in government, belief that opposing ideas are legitimate, and openmindedness (e.g., Borah, 2014; Mutz, 2015). If individuals do not voluntarily read uncivil news stories and do not engage with uncivil news content, perhaps scholars should be less worried about incivility’s potentially detrimental effects. Furthermore, journalists who know that citizens gravitate toward civility might be likely to break the persistent cycle of negative newsmaking when writing for digital spaces (Patterson, 1994).
On the other hand, taking into account the repeated research findings that show how powerful negativity is in various contexts (Rozin & Royzman, 2001), it is possible that participants were less attracted to civility and more repelled by incivility. Given Sobieraj and Berry’s (2011) evidence that outrageous incivility permeates news, the discovery that people did not actively engage with incivility signals a problem for news organizations. Citizens do not need much of an impetuous to avoid political news (Arceneaux & Johnson, 2013). If incivility pushes people away from news engagement, individuals may choose to leave a website or switch to nonpolitical news, two options they did not have in our project. Additional research should be conducted to determine whether the presence of uncivil news encourages people to find civil depictions of politics or pushes people away from political news entirely.
Another troubling finding is the relationship between incivility and partisanship. In Study 2, only perceptions that an out-group political party engaged in incivility prompted fewer clicks on the webpage. This result provides evidence that individuals treat mediated political party groups much like they act toward groups in face-to-face settings (Hartmann & Tanis, 2013; Tajfel & Turner, 1979) by reacting more negatively to their out-group’s bad behavior than their in-group’s bad behavior (Howard & Rothbart, 1980). This reaction to out-group incivility, together with Mutz’s (2015) finding that incivility decreases belief in the legitimacy of counter-attitudinal political arguments and Borah’s (2014) finding that incivility can increase attitude certainty, suggests that incivility in politics may be further polarizing partisans.
Unlike some researchers (e.g., Knobloch-Westerwick & Meng, 2009), however, we found no evidence that participants chose news that aligned with their partisan political beliefs. Partisan news was included in both studies but was not discussed in detail because in no case did partisanship predict engagement with news. The findings of the current project do, however, mirror Meffert et al.’s (2006) results, who also found that people avoided like-minded partisan news in a context where positive and negative news was present. We suspect that negativity in the news commands more attention than partisanship. Not all types of news coverage have similar effects when they are presented in combination with one another (e.g., Chong & Druckman, 2007). Perhaps positivity and negativity, or in this case civility and incivility, act as strong frames that overwhelm references to partisanship.
Given the experimental nature of this project, there are limitations to its validity. First, a social desirability bias may have prompted participants’ to choose civility. In Study 1, participants could, by comparing the available civil and uncivil news articles, guess that selecting the civil stories was more desirable. However, in Study 2, participants reacted similarly even without the uncivil stories for comparison.
In addition, participants in these studies were recruited through an online survey research firm. They had Internet access, but there was no guarantee that they represent typical online news users. We reran the analyses presented in this manuscript and included news use as a control variable. The pattern of results was similar.
Outside of a lab setting, the act of liking, sharing, or commenting on a news article publicizes individuals’ opinions about that article on their social media feeds. Here, we asked participants to interact with the news, but there was no public consequence to that interaction. Knowing that an interaction with the news might become public could prompt people to engage with different sorts of news as they manage their online presence. However, there is evidence from a large-scale content analysis suggesting that the most emailed stories on the New York Times website are positive and awe-inspiring (Berger & Milkman, 2012), providing evidence that similar patterns of attraction to positive information occur in outside the lab.
Finally, engagement with news might be inflated in experimental studies compared with typical news sites. One study found that between 1.4% and 5.8% of visitors to news websites clicked on buttons that linked to Facebook from those sites (Olmstead, Mitchell, & Rosenstiel, n.d.). In comparison, our 29% interaction rate with the buttons in Study 2 is quite high. However, the mean interaction rate of 0.47 button clicks per person that we found in Study 2 is on par with previously published research that found mean interaction rates with news comments of up to 1.22 button clicks per person (Stroud, Muddiman, & Scacco, 2016). Overall, field tests are necessary to determine whether the results replicate in online news settings.
Despite the ecological limitations, the studies in this project offer theoretical contributions to both negativity bias and incivility research. With respect to negativity bias, our findings make two substantial contributions. First, the project provides some of the first comparative evidence of the theoretical mechanism that drives negativity effects on news interactions: Perceiving incivility, our measure of automatic evaluative-categorization (Ito et al., 1998), led to behavioral effects while violations of news expectations (Soroka, 2014) did not. The studies in this project suggest that what Mutz (2007) calls a gut reaction to negativity is a better explanation for negativity’s effects than expectancy violations. Second, the results demonstrate that an attitudinal negativity bias does not always produce a behavioral negativity bias (Ito et al., 1998). Instead, negativity may be recognized in a news story and produce a backlash effect that prompts engagement with positive information.
This project also makes a considerable contribution to incivility research by encouraging scholars to look beyond politeness theory when conceptualizing both civility and incivility. Incivility studies, at times, conceptualize civility as the lack of impolite words (Thorson et al., 2010) and nonverbals (Mutz, 2015). We demonstrate that conceptualizing civility as the activity of encouraging respect and reciprocity among various political interests can further distinguish civility from incivility. In addition, as Mutz (2015) predicted, incivility as impoliteness alone has a less powerful effect in our text-based incivility context (see Study 1) than it does in an audiovisual context. Yet incorporating a public level of incivility focused on refusing to deliberate or work with political opponents (Uslaner, 1996) led to significant effects in the current project. Political (Rawls, 1993; Uslaner, 1996) and deliberative (Gutmann & Thompson, 1996) theory has acknowledged the importance of reciprocity and lack thereof. Media effects scholars interested in incivility in the news will benefit from doing the same.
If journalists have a responsibility to cover the state of political discourse, as Gutmann and Thompson (1996) suggest, then researchers have the responsibility to study the effects of such coverage on citizens. In the current project, participants were attracted to civil political discourse and pushed away from news that included both disrespectful impoliteness and refusal to work toward compromise with political opponents. Sometimes, as the current project shows, people want to read about politicians getting along.
Footnotes
Acknowledgements
The authors would like to thank Rachel Vliem and Alexandria Hellrung for their help with earlier versions of this project.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors disclosed receipt of the following financial support for the research, authorship and/or publication of this article: The authors received financial support for this project from the University of Wyoming College of Arts and Sciences.