Abstract
Drawing on the 2012 American National Election Studies (ANES) panel data, this study explores the influence of the consumption of partisan information sources on affective polarization and investigates the mechanism underlying this relationship. The results show that exposure to pro-party television sources strengthens affective polarization among partisans. The polarizing effects of pro-party sources are mediated by the discrete negative emotions (i.e., anger and fear) toward presidential candidates. The study discusses the impact of selective exposure on deliberative and participatory democracies.
For the past two decades, the share of American partisans who hold very unfavorable opinions toward the opposite party has more than doubled (Doherty, 2014). Partisans not only increasingly dislike their opponents but also impute negative traits, such as hypocritical and selfish, to the out-party members (Hmielowski, Beam, & Hutchens, 2016; Iyengar, Sood, & Lelkes, 2012). Such tendency of viewing out-party members negatively and in-party fellows positively is called affective polarization in recent literature (Iyengar et al., 2012).
Some researchers attribute this trend of political polarization to the high-choice media environment, in which Americans are facilitated to actively seek out likeminded information (Stroud, 2010; Sunstein, 2001). Indeed, the critical changes to the media system in 1996 (i.e., the passage of the Telecommunications Act of 1996) exacerbated the influence of television news use on affective polarization (Hmielowski et al., 2016). With the prevalence of partisan media outlets on cable news, partisans tend to consume a disproportionate amount of pro-party content (Mullainathan & Shleifer, 2005; Prior, 2007). Previous studies have found that the consumption of pro-party media sources leads viewers to evaluate the opponents more negatively (Levendusky, 2013).
Nevertheless, the mechanism underlying the relationship between selective exposure and affective polarization is rarely empirically examined. Research has suggested that the consumption of partisan information sources influences affective polarization through the induction of emotions (Garrett et al., 2014). Emotions are one of the best predictors of human attitudes because they often have profound influence on cognitive evaluations of a certain object or person (Frijda, Manstead, & Bem, 2000). This study moves beyond the dimensional view of emotions (e.g., positive vs. negative) by focusing on two discrete emotions: anger and fear. The discrete perspective has been found particularly useful for studying the mechanism underlying media effects because each emotion is considered qualitatively different and offers a more nuanced understanding of the communication processes (Dillard & Peck, 2001; Nabi, 2010). Along this line, an important goal of this study is to examine the mediating roles played by negative discrete emotions (i.e., anger and fear) in the relationship between partisan selective exposure and affective polarization.
This study employs the 2012 American National Election Studies (ANES) panel data to examine the relationship between the consumption of pro-party information sources and affective polarization. More importantly, this study investigates the mechanism underlying this relationship by testing the mediating roles of two discrete emotions (i.e., anger and fear) toward presidential candidates. The findings contribute to the understanding of the impact of selective exposure on deliberative and participatory democracies.
Selective Exposure and Affective Polarization
The contemporary high-choice media environment allows partisans to selectively consume political information sources that are consonant with their preexisting preferences (Iyengar & Hahn, 2009; Stroud, 2011). The stories on partisan media are framed to uphold a certain party’s political agenda and/or interpreted in a certain ideological direction (Arceneaux & Johnson, 2013; Baum & Groeling, 2008; Jamieson, Hardy, & Romer, 2007). Furthermore, the partisan programs on cable news usually provoke visceral responses among the audience by broadcasting a harsh narrative about the opposing side (Jamieson & Cappella, 2008; Sobrieraj & Berry, 2011). In this case, selective exposure to pro-party information is very likely to exacerbate affective polarization (Klapper, 1960; Levendusky, 2013).
Based on the social identity theory (Tajfel, 1978; Tajfel & Turner, 1979), affective polarization is defined as partisans’ tendency of viewing out-party members negatively and in-party fellows positively (Iyengar & Westwood, 2015). Compared with issue-based polarization, affective polarization is a “more diagnostic” and “more appropriate” indicator of mass polarization (Iyengar et al., 2012), because group-based affect is an ingrained human response for partisans (Billig & Tajfel, 1973). The act of identifying with a certain party is often accompanied by affective evaluations (Tajfel, 1978), including unfavorable views of the other side and favoritism toward one’s own party (Greene, 1999).
Partisan information sources are judged more credible and diverse among the in-party members, because they share a partisan identity with the audience (Levendusky, 2013; McGuire, 1985; Stroud, Muddiman, & Lee, 2014). In this case, the messages from these sources are particularly persuasive for the in-party members (Arceneaux & Johnson, 2013; Mackie, Worth, & Asuncion, 1990). Given that anchors and commentators of these partisan television programs usually explicitly criticize the opposing party (Jamieson & Cappella, 2008), in-party audience tends to trust their opinions and then consider the other side more unfavorable.
Furthermore, partisan media bolster one’s partisan identity by facilitating individuals to make comparisons favorable to their own party (Knobloch-Westerwick, 2014). On one hand, pro-party media confirm audience’s stereotypes of the opponents by portraying the out-party candidate as hypocritical and duplicitous; on the other hand, they also tend to promote the in-party candidate’s image through various ways (Iyengar et al., 2012; Jamieson & Cappella, 2008). Conceivably, such media coverage leads individuals to compare their in-party candidate’s positive traits with the out-party candidate’s negative ones, which would reinforce the salience of one’s partisan identity (Knobloch-Westerwick, 2014; Wood, 1989). The salience of a group identity is the basis of predicting the intergroup biases (Gaertner, Dovidio, Anastasio, Bachman, & Rust, 1993) and individuals tend to heighten differences between “us” and “them” to satisfy the need for positive distinctiveness of their own group (Greene, 1999). In politics, partisanship is considered the most prominent group identity for most Americans because parties are the groups directly competing for power in politics (Arceneaux & Johnson, 2013; Greene, 1999; Mason, 2015). Once the salience of partisan identity is enhanced by partisan media, individuals are likely to view out-party members more negatively and in-party fellows more positively (Garrett et al., 2014).
Indeed, previous studies have reported that media messages play a critical role in promoting affective polarization. Iyengar and his colleagues (2012) have shown that exposure to negative campaigns leads to partisans’ negative feelings for their opponents. In addition, Garrett et al. (2014) have found frequent use of pro-attitudinal websites and blogs is positively associated with affective polarization. Based on the aforementioned propositions and evidence, the following hypothesis is proposed:
Mediating Roles of Anger and Fear
Although previous studies have documented the positive association between pro-party source consumption and affective polarization (Garrett et al., 2014; Levendusky, 2013), the mechanism underlying this relationship has been rarely examined empirically. Garrett and his colleagues (2014) pointed out the intervening roles of emotion by suggesting that selective exposure could promote affective polarization through the mechanism of affective learning. The affective learning process suggests that the negative emotions displayed by messengers (e.g., anchors) are learned and mimicked through the emotional contagion effect (Hatfield, Cacioppo, & Rapson, 1992; Jamieson & Cappella, 2008). The effect refers to “the tendency to automatically mimic and synchronize expressions, vocalizations, postures, and movements with those of another person’s and, consequently, to converge emotionally” (Hatfield et al., 1992, pp. 153-154). This study focuses on negative emotions, because programs with a partisan slant spend more time attacking the other side than defending their own (Chalif, 2011; Jamieson & Cappella, 2008; G. Smith & Searles, 2014). Personalities on these programs often deliver information in a passionate manner, raising voice in outraged frustration and hurling affronts upon the opponents (Arceneaux & Johnson, 2013). Observing these verbal and nonverbal clues, the audience learns the personalities’ subjective emotions toward the political figures (Hatfield, Cacioppo, & Rapson, 1993).
In addition to the affective learning process, selective exposure could also induce negative emotions, such as anger and fear (Garrett et al., 2014). As proposed in the appraisal-tendency theory, these discrete emotions often arise following an automatic evaluation of the external stimuli (Keltner, Ellsworth, & Edwards, 1993; Lerner & Keltner, 2001; Smith & Ellsworth, 1985). Although anger and fear share a negative feeling, the consumption of pro-party sources elicits each of the emotions through very different mechanisms. Anger is usually induced when one perceives (a) others responsible for negative events and (b) a violation of certain standards (Carver & Harmon-Jones, 2009; Lerner & Keltner, 2000). Television programs with a partisan slant tend to blame the other side for national problems, and inform the audience about their opponent’s misconduct and/or the peculiar claims made by the other side (Arceneaux & Johnson, 2013). In this case, anger toward the out-party candidate is likely induced due to the opposing party’s responsibility for national problems and their elites’ violation of traditional standards for politicians. Fear, on the contrary, usually arises from (a) threatening stimuli in one’s environment and is associated with (b) a lack of certainty and personal control (Lazarus, 1991; Lerner & Keltner, 2000). The negative traits ascribed to opponents on partisan programs (Jamieson & Cappella, 2008) may lead audience to think about the opposing candidate as a threat to the country. If the opposing candidate were elected, the audience would feel uncertain about how the country’s policy would influence their well-being. Such perceived threats and uncertainty are likely to induce fear toward the out-party candidate.
These negative emotions, which are learned from and elicited by pro-party information sources, could subsequently exert a strong impact on affective polarization. Indeed, emotional response is an integrated part of politics and has powerful influences on political attitudes (Brader, 2005; Marcus, 1988). Although affective polarization and emotions are related with each other, it is worth noting that these are two separate constructs. Affect refers to favorable disposition toward a person or item, which is often measured through attitude scales (Batra, 1986; Fiore & Kim, 2007). As such, affective polarization is a manifestation of attitude strength in terms of favorability or hostility toward certain political objects, such as presidential candidates and political parties (Abelson, 1995; Hyun & Moon, 2016). Unlike affective polarization, emotions are short-lived and intense mental states representing evaluative and valenced reactions to certain external stimuli; they often arise automatically (Moors, 2010; Nabi, 1999). In this sense, emotions usually precede a conscious evaluation of relevant information and directly contribute to the formation of political attitudes, including favorability ratings of candidates and parties (Haidt, 2001; Marcus, Neuman, & MacKuen, 2000; Vasilopoulos, Marcus, & Foucault, 2018). For example, when people experience negative emotions such as anger, they tend to subsequently hold a hostile attitude toward out-group members (Van Zomeren, Spears, & Leach, 2008). Also, previous work in consumer behavior has documented that emotions elicited by retail design influence consumers’ favorability toward products, indicating that these two constructs can coexist (Holbrook, 1986).
According to the appraisal-tendency theory, the cognitive appraisals that evoke certain emotions would determine the emotion’s influence on political judgments and become “an implicit perceptual lens for interpreting subsequent situations” (Lerner, Gonzales, Small, & Fischhoff, 2003, p. 144). Anger toward the out-party candidate, which is induced by the attack on the other side, would increase the tendency to make punitive judgments of the opponents, such as negative ratings of the opposing candidate (Lerner, Goldberg, & Tetlock, 1998). Meanwhile, angry people may evaluate their in-party candidate more positively because anger is likely to mobilize individuals to defend their loved ones (Nabi, 1999). In terms of fear, because it arises from the perceived uncertainty about the leadership of the out-party candidate, fearful people would tend to vote against and give negative comments about the candidate (Ladd & Lenz, 2008). Given the existing evidence and the emotional nature of affective polarization, it is reasonable to expect that the negative emotions toward presidential candidates would exert effects over the divide between in- and out-party evaluations.
Previous research has empirically documented the mediating role of emotions in the effects of political information consumption. For example, Cho and Choy (2011) found negative emotions mediate the relationship between debate viewing and political conversations among partisans. The positive impact of pro-attitudinal news exposure on political participation is mediated by anger (Wojcieszak, Bimber, Feldman, & Stroud, 2016). Putting these propositions together, the following hypotheses are proposed:
Method
Data
This study employed the data set from the ANES 2012 Time Series Study, which consists of two waves of surveys (Wave 1: October 11, 2012-November 6, 2012; Wave 2: November 29, 2012-January 24, 2013). The sample size of this nationally representative panel study was 5,914. Because this study focuses on affective polarization among partisans (i.e., Democrats and Republicans), respondents who self-identify as Independents or affiliate with other parties are excluded from further analysis. Among all of the partisans, 239 participants of the first wave did not respond to the survey request of the second wave. Therefore, the final sample size is 3,413.
Measurement
Pro-party information source was measured by tapping respondents’ exposure to a total of 32 television programs. In the first wave of this study, respondents were asked whether they have heard anything about the presidential campaigns via television news programs, talk shows, or news analysis programs. As a result, 345 respondents who did not use these television sources for election information were not asked about the follow-up questions. Those who answered yes to this filter question were shown a list of 48 television programs, among which 16 were unrelated to political information sources (e.g., Dancing with the stars). Therefore, 32 television programs were included in this study and respondents were asked to identify “which of the following television programs do you watch regularly? Please check any that you watch at least once a month.” Those (N = 698) who did not consume any of the 32 sources were excluded from further analyses, because it is unknown whether they rely on pro-party, neutral, or antiparty television sources for political information. In this case, a total of 2,370 participants were included in the subsequent analyses.
According to the classification in previous studies (Dilliplane, 2011; Jocobson, 2015; Song, 2017), each of the 32 television sources was coded as neutral, leaning toward the Democrats, or leaning toward the Republicans, as shown in Table 1. Based on the slant of each program and respondents’ self-reported partisanship, this study produced a tally of the number of pro-party TV information sources that each respondent had watched. Finally, the variable of pro-party information source (M = 0.30, SD = 0.34, Range = 0 to 1) was calculated by dividing the tally by the total number of consumed television programs (Dilliplane, 2011). For example, if a Democrat watches five television programs in total and only one of them has a Democratic slant, then his or her pro-party information source value is 0.2.
Partisan Slant of TV Programs.
Affective polarization was measured by a combination of two items: polarization on candidates and polarization on parties (Iyengar et al., 2012). First, the favorability polarization on candidates was measured by the difference between the feeling thermometer index of Barack Obama and Mitt Romney, which ranges from 0 to 100. The Romney thermometer scores were subtracted from the Obama scores and then the absolute values were calculated based on the second wave data (M = 57.23, SD = 29.06, Range = 0-100). The second indicator of this dependent variable was the favorability polarization on parties. In the second wave, respondents used a 10-point scale to indicate how much they like or dislike the Democratic Party and the Republican Party. The favorability ratings of the Democratic Party were subtracted from the ratings of the Republican Party, and then the absolute values were calculated to assess the favorability polarization on parties (M = 5.19, SD = 2.95, Range = 0-10). Because these two items were measured with different scales, we first calculated the standard score (z score) of each item and then combined them to create the measurement of affective polarization (M = .49, SD = 1.71, Range = −3.12-3.49, Pearson’s r = .57, p < .001). Candidate polarization (M = 58.27, SD = 28.78, Range = 0-100) and party polarization (M = 48.93, SD = 29.44, Range = 0-100) are both measured in the first wave with a 100-point scale. In this case, we simply combined these two items and created a measurement of affective polarization to be included in the subsequent analyses as a control variable (M = 107.26, SD = 52.87, Range = 0-200, Pearson’s r = .65, p < .001). By controlling for lagged values of the dependent variable, this study is able to estimate the effect of pro-party sources on the changes in affective polarization between panel waves (Finkel, 1995).
Differential in anger was measured in the first wave by asking respondents to use a 5-point scale to indicate how often they feel angry with the presidential candidates, Barack Obama and Mitt Romney. Based on respondents’ self-report party identification, the responses to this question were then recoded to create two variables: anger toward the out-party candidate (M = 1.73, SD = 1.42, Range = 0-4) and anger toward the in-party candidate (M = 0.25, SD = 0.63, Range = 0-4). For example, Obama was coded as the in-party candidate for Democrats, but the out-party candidate for Republicans. Finally, the score of anger toward the in-party candidate was subtracted from the score of anger toward the out-party candidate (Cho & Choy, 2011). 1
Differential in fear was measured in the first wave by asking respondents to use a 5-point scale to indicate how often they feel afraid of the presidential candidates. Based on respondents’ self-report party identification, the responses to this question were then recoded to create two variables: fear toward the out-party candidate (M = 1.37, SD = 1.44, Range = 0-4) and fear toward the in-party candidate (M = 0.16, SD = 0.55, Range = 0-4). Finally, the score of fear toward the in-party candidate was subtracted from the score of fear toward the out-party candidate.
Control variables 2 in this study include age (M = 47.91, SD = 17.48), gender (male: 45.7%), education (M = 2.91, SD = 1.14, Range = 1 [less than high school credential] to 5 [graduate degree]), income (M = 14.49, SD = 7.99, Range = 1 [below 5,000] to 28 [250,000 or more]), race (White: 68.5%), party identification (Democratic: 56.1%, Republican: 43.9%), partisan strength (strong partisans: 55.5%, weak partisans: 44.5%), differential in pride (M = −1.68, SD = 1.63, Range = −4-4), using newspaper to follow campaigns (50.5%), using TV news to follow campaigns (87.1%), using radio news to follow campaigns (50.7%), using Internet to follow campaigns (46.2%), using social media to follow campaigns (number of days in a typical week: M = 1.58, SD = 2.57, Range = 0-7), antiparty TV sources (M = 0.10, SD = 0.20, Range = 0-1), and total number of consumed TV information sources (M = 4.09, SD = 4.10, Range = 0-27). The pairwise correlations of the key variables (i.e., independent variable, mediators, and dependent variables) are reported in Table 2.
Correlations Between Key Variables.
p < .001.
Results
This study employed ordinary least squares (OLS) regression models to test the hypotheses with panel analysis of lagged dependent variables. Given the large sample size of this data set, we decide to adopt a stricter alpha level (i.e., α = .01) than the conventional level of .05 so as to reduce the probability of making a Type I error. Several control variables were found significantly related to affective polarization. For example, strong partisans are more polarized than their weak counterparts. Also, consistent with the literature (Garrett et al., 2014), exposure to antiparty information sources reduces one’s level of affective polarization. Findings are summarized in Table 3.
Regression Analysis of Affective Polarization, Unstandardized (SE).
p < .10. *p < .05. **p < .01. ***p < .001. (listwise deletion).
To test the aforementioned predictions about the mediators in the relationship between the consumption of pro-party information sources and affective polarization, this study specified a multimediator path model using the PROCESS Macro and the Model 4 template offered by Hayes (2013). As depicted in Figure 1, this single theoretical model shows the indirect effects of pro-party television programs on affective polarization through anger and fear. Using the bootstrapping technique (Preacher & Hayes, 2004) with 10,000 bootstrap samples, this study analyzed the 99% bias-corrected confidence intervals (CIs) associated with the indirect effects of anger and fear.

Conceptual model of mediation from pro-party TV information sources to affective polarization through differential in anger and differential in fear.
As shown in Table 4, the relationship between the consumption of pro-party information sources and affective polarization is significantly mediated by the differential in anger (B = .03, SE = .01, CI = [.0117, .0661]) and the differential in fear (B = .02, SE = .01, CI = [.0024, .0500]). The CIs do not include zero, suggesting significant indirect relationships. Therefore, our analyses provide support for
Mediation of the Relationships Between Pro-Party Information Consumption and Affective Polarization Through Differential in Anger and Differential in Fear.
Note. Bootstrapping results are bias corrected and accelerated; 10,000 bootstrap samples; all of the control variables listed in Table 3 were also included into the equations of these mediation analyses, but not reported here due to space limitation. LLCI = lower limit confidence interval; ULCI = upper limit confidence interval.
Discussion
This study investigated the influence of selective exposure on affective polarization and explored the mechanism underlying this relationship. Employing the 2012 ANES panel data, this study examined the relationships between the consumption of pro-party sources and affective polarization. Furthermore, building on the appraisal theories of emotions, we investigated the mediating effects of two discrete emotions: anger and fear.
Consistent with previous research (Levendusky, 2013), the findings confirmed that the consumption of pro-party TV programs promotes partisan affective polarization. The results add evidence to somewhat pessimistic view about the influence of partisan media, in that affective polarization has rather harmful influences on democracy. Specifically, the increasingly negative ratings of out-party presidential candidates could lead to political cynicism and intolerance (Layman, Carsey, & Horowitz, 2006). We can hardly expect a Republican, who dislikes Obama, to trust and deliberate on the initiatives proposed by Obama. As a result, affective polarization may lead to attitudinal polarization (Levendusky, 2013). In addition, the hostile attitudes toward the opposing party are likely to reduce collaboration and compromise (MacKuen, Wolak, Keele, & Marcus, 2010), which are fundamental values of the American constitutional system (Gutmann & Thompson, 2010). With the ample presentations of entertainment programming on cable TV, some scholars argued that the polarizing effects of pro-party sources may be attenuated because most individuals would not tune in political programs (Arceneaux & Johnson, 2013). In such cases, however, democracy is also shrinking because the American electorate would be composed of the polarized and the uninformed.
The most prominent contribution of this study is that it unraveled the mechanism through which pro-party TV sources promote affective polarization. Building on existing theoretical propositions (Garrett et al., 2014; Lerner & Keltner, 2001), this study empirically examined how anger and fear mediate the effects of pro-party information consumption on affective polarization. The critical role of the negative discrete emotions supports the notion that affective polarization is conceptually based on emotion (Garrett et al., 2014). The conflict-based contents on partisan media and their outrageous way of delivering information elicit in-party members’ negative emotions (i.e., anger and fear) toward the other side, which in turn, polarize partisans’ evaluations of candidates and parties. Affective polarization could weaken the democracy in many respects, but the elicited negative emotions could possibly strengthen democracy. As suggested by the cognitive appraisal theories (Lerner & Keltner, 2001), anger mobilizes involvement in both low-cost and high-cost political activities, while fear only affects the former type of participation (Valentino, Brader, Groenendyk, Gregorowicz, & Hutchings, 2011). From the perspective of deliberative democracy, negative emotions are also instrumental. For example, anger encourages individuals to process political information more thoroughly and then come up with more elaborate opinions (Kim, 2016; Nabi, 2002).
The results on the mediating effect of anger and fear also suggest important directions for future research. First, following the appraisal theory, this study maintains that the consumption of pro-party information sources triggers anger and fear through different mechanisms (Keltner et al., 1993; Lerner & Keltner, 2000). Anger is induced by the blame toward the opposing party and its candidate, whereas fear is elicited by the perception that the election of the opposing candidate will lead to threat and uncertainty to this country. However, due to the limitation of the data set, we are unable to empirically examine such theoretical predictions associated with the different appraisals of the discrete emotions. Future research could contribute to the literature of emotions and political communication by investigating these cognitive appraisals underlying the induction of emotions. Second, while this study focuses on negative emotions, future studies could extend this line of research by exploring the roles of positive emotions, such as pride included in this study as a control variable. The results of Table 3 suggest that consumption of pro-party information sources is likely to trigger pride among in-party members. Indeed, pride occurs when individuals feel their social identity is enhanced by the televised achievements of the party that they identify with (Lazarus, 1991). With an enhanced partisan identity, individuals experiencing pride may subsequently strengthen the level of affective polarization (E. R. Smith & Mackie, 2008).
Some limitations need to be noted in the interpretation of the findings. First, some of the analyses rely on cross-sectional data, which do not allow us to draw causal conclusions. For example, when testing the mediating role of emotions, this study theoretically proposes the causal effects of pro-party information consumption on partisans’ emotions. Nevertheless, these variables are both measured in Wave 1 of the panel survey. To further investigate the chain of causality, future studies should employ more appropriate methods, such as controlled experiments. Second, as the independent variable of this study, the consumption of pro-party television programs is measured by the number of such programs that a person watches at least once a month. In this case, a person who watches a daily news program everyday is treated equally as another person who watches the same program only once a month. Future research could address this limitation by asking respondents about their consumption frequency of each television program. Finally, the measurement of emotions in this study is imperfect, though it is adopted by some recent studies (e.g., Hasell & Weeks, 2016). Although emotional response is conceptualized as an automatic and short-lived state, we have no direct measure of the emotional experience itself and rely on retrospective assessments in the survey. Nevertheless, emotions that fall within the scope of cognitive appraisal theories are often intense and “full-blown” states (Clore & Ortony, 2000). In this sense, we can expect that individuals may be able to recall the frequency of their emotional responses. Future studies could adopt more sophisticated measurement to replicate the findings of this article.
Despite such limitations, this study extends the understanding of how partisan selective exposure exerts influence on affective polarization through negative emotions (i.e., anger and fear). Given the importance of emotional responses in the realm of politics, the mechanisms explored in this study provide a more nuanced understanding of the effects of partisan media.
Footnotes
Acknowledgements
Special thanks to the anonymous reviewers at Journalism & Mass Communication Quarterly (JMCQ) and Dr. Jessica G. Myrick for their support and insights.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
