The Affective Underpinnings of Hostile Media Perceptions

Abstract

The hostile media effect (HME) refers to a process by which highly involved audiences tend to perceive media coverage as biased against their own views. In this process, issue involvement is usually treated as a cognitive construct, that is, as the extent to which the attitudinal issue under consideration is of personal importance. Although Vallone, Ross, and Lepper raised the issue of affective involvement in their seminal study, hardly any research has tried to disentangle the effects of cognitive and affective involvement. Thus, the aim of this article is to clarify whether the HME is triggered by cognitive and/or affective involvement. Data from three survey studies demonstrate that affective involvement—measured as emotional arousal or as the experience of concrete emotions—can explain the HME over and beyond cognitive involvement. Implications of these findings for future HME research are discussed.

Keywords

hostile media effect involvement affective arousal

In almost all Western democracies, allegations of news media bias are on the rise. Numerous studies both in the United States and in Europe revealed that positive perceptions of the news media—such as judgments of trust, credibility, or fairness—have been declining over the past few decades (e.g., Donsbach, Rentsch, & Mende, 2009; Gronke & Cook, 2007). This trend is troubling for modern democracy because such news media distrust has important consequences for political behavior and democratic citizenship. In fact, bias perceptions and news media distrust are associated with a waning of political participation (Moy, Torres, Tanaka, & McCluskey, 2005), a resistance to agenda setting (Tsfati, 2003), a discounting of campaign news (Ladd, 2010), collective misjudgments of public opinion (Choi, Yang, & Chang, 2009; Gunther & Chia, 2001), an increase in sectorial and nonmainstream news consumption (Tsfati & Peri, 2006), an increased minority alienation (Tsfati, 2007), an unwillingness to accept democratic decisions (Tsfati & Cohen, 2005), and less confidence in democracy (Tsfati & Cohen, 2005).

Although biases in the news may be real (see Entman, 2007), a growing body of research suggests that the extent to which news content is objectively biased is less important than individual perceptions of news media impartiality. There is, for instance, an abundance of anecdotal evidence that conservatives blame the press for being liberal while liberals perceive the press as favoring the conservatives. Research on the hostile media phenomenon or hostile media effect (HME), in particular, has revealed that involved audience members tend to perceive a news bias even when news content is well balanced (Gunther & Schmitt, 2004; Schmitt, Gunther, & Liebhart, 2004; Vallone, Ross, & Lepper, 1985). First described by Vallone et al. (1985), the HME has been demonstrated in a considerable number of studies examining various controversial topics, using laboratory data (Gunther, Miller, & Liebhart, 2009; Schmitt et al., 2004), survey methods (e.g., Eveland & Shah, 2003; Tsfati, 2007; Tsfati & Peri, 2006), exploring mechanisms (e.g., Gunther et al., 2009; Schmitt et al., 2004), and consequences (e.g., Tsfati & Cohen, 2005).

To explain the HME, scholars have mainly focused on individual-level variables such as issue involvement. As Choi et al. (2009) and Gunther et al. (2009) have recently observed, however, previous research has not been sensitive to the possible variety of involvement conceptualizations. Issue involvement is usually treated as a cognitive construct, that is, the extent to which the attitudinal issue under consideration is of personal importance (see, for a review, Choi et al., 2009). However, in contrast to this ongoing research practice, Vallone et al. (1985) theorized in their original article that “perceptions of hostile bias are difficult to document unless subjects are intellectually and affectively engaged by the matters being covered in the media” (Vallone et al., 1985, p. 582; emphasis added). Thus, although Vallone et al. raised the issue of affective or emotional involvement in their seminal study, hardly any research has tried to disentangle cognitive and affective involvement when explaining bias perceptions.

Based on this shortcoming, the aim of this article is to clarify whether the HME is also triggered by affective involvement. Data from three survey studies demonstrate that affective involvement—measured as emotional arousal or as the experience of concrete emotions—can explain the HME over and beyond several types of cognitive involvement. Study 1 tests this notion with an online survey controlling for various state and trait notions of involvement. Study 2 replicates the findings of Study 1 using representative survey data and additionally controlling for processing involvement. Finally, Study 3 combines a panel survey with an extensive content analysis of the news media. This design was chosen, first, to address questions of causal order and, second, to control for the objective news hostility that respondents were exposed to.

The HME

In their landmark study, Vallone et al. (1985) exposed pro-Israeli and pro-Arab American students to a television broadcast about the 1983 Beirut massacre. Although their news story was well-balanced and objective, both groups perceived it as more favorable to the opposing side. Vallone et al. called the observed reaction the “hostile media phenomenon.” This phenomenon has been replicated for a variety of topics such as genetically modified foods (Schmitt et al., 2004), general campaign news (Huge & Glynn, 2010), ethnic conflict (Matheson & Dursun, 2001), interreligious conflict (Ariyanto, Hornsey, & Gallois, 2007), Native American interests (Gunther et al., 2009), or social security polices (Hwang, Pan, & Sun, 2008).

There is good evidence that the HME can be erased when the identical article is presented in a nonmedia context. For instance, in Gunther and Schmitt’s (2004) study, the HME vanished when the news article was attributed to a student but it appeared when respondents thought the source of the article was a news journalist. Gunther and Liebhart (2006) argue that the expected reach of a medium and the characterization of the source as a journalist are fundamental preconditions of the HME. The reason is, as these authors have put it, that “a low-reach channel, one that has little or no apparent outside audience, is viewed with a benign eye” (p. 450).

In most experimental HME research, scholars have worked with fair, objective, and balanced news stories. However, recent research has extended the HME to unbalanced news content. The so-called relative HME holds that involved individuals will evaluate news content to be biased even if the coverage is slanted. For instance, Gunther, Christen, Liebhart, and Chia (2001) exposed animal-rights activists and primate researchers to strongly slanted news articles. They found that both groups saw the slant as relatively more unfavorable to their own position.

In recent years, informed by a more sophisticated understanding of cognitive processes, some scholars have begun to reemphasize the psychological mechanisms behind the HME. In their seminal study, Vallone et al. (1985) proposed selective recall and different standards as underlying mechanisms. According to the different standards explanation, involved individuals “could essentially agree about the nature of the stimulus (i.e., its content and valence) but disagree about the appropriateness of the content and valence in light of their differing views about the larger truth that the stimulus was designed to portray” (Vallone et al., 1985, p. 579). Selective recall, in contrast, suggests that involved individuals recall counterattitudinal information better than attitude-consistent information. Giner-Sorolla and Chaiken (1994) were the first to raise the awareness for an additional mechanism, selective characterization. Selective categorization makes “viewers with opposite attitudes recall identical items (e.g., images, facts, or arguments), but classify a predominance of individual items as hostile to their own side” (p. 166). When comparing these three mechanisms, Schmitt et al. (2004) found experimental evidence only for the selective categorization mechanism but not for the other two.

The Role of Involvement

The HME refers to “the tendency for people who are highly involved in an issue to see news coverage of that issue as biased” (Gunther et al., 2001, p. 296, emphasis added). Thus, audience involvement is a pivotal antecedent that makes individuals susceptible to these perceptual biases. Despite its importance, however, “the role of involvement in HME has not been elaborated in the literature” (Choi et al., 2009, p. 55). In fact, involvement is often used interchangeably with the notion of partisanship (see, for this observation, Gunther et al., 2009). Furthermore, there is a myriad of different conceptualizations and operationalizations of these variables, not only in HME research but also in the more general persuasion literature (Wirth, 2006). Some HME studies used attitude measures, defining highly involved individuals as those who hold strong or extreme attitudes (e.g., Hwang et al., 2008). Others used indicators such as fanship (Arpan & Raney, 2003), ideology or ideology strength (Hwang et al., 2008), issue opinion (Chia, Yong, Wong, & Koh, 2007; Gunther & Chia, 2001), personal importance (Gunther & Christen, 2002), engagement in political activities (Eveland & Shah, 2003), group membership (Ariyanto et al., 2007; Gunther et al., 2009; Vallone et al., 1985), group identification (e.g., Matheson & Dursun, 2001), party identification (Dalton, Beck, & Huckfeldt, 1998; Eveland & Shah, 2003; Huge & Glynn, 2010), or knowledge (Vallone et al., 1985). Taken as a whole, it becomes clear that one overarching pattern resonating in HME research is that people must be somehow involved in the issue at hand. However, the conceptualization of involvement, as Gunther et al. (2009, p. 751) have put it in a recent article, is usually rather murky. In general terms, it can be concluded that involvement—within HME research—refers to the intensity of cognitive activity devoted to an issue. This activity manifests in several outcomes, such as strong attitudes, high knowledge, strong group membership/identification, or strong party affiliation. Nevertheless, as Gunther et al. noted, it is “not clear whether either of these dimensions, group identification or extremity of an attitude, is independently sufficient to give rise to the contrast bias” (p. 751).

Only a few scholars have recently begun to examine the relative impact of several involvement dimensions (Choi et al., 2009; Gunther et al., 2009). In line with the more general persuasion literature (Johnson & Eagly, 1989), Choi et al. (2009) have distinguished between value-relevant and outcome-relevant involvement. For value-relevant involvement (or ego involvement), an individual’s attitude is serving a value-expressive function. Outcome-relevant involvement, in contrast, is defined in terms of “the relevance of an issue to one’s currently important goals or outcomes” (Johnson & Eagly, 1989, p. 292.). Comparing these two types of involvement, there is some evidence that value-relevant involvement is a better predictor of the HME than outcome-related involvement (Choi et al., 2009; Gunther et al., 2009).

Beside the notions of value-relevant and outcome-related involvement, however, there are other forms of involvement that might be potentially important in explaining the HME (Gunther et al., 2009). Interestingly, Vallone et al. (1985) themselves have highlighted one such additional dimension, affective involvement. They stated in clear terms that future scholarship should strive to “disentangle the role of affective involvement and knowledge in producing these phenomena” (p. 584). In line with this call, Gunther et al. (2009) have recently reminded HME scholars that there are “many potentially important dimensions of involvement, including issue importance, knowledge, confidence in one’s opinion, and affect” (p. 762, emphasis added). In other words, almost 25 years later, Gunther et al. have repeated the original call by Vallone et al., urging HME scholarship “to measure and analyze [involvement dimensions] more extensively in future research” (Gunther et al., 2009, p. 762).

Affective Involvement

There are grounds to believe that affective involvement can explain the HME over and beyond cognitive accounts. As outlined above, recent studies revealed that value-relevant involvement is a stronger predictor of the HME than outcome-relevant involvement. Yet the overwhelming focus of all these theoretical accounts is largely cognitive, treating involvement as a fairly dispassionate, essentially rational construct. Therefore, we introduce the notion of affective involvement in the next section. After that, we outline its relevance for the HME.

Definition and Scope

It is well established in the persuasion literature that affective involvement is a distinct and useful concept with unique antecedents and consequences (see, for a review, Wirth, 2006; see also Krosnick, Boninger, Chuang, Berent, & Carnot, 1993; Perse, 1990). Rothschild and Ray (1974), for instance, distinguished three main components of involvement, cognitive, affective, and conative. Affective involvement generally refers to emotional feelings associated with an attitude object (Wirth, 2006). Similarly, Chaudhuri and Buck (1995) defined this type of involvement as the experienced intensity of specific emotions (see Perse, 1990). Also research on the various dimensions of attitude strength clearly distinguishes cognitive (i.e., extremity, certainty, importance) from affective indicators (i.e., “the emotional reaction provoked by the attitude object”, Krosnick et al., 1993, p. 1132).

Along similar lines, the more general political communication literature is increasingly separating affective from cognitive mechanisms. Abelson, Kinder, Peters, and Fiske (1982), for instance, were one of the pioneers to stress that affective reactions can be independent of cognitions (see also Ottati, Steenbergen, & Riggle, 1992). Other recent studies demonstrated that emotions have a genuine and distinct influence on political judgments, though cognitions and emotions typically interact in this process (e.g., LeDoux, 1999; Marcus & MacKuen, 1993). Such corroboration for the influences of emotion on political cognition comes both from survey data (e.g., Sniderman, Brody, & Tetlock, 1991) and from experimental designs (e.g., Brader, 2005). The lesson drawn from all these studies is to conceptualize involvement as consisting of two independent but intertwined processes, cognitive and affective (Perse, 1990; Way & Masters, 1996; Wirth, 2006).

When we look at prior HME research, it can be assumed—for all those various issues that have been studied so far—that individuals vary in the extent to which they react emotionally when confronted with these issues. In fact, there is compelling evidence that issues are likely to elicit (especially negative) affective reactions (Brader, 2005; Marcus & MacKuen, 1993). These affective reactions can become chronically accessible and may be activated by the mere presentation of the attitude object (i.e., the issue). Such an automatic activation of affects is similar to the classic notion of associative network theory (e.g., Price & Tewksbury, 1997). More specifically, Bower and Forgas (2001) theorized that associative networks consist of an element that is commonly overlooked: affective nodes. It follows, as Morris, Squires, Taber, and Lodge (2003) stated, that the “emotional evaluation is stored with the concept—that is, political terms are indeed ‘hot’” (p. 742).

Affective Involvement and Bias Perceptions

Based on these insights, the key argument of this article is that people can be affectively aroused when they think about political issues. Following this reasoning, it can be asserted that affective involvement can foster the HME independent from and in addition to cognitive forms of involvement. That is, when people are cognitively involved with an issue, they might not necessarily be affectively involved. They might hold extreme attitudes about an issue, identify with an issue-relevant in-group, or think that the issue is central to future outcomes or to their cherished values. Still people can differ in the extent to which they are affectively aroused in this context. Some people might be cognitively involved but approach the issue in nonaffective, rather rational terms. Others, similarly cognitively involved, may feel strong emotions and experience intensive affective reactions. It is precisely this difference that justifies the separation of cognitive from affective involvement.

No research so far has investigated this point. However, some research from related fields suggests that affective states can lead to biases in information processing. For instance, Gino and Schweitzer (2008) found that individuals in a neutral emotional state were more trusting and more accurate in their judgments than people who felt incidental anger. Ortony, Clore, and Collins (1988) explained that some emotions (i.e., anger, disgust) increase subjective feelings of certainty, which in turn prompts biased information processing. Other research in the affective priming domain demonstrated that individuals experiencing negative affects tend to interpret the same information more negatively than people in positive moods (Bower & Forgas, 2001). All of this suggests that especially negative affective reactions foster perceptual biases.

Based on the historical relevance of affective involvement in HME research (see Vallone et al., 1985) and the evidence we have from the more general persuasion and cognitive psychology literature, it can be hypothesized that affective involvement can instigate the HME even when various measures of cognitive involvement are controlled for. This leads to our main hypothesis:

Hypothesis 1: Affective involvement predicts perceptions of news media bias, after controlling for cognitive involvement.

Overview of Studies

The main hypothesis was tested with three survey studies. The studies differ in context, issue, and year in which they were conducted. With the use of survey data, we follow the well-established inference logic in HME research to correlate indicators of involvement with perceptions of bias (see Eveland & Shah, 2003; Gunther & Chia, 2001; Huge & Glynn, 2010; Lee, 2010; Tsafti, 2007; Tsfati & Cohen, 2005). In all three studies, we expected to find a significant positive effect of affective involvement on bias perceptions. Study 1 uses online survey data about a contested issue. Study 2 replicates the findings of Study 1 with a nationally representative telephone survey about an immigration issue, also controlling for processing involvement as an additional cognitive involvement aspect. Both studies unequivocally demonstrate that affective involvement predicts perceptions of media slant after controlling for several operationalizations of cognitive involvement. However, as it is perilous to infer causal relationships from cross-sectional data, Study 3 uses a two-wave panel survey on an economic issue to investigate whether affective involvement determines hostile media perceptions or—which is equally plausible (see Hwang et al., 2008)—whether hostile media perceptions foster affective involvement. In addition, Study 3 applies a unique and hitherto unknown design to HME research: By combining an extensive content analysis of those media sources the respondents were exposed to, the objective degree of unfavorable media content was controlled for.

Specification of Data Analysis Models

To avoid underspecified regression models, and to make the studies comparable, we have included the same list of controls in all three studies. Besides demographics, we controlled for TV and newspaper use because the manifestation of rather enduring hostile perceptions (as measured in surveys) depends on the intensity of exposure (see Tsfati & Cohen, 2005).

When it comes to cognitive involvement, the aim was to operationalize as many different facets as possible. To begin with, controlling for a person’s own opinion appeared warranted because the HME relates one’s actual position on an issue. In fact, many studies find correlations between issue attitudes and biased perceptions (see, for example, Chia et al., 2007; Gunther & Chia, 2001). Second, using attitude strength as the overarching theoretical framework, attitude extremity, attitude importance, and attitude certainty were included in all three studies (see for these variables in HME research, for example, Gunther & Christen, 2002; Gunther & Schmitt, 2004; Hwang et al., 2008). Third, following the involvement operationalizations by Eveland and Shah (2003), Hwang et al. (2008), or Huge and Glynn (2010), ideology strength was added. The idea behind ideology strength is that both strong conservatives and strong democrats will perceive more bias than people with weaker ideological convictions.

Finally, we took the perspective of social psychological dual-process models, the elaboration likelihood model (Petty & Cacioppo, 1990), and the heuristic systematic model (Chen & Chaiken, 1999). In these prominent models, involvement is usually understood as the motivation to process arguments (i.e., in the central/systematic route). This processing motivation can be conceptualized at the trait or at the state level. When it comes to the trait level, researchers within this domain usually measure high involvement as individuals’ need for cognition (NFC, see Cacioppo, Petty, & Kao, 1984). The rationale behind including NFC is that some people are generally more cognitively involved than others. This general tendency unfolds its impact across issues, and, hence, there should be some trait variance in explaining the HME. At the state level, involvement refers to the issue-specific cognitive engagement. It is usually operationalized as systematic processing involvement (Schemer, Matthes, & Wirth, 2008; Trumbo, 1999). Thus, NFC (Studies 1, 2, and 3) and processing involvement (only Studies 2 and 3) were controlled for.

To reiterate, the aim of this modeling strategy was neither to clarify the relations among all those different cognitive involvement variables nor to discuss here which of these indicators is most appropriate on theoretical grounds. In contrast, the premier goal was to control all potentially relevant operationalizations of involvement. The intuitive logic behind this strategy is that a broad variety of involvement indicators maximizes the variance and explanatory power of this concept. Then, if affective involvement triggers slant perceptions even though this list of variables is controlled, we can conclude that the HME is, in part, driven by affective processes.

Study 1

Method

The first study was administered through a web survey pertaining to the World Economic Forum’s (WEF) annual meeting, a major global event that took place at Switzerland, from January 23 to January 27, 2008. The WEF, as a highly controversial event, was accompanied by public protest and outrage. Thus, the issue is suited for an HME. Respondents were solicited by placing announcements on a variety of newsgroups, blogs, text, and banner ads on newspaper websites, homepages of politicians and political activists, and on official sites by the local state government. Special attention was paid to recruiting both opponents and proponents of the WEF. The survey was available online from January 23 to February 14, 2008. In order to increase data quality, respondents who did not complete the survey from beginning to end and respondents who completed the survey more quickly than pilot testing suggested was possible were excluded. The final sample included 1,096 respondents (M_age = 41.33, SD = 17.26, range = 18 to 86 years; 26% female). Respondents with higher educational degrees were overrepresented (35% college degree, 27% high school degree necessary for college).

Measures

If not indicated otherwise, all items were measured on a 5-point scale (1 = do not agree at all to 5 = fully agree), and all indices were one-dimensional as revealed by principal components analysis (nonorthogonal rotation, parallel analysis; see Fabrigar, Wegener, MacCallum, & Strahan, 2009). The items are detailed in the appendix. The dependent variable, hostile media perceptions, was measured with an index of five items tapping opinion-hostile news coverage (α = .72; M = 3.54, SD = 0.75). These are well-established in the HME literature (e.g., Choi et al., 2009; Eveland & Shah, 2003; Hwang et al., 2008; Tsfati, 2007). Also, following standard procedures (see Krosnick et al., 1993; Perse, 1990; Wirth, 2006), affective involvement was operationalized with four items as the felt intensity of emotions (α = .83; M = 3.13, SD = 0.90). For additional explorative purposes, specific emotions were also measured. People were asked to what extent they experienced “anxiety,” “anger,” “sadness,” “concern,” “joy,” or “hope” when thinking about the WEF. When submitted to principal components analysis, these items formed two factors, negative emotions and positive emotions. Thus, two indices were added (negative emotions, four items: α = .84; M = 2.84, SD = 1.17; positive emotions, two items: α = .80; M = 2.23, SD = 1.21).

As for the controls, sex (2 = female, 1 = male), age, highest education as well as frequency (1 = very seldom to 5 = very often) of TV use (M = 3.47, SD = 1.32) and newspaper use (M = 4.16, SD = 0.96) were measured. Issue attitude was controlled with one item (1 = do not agree at all, 10 = fully agree; M = 4.62, SD = 3.23), importance was assessed with two items (α = .68; M = 2.98, SD = 1.11), and attitude certainty was measured with one standard single item (see Matthes, Morrison, & Schemer, 2010; M = 3.92, SD = 0.93). Extremity (M = 2.55, SD = 1.36) was operationalized as the absolute value of the difference between participants’ own attitude and the scale’s midpoint (see Lavine, Borgida, & Sullivan, 2000). Finally, NFC was assessed with a single item taken from Cacioppo et al. (1984; M = 3.56, SD = 1.07).

Data analysis

Data were analyzed by a block-wise, hierarchical OLS regression. Sex, age, and education were included in the first block, TV and newspaper use in the second, issue attitude and cognitive involvement in the third, and, finally, affective involvement in the forth.

Results

The results are depicted in Table 1. As can be seen, age (β = –.116, p < .001) and TV use (β = –.121, p < .001) were significant predictors of media bias. In line with prior research on the HME, issue attitude (β = –.333, p < .001), extremity (β = .147, p < .001), certainty (β = .102, p < .01), ideology strength (β = .142, p < .01), and NFC (β = .066, p < .05) were all related to perceptions of news media slant.¹

Table 1.

Regression Coefficients for Predicting Hostile Media Perceptions (Study 1).

Predictors	Unstandardized coefficients	Standard error	Standardized coefficients	p value
Block 1: Demographics
Sex	.079	.057	.044
Age	−.005	.001	−.116	***
Education	−.004	.024	−.005
Incremental R² = .014**
Block 2: Media use
Frequency TV news	−.069	.018	−.121	**
Frequency newspaper use	−.046	.024	−.059
Incremental R² = .020**
Block 3: Cognitive involvement
Issue attitude	−.077	.008	−.333	***
Attitude extremity	.079	.026	.147	***
Attitude certainty	.082	.018	.102	**
Importance	.014	.021	.014
Ideology strength	.174	.052	.142	**
Need for cognition	.046	.020	.066	*
Incremental R² = .165***
Block 4: Affective involvement
Affective involvement	.050	.024	.071	*
Incremental R² = .003*
Total R² = .202
Total R²_Adjust=193

Note: *p < .05. **p < .01. ***p < .001.

In Hypothesis 1, it was asserted that affective involvement should significantly predict hostile media perceptions even when various facets of cognitive involvement are controlled for. As can be seen in Table 1, this was the case. Although the effect is comparatively small, perceptions of media hostility were more pronounced when individuals were affectively engaged in the issue at hand (β = .071, p < .05).² This effect remains robust when each of the cognitive involvement indicators is excluded from the model. Moreover, affective involvement was only weakly or moderately correlated with issue attitude (r = –.335, p < .001), extremity (r = .081, p < .01), certainty (r = .167, p < .01), ideology strength (r = .249, p < .001), importance (r = .342, p < .001), and NFC (r = .047, ns). This lends further credence to our reasoning that affective involvement and cognitive involvement are related but distinct concepts.

In an explorative fashion, the Affective Involvement Index was exchanged for the index measuring specific emotions (not shown in Table 1). For negative emotions, the results remain the same: After controlling for cognitive involvement, negative emotions were a positive predictor of bias (β = .116, p < .01). The opposite was true for positive emotions (β = –.113, p < .05).³

Discussion

Study 1 demonstrated that affective involvement explained the HME even though various facets of cognitive involvement were controlled. Additional analysis suggested that also concrete negative emotions foster bias perceptions. It is important to stress that, as a whole, the effects of cognitive involvement were rather strong, explaining the largest amount of variance. Thus, there are grounds to assume that cognitive involvement was adequately operationalized in this study.

Some control variables were also significant. Older people were less likely to perceive news slant. This might be explained by a higher amount of political sophistication that older citizens have compared with younger citizens. The negative effect of TV use suggests that frequent exposure makes it more likely that respondents might abandon their impressions of new slant.

Taken together, the findings of Study 1 confirm the hypothesized role of affective involvement. However, some caution is warranted in drawing this conclusion. On the theoretical side, we have argued that all potential factors of cognitive involvement need to be controlled for. However, we were not able to test the effects of systematic processing involvement, the issue-specific component of involvement as conceptualized by dual-process theories (Chen & Chaiken, 1999; Petty & Cacioppo, 1990). Although a single NFC item was included to account for the trait variance of this construct, we were unable to control its issue-specific variance.

On the methodological side, Study 1 can potentially be criticized for its sampling methodology. Although web-based surveys are largely accepted in the scientific community, it is fair to say that representative telephone surveys working with random sampling provide more reliable and more valid results. Finally, although we have used standard measures for almost all cognitive involvement variables, the operationalization of NFC with a single item was not optimal. Therefore, the study needs to be replicated with a better sample and working with a multiple-item NFC Index. Besides, as has often been noted, no single study can be considered definitive with respect to some research issue or question. Thus, replicating the findings of Study 1 with the same list of predictors would lend more confidence to our reasoning.

Study 2

The second study was a nationally representative sample (telephone interviews, random quota; N = 500; age: M = 49.25, SD = 17.13; 53% female). The sample showed sufficient variety in educational levels. The survey dealt with the issue of asylum policy. Asylum policy is suited to study the HME as it is a contested issue with an ongoing heated debate in Europe.

Measures and Data Analysis

Measures and measurement procedures were similar to Study 1 (i.e., 5-point scale; one-dimensional indices). Hostile media perceptions were measured with six items (α = .74; M = 3.63, SD = 0.63; see Kohring & Matthes, 2007) tapping the degree to which respondents felt the reported information was true and covered all relevant aspects of an issue (see, for similar measures, Lee, 2010). Two items referred to television, two to newspapers, and two to radio;⁴ and they all referred to the issue of asylum policy and the media that respondents were exposed to. Affective involvement was again measured with four items, three of which were identical to Study 1 (α = .73; M = 3.36, SD = 0.94).

Again, sex (2 = female, 1 = male), age, highest education as well as TV use (M = 3.62, SD = 1.09) and newspaper use (M = 4.03, SD = 1.05) were controlled. One issue attitude item (M = 3.38, SD = 1.46), two importance items (α = .76; M = 4.37, SD = 0.84), one attitude certainty item (M = 3.87, SD = 1.18), attitude extremity (M = 1.72, SD = 0.45), and a three-item NFC Index taken from Cacioppo et al. (1984; α = .71; M = 3.49, SD = 0.89) were modeled. Finally, not present in Study 1, a five-item index of processing involvement taken from Schemer et al. (2008) was added (α = .75; M = 3.64, SD = 0.81). Data analysis was identical to Study 1.

Results

Results of Study 2 are depicted in Table 2. Again, TV use was negatively related to bias perception. When it comes to cognitive involvement, only issue attitude (β = .148, p < .001) and processing involvement (β = .128, p < .05) were significant predictors. The other indicators were unrelated to bias perceptions. Some effects pointed even to the opposite direction, though not statistically significant. More germane to our central argument, affective involvement had a positive impact (β = .112, p < .05) on respondents’ feeling that the news media was slanted. It follows that there is clear evidence for Hypothesis 1.

Table 2.

Regression Coefficients for Predicting Hostile Media Perceptions (Study 2).

Predictors	Unstandardized coefficients	Standard error	Standardized coefficients	p value
Block 1: Demographics
Sex	.046	.065	.039
Age	−.001	.002	−.027
Education	.003	.010	.018
Incremental R² = .002
Block 2: Media use
Frequency TV news	−.079	.029	−.143	**
Frequency newspaper use	−.042	.031	−.073
Incremental R² = .026**
Block 3: Cognitive involvement
Issue attitude	.061	.022	.148	**
Attitude extremity	−.030	.087	−.021
Attitude certainty	−.040	.031	−.078
Importance	−.071	.044	−.098
Ideology strength	−.082	.092	−.047
Need for cognition	.002	.040	.003
Processing involvement	.96	.046	.128	*
Incremental R² = .041*
Block 4: Affective involvement
Affective involvement	.072	.037	.112	*
Incremental R² = .010*
Total R² = .079
Total R²_Adjust= .045

Note: *p < .05. **p < .01. ***p < .001.

Affective involvement was again only weakly or moderately related to issue attitude (r = .032, ns), extremity (r = .096, p < .05), certainty (r = .147, p < .01), ideology strength (r = .112, p < .05), importance (r = .342, p < .001), NFC (r = .095, p < .05), and processing involvement (r = .297, p < .001). The effect of affective involvement remained robust when each of these indicators was excluded from the regression model.

Discussion

Albeit the effects of cognitive involvement were smaller than in Study 1, Study 2 validated the findings of the first study. One reason for the loss of explanatory power of cognitive involvement might lie in Study 2’s operationalization of biased perceptions. Unlike Study 1, these were measured rather indirectly in terms of credibility and completeness judgments. Though these facets have been found to be strongly related to bias perceptions (see Arpan & Peterson, 2008; Choi et al., 2009; Tsfati & Cohen, 2005)—and some scholars do not differentiate between the two (see Meyer, 1988; West, 1994)—it can be speculated that the effects of cognitive involvement might be more pronounced when using bias perceptions as the dependent variable.

Despite this compelling evidence for our argument, the results of both studies should still be interpreted with caution. There are essentially three major drawbacks. First, the effect of the target independent variable was rather small, raising questions about the robustness of this effect. Second, and more importantly, there might be other explanations for the relationship between affective involvement and bias perceptions. Some recent research suggests that perceptions of media bias can lead to negative emotional reactions and feelings of media indignation (Hwang et al., 2008). Thus, it is equally plausible to assume that affective involvement is not the cause but rather the consequence of hostile media perceptions (see Munro & Ditto, 1997). Even more fundamentally, one could argue that bias perceptions might cause cognitive involvement. Clearly, the cross-sectional design of Studies 1 and 2 is unable to rule out this rivaling explanation. What is needed, therefore, is a replication of these effects in a panel setting. Third, survey studies on the HME are plagued by their failure to control the news coverage that people were exposed to. That is, experimental HME studies can expose subjects to either balanced or unbalanced news stories. In real-world studies, however, there is an extensive self-selection of news sources. Because of that it is conceivable that the observed relationships are the result of true biases in the news content that people were exposed to (e.g., through self-selection or other potential confounding variables).

Therefore, Study 3 was designed to replicate findings of Studies 1 and 2 using a panel design and controlling for the news content that people were exposed to. This is accomplished by matching a content analysis and a panel survey at the individual level (see Kepplinger, Brosius, & Staab, 1991). The advantage of this procedure is that bias perceptions can be regressed on the specific news content—that is, objective slant—that people were exposed to.

Study 3

Study 3 dealt with a reform of corporate taxation in 2008. The reform was accepted in a popular vote with a majority of 50.5%. As with any popular vote or referendum, proponents and opponents were mobilized throughout the debate, and there was a conflict between both camps.

A panel survey was combined with a content analysis of news media. The idea was to demonstrate a cross-lagged effect of affective involvement at t₁ on bias perceptions at t₂, while also examining the opposite causal relationship. The theoretical model is depicted in Figure 1. Paths a and b denote the autoregressive effects, path f signals the effect of bias perceptions on affects, and path e—equivalent to Hypothesis 1—the impact of affective involvement on bias. The controls including cognitive involvement were measured at t₁, explaining bias and involvement at t₂. The reason for this was to give cognitive and affective involvement the same statistical weight in the model. As will be apparent below, Hypothesis 1 predicts a cross-lagged effect of affective involvement at t₁ on bias at t₂. Thus, the same effect must be predicted for cognitive involvement (and all other controls). If path e is significant despite the paths for cognitive involvement, Hypothesis 1 can be confirmed. As for all exogenous variables, correlations between the controls and involvement at t₁ were allowed.

Figure 1.

Theoretical structural equation model.

The key idea of an autoregressive panel model is to demonstrate a cross-lagged influence while controlling for autoregressive effects. The major premises for a causal interpretation are the temporal order of variables and the control of other variables. As the autoregressive effects a and b are controlled for simultaneously, and there is a clear temporal order, the effects f and e can be interpreted as causal effects (see Burkholder & Harlow, 2003; Jöreskog, 1979).

Method

Survey data and measures

The sample of the two-wave panel survey (N₁ = 1,251, N₂ = 1,001) was representative for French- and German-speaking eligible voters of Switzerland. It was recruited by random digit dialing (random quota, computer-assisted telephone interviewing [CATI]) in January (Wave 1) and March 2008 (Wave 2; age: M = 50.02, SD = 16.3; 50.3% female). Measurement procedures were identical to Studies 1 and 2 (i.e., 5-point scale; one-dimensional indices). Hostile media perceptions were measures with three items (Wave 1: α = .69; M = 3.45, SD = 0.88; Wave 2: α = .74; M = 3.39, SD = 0.74) similar to Study 2.⁵ Items referred to the issue under investigation and the news media that respondents were exposed to. Affective involvement was measured with three items (Wave 1: α = .52; M = 3.18, SD = 0.94; Wave 2: α = .61; M = 3.17, SD = 0.98).⁶

Sex, age, highest education, TV use (M = 3.82, SD = 1.21), and newspaper use (M = 3.96, SD = 1.10) were controlled. Again, one issue attitude item (1 = do not agree, 10 = totally agree; M = 5.83, SD = 2.40), two importance items (α = .61; M = 3.72, SD = 1.00), one certainty item (M = 3.45, SD = 1.26), extremity (M = 1.72, SD = 0.45), a three-item NFC Index taken from Cacioppo et al. (1984; α = .61; M = 3.43, SD = 1.15), and a three-item index of processing involvement taken from Schemer et al. (2008)⁷ were modeled (α = .78; M = 3.34, SD = 1.15).

Content analysis of TV and newspaper coverage

Every news item about the reform that appeared in 25 newspapers and six TV shows between the first- and the second-panel wave (N = 505 stories, N = 3,059 arguments) was sampled. Unit of analysis was the argument. An argument can be defined as a verbalization of a specific point of view in which a claim is expressed with a certain evaluation. As with any referendum or vote, there are mainly two sides, the pro and the contra side. Thus, for the present article, all arguments were simply categorized into “pro” or “contra.” The pro position argued that the reform would help small and mid-sized companies, that the competitive capability of Switzerland would be increased by the reform, and that the reform would bring jobs and new investments. The contra position suggested that the reform would harm the old and survivors’ insurance contribution, that the Swiss federation and the cantons would lose tax money, and that the tax system would be further complicated. For every news item, it could be examined whether there were more pro or more contra arguments. Coding of single arguments yielded a reliability of Scott’s pi = .77 (two coders, random selection of five articles; for a critical discussion, see Krippendorff, 2011). Reliability at the general pro/contra level was perfect (Scott’s pi = 1).

Data matching and opinion-hostile coverage

The data of the content analysis and the panel survey were matched at the individual level. Similar to Kepplinger et al. (1991), the occurrence of pro and contra arguments was counted for every media outlet in the sample. Then, every survey participant was assigned a value representing the frequency with which he or she was confronted with pro or contra arguments. This matching was based on the specific news media use patterns that the survey respondents reported.⁸

Thus, two variables were added to the survey, number of received pro (M = 87.71, SD = 58.45) and number of received contra arguments (M = 154.71, SD = 102.00). Then, received contra arguments were subtracted from the pro, leading to a new variable (i.e., received pro arguments; see, for example, Beck, Dalton, Greene, & Huckfeldt, 2002). This variable was combined with the issue opinion. As a result, n = 240 respondents received attitude-consistent (i.e., were pro/contra and dominantly received pro/contra) and n = 203 respondents were exposed to counterattitudinal media content (i.e., were pro/contra and received contra/pro).⁹

Data analysis

Data were analyzed using a cross-lagged autoregressive panel analysis. Missing values were addressed with full information maximum likelihood (Enders & Bandalos, 2001). Indices were modeled as latent variables (see Figure 1). Autocorrelations of measurement errors over time and between exogenous variables were allowed. Both for affective involvement and news bias, partial metric measurement invariance over time was checked and secured for two items. Equivalence over time is crucial for such models as changes in the reliability of a measure must be separated from the stability of a construct (Hertzog & Nesselroade, 2003).

Results

The fit of the structural equation model was satisfactory but not excellent (χ²/df = 2.48; Comparative Fit Index [CFI] = .93; root mean square error of approximation [RMSEA] = .034; the p value for the test of closeness of fit [PCLOSE] = 1.00; see Browne & Cudeck, 1993).¹⁰ Factor loadings for all trust and bias items were larger than λ = .40 (not shown). As Table 3 shows, older (γ = –.064; p < .10) and more educated respondents (γ = –.99; p < .01) were less likely to perceive bias than their younger and less educated counterparts. When it comes to cognitive involvement, only processing involvement (γ = .164; p < .01) and attitude extremity (γ = .059; p < .10) were noteworthy predictors of hostile media perceptions. However, in line with Hypothesis 1, affective involvement at t₁ was a significant positive predictor (γ = .234; p < .01) of bias perceptions at t₂ (path e, Figure 1). Thus, there was a causal effect of affective involvement on bias. Surprisingly, opinion-hostile media coverage—that is, objective news bias—was not related to bias perceptions.

Table 3.

(Structural Equation Modeling) Path Coefficients for Predicting Hostile Media Perceptions and Affective Involvement at Panel Wave 2 (Study 3).

Predictors	Bias perceptions t₂		Affective involvement t₂
Predictors	Standard coefficients	Standard error	Standard coefficients	Standard error
Bias perceptions t₁	.492**	.061	.078	.071
Affective involvement t₁	.234**	.105	.881***	.182
Sex	.023	.043	−.001	.055
Age	−.064*	.015	−.029	.019
Education	−.099**	.001	.027	.001
Frequency TV news	−.001	.018	−.029	.023
Frequency newspaper use	.048	.021	−.036	.027
Issue attitude	.044	.016	−.026	.024
Attitude extremity	.059*	.016	−.061	.020
Attitude certainty	.001	.018	−.015	.023
Importance	.034	.034	.096	.044
Ideology strength	.023	.041	.072*	.052
Need for cognition	.013	.045	.099*	.057
Processing involvement	.164**	.127	.130**	.160
Objective opinion-hostile coverage t₁ − t₂	.001	.064	.184***	.084
Squared multiple correlation (explained variance)	.416		.743

p < .10. **p < .01. ***p < .001.

As stated above, it was theorized that the opposite effect is equally plausible. However, this was not the case. Perceptions of bias at t₁ were not significantly related to affective involvement at t₂, although the effect pointed into the expected direction. Ideology strength (γ = .072; p < .10), NFC (γ = .099; p < .10), and processing involvement (γ = .130; p < .01) increased affective involvement. In contrast to the prediction of news bias, exposure to objectively measured opinion-hostile coverage had an impact on (γ = .184; p < .001) affective involvement.

Discussion

The aim of Study 3 was to replicate and validate Studies 1 and 2 with a more appropriate design and with the inclusion of objectively measured exposure to slant. Results confirm the findings of both prior studies: People who were affectively involved perceived more bias compared with less involved respondents. Cognitive involvement, in contrast, was only able to exert some moderately sized effects. As in Study 2, one reason for this might lie in the measurement of hostile media perceptions. We will revisit this aspect in the General Discussion section.

Results from Study 3 extend our understanding of the link between affective involvement and news bias perceptions. Contrary to our expectations and contrary to some prior research (i.e., Hwang et al., 2008), hostile media perceptions did not fuel affective involvement over time. There are at least two possible explanations for this finding: First, it is worth highlighting that the autoregressive effect (i.e., the stability coefficient) for affective involvement was larger than the one for bias perceptions. This points to the fact that affective involvement is the more stable and persisting construct. As always in autoregressive models, it is nearly impossible to find cross-lagged effects for a dependent variable that remains stable over time. In such a case, all variance is explained by the autoregressive effect and literally nothing is left for other variables. Second, this finding might be explained by the operationalization of affective involvement. In contrast to Hwang et al. (2008), affective involvement was not measured as feelings of media indignation but as issue-specific emotional reactions. It can be speculated that judgments of media indignation might be less stable over time than issue-specific affective involvement. All of this suggests that the concepts of media indignation and affective involvement need to be carefully separated in future studies. As always, more research is needed to resolve this issue.

Perhaps even more importantly, it was observed that objectively received opinion-hostile coverage did not directly affect bias perceptions. However, there was an indirect effect of objective news bias on subjective bias perceptions, mediated through affective involvement. This finding signals that the relationship between bias, affective involvement, and news content is more complicated than previously thought. If opinion-hostile media content fosters affective involvement and affective involvement, in turn, leads to perceptions of bias, then affective involvement might also help to explain mechanisms other than the HME. Still, it is important to stress that affective involvement is not fully explained by objective news bias. The HME is about biases in information processing. Some variance of hostile media perceptions, however, might also be explained by real, objectively measured news biases (through affective involvement).

General Discussion

In their seminal paper, Vallone et al. (1985) had speculated that involvement might have an affective component and that this component may be, in part, responsible for the HME. Surprisingly, this theoretical idea has been—until very recently (Gunther et al., 2009)—almost entirely forgotten in extant HME research. Twenty-five years later, the present research set out to reintroduce this important variable to HME scholarship. Based on the conceptualization of affective involvement used in the more general (cognitive) psychology and persuasion literature, it was shown that affective involvement is able to explain the perceptions of bias even when prominent types of cognitive involvement are controlled for.¹¹ This leads us to conclude that Vallone et al. were right when they speculated that the HME increases when individuals are “intellectually and affectively engaged by the matters being covered in the media” (p. 582).

This conclusion, of course, hinges on the acceptance of our conceptualization of the key dependent and independent variables. To begin with, the operationalization of affective involvement chosen in the present research is well grounded in the involvement literature. However, because the present studies were parts of larger projects, the items for affective involvement did not remain constant throughout the studies. In Study 1, we measured affective involvement as the intensity of felt emotions when confronted with an issue. In addition, a measure tapping specific (negative) emotions was used and yielded the same results. In Study 2, an affect intensity measure was used. In Study 3, again, items referred to specific negative emotional issue aspects. Taken together, one obvious implication of the present results is that especially negative affective arousal is likely to foster perceptions of media bias. Negative emotional responses such as, for instance, anger have been found to spur action tendencies, foster simpler cognitive processing rules, and depress information search (e.g., Lerner & Tiedens, 2006). It remains to be tested with more convincing designs, however, if all specific negative emotions (e.g., anxiety) are relevant in this context.

Also, the bias perception measures slightly varied from study to study. One might argue that some measures applied in the present research might not be relevant for HME research. In fact, Studies 2 and 3 operationalized hostile media perceptions with judgments about truthfulness and completeness. In defense of the generality of the measures that were used, we cited a variety of studies under which similar results have been obtained. News hostility perceptions tend to be highly correlated with credibility items (e.g., Gaziano & McGrath, 1986; Meyer, 1988; West, 1994), and, more importantly, hostile media perceptions have been found to be a direct predictor of news credibility (Arpan & Peterson, 2008; Choi et al., 2009; Tsfati & Cohen, 2005). It follows that a considerable amount of variance of the measures used in Studies 2 and 3 can most likely be explained by perceptions of bias. Although other news characteristics might also shape credibility judgments, there are strong grounds to argue that especially bias is relevant in this context.

Although the results of the three studies are fairly similar, their comparability can of course be questioned. In fact, all studies have worked with different issues and have used different measures. Thus, the present research cannot be considered as truly cumulative, in a way that one study is built upon the insights of a preceding study. However, it can nevertheless be argued that the three studies allow making inferences beyond each individual study. First, we can express greater confidence in our findings when the results from different measures tend to be similar. In fact, all the different measures that were used in the present research have been applied in prior HME research as well. Therefore, the present findings can be generalized to a broader field of HME research compared with a situation where one single way of measurement remains constant throughout three studies. This point reaffirms the importance of combining measures, measures that complement each other and thereby tell a coherent story. Second, the same is true for working with different issues. A public event about global problems, an immigration issue, and a national economic issue are completely different instances where an HME could occur. Of course, no inferences whatsoever can be drawn from the present findings about which issues are most suitable for an HME. However, it can be concluded that affective involvement is an important aspect across these different issues. Again, this adds to the generalizability of the present findings compared with a situation where the issue remains constant across studies.

The aim of this article was not to clarify the concept of cognitive involvement. Nevertheless, we were able to include most of the cognitive involvement operationalizations that have been used in prior HME research. Some more recent studies, however, have worked with more direct measures of value-relevant involvement (Choi et al., 2009). Unfortunately, such items were not available when the three studies were designed. At least, there are grounds to assume that group identification items—as measured in Study 1 (see Note 1)—were able to account for some variance in value-relevant involvement. Moreover, value-relevant involvement should be highly correlated with the cognitive involvement measures that were applied here. Again, this reminds us that the role of cognitive involvement is by no means clarified. Cognitive involvement remains a slippery concept that necessitates more theoretical effort.

Theoretical and Methodological Implications

In considering the broader implications of our findings, especially three things become apparent. First, the findings point to the necessity to broaden our understanding of how cognitive and affective involvement are related to each other. Two scenarios—equally plausible on theoretical grounds—wait for scholarly attention and empirical investigation. On one hand, it is conceivable that various types of involvement are in a hierarchical relationship. That is, only when lower order types are present, higher order manifestations become relevant to hostile media perceptions. To give an example, it could be argued that importance is a lower order and affective intensity is a higher order form of involvement. This would imply that affective intensity does only matter to the HME when importance is given. Put more simply, how can you be affectively aroused when you think the issue is unimportant? On the other hand, it has been shown in cognitive psychology that affective reactions can be independent from cognitive ones. That is, people might be affectively aroused for other reasons than their cognitive involvement. In the present studies, however, affective involvement explained the HME additionally to cognitive involvement. This suggests that cognitive involvement is a necessary condition to trigger the HME. However, the types of cognitive involvement may also vary in their relevance. Some types might be necessary conditions of an HME, others sufficient conditions.

In all three studies, additional analysis demonstrated—as could be expected in a regression logic—that the effects of affective involvement increased when the cognitive variables were excluded. Although this informs us that both types share a common variance, we do not learn much about the precise relationship between the two. As a consequence, in order to answer the question about involvement hierarchy (and independence), we need to carefully distinguish various involvement types (Gunther et al., 2009), and we need to apply elaborate experimental designs manipulating these types in controlled settings. In real-world studies, as the ones reported here, all these things tend to correlate and it is notoriously difficult to address the underlying psychological processes.

The second implication of the present findings is to consider more carefully the role of time in HME research. The HME effect is usually tested with rather static designs working with existing group membership or well-established social and political issues. Unfortunately, as elegant these designs are, they invariably leave several interesting questions unanswered. In particular, they are unable to explain the formation, endurance, and long-term consequences of hostile media phenomena. Thus, more systematic attempts to trace hostile media perceptions over time are needed. Especially, longitudinal experiments or panel studies employing three waves (or more) are warranted.

Finally, we believe that the design used in Study 3 can be a useful starting point for drawing a more coherent picture about real-world HME. Space considerations allow only an enumeration of a few of the myriad research questions ready to be explored: Combining content analysis with survey data, for instance, enables a test of how hostility perceptions, media use, and objective media content relate to each other. That is, the design makes it possible to test the consequences of hostile media perceptions on subsequent media use and content perception. Or, one could look at other content characteristics (i.e., negative depiction of positions) and trace their effects on bias perceptions. Similarly, this design allows us to test how citizens react to objective bias (in comparison with biased perceptions). In light of all this, it is hoped that the present research provides a first step on this promising avenue.

Footnotes

Appendix

Declaration of Conflicting Interests

The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The research conducted in this article was supported by a grant from the Swiss National Science Foundation as a part of the National Center of Competence in Research (NCCR) “Challenges to Democracy in the 21st Century.”

Notes

Author Biography

Jörg Matthes (PhD, University of Zurich, 2007) is full professor at the Department of Communication, University of Vienna, Austria. His research focuses on public opinion formation, media effects, advertising, and empirical methods.

References

Abelson

R. P.

Kinder

D. R.

Peters

M. D.

Fiske

S. T.

(1982). Affective and semantic components in political person perception. Journal of Personality and Social Psychology, 42, 619-630.

Ariyanto

Hornsey

M. J.

Gallois

(2007). Group allegiances and perceptions of media bias: Taking into account both the perceiver and the source. Group Processes & Intergroup Relations, 10, 266-279.

Arpan

L. M.

Peterson

(2008). Influence of source liking and personality traits on perceptions of bias and news source selection. Media Psychology, 11, 310-329.

Arpan

L. M.

Raney

A. A.

(2003). An experimental investigation of news source and the hostile media effect. Journalism and Mass Communication Quarterly, 80, 265-281.

Beck

P. A.

Dalton

R. J.

Greene

Huckfeldt

(2002). The social calculus of voting: Interpersonal, media, and organizational influences on presidential choices. American Political Science Review, 96, 57-73.

Bower

G. H.

Forgas

J. P.

(2001). Mood and social memory. In Forgas

J. P.

(Ed.), Handbook of affect and social cognition (pp. 95-120). Mahwah, NJ: Erlbaum.

Brader

(2005). Striking a responsive chord. How political ads motivate and persuade voters by appealing to emotions. American Journal of Political Science, 49, 388-405.

Browne

M. W.

Cudeck

(1993). Alternative ways of assessing model fit. In Bollen

K. A.

Long

J. S.

(Eds.), Testing structural equation models (pp. 136-162). Newbury Park, CA: Sage.

Burkholder

G. J.

Harlow

L. L.

(2003). An illustration of a longitudinal cross-lagged design for larger structural equation models. Structural Equation Modeling, 10, 465-486.

10.

Cacioppo

J. T.

Petty

R. E.

Kao

C. F.

(1984). The efficient assessment of need for cognition. Journal of Personality Assessment, 48, 306-307.

11.

Chaudhuri

Buck

(1995). Affect, reason, and persuasion: Advertising strategies that predict affective and analytic-cognitive responses. Human Communication Research, 21, 422-441.

12.

Chen

Chaiken

(1999). The heuristic-systematic model in its broader context. In Chaiken

Trope

(Eds.), Dual-process theories in social and cognitive psychology (pp. 73-96). New York, NY: Guilford.

13.

Chia

S. C.

Yong

S. Y. J.

Wong

Z. W. D.

Koh

W. L.

(2007). Personal bias or government bias? Testing the hostile media effect in a regulated press system. International Journal of Public Opinion Research, 19, 313-330.

14.

Choi

Yang

Chang

J. J. C.

(2009). Elaboration of the hostile media phenomenon: The roles of involvement, media skepticism, congruency of perceived media influence, and perceived opinion climate. Communication Research, 36, 54-75.

15.

Dalton

R. J.

Beck

P. A.

Huckfeldt

(1998). Partisan cues and the media: Information flows in the 1992 presidential election. American Political Science Review, 92, 111-126.

16.

Donsbach

Rentsch

Mende

(2009). The ethics gap. Why Germans have little esteem and no trust in journalists. Paper presented at the 60th Annual Conference of the International Communication Association, Chicago, IL.

17.

Enders

C. K.

Bandalos

D. L.

(2001). The relative performance of full information maximum likelihood estimation for missing data in structural equation models. Structural Equation Modeling, 8, 430-457.

18.

Entman

R. M.

(2007). Framing bias: Media in the distribution of power. Journal of Communication, 57, 163-173.

19.

Eveland

W. P.

Jr. Shah

D. V.

(2003). The impact of individual and interpersonal factors on perceived news media bias. Political Psychology, 24, 101-117.

20.

Fabrigar

L. R.

Wegener

D. T.

MacCallum

R. C.

Strahan

E. J.

(1999). Evaluating the use of exploratory factor analysis in psychological research. Psychological Methods, 4, 272-299.

21.

Gaziano

McGrath

(1986). Measuring the concept of credibility. Journalism Quarterly, 63, 451-462.

22.

Giner-Sorolla

Chaiken

(1994). The causes of hostile media judgments. Journal of Experimental Social Psychology, 30, 165-180.

23.

Gino

Schweitzer

(2008). Blinded by anger or feeling the love: How emotions influence advice taking. Journal of Applied Psychology, 93, 1165-1173.

24.

Gronke

Cook

T. E.

(2007). Disdaining the media: The American public’s changing attitudes toward the news. Political Communication, 24, 259-281.

25.

Gunther

A. C.

Chia

(2001). Predicting pluralistic ignorance: The hostile media percept and its consequences. Journalism and Mass Communication Quarterly, 78, 688-701.

26.

Gunther

A. C.

Christen

C. T.

(2002). Projection or persuasive press? Contrary effects of personal opinion and perceived news coverage on estimates of public opinion. Journal of Communication, 52, 177-195.

27.

Gunther

A. C.

Christen

C. T.

Liebhart

J. L.

Chia

(2001). Congenial public, contrary press, and biased estimates of the climate of opinion. Public Opinion Quarterly, 65, 295-320.

28.

Gunther

A. C.

Liebhart

J. L.

(2006). Broad reach or biased source? Decomposing the hostile media effect. Journal of Communication, 56, 449-466.

29.

Gunther

A. C.

Miller

Liebhart

J. L.

(2009). Assimilation and contrast in a test of the hostile media effect. Communication Research, 36, 747-764.

30.

Gunther

A. C.

Schmitt

(2004). Mapping boundaries of the hostile media effect. Journal of Communication, 54(1), 55-70.

31.

Hertzog

Nesselroade

J. R.

(2003). Assessing psychological change in adulthood: An overview of methodological issues. Psychology and Aging, 18, 639-657.

32.

Huge

Glynn

(2010). Hostile media and the campaign trail: Perceived media bias in the race for governor. Journal of Communication, 60, 165-181.

33.

Hwang

Pan

Sun

(2008). Influence of hostile media perception on willingness to engage in discursive activities: An examination of mediating role of media indignation. Media Psychology, 11, 76-97.

34.

Johnson

B. T.

Eagly

A. H.

(1989). The effects of involvement on persuasion: A meta-analysis. Psychological Bulletin, 106, 290-314.

35.

Jöreskog

K. G.

(1979). Statistical estimation of structural models in longitudinal-developmental investigations. In Nesselroade

J. R.

Baltes

P. B.

(Eds.), Longitudinal research in the study of behavior and development (pp. 303-351). New York, NY: Academic Press.

36.

Kepplinger

H. M.

Brosius

H.-B.

Staab

J. F.

(1991). Opinion formation in mediated conflicts and crises: A model of cognitive-affective media effects. International Journal of Public Opinion Research, 3, 132-156.

37.

Kohring

Matthes

(2007). Trust in news media. Development and validation of a multidimensional scale. Communication Research, 34, 231-252.

38.

Krippendorff

(2011). Agreement and information in the reliability of coding. Communication Methods and Measures, 5, 93-112.

39.

Krosnick

J. A.

Boninger

D. S.

Chuang

J. C.

Berent

M. K.

Carnot

C. G.

(1993). Attitude strength: One construct or many related constructs? Journal of Personality and Social Psychology, 65, 1132-1151.

40.

Ladd

J. M.

(2010). The role of media distrust in partisan voting. Political Behavior, 32, 567-585.

41.

Lavine

Borgida

Sullivan

J. L.

(2000). On the relationship between attitude involvement and attitude accessibility: Toward a cognitive-motivational model of political information processing. Political Psychology, 21, 81-106.

42.

LeDoux

J. E.

(1999). The emotional brain: The mysterious underpinnings of emotional life. London, UK: Phoenix.

43.

Lee

(2010). Why they don’t trust the media: An examination of factors predicting trust. American Behavioral Scientist, 54, 8-21.

44.

Lerner

J. S.

Tiedens

L. Z.

(2006). Portrait of the angry decision maker: How appraisal tendencies shape anger’s influence on cognition. Journal of Behavioral Decision Making, 19, 115-137.

45.

Marcus

G. E.

MacKuen

M. B.

(1993). Anxiety, enthusiasm, and the vote. The emotional underpinnings of learning and involvement during presidential campaigns. American Political Science Review, 87, 672-685.

46.

Matthes

Morrison

K. R.

Schemer

(2010). A spiral of silence for some: Attitude certainty and the expression of political minority opinions. Communication Research, 37, 774-800.

47.

Matheson

Dursun

(2001). Social identity precursors to the hostile media phenomenon: Partisan perceptions of coverage of the Bosnian conflict. Group Processes & Intergroup Relations, 4(2), 116-125.

48.

Meyer

(1988). Defining and measuring credibility of newspapers: Developing an index. Journalism Quarterly, 65, 567-574, 588.

49.

Morris

J. P.

Squires

N. K.

Taber

C. S.

Lodge

(2003). Activation of political attitudes: A psychophysiological examination of the hot cognition hypothesis. Political Psychology, 24, 727-745.

50.

Moy

Torres

Tanaka

McCluskey

M. R.

(2005). Knowledge or trust? Investigating linkages between media reliance and participation. Communication Research, 32, 59-86.

51.

Munro

G. D.

Ditto

P. H.

(1997). Biased assimilation, attitude polarization, and affect in the processing of stereotype-relevant scientific information. Personality and Social Psychology Bulletin, 23, 636-653.

52.

Neuwirth

(2000). Testing the spiral of silence model: The case of Mexico. International Journal of Public Opinion Research, 12, 138-159.

53.

Ortony

Clore

G. L.

Collins

(1988). The cognitive structure of emotions. Cambridge, UK: Cambridge University Press.

54.

Ottati

V. C.

Steenbergen

M. R.

Riggle

(1992). The cognitive and affective components of political attitudes: Measuring the determinants of candidate evaluations. Political Behavior, 14, 423-442.

55.

Perse

E. M.

(1990). Involvement with local television news: Cognitive and emotional dimensions. Human Communication Research, 16, 556-581.

56.

Petty

R. E.

Cacioppo

J. T.

(1990). Involvement and persuasion: Tradition versus integration. Psychological Bulletin, 107, 367-374.

57.

Preacher

K. J.

Hayes

A. F.

(2008). Asymptotic and resampling strategies for assessing and comparing indirect effects in multiple mediator models. Behavior Research Methods, 40, 879-891.

58.

Price

Tewksbury

(1997). News values and public opinion: A theoretical account of media priming and framing. In Barnett

Boster

F. J.

(Eds.), Progresses in the communication sciences (pp. 173-212). Greenwich, CT: Ablex.

59.

Rothschild

M. L.

Ray

M. L.

(1974). Involvement and political advertising effect: An exploratory experiment. Communication Research, 1, 264-284.

60.

Schemer

Matthes

Wirth

(2008). Toward improving validity and reliability of information processing measures in surveys. Communication Methods and Measures, 2(2), 1-33.

61.

Schmitt

K. M.

Gunther

A. C.

Liebhart

J. L.

(2004). Why partisans see mass media as biased. Communication Research, 31, 623-641.

62.

Sniderman

P. M.

Brody

R. A.

Tetlock

P. E.

(1991). Reasoning and choice: Explorations in political psychology. Cambridge, UK: Cambridge University Press.

63.

Trumbo

C. W.

(1999). Heuristic-systematic information processing and risk judgment. Risk Analysis, 19, 391-400.

64.

Tsfati

(2003). Does audience skepticism of the media matters in agenda setting? Journal of Broadcasting and Electronic Media, 47, 157-176.

65.

Tsfati

(2007). Hostile media perceptions, presumed media influence and minority alienation: The case of Arabs in Israel. Journal of Communication, 57, 632-651.

66.

Tsfati

Cohen

(2005). Democratic consequences of hostile media perceptions: The case of Gaza settlers. Harvard International Journal of Press/Politics, 10, 28-51.

67.

Tsfati

Peri

(2006). Mainstream media skepticism and exposure to extra-national and sectorial news media: The case of Israel. Mass Communication & Society, 9, 165-187.

68.

Vallone

R. P.

Ross

Lepper

M. R.

(1985). The hostile media phenomenon: Biased perception and perceptions of media bias in coverage of the Beirut massacre. Journal of Personality and Social Psychology, 49, 577-585.

69.

Way

B. M.

Masters

R. D.

(1996). Emotion and cognition in political information-processing. Journal of Communication, 46, 48-65.

70.

West

M. D.

(1994). Validating a scale for the measurement of credibility: A covariance structure modeling approach. Journalism Quarterly, 71, 159-168.

71.

Wirth

(2006). Involvement. In Bryant

Vorderer

(Eds.), Psychology of entertainment (pp. 199-213). Mahwah, NJ: Erlbaum.