3

Communication Research 2017, Vol. 44(8) 1075 –1098

© The Author(s) 2015 Reprints and permissions:

sagepub.com/journalsPermissions.nav DOI: 10.1177/0093650215594234

journals.sagepub.com/home/crx

Article

Toward a Cognitive-Affective Process Model of Hostile Media Perceptions: A Multi-Country Structural Equation Modeling Approach

Jörg Matthes1 and Audun Beyer2

Abstract This article develops and tests a theoretical cognitive-affective process model of the hostile media effect (HME). To explain the HME, scholars have mainly focused on cognitive involvement, that is, the extent to which an issue is of personal importance. In addition, we introduce the notion of affective involvement and hypothesize three distinct routes responsible for a HME: a cognitive, an affective, and a cognitive- affective route. Simultaneously collected representative survey data from the United States, Norway, and France employing country-invariant measures provide clear evidence that the three routes each and independently drive the HME. Theoretical and methodological implications of these findings are discussed.

Keywords hostile media effect, news trust, bias perception, affective priming, measurement invariance

Public opinion research across the globe has observed an increase of public distrust in mainstream news media (Donsbach, Rentsch, & Mende, 2009; Gronke & Cook, 2007; Tsfati & Ariely, 2014). In explaining this phenomenon, scholars have argued that news distrust can be understood as a situational response following from high issue involve- ment. This theoretical perspective is not new, however. It was introduced by Vallone,

1University of Vienna, Austria 2University of Oslo, Norway

Corresponding Author: Jörg Matthes, Professor, Department of Communication, University of Vienna, Waehringerstr. 29, 1090 Vienna, Austria. Email: [email protected]

594234 CRXXXX10.1177/0093650215594234Communication ResearchMatthes and Beyer research-article2015

1076 Communication Research 44(8)

Ross, and Lepper (1985), who called it the hostile media phenomenon or hostile media effect (HME). The question of what prompts highly involved individuals to believe that the news media is biased against their own views has piqued the interest of many communication scholars. As key antecedents of hostile media perceptions (HMPs), HME researchers have focused on individual-level variables such as involvement, that is, the extent to which a political issue is of personal importance (Ariyanto, Hornsey, & Gallois, 2007; Choi, Yang, & Chang, 2009; Gunther, Christen, Liebhart, & Chia, 2001; Gunther, Miller, & Liebhart, 2009).

Yet in their seminal article on the HME, Vallone et al. (1985) made a clear point that the HME is not entirely driven by cognitive processes. More specifically, the authors stated, “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). Surprisingly, this assumption has— until very recently (Gunther et al., 2009; Matthes, 2013)—almost entirely been forgot- ten in extant HME research. As one exception, Matthes (2013) reported three studies demonstrating that affective involvement is able to explain the perceptions of media bias even when prominent types of cognitive involvement are controlled for. Thus, there seems to be a kind of affective basis underlying the HME.

The present article follows the footsteps of Matthes’s (2013) study but differs in three important aspects. First, instead of treating cognitive and affective involvement as (correlated) independent variables, we aim to shed light on the complex relationship between cognitive involvement, affective involvement, and bias perceptions. We introduce a cognitive-affective process model that distinguishes a cognitive route, an affective route, and a cognitive-affective route to HMPs. This enables hostile media scholars—for the first time—to separate cognitive from affective mechanisms and estimate their unique contributions. Second, in contrast to the most prior research on the HME, we test our conjectures using representative survey data from three coun- tries, in three languages, using identically worded questions on a globally important topic, rather than employing single-country student samples. What is more, we use a structural equation modeling (SEM) approach that rigorously checks the equivalence of measurement constructs across countries—which is one of the biggest challenges in cross-cultural research (Steenkamp & Baumgartner, 1998)—in order to rule out the possibility that items have different meanings in different countries leading to errone- ous conclusions about substantial relationships. Third, instead of using Matthes’s broad measures of affect intensity, the present research distinguishes positive from negative emotions, therefore exploring the role of emotional valence in the HME. That is, we theorize that negative as well as positive emotions are relevant to HMPs.

The HME

In the landmark study by Vallone et al. (1985), pro-Israeli and pro-Arab American students were exposed to a television broadcast about the 1983 Beirut massacre. Despite the news story being well balanced and objective, each group considered it as more favorable to the opposing side. Vallone et al. dubbed the observed reaction the

Matthes and Beyer 1077

hostile media phenomenon or effect. The HME refers to a process in which highly involved individuals rate identical news texts as biased against their own side, even though the texts are completely neutral and impartial. Research on the HME has intro- duced a whole new perspective to the study of bias perceptions in particular and news media trust judgments in general. The key contribution has been to move scholarly attention away from content factors or communicator characteristics underlying trust judgments. Rather, this research has brought attention to the subjective perceptual processes of audience members. The HME has spurred substantial empirical research endeavors, the effect has been replicated for a variety of topics (Ariyanto et al., 2007; Huge & Glynn, 2010; Hwang, Pan, & Sun, 2008), in several countries such as the United States (Gunther et al., 2009; Schmitt, Gunther, & Liebhart, 2004), Israel (Tsfati, 2007), the United Kingdom (Hartmann & Tanis, 2013), or Switzerland (Matthes, 2013), both inside (Gunther et al., 2009; Schmitt et al., 2004) and outside the labora- tory (e.g., Tsfati, 2007; Tsfati & Peri, 2006). Early HME research has employed objec- tively fair and balanced news stories. More recent research, however, has extended the HME to unbalanced news content (Gunther et al., 2001). As another important feature of the HME, it has been shown that the expected reach of a medium and the character- ization of the source as a journalist are fundamental preconditions for the HME to occur (Gunther & Liebhart, 2006).

Involvement and the HME

The HME is defined as the “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). Surprisingly, however, “the role of involvement in HME has not been elaborated in the literature” (Choi et al., 2009, p. 55). In fact, there is no agreed- upon definition of involvement (Gunther et al., 2009). There is a huge array of differ- ent involvement operationalizations in the HME literature, including attitude strength or extremity (e.g., Hwang et al., 2008), ideology or ideology strength (Hwang et al., 2008), issue opinion (Chia, Yong, Wong, & Koh, 2007; Gunther & Chia, 2001), per- sonal importance (Gunther & Christen, 2002), group membership (Ariyanto et al., 2007; Gunther et al., 2009; Vallone et al., 1985), party identification (Huge & Glynn, 2010), or knowledge (Vallone et al., 1985). Recent evidence suggests that especially value-relevant involvement drives HMPs (Choi et al., 2009; Gunther et al., 2009).

No matter which involvement measure has been used, what all of them share is their cognitive nature. In response to this observation, Gunther et al. (2009) have 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). This idea is not a new one. On the contrary, Vallone et al. stated in clear terms as early as 1985 that future scholarship should strive to “disentangle the role of affective involvement and knowledge in producing these phenomena” (p. 584). Following the footsteps of Vallone et al. (1985) and Gunther et al. (2009), Matthes (2013) conceptualized affective involvement as a unique predic- tor of HMPs, over and beyond several measures for cognitive involvement.

1078 Communication Research 44(8)

Drawing on the general persuasion literature (Rothschild & Ray, 1974), Matthes (2013) distinguished three components of involvement—cognitive, emotional, and cona- tive—of which the first two matter for the HME. Affective involvement can be indicated by the emotional intensity associated with an attitude object (Chaudhuri & Buck, 1995) or by the experience of specific emotions (see Perse, 1990). The separation of cognitive from affective involvement also is in line with the more general political psychology lit- erature. In fact, Abelson, Kinder, Peters, and Fiske (1982) as well as Ottati, Steenbergen, and Riggle (1992) made a clear case that affective reactions can be independent of cogni- tions. Furthermore, there is a bulk of empirical evidence that emotions have a genuine and distinct influence on political judgments, which has been demonstrated in the laboratory (e.g., Brader, 2005) or by using survey data (e.g., Sniderman, Brody, & Tetlock, 1991). Translated to HME research, it can be concluded that involvement should be conceptual- ized as consisting of two independent but intertwined components, cognitive and affec- tive (Matthes, 2013; Perse, 1990; Way & Masters, 1996; Wirth, 2006).

Based on this theorizing, Matthes (2013) tested the role of affective involvement in HMPs. In three studies using survey data, the author operationalized affective involve- ment either as the intensity of emotion (i.e., general emotional arousal when con- fronted with issues) or as negative (“anxiety,” “anger,” and “sadness”) and positive (“concern,” “joy,” and “hope”) emotion. Findings revealed that affectively involved individuals are more likely to see the news media as biased than less affectively involved respondents, even if a multitude of cognitive involvement dimensions are controlled for. Furthermore, it was found that both positive and negative emotions could predict the HME. While negative emotions led to an increase in bias percep- tions, the opposite occurred for positive emotions. Ultimately, this research shows that there is an affective base underlying the HME: The HME increases when people are emotionally involved in addition to cognitive involvement.

This emerging literature raises several challenging research questions that have not been adequately answered so far. Although there are strong grounds to assume that affective involvement can explain the HME over and beyond cognitive involvement, it remains unclear how cognitive and affective involvement play together. There has been a huge debate in psychological research as to whether cognitions drive emotions or emotions drive cognitions (Lazarus, 1991; Zajonc, 1984). Irrespective of which side one takes in this discussion, the relationship between cognitive and affective involve- ment as drivers of the HME deserve more research efforts. Interestingly, as one first hint, Matthes (2013) observed a significant effect of cognitive involvement on affec- tive involvement in a cross-lagged panel model. While this gives us some indication that the two components are related, we need a theoretical model that is able to clarify the relations between cognitive and affective involvement.

A Cognitive-Affective Process Model of the HME

A model that incorporates cognitive as well as affective involvement should be con- ceptualized to answer four interrelated questions. First, similar to conventional HME research, it needs to explain why highly cognitively involved audience members react

Matthes and Beyer 1079

with more perceived media bias compared with less cognitively involved members. This is a fundamental pillar of HME research. Second, it needs to explain how affective involvement can impel bias perceptions, in addition to cognitive involvement. Third, it should incorporate positive and negative emotions and attempt to explain how positive versus negative emotions matter in this process. Fourth, and finally, it needs to clarify how cognitive involvement and affective involvement are intertwined when explaining slant assessments. This article aims to provide a first attempt at arriving at such a theo- retical cognitive-affective process model. As the key idea, we distinguish three routes to the HME: a cognitive route, an affective route, and a cognitive-affective route.

Cognitive Route

Almost all prior research has focused on the cognitive route. This cognitive account can be traced back to Vallone et al. (1985) who have proposed “selective recall” and “different standards” as underlying mechanisms. Giner-Sorolla and Chaiken (1994) proposed a “selective categorization” account of HMPs. 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). Schmitt et al. (2004) designed an experiment comparing these three cognitive mechanisms. The results suggest that the selective categorization mecha- nism can best explain the HME. A cognitive route to HMPs assumes a direct effect of cognitive involvement (independent of affective involvement). It refers to the basic hypothesis of the HME that is formulated as follows:

Hypothesis 1: People high in cognitive involvement will perceive a higher news media bias than people low in cognitive involvement (when affective involvement is controlled).

Affective Route

The cognitive route holds that highly involved individuals classify the information they recall from the news as hostile to their own views. As Matthes (2013) states, however, when people are cognitively involved with an issue, they may not necessar- ily be affectively involved; some people might be cognitively involved but approach the issue in non-affective, rather rational terms.

We propose affective priming theory as an account to explain affective HMEs. Affective priming refers to a process in which emotions, mood, or affect are temporar- ily more available in memory and therefore inform subsequent judgment formation (Baumgartner & Wirth, 2012; Erisen, Lodge, & Taber, 2013; Forgas & Bower, 1987; Kühne, Schemer, Matthes, & Wirth, 2011; Wirth, Schemer, & Matthes, 2010). Such an affective priming mechanism can be explained by the functioning of human memory (Forgas & Bower, 1987). When people experience positive affect, affectively congru- ent information is automatically activated in recall and therefore is more accessible in memory. It follows that positive (negative) emotions automatically facilitate the recall

1080 Communication Research 44(8)

of information that is congruent to these positive (negative) emotions. Bower and Forgas (2001) give an explanation for this affective priming process: Based on the associative network of human memory, they theorize that humans’ associative net- work has two emotion nodes, a positive one (positive affect) and a negative one (nega- tive affect). As Kühne et al. (2011) explain, “If an individual experiences an emotional reaction in response to a given situation, the corresponding emotion node is activated and acts as the source of an impulse travelling the associative paths to affectively con- gruent concepts” (p. 490). As a consequence, positive (negative) emotions will acti- vate the node for positive (negative) affect, which makes it more likely that positively (negatively) valenced concepts are activated.

There is an abundance of evidence for affective priming effects, as for instance in psychology (Bower & Forgas, 2001; Forgas & Bower, 1987), in political science (Erisen et al., 2013), or in communication (Baumgartner & Wirth, 2012; Chang, 2014; Kühne et al., 2011; Wirth et al., 2010). More specifically, research suggests that affect can shape perceptions of fairness (Mullen, 2007), credibility perceptions (Wirth et al., 2010), mem- ory content (Adolphs & Damasio, 2001), or impression and attitude formation (for a review, see Forgas, 1995). It follows that “people’s affective states have the potential to influence how fair or unfair they determine an event to be” (Mullen, 2007, p. 22).

Applied to the HME, this means that affective involvement can be positive or nega- tive. When there is positive affective involvement, people are more likely to activate affect-congruent information when they judge the trustworthiness of news items. Because bias refers to clearly negative attributes of news stories, positive affective involvement should dampen the HME. By the same token, when negative affective involvement is high, affect-congruent categories become easily and rapidly available when judging news content. As a consequence, individuals will rate the news item as less trustworthy (compared with positive affective involvement). There is some first evidence for such an affective route. Matthes (2013) showed that affect explains the HME even when all key dimensions of cognitive involvement were controlled for. That is, there must be a second mechanism at work other than the cognitive. Importantly, Dunn and Schweitzer (2005) demonstrated that persons who are in a positive affective state place more trust in communicators than in persons who are in a negative state. Also the study by Wirth et al. (2010) gave direct and unequivocal evidence that posi- tive affective states lead to an increase in subjective perceptions of news credibility and trust compared with negative affective states. Therefore, what can be derived from affective priming theory is the prediction that positive affective involvement should reduce the HME whereas negative involvement should foster the HME. However, in order to interpret these relations as an affective pathway to the HME, it is important to stress that the cognitive route must be statistically controlled. In more technical terms, the affective route accounts for the variance that is not explained by the cognitive route. This leads to the following hypotheses:

Hypothesis 2: People high in positive affective involvement will perceive less news media bias than people low in positive affective involvement (when cognitive involvement is controlled).

Matthes and Beyer 1081

Hypothesis 3: People high in negative affective involvement will perceive more news media bias than people low in negative affective involvement (when cogni- tive involvement is controlled).

Cognitive-Affective Route

So far, we have explained how cognitive and affective involvement can predict the HME. In line with previous scholarship on the relationship between affect and cogni- tion, however, it would seem important to know how the affective route and the cogni- tive route play together. There are essentially two answers to this question, both of which received tremendous scientific attention and empirical support. First, affect drives cognition, and second, cognition drives affect. This debate can be traced back to the question of a primacy of affect which has been suggested by Zajonc (1984) and disputed by Lazarus (1991).

Appraisal theory would suggest that cognition precedes affect because emotions derive from a cognitive evaluation of an attitude object, that is, from an appraisal. An appraisal can be defined as “an evaluation of the significance of what is happening in the person-environment relationship” (Lazarus, 1991, p. 87). In the context of a HME, this means that cognitive involvement would impact affective involvement. Rephrased, cognitive involvement is theorized to be a predictor of affective involvement. To give an example, there are grounds to assume that people who are highly cognitively involved in an issue are more likely to react with emotions compared with less involved people. Appraisal theorists would argue that there can be no emotional involvement when there is no cognitive involvement in the first place. As a consequence of an increase in cognitive involvement, the affective route (i.e., affective priming) will gain more weight. Put formally, affective involvement is theorized to mediate the relation- ship between cognitive involvement and bias perceptions.

Zajonc (1984), in contrast, introduced the idea that affective processes can heavily drive cognition. Such a “primacy of affect” account holds that affective processes can impact, for instance, decision making in the absence of any conscious cognitions (LeDoux, 1999; McDermott, 2004). Such a model would lead us to predict that people are affectively involved in the first place, which then determines their cognitive involvement. Thus, affective involvement may increase cognitive involvement, whereas the latter will lead to bias perceptions. That is, the impact of selective catego- rization on bias perceptions is increased due to the effect of affective involvement on cognitive involvement. Formally, this model would treat cognitive involvement as a mediator for the relationship between affective involvement and bias.

Whether there is a primacy of affect or a primacy of cognition cannot be sorted out in the present research (but see Ellsworth, 2013; LeDoux, 2012; Mulligan & Scherer, 2012; Nadeau, Niemi, & Amato, 1995). However, what we learn from this debate is that it is not enough to distinguish a cognitive from an affective route. We need to model their interrelationship. In our model, we prefer to follow an appraisal approach that predicts cognitive involvement to be a cause of affective involvement. We think it is plausible to argue that―in the context of the HME―it is more reasonable to assume

1082 Communication Research 44(8)

that cognitive involvement is a precondition for affective involvement rather than assuming that affective involvement is present before there is any cognitive involve- ment. In fact, in the context of news topics, it is rather unlikely that people experience topic-related affect when they are not cognitively involved at all, for example, when an issue is not relevant to them. Even when it comes to examples such as human suffering in faraway places, we would argue that people need to see some relevance in a topic before they experience emotions. Based on this premise, a number of hypotheses now can be derived. Assuming that cognitive involvement drives affective involvement, we hypothesize that this concerns both positive and negative affect. In other words, when individuals are highly cognitively involved, it becomes more likely that they will expe- rience positive and negative affect compared with less cognitively involved people.

Hypothesis 4: The relationship between cognitive involvement and bias percep- tions will be mediated by positive affect in a way that cognitive involvement increases positive affect, which, in turn, decreases bias. Hypothesis 5: The relationship between cognitive involvement and bias percep- tions will be mediated by negative affect in a way that cognitive involvement increases negative affect, which, in turn, increases bias.

Our five hypotheses are visualized in the cognitive-affective process model of HMEs that is depicted in Figure 1. It is important to note that the cognitive-affective route by no means suppresses the single cognitive and single affective paths. As Figure 1 shows, the direct effect of cognitive involvement on HMPs is called the cognitive route. This effect is based on selective categorization, as demonstrated by Giner-Sorolla and Chaiken

Figure 1. Cognitive-affective process model of hostile media perceptions.

Matthes and Beyer 1083

(1994) as well as Schmitt et al. (2004). Likewise, the direct effect of affective involve- ment (positive and negative) on perceived slant is called the affective route, explained by an affective priming process (Dunn & Schweitzer, 2005; Forgas & Bower, 1987; Kühne et al., 2011; Wirth et al., 2010). Finally, the cognitive-affective route refers to a process in which cognitive involvement serves as a precondition for emotional involvement, whereas the latter then impacts bias perceptions. We do not expect to observe a full mediation. A full mediation would mean that the direct effect of cognitive involvement vanishes when we include the mediator. This would imply that the HME was entirely explained by the mediating role of affect. In contrast to this view, we theorize that all effects exert their influence simultaneously. That is, there is an effect by the affective route although the cognitive route is controlled. Likewise, there should be an effect by the cognitive route despite the explanatory power of emotions. And, there should be an indirect effect of cognitive involvement on bias, mediated by affective involvement.

The interplay of affect and cognitions could also be modeled as moderation instead of mediation (see Kühne et al., 2011). However, in our case, a moderated effect would not make sense on theoretical grounds. It seems odd to assume that affective priming should be stronger when people have a tendency to engage in selective categorization due to high cognitive involvement. Likewise, there is no theoretical basis for the assumption that selective categorization due to high cognitive involvement increases when affective priming is at work. It follows that we understand the effects of cogni- tive and affective involvement as additive, rather than interactive. Rephrased, the HME can be explained by selective categorization on the cognitive side, and addition- ally by affective priming on the affective side. We can say so because our model assumes the effects of either side under the condition that the other side is controlled. However, to clarify this question, we formulate the following research question:

Research Question 1: Do affective and cognitive involvement interact in their effects on HMPs?

A Multi-Country SEM Approach

We designed a test of our model under real-world conditions with representative sur- vey data. Unlike all prior research, however, we use multi-country data of more than 3,000 respondents from the United States, Norway, and France. The surveys were conducted simultaneously on the issue of (illegal) immigration. We have chosen the same issue in all three countries because multiple issues would complicate the inter- pretation of potential country differences (see for a similar approach, Lee, Detenber, Willnat, Aday, & Graf, 2004).

Issue

The United States has been a country largely based on immigration for centuries, France is a former colonial power, and in Norway, net immigration started to rise as late as in the 1970s. The three countries also differ in their media and political systems. There are genuine and popular anti-immigrant parties in Norway and France but not in

1084 Communication Research 44(8)

the United States. Immigrants have had different status in the countries, with largely economic immigrants in the United States and asylum seekers in France and Norway (see Beyer & Matthes, 2015; Thorbjørnsrud, 2015).

Despite these differences, the issue is perfectly suited for our research purpose. The issue of immigration is an equally important topic in the three countries. There is a popular resistance to immigration, and the share of the immigrant population is fairly similar in all three countries. It is also a contested and polarized issue, so there are opponents and supporters of immigration in Norway, France, and the United States. Since there are no theoretical grounds to predict country differences, testing our hypotheses simultaneously in three countries with the same issue greatly enhances the generalizability of our findings.

Measurement Invariance

The key challenge of such multi-country, multi-language research is to ascertain that items have similar or identical meanings in the different countries. In fact, there are established methodological tools available for this test (Davidov, Schmidt, & Billiet, 2011; Vandenberg & Lance, 2000). Our hypotheses and our research question formu- lated above require that the concepts of trust/bias, cognitive involvement, and affec- tive involvement have the same substantial meaning in the three countries. This is generally called measurement invariance. Only when we secure measurement invari- ance, we can draw valid conclusions about substantial differences (or similarities) in different countries. Without evidence of measurement invariance, the observed rela- tions may stem from measurement idiosyncrasies and may not reflect real differences. As Medina et al. (2009) have put it, “Practitioners must be careful not to assume that the availability of cross-national survey data implies the appropriateness of cross- national comparisons” (p. 334).

There are three forms of measurement invariance (Davidov et al., 2011). Configural invariance means that the items load on the same factors in all the countries. It is assumed that a construct has a similar meaning across countries because the same items define the construct. Metric invariance assumes that the factor loadings are equal across countries. This means that if there is a one-unit change in the latent construct on one country, we can assume the same change in another. Metric invariance allows compari- son of relationships between variables across countries. Restricting factor loadings to be equal across countries is a conservative and often unrealistic assumption. Scholars therefore regard it as sufficient when at least two loadings are identical across countries (e.g., Steenkamp & Baumgartner, 1998). This is called partial metric invariance. Finally, if we are to compare means across countries, we need to establish scalar invariance, that is, equal intercepts in addition to equal loadings (Davidov et al., 2011).

A SEM approach testing measurement invariance has important advantages over traditional regression models. First, besides checking basic psychometric properties in SEM, we can ascertain that constructs have the same meaning in each country. This is a fundamental precondition for any meaningful comparison. Second, if we compare effects across countries, we can test if these are significantly different from each other by multi-group nested model comparison. Taken together, a SEM approach allows us

Matthes and Beyer 1085

to test the universal stability of our cognitive-affective model in completely different contexts and countries.

Method

Data

We conducted three simultaneous surveys in the United States (N = 1,026), in Norway (N = 1,048), and in France (N = 1,034). Respondents were recruited from the online access panels of IPSOS, a global market research company. Respondents were provided with an incentive. Quota sampling was applied with respect to age, gender, and education in each country (United States: 60% female, Mage = 53.01 [SD = 15.36], education—66% some kind of college degree; Norway: 53% female, Mage = 53.28 [SD = 15.10], educa- tion—54.1% some kind of college; France: 58.3% female, Mage = 46.82 [SD = 17.10], education—18.4% college degree). IPSOS reported a response rate (RR) for online data collection: RR = 0.29 (Norway), RR = 0.22 (France), and RR = 0.07 (United States).

Measures

Items were professionally translated by IPSOS and measured on 7-point scales. Items are listed in the appendix. Two alternative measures for the dependent variable were employed. Both were used in prior research on the HME. First, we used three key items measuring trust in news media taken from Kohring and Matthes (2007). Affective involvement was measured with three items for negative and two items for positive emo- tions. Items were taken from Matthes (2013). Respondents were given a list of emotions and they were asked to indicate to what extent they experienced these emotions with respect to current media coverage about illegal immigration. Comparable measures have been used in prior research (see Watson & Clark, 1997). A principal components analysis (PCA) confirmed these two factors. Cognitive involvement was measured with two items tapping personal issue importance based on Gunther and Christen (2002). Means, standard deviations, and Cronbach’s alpha for these items are depicted in Table 1.

In addition to these key measures and standard demographics (gender: 1 = male, 2 = female; education: 1 = high, 2 = low), we have also controlled left-right political ideology, political inefficacy, political interest, issue opinion, and attitude extremity (based on issue opinion) with single items. We chose these controls to rule out the pos- sibility that important aspects of cognitive involvement are missed out. Also, the con- trols are understood as an exploratory effort to specify the circumstances when one or the other dimension of involvement is more likely to operate.

Data Analysis

The full model is depicted in Figure 2. We analyzed the data with full information maximum likelihood (FIML). This method produces more reliable estimated values (Enders & Bandalos, 2001). To evaluate model fit, the following criteria were used: confirmatory fit index (CFI), root mean square error of approximation (RMSEA), and

1086 Communication Research 44(8)

Figure 2. Theoretical structural equation model. Note. The effects of the controls age, sex, education, issue opinion, opinion extremity, political interest, inefficacy, and left-right political ideology on all other variables are omitted in this figure for clarity reasons.

the p of close fit (PCLOSE). In order to check measurement invariance, trust, positive as well as negative affective involvement, and cognitive involvement were modeled as latent variables.

Results

Measurement Invariance

We set the factor loadings of the “importance of illegal immigration” item, the “all important information” trust item, the anger item, and the contentment item to 1 for each country (for identification purposes) while freely estimating the remaining factor loadings. This gives us a reference model against which model comparisons can be made (CFI = 0.96, RMSEA = 0.03, PCLOSE = 1.00). The fit of this model is good.

Table 1. Means, Standard Deviations, and Reliability for the Dependent and Independent Variables by Country.

Norway United States France

M SD α M SD α M SD α

Trust in news media 3.81 1.29 .68 3.52 1.41 .73 3.47 1.35 .75 News media bias 0.95 0.87 — 1.42 1.04 — 1.46 1.09 — Negative affective involvement 3.23 1.43 .79 3.47 1.56 .77 3.79 1.59 .73 Positive affective involvement 2.64 1.23 .67 2.50 1.36 .67 3.44 1.35 .72 Cognitive involvement 4.78 1.33 .84 4.93 1.61 .92 5.42 1.48 .90

Matthes and Beyer 1087

In order to interpret the substantial relations among affective involvement, cogni- tive involvement, bias, and news media trust, we need to secure equal meaning of these constructs across countries. This is achieved by testing metric invariance. To assume full metric invariance, we constrained all factor loadings to be equal across all three countries. As a result, global fit measures indicate a significant decrease in model fit, not supporting measurement invariance (Δχ2 = 32.82, p < .001). We can thus not compare the effects of cognitive and affective involvement across countries with a fully invariant model. As a consequence, we modeled the less strict assumption of partial invariance by constraining the factor loading of at least one item per construct to be equal in all countries while allowing the loadings for the remaining indicators to differ across countries. In order to find the model with the largest number of invariant items, a series of iterations was done. As the optimal solution, we set the following items to be equal in all three countries: the anxiety item, the hope item, the “frequency is adequate” trust item, and the “importance of immigration” item. The other items were only invariant for two of the three countries: The fear item as well as the “differ- ent points of view” trust item were set to be equal for Norway and the United States but not for France. Compared with a model with no constraints, the partially invariant model did not significantly decrease model fit (Δχ2 = 14.50, p = .15; CFI = 0.96, RMSEA = 0.03, PCLOSE = 1.00).

Substantial Relationships

Having established partial metric invariance, we can now test the relationships between the two types of involvement and news media trust. The results are depicted in Table 2. The first hypothesis stated a direct effect of cognitive involvement on bias perceptions and trust, respectively (i.e., the cognitive route). As can be seen, there was a positive effect of cognitive involvement on news media bias in all three countries (Norway: b = .08, p < .01; United States: b = .07, p < .05; France: b = .12, p < .001). For trust in news media, there was only a significant negative effect in Norway (b = −.10, p < .05). The effect was very close to significance in France (b = −.08, p = .05) but not signifi- cant in the United States (b = −.01, n.s.). Overall, these findings confirm Hypothesis 1, especially when using news media bias as the dependent variable.

The affective route to the HMP was described in Hypotheses 2 and 3. Findings revealed that positive affect exerted a significant impact on trust in news media (Norway: b = .37, p < .001; United States: b = .68, p < .001; France: b = .58, p < .001) and news media bias (Norway: b = −.28, p < .001; United States: b = −.28, p < .001; France: b = −.36, p < .001). With only one exception (i.e., France), negative affect was also a significant predictor of news media trust (Norway: b = −.18, p < .001; United States: b = −.27, p < .001; France: b = .00, n.s.) and news media bias (Norway: b = .11, p < .01; United States: b = .19, p < .001; France: b = .21, p < .001). This supports Hypotheses 2 and 3: Positive affect resulted in more trust and less bias, whereas nega- tive affect had precisely the opposite effect.

Now that we have shown that cognitive involvement and affective involvement predict bias and trust, we can investigate how cognitive and affective involvement is

1088 Communication Research 44(8)

Table 2. Structural Equation Modeling Unstandardized Path Coefficients.

Cognitive involvement

Positive affective

involvement

Negative affective

involvement Trust in news

media News media

bias

b SE b SE b SE b SE b SE

Norway Age .00 0.00 .01*** 0.00 .00 0.00 .01** 0.00 .00 0.00 Sex .21** 0.08 −.26** 0.08 .02 0.06 .02 0.08 −.06 0.06 Education .05 0.08 −.03 0.08 .02 0.06 .00 0.08 .03 0.06 Issue opinion .19*** 0.34 −.08* 0.04 .07* 0.03 .04 0.03 −.08** 0.03 Opinion extremity .00 0.09 −.03 0.09 −.03 0.07 .03 0.08 .00 0.07 Political interest .41 0.03 −.07* 0.03 −.03 0.03 −.01 0.03 .04 0.03 Inefficacy .07* 0.03 .11*** 0.03 .09*** 0.02 .01 0.03 .01 0.02 Left-right ideology .08* 0.04 .02 0.04 .04 0.03 −.06* 0.03 .01 0.03 Cognitive involvement — — .07† 0.04 .24*** 0.03 −.10* 0.04 .08** 0.03 Positive affective

involvement — — — — — — .37*** 0.06 −.28*** 0.04

Negative affective involvement

— — — — — — −.18*** 0.06 .11** 0.04

United States Age .01*** 0.00 −.01*** 0.00 −.01* 0.00 −.01* 0.03 .00 0.00 Sex .21* 0.09 −.17* 0.09 −.10 0.08 .08 0.10 −.10 0.07 Education .08 0.11 .09 0.10 −.07 0.09 −.17 0.12 .00 0.08 Issue opinion .26*** 0.03 −.07* 0.03 .03 0.03 .08* 0.04 −.04 0.03 Opinion extremity .11 0.10 −.09 0.09 −.03 0.08 .03 0.11 .11 0.08 Political interest .30*** 0.03 .07** 0.03 −.04 0.02 .00 0.03 .09*** 0.02 Inefficacy .06* 0.03 .10*** 0.02 .10*** 0.03 −.01 0.03 .01 0.02 Left-right ideology .17*** 0.03 −.03 0.03 .05 0.03 −.05 0.04 .05† 0.03 Cognitive involvement — — −.04 0.03 .25*** 0.03 −.01 −0.05 .07* 0.03 Positive affective

involvement — — — — — — .68*** 0.09 −.28*** 0.05

Negative affective involvement

— — — — — — −.27*** 0.07 .19*** 0.05

France Age .01*** 0.00 .01*** 0.00 .00 0.00 .00 0.00 .00 0.00 Sex .00 0.01 −.03 0.08 .03 0.07 .17 0.09 −.02 0.07 Education .01 0.10 .20* 0.09 .17* 0.07 .02 0.10 .03 0.08 Issue opinion .19*** 0.04 −.08* 0.04 .06* 0.03 .07 0.04 −.02 0.03 Opinion extremity .19 0.11 −.27*** 0.10 −.03 0.08 .09 0.11 .15 0.09 Political interest .18*** 0.03 .00 0.02 .03 0.02 .07** 0.03 .07*** 0.02 Inefficacy .09*** 0.03 .08** 0.02 .06*** 0.02 .03 0.03 −.02 0.02 Left-right ideology .13*** 0.03 −.02 0.03 .11** 0.03 −.04 0.04 .00 0.03 Cognitive involvement — — −.06 0.03 .11* 0.05 −.08†† 0.04 .12*** 0.03 Positive affective

involvement — — — — — — .58*** 0.06 −.36*** 0.05

Negative affective involvement

— — — — — — .00 0.06 .21*** 0.05

†p = .06. ††p = .05. *p < .05. **p < .01. ***p < .001.

Matthes and Beyer 1089

related. We assumed an effect from cognitive to affective, rather than from affective to cognitive involvement. Therefore, we tested the indirect effects of cognitive involve- ment on bias and trust mediated by affective involvement (Hypotheses 4 and 5). See Table 2 for the effects of cognitive involvement on affective involvement. Taking negative involvement as a mediator, results confirm the hypothesized indirect effect on news media trust for Norway (bindirect = −.04, p < .01) and the United States (bindirect = .07, p < .001) but not for France (due to the lack of a negative affective route). The indirect effect on news media bias was statistically significant in all countries (Norway: bindirect = .03, p < .01; United States: bindirect = .05, p < .001; France: bindirect = .02, p < .01). The pattern was less clear when positive affective involvement was modeled as a mediator. Here, the indirect effect was statistically significant for Norway (news media trust: bindirect = .03, p < .01; news media bias: bindirect = .02, p < .01) but not for the United States and France (again, due to the absence of direct effect of cognitive involvement on positive affect). Overall, the results clearly support Hypothesis 5. Hypothesis 4, however, only finds support in Norway.

In addition to these hypothesis tests, some other interesting patterns emerged from the structural equation model. First, cognitive involvement was well explained by almost all exogenous variables in all three countries, as for instance by sex, issue opin- ion, political interest, inefficacy, and political ideology (see Table 2 for the coefficients in each country). In total, 26% of cognitive involvement could be explained for Norway (United States: 35%; France: 21%). Furthermore, Table 2 suggests that there were more significant predictors for positive affective involvement than for negative affective involvement. For instance, political interest and sex explained positive affec- tive involvement but not negative affective involvement (except for France, see Table 2). However, the overall explained variance was higher for negative than for positive affective involvement (positive—Norway: 9%, United States: 7%, France: 10%; nega- tive—Norway: 18%, United States: 20%, France: 17%). It follows that, first, the pre- dictors for cognitive and affective involvement differed, and second, that we were better able to explain negative affective involvement compared with positive.

When looking at Table 2, it also becomes apparent that trust and bias were primar- ily explained by cognitive and affective involvement. The more general variables we controlled do not seem to be relevant. Nevertheless, the explained variance of news media trust (Norway: 19%, United States: 25%, France: 24%) and news media bias (Norway: 11%, United States: 15%, France: 23%) was sufficiently high. When look- ing at all dependent variables in all countries, it was striking that inefficacy had a posi- tive effect on all dimensions of involvement.

Finally, we asked in Research Question 1 whether cognitive and affective involve- ment interact. That is, rather than treating the effects of both dimensions as additive, it might occur that the effect of one dimension is only positive, when the other dimen- sion is also positive. In an exploratory fashion, we modeled the interactive effects between cognitive and positive affective involvement as well as cognitive and nega- tive affective involvement in path analyses using manifest variables. Starting with Norway, there was no interaction between cognitive involvement and positive affec- tive involvement (trust: b = .05, n.s.; bias: b = .01, n.s.), as well as between cognitive

1090 Communication Research 44(8)

involvement and negative affective involvement (trust: b = −.03, n.s.; bias: b = .01, n.s.). A similar pattern emerged for France both for the Cognitive × Positive affect interaction (trust: b = .05, n.s.; bias: b = .00, n.s.) and for the Cognitive × Negative affect interaction (trust: b = −.03, n.s.; bias: b = .00, n.s.). For the U.S. data, the involve- ment types did not interact when explaining trust (Cognitive × Positive: b = .04, n.s.; Cognitive × Negative: b = −.01, n.s.). When it comes to bias, Cognitive × Positive affect interaction was not significant (b = .03, n.s.). However, a Cognitive × Negative affect interaction (b = .03, p < .05) indicated that the effects of cognitive involvement grow stronger with rising negatively affective involvement.

According to our model, cognition precedes affect. However, the alternative model (i.e., affect precedes cognition) is also plausible (see Nadeau et al., 1995). In an addi- tional analysis, we therefore tested a model in which the impact of affective involve- ment on trust is mediated by cognitive involvement. We found that negative affective involvement significantly predicts cognitive involvement in Norway (b = .38, p < .001), the United States (b = .50, p < .001), and France (b = .28, p < .001). Positive affective involvement, however, does predict cognitive involvement in the United States (b = −.30, p < .01) and France (b = −.15, p < .01) but not in Norway (b = −.01, n.s.). In turn, cognitive involvement negatively explains trust in most cases (Norway: b = −.10, p < .05; United States: b = −.00, n.s.; France: b = −.07, p = .054) and bias (Norway: b = .08, p < .05; United States: b = .07, p < .05; France: b = .12, p < .001). Also the direct effects of affective involvement on trust remained significant in five of six paths (positive: Norway: b = .37, p < .001; United States: b = .68, p < .001; France: b = .57, p < .001; negative: Norway: b = −.18, p < .01; United States: b = −.27, p < .001; France: b = .00, n.s.). These analyses demonstrate that the affect-precedes- cognition sequence is equally valid.

Discussion

Although originally proposed by Vallone et al. (1985), affective involvement has almost been forgotten in HME research. The idea behind affective involvement is that there can be an emotional arousal in citizens in addition to their cognitive involve- ment. Furthermore, such arousal can be negative or positive. However, despite the relevance of affective involvement for the perceptions of media bias, our knowledge about the role of affect regarding the HME has remained sparse. In response to this― rather than simply controlling cognitive involvement while estimating the effects of affective involvement (see Matthes, 2013)―we have developed and tested a theoreti- cal model of cognitive-affective HMPs.

The key idea of our model is that several routes can produce the HME. First, there is a cognitive route that has been the focus of scholarly attention in almost all prior research. The cognitive route holds that bias perceptions are created based on a selec- tive categorization process due to high cognitive involvement. Such an understanding of the HME can best be described as a cognitive mistake: Citizens erroneously catego- rize information in ways that make them believe that the media are hostile to them. In addition to the cognitive route, we have suggested an affective route to HMPs. The

Matthes and Beyer 1091

affective route does not assume that individuals make mistakes while categorizing (objectively neutral) information. Rather, it is assumed that positive and affective states can lead to judgments about trust and bias due to a process called affective prim- ing. When people experience positive emotions, the availability of positive informa- tion in memory is higher than when they experience negative emotions. As a consequence, individuals are more likely to see positive things in the objects they evaluate. To give an example, when citizens are confronted with a political issue in the news, this automatically activates positive and/or negative affect. Positive and/or neg- ative affect is transferred to judgments about news media content. Thus, the affective route assumes a transfer of affective states to perceptions of credibility and bias as well as to judgments about trust. Of course, this is an entirely new way of understanding HMEs. In defense of the generality of such an affective priming mechanism, we could cite literally hundreds of studies that have obtained similar results in different contexts (for a review, see Kühne et al., 2011). When people experience affect in response to political issues―and we know they do (Brader, 2005)―this can lead to changes in how “the bringer of a message” is evaluated. In short, affect can impact perceptions of credibility and bias (Dunn & Schweitzer, 2005; Wirth et al., 2010).

We have stressed throughout this article that the interpretation of one route hinges on the fact that the other route is statistically (and conceptually) controlled. So, if we say that selective categorization drives the HME based on high cognitive involvement, and if we still see additional variance that is explained by affective involvement, then the impact of affective involvement must refer to a unique and distinct process. However, this is not to say that cognitive and affective dimensions of involvement are entirely unrelated. Therefore, we have proposed a third route to HMPs, the cognitive- affective route.

The cognitive-affective route is built on the idea that cognitive involvement drives affective involvement. That is, without being somehow cognitively involved, people are unlikely to experience affect (more specifically, topically related affect). However, the unique contribution of the cognitive-affective route is that it puts the effects of cognitive involvement into a whole new light. In addition to the “classic” effect of cognitive involvement, we assume that cognitive involvement may also impact the likelihood of affective reaction in response to news content. This can lead to two dif- ferent outcomes. First, when cognitive involvement fosters positive affect (and the affective route also holds), then cognitive involvement actually decreases bias percep- tion. Ironically, this contradicts the original idea of the HME. However, in line with theorizing on indirect effects, this is not an unlikely scenario. Second, when cognitive involvement leads to negative effects, it accelerates the HME. This is done by an addi- tive impact of the direct effect and the indirect effect.

We have tested this model using data from three countries, the United States, France, and Norway using established measures for cognitive and affective involve- ment as well as two different outcome variables, a more specific bias measure and a more general trust measure. By conducting this research in three countries and in three different languages, we have added a serious methodological challenge to this research, the question of measurement invariance. Only when invariance is established, can we

1092 Communication Research 44(8)

compare the path coefficients that we observe in the different countries. A three- country, partially invariant measurement model was specified and fitted to the data. Therefore, in terms of methodological contribution, this study adds a cross-country survey approach to the arsenal of tools available to HME scholars.

Findings generally support the basic pillars of our model. There was strong and consistent evidence for a cognitive and an affective route. As far as the cognitive- affective route is concerned, the evidence was more compelling for negative than for positive affect. The positive cognitive-affective route was only present for the Norwegian data, but not for the French and the U.S. data. This may signal that nega- tive affect may be more important to HMPs than positive affect, possibly due to a higher subjective intensity of negative affect. Of course, more empirical evidence is needed to back up this claim. Finally, the high amount of explained variance for the two outcome variables lends additional credence to the predictions of our model.

Alternative Explanations

It is important to address some alternative explanations. One could argue that affective involvement simply fosters the (cognitive) process of selective categorization that is driven by cognitive involvement. Toward that end, this could mean that affective prim- ing does not explain the HME. If this explanation holds, the effect of affective involve- ment on bias and trust should disappear once cognitive involvement is treated as a mediator. As described above, we also tested a model in which cognitive involvement served as a mediator for the relationship between affective involvement and bias per- ceptions. In all models, however, a strong and consistent direct path from affective involvement to bias and trust remained highly significant. Thus, this explanation can be ruled out.

Second, as we have asked in our research question, one could argue that the two forms of involvement are in a moderated rather than a mediated relationship. We argued on theoretical grounds that interactive effects of both involvement types lack base. Our findings supported this assumption. Of 12 possible interactions, only one was statistically significant. It follows that the effects of cognitive and affective involvement are additive rather than interactive. This also makes sense in terms of affective priming theory (Kühne et al., 2011). Unlike other processes of affective influence on attitudes, affective priming does not primarily occur under the condition of low cognitive involvement. In fact, the bulk of empirical evidence suggests that affective priming can easily occur even if people are highly cognitively involved (Forgas, 1995). However, in the context of the HME, more evidence is needed to prove and validate this claim.

Limitations and Future Research

As with all non-experimental research, causality is a big issue. Needless to say, we are unable to prove the causal order of cognitive involvement, affective involvement, and bias perceptions. Still, a survey-based approach to HMPs has important advantages in

Matthes and Beyer 1093

terms of external validity, and as in our case, also in terms of international generaliz- ability. Of course, it is reasonable to assume that, in the real world, affect and cognition are related in multi-directional ways. Similar to Slater’s (2007) reinforcing spiral assumption, it can be theorized that cognitive involvement determines affective involvement, which then, in turn, influences cognitive involvement. Ultimately, panel data or complex longitudinal experimental designs are needed to model these assumptions.

To be clear, our notion of cognition as a precondition of affect cannot be clarified or tested in the present research or in any other research involving survey measures. In contrast to our model, primacy of affect (Zajonc, 1984) means that affective responses chronologically precede cognitive ones in attitude formation (see Edwards & von Hippel, 1995). In line with this, Nadeau et al. (1995), for instance, demonstrate that people prioritize those issues that make them anxious.

Cross-sectional surveys, however, measure both concepts simultaneously. Hence, we cannot prove that our theoretical model is more correct than the opposite one. Furthermore, the topic of the present investigation, illegal immigration, has been very prominent in all three countries for decades. So even if there was a primacy of affect (or of cognition), it is far too late to test this given this issue cycle. Both con- cepts will be strongly interrelated for this topic. For the present investigation, the key message is that neither affects nor cognitions alone impact the HMP. Affects and cognitions are related, which clearly increases our understanding of how HMPs are formed.

Yet, one argument we can offer in favor of the cognitive-affective route is that it is unlikely to experience strong effects when people believe that an issue is unimportant (i.e., low cognitive involvement). In other words, why should I react with strong emo- tions if I am not interested in an issue? For instance, research by Ladd and Lenz (2008) suggests that citizens only experience emotions toward candidates they strongly eval- uate in the first place (i.e., cognitions precede emotions). However, experimental research is needed to address this question in the context of the HMP.

Related to this, an examination of our model in an experimental setting would fur- ther test the internal validity of the relationships we found. In the lab, we could also explore the role of concrete emotions, that is, separating the effects of, for instance, anger from those of fear. This is, however, hardly possible using real-world survey data because concrete emotions often merge, especially when it comes to the issue of illegal immigration. For instance, when people experience anger in response to immigration, they are also likely to experience fear. Laboratory studies could manipulate specific forms of cognitive involvement as well as single emotions. This would also allow us to separate an affective priming mechanism―that is built upon the notion of positive and negative affect―from other potential affect-based theoretical explanations. Furthermore, experiments would enable us to extend our theoretical model to arrive at more fine- grained predictions about the role of specific emotions or specific relations between emotions and cognitive variables. On the methodological side, future research should use more items for all constructs. The measures for affect should be more specific, and there is a need to include several items per single emotion. In addition, there is a need to

1094 Communication Research 44(8)

go beyond self-reports when it comes to the measurement of affect. It would also be worthwhile to measure trust and bias with respect to individual news outlets. Yet, these are only some of the challenges with which researchers who follow the lead of our cognitive-affective approach to HMEs will have to contend.

Appendix

Trust in News Media

Do you think that . . .

the frequency with which illegal immigration is covered by news media is adequate? all the important information regarding the topic of illegal immigration is provided in the news? the news coverage about illegal immigration includes different points of view?

Bias Perception

Do you think that the portrayal of immigration in the news is biased against illegal immigrants, strictly neutral, or biased in favor of illegal immigrants? (1 and 7 recoded to 3 = strong bias; 2 and 6 recoded as 2 = moderate bias; 3 and 5 recoded as 1 = weak bias, and 4 recoded as 0 = no bias).

Affective Involvement

I experience . . . . . . ”fear,” “anger,” “anxiety,” “hope,” “contentment” (1 = not at all to 7 = very much).

Cognitive Involvement

How important are issues concerning immigration to you? How important is the issue of illegal or unauthorized immigration to you?

Interest

Please indicate on a scale from 1 to 7 how interested you are in politics?

Inefficacy

How often does politics seem so complicated that you can’t really understand what is going on? Please indicate your answer on a scale from 1 (never) to 7 (frequently).

Matthes and Beyer 1095

Ideology

Here is a 7-point scale on which the political views that people might hold are arranged from extremely liberal to extremely conservative. Where would you place yourself on this scale?

Issue Opinion

Please indicate on a scale from 1 to 7 where you would place yourself with regard to this issue; 1 indicates that immigration should not be limited, while 7 indicates that immigration should be limited as much as possible.

Declaration of Conflicting Interests

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

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The research conducted in this article was supported by a grant from the Norwegian Research Council (Project “Mediation of Migration: Media Impacts on Norwegian Immigration Policy, Public Administration and Public Opinion”).

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.

Adolphs, R., & Damasio, A. R. (2001). The interaction of affect and cognition: A neurobiologi- cal perspective. In J. P. Forgas (Ed.), Handbook of affect and social cognition (pp. 27-49). Mahwah, NJ: Lawrence Erlbaum.

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

Baumgartner, S. E., & Wirth, W. (2012). Affective priming during the processing of news arti- cles. Media Psychology, 15, 1-18.

Beyer, A., & Matthes, J. (2015). Public perceptions of the media coverage of irregular immi- gration. Comparative insights from France, the United States and Norway. American Behavioral Scientist, 59, 839-857.

Bower, G. H., & Forgas, J. P. (2001). Mood and social memory. In J. P. Forgas (Ed.), Handbook of affect and social cognition (pp. 95-120). Mahwah, NJ: Lawrence Erlbaum.

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

Chang, C. (2014). The influence of ambivalence toward a communication source: Media context priming and persuasion polarization. Communication Research, 41, 783-808. doi:10.1177/0093650213495408

1096 Communication Research 44(8)

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

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.

Choi, J., Yang, M., & 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.

Davidov, E., Schmidt, P., & Billiet, J. (Eds.). (2011). Cross-cultural analysis: Methods and applications. New York, NY: Routledge.

Donsbach, W., Rentsch, M., & Mende, A. (2009, May). 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.

Dunn, J., & Schweitzer, M. E. (2005). Feeling and believing: The influence of emotion on trust. Journal of Personality and Social Psychology, 88, 736-748.

Edwards, K., & von Hippel, W. (1995). Hearts and minds: The priority of affective and cogni- tive factors in person perception. Personality and Social Psychology Bulletin, 21, 996-1011.

Ellsworth, P. C. (2013). Appraisal theory: Old and new questions. Emotion Review, 5, 125-131. 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.

Erisen, C., Lodge, M., & Taber, C. S. (2013). Affective contagion in effortful political thinking. Political Psychology, 35, 187-206.

Forgas, J. P. (1995). Mood and judgment: The affect infusion model (AIM). Psychological Bulletin, 117, 39-66.

Forgas, J. P., & Bower, G. H. (1987). Mood effects on person perception judgments. Journal of Personality and Social Psychology, 53, 53-60.

Giner-Sorolla, R., & Chaiken, S. (1994). The causes of hostile media judgments. Journal of Experimental Social Psychology, 30, 165-180.

Gronke, P., & Cook, T. E. (2007). Disdaining the media: The American public’s changing atti- tudes toward the news. Political Communication, 24, 259-281.

Gunther, A. C., & Chia, S. (2001). Predicting pluralistic ignorance: The hostile media percep- tion and its consequences. Journalism & Mass Communication Quarterly, 78, 688-701.

Gunther, A. C., & Christen, C. (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.

Gunther, A. C., Christen, C., Liebhart, J., & Chia, S. (2001). Congenial public, contrary press, and biased estimates of the climate of opinion. Public Opinion Quarterly, 65, 295-320.

Gunther, A. C., & Liebhart, J. L. (2006). Broad reach or biased source? Decomposing the hostile media effect. Journal of Communication, 56, 449-466.

Gunther, A. C., Miller, N., & Liebhart, J. L. (2009). Assimilation and contrast in a test of the hostile media effect. Communication Research, 36, 747-764.

Hartmann, T., & Tanis, M. (2013). Examining the hostile media effect as an intergroup phe- nomenon: The role of ingroup identification and status. Journal of Communication, 63, 535-555.

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

Matthes and Beyer 1097

Hwang, H., Pan, Z., & Sun, Y. (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.

Kohring, M., & Matthes, J. (2007). Trust in news media: Development and validation of a mul- tidimensional scale. Communication Research, 34, 231-252.

Kühne, R., Schemer, C., Matthes, J., & Wirth, W. (2011). Affective priming in political cam- paigns. How campaign-induced emotions prime political opinions. International Journal of Public Opinion Research, 23, 485-507.

Ladd, J. M., & Lenz, G. S. (2008). Reassessing the role of anxiety in vote choice. Political Psychology, 29, 275-296.

Lazarus, R. S. (1991). Emotion and adaptation. New York, NY: Oxford University Press. LeDoux, J. E. (1999). The emotional brain: The mysterious underpinnings of emotional life.

London, England: Phoenix. LeDoux, J. E. (2012). A neuroscientist’s perspective on debates about the nature of emotion.

Emotion Review, 4, 375-379. Lee, W., Detenber, B. H., Willnat, L., Aday, S., & Graf, J. (2004). Public outspokenness in two

cultures: A test of the spiral of silence theory in Singapore and the US. Asian Journal of Communication, 14, 205-226.

Matthes, J. (2013). The affective underpinnings of hostile media perceptions: Exploring the dis- tinct effects of affective and cognitive involvement. Communication Research, 40, 360-387.

McDermott, R. (2004). The feeling of rationality: The meaning of neuroscientific advances for political science. Perspectives on Politics, 2, 691-704.

Medina, T. R., Smith, S. N., & Long, J. S. (2009). Measurement models matter: Implicit assump- tions and cross-national research. International Journal of Public Opinion Research, 21, 333-361.

Mullen, E. (2007). The reciprocal relationship between affect and perceptions of fairness. In K. Tornblom & R. Vermunt (Eds.), Distributive and procedural justice: Research and social applications (pp. 15-37). Burlington, VT: Ashgate.

Mulligan, K., & Scherer, K. R. (2012). Toward a working definition of emotion. Emotion Review, 4, 345-357.

Nadeau, R., Niemi, R. G., & Amato, T. (1995). Emotions, issue importance, and political learn- ing. American Journal of Political Science, 39, 558-574.

Ottati, V. C., Steenbergen, M. R., & Riggle, E. (1992). The cognitive and affective compo- nents of political attitudes: Measuring the determinants of candidate evaluations. Political Behavior, 14, 423-442.

Perse, E. M. (1990). Involvement with local television news: Cognitive and emotional dimen- sions. Human Communication Research, 16, 556-581.

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

Schmitt, K. M., Gunther, A. C., & Liebhart, J. L. (2004). Why partisans see mass media as biased. Communication Research, 31, 623-641.

Slater, M. D. (2007). Reinforcing spirals: The mutual influence of media selectivity and media effects and their impact on individual behavior and social identity. Communication Theory, 17, 281-303.

Sniderman, P. M., Brody, R. A., & Tetlock, P. E. (1991). Reasoning and choice: Explorations in political psychology. Cambridge, UK: Cambridge University Press.

Steenkamp, J.-B. E. M., & Baumgartner, H. (1998). Assessing measurement invariance in cross-national consumer research. Journal of Consumer Research, 25, 78-90.

1098 Communication Research 44(8)

Thorbjørnsrud, K. (2015). Framing irregular immigration in Western media. American Behavioral Scientist, 59, 771-782.

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

Tsfati, Y., & Ariely, G. (2014) Individual and contextual correlates of trust in media across 44 countries. Communication Research, 41(6), 760-782.

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

Vallone, R. P., Ross, L., & 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.

Vandenberg, R. J., & Lance, C. E. (2000). A review and synthesis of the measurement invari- ance literature: Suggestions, practices, and recommendations for organizational research. Organizational Research Methods, 3, 4-70.

Watson, D., & Clark, L. A. (1997). Measurement and mismeasurement of mood: Recurrent and emergent issues. Journal of Personality Assessment, 68, 267-296.

Way, B. M., & Masters, R. D. (1996). Emotion and cognition in political information-process- ing. Journal of Communication, 46, 48-65.

Wirth, W. (2006). Involvement. In J. Bryant & P. Vorderer (Eds.), Psychology of entertainment (pp. 199-213). Mahwah, NJ: Lawrence Erlbaum.

Wirth, W., Schemer, C., & Matthes, J. (2010). Trivializing the news? Affective context effects of commercials on the perception of television news. Mass Communication and Society, 32, 139-156.

Zajonc, R. B. (1984). On the primacy of affect. American Psychologist, 39, 117-123.

Author Biographies

Jörg Matthes (PhD, University of Zurich) is a professor at the Department of Communication, University of Vienna. His research focuses on public opinion formation, media effects, advertis- ing research, and empirical methods.

Audun Beyer (PhD, University of Oslo) is an associate professor at the Department of Media and Communication, University of Oslo. His research focuses on public opinion formation, media and migration, and empirical methods.