That's Not the Way It Is: How User-Generated Comments on the News Affect Perceived Media Bias





This study investigated if user-generated comments on Internet news sites affect readers' inferences about public opinion, and subsequently, their perceptions of media bias, and how ego-involvement moderates such effects. Supporting the notion that hostile media perception (HMP) stems from defensive processing, those who read others' comments discordant (vs. concordant) with their own opinion believed that the public was against their position and perceived the news report to be more hostile and partial in its coverage, but such effects were limited to those with higher ego-involvement. Readers' comments also had a direct effect on HMP among more involved individuals, without altering their perceptions of public opinion, suggesting that people might misattribute the opinions expressed in others' comments to the news article.

Various Internet-based news services have caused a wide array of changes in the way news is produced, disseminated, and consumed, but what seems to distinguish online journalism most clearly from its broadcast or print counterparts is the ability for news consumers to contribute comments. In South Korea, for example, readers' comments have become a signature characteristic of online news portals, although specific features slightly vary across news web sites. Not only are readers' comments widely available on Internet news sites, but they are also frequently read: 84.3% of Internet news users reported reading others' comments at least once a week (Na & Rhee, 2008). Online newspapers even rank the “most commented” articles and present them along with the “most viewed” articles (e.g.,, treating the aggregation of responses as an indicator of the importance of the news article.

Thus far, researchers have found that user-generated comments significantly alter not only other readers' perceptions of what public opinion on a specific issue is, but their own personal opinion about the news issue, independently or in conjunction with other factors (Kim & Sun, 2006; Lee & Jang, 2010). These studies tend to treat news articles and users' comments as competing sources of influence on individuals' perceptions of reality. Yet notions that one's judgments are susceptible to social influence prompted by cues in the context within which an object or issue appears (Sherif & Hovland, 1961; Petty & Wegener, 1993) suggest that user-generated comments about a news topic that are juxtaposed to the news article they address might affect readers' very perception of the news itself. Rather than assume that news stories and user-generated comments operate independently on readers' perceptions, the present experiment focused on how user-generated comments on Internet news sites affect responses to the accompanying news story in terms of readers' perceptions of the bias in the news story, specifically, in terms of readers' hostile media perception (HMP), which refers to the tendency to perceive news coverage as biased against one's own view (e.g., Gunther, Christen, Liebhart, & Chia, 2001; Gunther & Schmitt, 2004; Vallone, Ross, & Lepper, 1985). By examining how individuals' perceptions of media bias vary depending on other users' comments and the congruent or incongruent relation of those comments to the readers' own opinion, the current research aimed to extend the relevant literature in the following respects.

First, by incorporating readers' perceptions of public opinion as a potential mediator of the effects of user-generated comments on HMP, this study provided a critical test of the two competing explanations for HMP. On the one hand, if readers assess media bias based on how discrepant the news report is from what they believe others think (Eveland & Shah, 2003), then exposure to other users' comments that are concordant with their personal viewpoints might convince them that the public is on their side and that the news story does not accurately represent real public sentiment, thereby amplifying readers' perceptions that the media are biased. On the other hand, if HMP occurs because people, especially partisans, presume that the news media will exert significant influence on a broad audience, and their concern about undesirable media influence on those others prompts partisans to view the news in a defensive manner (Gunther & Liebhart, 2006; Gunther, Miller, & Liebhart, 2009), then encountering other users' comments that support their own position should alleviate this concern, and thus, attenuate HMP. By systematically varying the valence of other users' comments and examining their influence on individuals' inferences about public opinion and their perception of the news, this study attempted to elucidate the psychological mechanism underlying HMP.

Second, the present research explored the possibility that user-generated comments might exert a direct impact on individuals' perception about a news story because readers might find it difficult to sort out which information came from what source, and may mistakenly attribute the opinions expressed in users' comments to the news article. Although they did not directly examine this possibility, Kim and Sun's (2006) finding that individuals' evaluations of news quality were affected by the quality of user-generated comments associated with the news suggests an assimilation bias, which demands further investigation.

Third, the current research addressed how an individual's ego-involvement, which has mostly been considered as a necessary condition for, or a factor that facilitates HMP (Choi, Yang, & Jang, 2009; Gunther et al., 2001), might moderate the effects of user-generated comments. For example, the aforementioned confusion over the information source and ensuing misattribution might depend more or less on the extent to which news readers are motivated to process the news, and information associated with the news, systematically (Chaiken & Maheswaran, 1994; Chen & Chaiken, 1999). Likewise, the tendency to infer public opinion from several unknown users' reactions, and/or the susceptibility to their influence, might also hinge on how involved readers are in the news object. By incorporating readers' ego-involvement toward the news topic as a potential moderator, this study therefore explored the boundary conditions for the effects of user-generated comments on readers' perceptions of public opinion and media bias in a participatory web environment.

Direct and Indirect Effects of Readers' Comments on Perceived Media Bias

Although comment-writing is still concentrated among the most opinionated news consumers, Lee and Jang (2010) found that readers nonetheless infer public opinion on the basis of several unknown users' posts. Specifically, when exposed to user-generated comments opposing the position that was advocated in a news article, readers inferred that public sentiment was more discrepant from the news article's position than did those who read only the article. When other users' reactions were presented in the aggregated form of approval versus disapproval ratings, however, the readers' perceptions of public opinion were not altered. Others' comments also exerted significant influence on the readers' personal opinion whereas the approval ratings did not, resonating well with the relative advantage of exemplars over statistical summary in persuasion (Zillmann, 2002), but such effects were confined to those with lower need for cognition.

Unlike Lee and Jang's (2010) study, which compared the relative influence of the news article and users' comments on readers' perceptions of public opinion as well as on their personal opinions, the present study explored how user-generated comments might bias readers' interpretations of the news itself. In so doing, this study focused on hostile media perception, which refers to the well-documented finding that individuals who are strongly involved in a controversial issue tend to see media coverage of that issue as being biased against their own position or in favor of their opponent's (Gunther et al., 2009; Schmitt, Gunther, & Liebhart, 2004; Vallone et al., 1985). Measuring the effect of user-generated comments that accompany news items on news readers' inferences about the nature of public opinion can help adjudicate between two competing theoretical explanations for HMP, since these approaches—the distorted standards account vs. the defensive processing account—make opposite predictions about how user-generated comments affect HMP.

First, the distorted standards account proposes that people judge media bias in terms of how discrepant media coverage is from what they believe that public opinion is (Eveland & Shah, 2003). Because people do not normally have an objective standard against which to evaluate media bias, they tend to infer what public sentiment is from what they hear in their discussions with other people they know. More often than not, however, this “interpersonally generated reality” yields “distorted standards” (p. 113) by which people assess the extent of media bias, since their interpersonal network typically consists of like-minded others, leading them to overestimate the amount of public support for their position. Consequently, the more often people converse with ideologically similar others (i.e., have “safe discussion”), the more they judged the news media to be inconsistent with the public's opinion, and thus, biased (Eveland & Shah, 2003).

In contrast, the defensive processing account posits that HMP stems from partisans' concern about the negative influence media exerts on other people. To test this notion, Gunther and Schmitt (2004) presented two different news media outlets to readers, with one medium having the potential to reach a systematically greater number of people than the other (a newspaper article vs. a student essay). They found that the hostile bias occurred only when considering the newspaper article. Based on these findings (see also Gunther & Liebhart, 2006), they argued that “the mass media context might stimulate partisan concerns about the potentially misleading influence of information on a broad audience” (Gunther et al., 2009, p. 749), which puts partisans in a defensive mode and leads them to see objectively neutral or balanced content as though it is unfavorable and misleading. In this view, it is the anticipation that media will change the public's opinion away from one's own position that triggers biased processing of a news story.

Although these two accounts hypothesize different cognitive processes leading to HMP, both approaches assume that people cannot accurately assess the extent to which the public supports their own position. However, if user-generated comments on Internet news sites appear to readers to be a reflection of public opinion (Lee & Jang, 2010), the comments might either amplify or attenuate HMP. If readers evaluate media bias in terms of how discrepant a news report is from what they see as being the public's sentiment (Eveland & Shah, 2003), exposure to others' comments supporting their own view will amplify HMP. On the other hand, if HMP is the outcome of anticipated negative media influences on a generalized audience (Gunther et al., 2009), exposure to comments that support the readers' position should alleviate the readers' concern about undesirable media influence, and hence, attenuate HMP.

In fact, a recent study (Lee, 2011) examined how readers' exposure to a news story and other users' comments about animal testing altered reader's perceptions of public opinion and their HMP, compared to readers who read the same news story but no user-generated comments. When the comments led readers to perceive that public opinion opposed their own personal opinion on animal testing, they exhibited stronger HMP than when readers sensed that public opinion favored their own position. These results lend tentative support to the defensive processing account. However, several aspects of that study raise questions about the validity and generalizability of the results, including (a) the participants' opinions were measured after exposure to the users' comments, inviting potential contamination, (b) only the internal relationship between perceived public opinion and HMP was tested, without attempting to establish the hypothetical link between others' comments and readers' perceptions of news bias, and (c) a single set of comments was used focusing an issue of little social relevance. To address these limitations and reduce theoretical ambiguity, the current research (a) randomly assigned the participants to read comments that were either congruent or incongruent with their pre-existing opinions about a controversial public issue, and (b) validated the mediation model in its entirety connecting user-generated comments and HMP as a result of comments' effects on perception of public opinion.

H1: Readers who view other users' comments that are concordant with their own opinion perceive public opinion to be more congenial than readers who view discordant comments.

The distorted standards account suggests,

H2a: Perceived congeniality of public opinion is positively associated with HMP.

In contrast, the defensive processing account suggests,

H2b: Perceived congeniality of public opinion is negatively associated with HMP.

In addition, user-generated comments might also have a direct effect on HMP. That is, those who read user-generated comments that are incongruent with their own opinions might display stronger HMP, not because such comments affect their perception of public opinion and put them in a defensive mode (H2b), but because they simply fail to distinguish what they read in the comments from what was said in the news article, i.e., misattribution occurs. In related research, Kim and Sun (2006) found that readers who saw poor-quality comments (operationalized by vulgar language and the lack of sound reasoning) accompanying a news article rated the quality of the news article itself more negatively, compared to readers who saw the identical article accompanied by good-quality comments (presenting sensible, relevant arguments in a more civilized manner). In the aforementioned study, Lee (2011) also found that those who read others' comments opposing animal testing believed that the news report the comments accompanied was more biased in the same direction as the comments than did those who read only the news article. Still, it remains unclear (a) to what extent such effect stemmed from readers' mere confusion about whether the commenters or the media were the source of the opinion (a direct effect), rather than defensive processing (an indirect effect), and (b) how the effect might vary depending on the congruity of users' comments with the reader's own position. Thus, the following research question was proposed.

RQ1: Do user-generated comments have a direct impact on HMP, not mediated by readers' perceptions of public opinion?

The effects of user-generated comments, however, might be either more or less pronounced depending on an individual's ego-involvement in the topic of the news story. One possibility is that very involved individuals might show greater interest in and pay closer attention to the issue-relevant information, by virtue of its pertinence to their personal values, beliefs, and identity (Cho & Boster, 2005). Therefore, they might process other readers' comments more carefully when they concerned an issue about which they deeply care, and utilize the content in their subsequent judgments about public opinion (H1). As such, potential confusion over the source of information (H3) might be less characteristic of more involved individuals who engage in greater systematic message processing (Chen & Chaiken, 1999). An alternative possibility is that individuals who are very involved in a topic already possess relatively stable beliefs about public sentiment regarding that topic, and thus, their perceptions of public opinion might be less susceptible to minimal external cues, such as the comments generated by unknown other users.

Second, ego-involvement in a topic might affect the way perceptions of public opinion influence HMP. On the one hand, defensive processing might be more pronounced among highly involved individuals who take the issue more seriously than do those who care less about it. Since readers who are relatively uninvolved in an issue do not have much to be defensive about in the first place, even if they come to perceive unfavorable public sentiment, they might not be alarmed enough that they would see neutral media content as being biased against them. On the other hand, congenial public opinion inferred from other users' comments might not sufficiently alleviate strongly involved readers' deep-rooted concern about impending negative media influence upon the public, thereby failing to attenuate HMP. With such diverse theoretical possibilities concerning the specific roles ego-involvement might play in the hypothesized process, the following research question was proposed.

RQ2: How does an individual's ego-involvement in a news topic moderate the relationships among user-generated comments, perceived public opinion, and HMP?



An online survey company in South Korea recruited a total of 240 participants to take part in a web-based experiment. Because it is imperative to define HMP in reference to an individual's existing opinion on a topic, individuals who had no opinion or held a neutral position on the study's focal issue (N = 26, 10.8%) were excluded, as was done in previous studies (e.g., Choi et al., 2009; Gunther & Liebhart, 2006). Data from 214 participants (113 men, 101 women; age M = 39.42, SD = 11.01) were retained for analyses.


To enhance ecological validity, participants were instructed to access the study web site at their convenience from a location where they felt most comfortable. After indicating consent to participate, they completed a pretest questionnaire that measured their ego-involvement, position on the focal issue, and the frequency of interpersonal communication with those holding each position on the issue (pro vs. con). They then viewed a news article as well as users' comments that were either congruent or incongruent with their own position on the topic. Both the news article and the comments were created by modifying previously published news stories and associated comments on news portals. To make the news article as neutral and balanced as possible, care was taken to include an equal number of arguments on each side of the issue.

The news topic concerned the ban on corporal punishment in grade schools initiated by the newly elected liberal educational superintendent of Seoul, South Korea. The ordinance prohibits teachers from inflicting physical punishment on students for their misbehavior and restricts teachers to other disciplinary measures, such as suspension, community service, and parent-teacher counseling. The news story indicated that proponents of the ban welcomed the ordinance as enhancing students' human rights, and that they claim that teachers often abuse corporal punishment simply out of anger or frustration, which tends to provoke resentment and further misbehavior from students. The news story also presented the arguments of the opponents of the ban that the prohibition, however moral it may appear, is out of touch with reality and that with teachers' authority continuously and substantially declining, corporal punishment could be the last resort to maintain control in the classroom.

For user-generated comments, a total of eight postings appeared immediately below the news article. Two of the comments were more prominent than the others, appearing in a colored box signifying that they had garnered the most reactions (agree vs. disagree) from other site visitors. To make them look more realistic, one minority opinion was included, with other comments unanimously endorsing the same position, either supporting (word count M = 40.29, SD = 13.65) or opposing (word count M = 40.43, SD = 13.58) the ban on corporal punishment. Examples of proban comments are: “You keep talking about educational authority, but educational authority does not come from physical punishment. Authority can only be established when teachers live respectable lives consistent with their teaching. Honestly, I've never seen a teacher using corporal punishment who deserved any respect.” “This policy is totally awesome…The superintendent is doing an amazing job. Gotta say, it's absolutely the way to go and will make life worth living for students.” Examples of antiban comments are: “You keep talking about human rights, but you should also consider educational authority. Proper education can be guaranteed only when teachers are given the respect they deserve. There are a number of students these days who cannot be controlled without using corporal punishment,” “This policy is totally absurd…The superintendent has no idea what he's doing. Gotta say, he's lost touch with reality, and life is gonna be a living hell for teachers.” Once the participants finished reading the news article and users' comments, they completed a questionnaire indicating their perceptions of the news story's position, public opinion, and the comments they had read.

Pretest Measures

To indicate their pre-existing opinion on the focal issue, participants signaled how much they agreed with the following statement (−5 = Strongly disagree, 5 = Strongly agree): “Corporal punishment in grade schools should be banned” (M = −1.13, SD = 3.15). The participants were then asked how strongly they agreed with each of the following statements, measuring ego-involvement (Cho & Boster, 2005): “My position on the ban on corporal punishment has a lot to do with my beliefs about how life should be lived,” “My position on (the issue) reflects who I am,” “The values that are most important to me are what determine my stand on (the issue)” (1 = Strongly disagree, 7 = Strongly agree). Scores were averaged to create the ego-involvement index (α = .94, M = 4.24, SD = 1.52).

To control for the influence of one's interpersonal network, participants indicated how often they talked with others about supporting or opposing the ban on corporal punishment (1 = Rarely, 7 = Frequently). Responses were recoded in reference to the participant's own position, such that the frequency of conversations with people who agree with the participant constituted a measure of safe discussion (M = 3.60, SD = 1.74) and frequency of conversations with people who disagreed with a participant comprised the dangerous discussion variable (M = 3.26, SD = 1.66) (Eveland & Hively, 2009). So if a participant favored the ban, the frequency of conversations with others who support the ban was his or her safe discussion score, whereas conversations with people who oppose the ban constituted the dangerous discussion score.

Posttest Measures

Scores for HMP were computed as follows. After reading the news article and the users' comments that were either congruent or incongruent with their own opinions, participants were prompted to indicate to what extent they thought the news article agreed with the statement, “Corporal punishment in grade schools should be banned” (−5 = Strongly disagree, 5 = Strongly agree; M = 0.10, SD = 1.84). Retaining the absolute value of the original score, different signs were assigned depending on which side of the scale the perceived news position fell. That is, if the participant perceived the news report to be on the opposite side of his or her own position, the plus sign was given (indicating hostile media perception), whereas the minus sign was assigned when the participant saw the news report as supporting his or her own viewpoint (i.e., friendly media perception) (M = −0.22, SD = 1.83). In addition, as a supplementary measure of HMP, participants were also asked to assess media bias explicitly by indicating how impartial they thought the news report was in dealing with the issue (1 = Not at all impartial, 11 = Very impartial; M = 5.82, SD = 1.64).

For readers' perception of public opinion, participants indicated the extent to which they thought the general public would agree with the statement, “Corporal punishment in grade schools should be banned” (−5 = Strongly disagree, +5 = Strongly agree; M = −0.08, SD = 2.41). Congeniality of public opinion was defined as the degree to which perceived public opinion deviates from the neutral position in either a congenial (+) or incongenial (−) direction, relative to the participant's own opinion. For example, if one strongly disapproved the ban (−5) but perceived the general public to be moderately in favor of it (+3), her perceived congeniality of public opinion score was −3 (M = 0.87, SD = 2.25).


Manipulation Checks

To assess the objective neutrality of the news article, and because user-generated comments can influence individuals' perceptions of news bias, a separate group of 38 individuals (13 men, 25 women: age M = 36.89, SD = 10. 39) who were identified as neither supporting nor opposing the ban, read the news article without any accompanying comments. They indicated how much they thought the news article favored the ban on corporal punishment in grade schools (−5 = Strongly against, +5 = Strongly in favor of) (M = 0.16, SD = 1.00). A one-sample t-test showed that the perceived tone of the news article was not significantly different from the scale midpoint (0.00), t (37) = 0.97, p = .34, attesting to the neutrality of the news article.

To ensure that the valence of the user-generated comments was recognized, participants in the main experiment indicated how much the comments agreed with the statement, “Corporal punishment in grade schools should be banned” (−5 = Strongly disagree, +5 = Strongly agree). An independent-samples t-test showed that the comments that were designed to support the ban (M = 1.63, SD = 2.48) were perceived to be significantly more favorable toward the ban than those designed to oppose it (M = −1.62, SD = 2.17), t (212) = −10.01, p < .001.

Hypothesis Tests

H1 predicted that people would infer the general public sentiment from the opinions expressed in user-generated comments. Because it is possible that the participants' pre-existing opinion might interact with the valence of user-generated comments, a 2 (congruent vs. incongruent comments) × 2 (pretest antiban vs. proban opinion) ANOVA was computed on perceived congeniality of public opinion. Results showed that those exposed to other readers' comments congruent with their own opinion believed that public opinion was significantly more congenial (M = 1.52, SD = 1.94) than those presented with the comments countering their stance (M = 0.24, SD = 2.35), F (1, 210) = 17.36, p < .001, ηp2 = .08. No significant interaction was found between comment valence and pre-existing opinion, F < 1. Thus, H1 was supported.

To determine whether perceived congeniality of the public opinion amplifies (H2a) or attenuates HMP (H2b), and if user-generated comments have a direct effect on HMP over and above the indirect effect through perceived public opinion (RQ1), simple mediation tests were conducted (Preacher & Hayes, 2008). First, the indirect effect of comments' congruity with a reader's existing opinion on HMP through readers' assessment of public opinion was statistically significant, b = −0.23, Z = −2.61, p = .009 (95% bias corrected 5000 bootstrap confidence interval: -.486 to -.066). Specifically, exposure to other readers' comments concordant with one's own position led to perceptions of more congenial public opinion, b = 1.29, t = 4.35, p < .001, and the more congenial participants thought that public opinion was, the less hostile their perception of the news story was, b = −0.18, t = −3.23, p = .001. Therefore, H2b was supported. Other readers' comments also had a significant direct effect on HMP, suggesting the possibility of misattribution, b = −0.60, t = −2.40, p = .02. When the same analysis was conducted for perceived impartiality of the news, however, neither the direct effect, b = .25, t = 1.07, p = .29, nor indirect effect was statistically significant, b = 0.08, Z = 1.16, p = .25.

To elucidate how ego-involvement moderates the direct and indirect effects of user-generated comments on perception of the news (RQ2), a moderated mediation test was conducted using the MODMED macro (Preacher, Rucker, & Hayes, 2007). Because safe discussion was previously found to have a significant effect on HMP (Eveland & Shah, 2003), safe and dangerous discussions were included as covariates in the analysis. With no significant interaction between ego-involvement and users' comments on perception of public opinion, b = −0.14, t = −0.69, p = .49, ego-involvement was considered as a potential moderator that affected only the relationship between perceptions of congenial public opinion (the mediator) and HMP (the dependent variable). Results showed a significant interaction between congenial public opinion perception and ego-involvement on HMP, b = −0.06, t = −1.99, p = .048. Specifically, the conditional indirect effect of users' comments on HMP through congenial public opinion perception was statistically significant for more involved individuals (+1 SD from the mean), b = −0.35, Z = −2.95, p = .003, but not for less involved individuals (−1 SD from the mean), b = −0.12, Z = −1.18, p = .24. That is, reading others' comments supporting their own position led the participants to perceive greater public support, which in turn, made them perceive less hostility in the news report, but such effects emerged only for more involved individuals. Meanwhile, safe discussion amplified HMP, b = 0.31, t = 3.17, p = .002, while dangerous discussion attenuated it, b = −0.23, t = −2.38, p = .02.

To better understand the interaction, participants' scores were dichotomized using a median split (n = 105, M = 3.01, SD = 1.02 for the lesser-involvement group; n = 109, M = 5.45, SD = 0.86 for the greater-involvement group) and simple mediation tests were performed for each group, separately. For the less involved group, although the exposure to other readers' comments concordant (vs. discordant) with one's own position induced more congenial perceptions of public opinion, b = 1.10, t = 2.83, p = .006, perceived congeniality of public opinion did not significantly affect HMP, b = −0.13, t = −1.39, p = .17. Not only did others' comments have no significant indirect effects on HMP, b = -0.14, Z = -1.26, p = .21, but they exerted no significant direct effect, b = −0.46, t = −1.23, p = .22. For the more ego-involved group, however, user-generated comments congruent with one's viewpoint induced the perceptions of more favorable public opinion, which in turn, lowered perceived hostility in the news report, b = −0.32, Z = −2.33, p = .02 (see Figure 1). In addition, others' comments had a significant direct effect on HMP, not mediated by perceptions of public opinion.

Figure 1.

Direct and Indirect Effects of Readers' Comments on HMP and Perceived News Impartiality: High-Involvement Group

A moderated mediation test on perceived impartiality of the news story also yielded a significant interaction between ego-involvement and congenial perceptions of public opinion, b = 0.10, t = 3.61, p < .001. The conditional indirect effect of readers' comments on perceived news impartiality was significant for more involved individuals (+1 SD from the mean), b = 0.26, Z = 2.56, p = .01, but not for less involved individuals (−1 SD from the mean), b = −0.14, Z = −1.38, p = .17. To decompose the interaction, simple mediation tests were conducted for greater- and lesser-involvement groups. For the low-involvement group, as was the case for HMP, readers' comments had no significant direct, b = 0.14, t = 0.46, p = .65, or indirect effect, b = −0.12, Z = −1.35, p = .18. By contrast, for the high-involvement group, exposure to congruent comments elicited ratings of greater impartiality of the news, through more congenial public opinion perception, b = 0.27, Z = 2.05, p = .04. Put differently, all readers got a sense of what public opinion was on the basis of what the user-generated comments said, but when readers who considered the news topic to be more important to them inferred from the comments that public opinion was on their side, they saw significantly less bias in the news story. Unlike for HMP, however, there was no significant direct effect of readers' comments on perceived news impartiality. Taken together, the results indicated that the defensive processing mechanism (H2b) and misattribution (RQ1) noted earlier occurred mostly among those with greater ego-involvement in the news topic.


The present study investigated how user-generated comments, presented simultaneously with a news article on an Internet news portal, shape readers' perceptions of public opinion as well as their evaluations of news coverage. In so doing, two competing hypotheses were examined to elucidate the effects of certain Web 2.0 messages on the psychological processes underlying HMP. Despite their limited value as a nonrepresentative sample, the presence of several unknown readers' comments serve readers as a proxy for public opinion; user-generated comments accompanying news stories significantly altered the participants' beliefs about what other members of the society think. Moreover, such effects were found for participants who were both more and less ego-involved in the news topic, underscoring the power of exemplars in social judgments (Zillmann, 2002). By changing the participants' beliefs about the amount of public support, users' comments exerted indirect effects on their evaluation of the news, as those who perceived that public opinion was like their own opinion attributed less hostility and partiality to the news story, albeit only among readers with greater ego-involvement. User-generated comments also had a significant direct effect on HMP among more involved individuals, suggesting that readers experience confusion over whether the news story or its user commenters lent a position to the readers' inferences.

Theoretical Implications

While investigating how readers' comments on Internet news sites shape individuals' evaluations of the news article, this study assessed critically two competing explanations for HMP. According to the defensive processing account, those with greater ego-involvement in an issue (i.e., partisans) tend to worry about the possible negative influence of the news on a broad audience, and thus, exaggerate the proportion of unfavorable content they see in otherwise neutral news reports. The findings that (a) perceived congeniality of public opinion, as inferred from other users' supportive comments, significantly suppressed HMP and enhanced perceived impartiality of the news story and (b) such effects were found only among more involved individuals comport well with this explanation.

Considering that users' comments were positioned below the news article, and thus, most likely viewed after the news story was read, it merits note that the process this study examined is slightly different from, albeit compatible with, the one suggested in the defensive processing account. That is, the defensive processing account argues that HMP stems from anticipated undesirable influence on the public, whereas readers' comments countering one's position represent realized unfavorable public opinion. Rather than directly examining if partisans' anticipation of negative media influence indeed facilitates HMP, the current study investigated if provision of a reality cue, either supporting or refuting their presumption about undesirable media influence on others, would prompt readers to adjust their evaluations of the news story. Specifically, it was reasoned that other readers' comments concordant with one's position might prove that partisans' concern about the public's vulnerability to undesirable media influence is unwarranted (i.e., correction) and thus attenuate HMP, whereas others' comments opposing their viewpoint might validate their concern (i.e., corroboration), thereby amplifying HMP. The results suggest that fear of losing ground, at least in part, accounts for the hostile media bias, and that user-generated comments can either ameliorate or aggravate such fear, ultimately influencing news readers' evaluations of a news story.

On the other hand, little support emerged for the rival explanation, the distorted standards account. This approach posits that people infer from conversations with like-minded others that the public favors their own position, and compare news stories against this biased anchor (Eveland & Shah, 2003). Safe discussion was positively associated with HMP in the present study as it did elsewhere (Eveland & Shah, 2003), but it had no significant effect on perceived public opinion, the supposed mediator (see Table 1). More important, the more favorable to their own position participants perceived public opinion to be, the less hostile and more impartial they perceived the news report to be, directly contradicting the distorted standards explanation.

Although ego-involvement has long been considered either a necessary condition for HMP or a factor that amplifies it (Gunther et al., 2001), another look at the current data suggests that ego-involvement might play a role only under certain conditions. Specifically, when the participants thought that the public was neutral or opposed their own position, ego-involvement did foster HMP, r (95) = .31, p = .003, but it had no significant effect on HMP among those who believed the public was on their side, r (119) = .06, p = .51; Zdiff = 1.81, p = .07. Likewise, more involved individuals attributed stronger partiality to the news report than did less involved ones, but only when they sensed either neutral or unfavorable public opinion, r (95) = -.28, p = .007. When they believed that they were in the majority, ego-involvement had little to do with their assessment of how impartial the news report was, r (119) = .08, p = .42; Zdiff = 2.56, p = .001. Taken together, these results indicate that by telling readers what public sentiment is like (regardless of its accuracy), user-generated comments may create a situation in which readers' ego-involvement may or may not facilitate biased perceptions of the news, depending on whether the public is with them or against them. In keeping with the defensive processing account, the belief that they are on the winning side seems to allow bias-prone partisans to see a news report as is.

Extending previous research that examined how ego-involvement alters the extent to which people perceive media as biased (as a main effect), this study focused on how involvement moderates the effects of user-generated comments on individuals' perceptions of popular opinion as well as their evaluations of a news story (as an interaction effect). Of particular interest is the finding that less involved individuals' perceptions of public opinion had no significant effects on their HMP or perceived impartiality of the news. Although they, too, inferred public opinion from users' comments, they did not seem to consider it relevant when judging news bias. By contrast, those more involved in the issue seemed to take many things into consideration when evaluating the news, such as (perceived) public sentiment, which ironically led to perceptions that the news was more biased, an assessment which was inaccurate.

Likewise, strongly involved individuals assimilated position expressed in the news to the positions advocated in users' postings, whereas no corresponding tendency was observed for those less involved in the issue. Given that the news article and readers' comments were authored by independent entities, participants should not have judged the news' position differently based on what was said in the users' comments. Yet such perceptual distortion occurred among those individuals who would otherwise be expected to process issue-relevant information more thoroughly. One potential explanation for this seemingly counterintuitive result is that individuals with low involvement might not have heeded others' postings carefully, and thus, were relatively insulated from their effects. However, the finding that individuals both with lesser or greater involvement inferred public opinion from other readers' comments makes this notion unlikely. Another possibility pertains to biased systematic processing. User-generated comments might have biased strongly involved individuals' processing of the news article, making the comments-consonant news content better stored and more readily retrieved when making judgments. Put differently, those attaching importance to the issue might have processed others' comments more thoroughly, during which related arguments contained in the news article were reactivated and integrated into memory, whereas their less involved counterparts went through rather cursory processing of the stimuli (Lang, 2000). Future research should examine this possibility by measuring these cognitive subprocesses.

Limitations and Future Directions

Albeit perceptually most prominent, users' comments are not the only, or even the most commonly utilized, form of user feedback on Internet news sites. For example, people can indicate whether they agree/disagree or like/dislike a news article and others' comments. Considering that only users' comments influenced individuals' perceptions of public opinion, with no effect due to aggregated approval ratings of the news article (Lee & Jang, 2010), the effects of user-generated comments observed herein might not generalize to different forms of audience feedback. Therefore, for a more comprehensive understanding of how users' participation in news publication alters readers' interpretations of and reactions to news, future research should employ various forms of user feedback and explore if there are any qualitative differences in their respective effects.

It is possible that the direct effect of other users' comments on HMP might not indicate cognitive confusion over who said what, as it simply indicates that a significant portion of the effects others' comments had on HMP was not mediated by perceived public opinion. Because this study was primarily concerned with the role of perceived public opinion in the hostile media phenomenon, no other potential mediators were considered. Future research might nevertheless benefit from considering different cognitive paths through which user-generated comments affect individuals' interpretations of the news. Directly measuring how well readers correctly identify the source of information they have just viewed—the news article or other users' comments—can also help to clarify the nature of the direct effect observed in this study.

Readers' perceptions of media bias were measured in two different ways, one assessing how unfavorable participants perceived the news story was in relation to their own viewpoint, and the other asking how impartial they thought the news story was. Although users' comments had the same indirect effects on both of these variables, they exerted a significant direct impact only on the former. While the lack of direct effect underscores the pivotal role of perceived public opinion in HMP, such divergent findings might also suggest a potentially interesting measurement issue: implicit versus explicit judgments of media bias. When prompted explicitly to assess how biased the news story was—a serious charge against any news producer—participants might have become more conscious about their judgments than when simply indicating what position they thought the news story was promoting, with no explicit reference to the term, “bias.” As such, they might have been less susceptible to an assimilation bias or misattribution in their evaluation of the news article's (im)partiality, due to greater (retroactive) elaboration over the news content. Albeit speculative, it seems worthwhile to ask if implicit and explicit measures of perceived media bias yield different outcomes, and if so, how.


With more and more Internet news sites incorporating diverse user feedback systems in the hopes of attracting more visitors and enhancing their credibility, it seems increasingly crucial to understand how people process and integrate technology-mediated messages from unknown others to make sense of their social world. Unlike offline discussions in which biased sampling of public opinion often takes place through safe discussion, the valence of users' comments one encounters browsing the web is rather random and beyond one's control, accosting readers with a sometimes surprising view of popular opinion, yet rendering Internet news sites to researchers as a naturally occurring and ecologically valid experimental setting. Also, it might seem more reasonable to extrapolate public opinion from impersonal others' comments than from one's conversations with friends and family, whose opinions are likely to be similar to one's own. At the same time, the juxtaposition of a news story and users' comments might make individuals' news perceptions more amenable to the assimilation bias, which is less likely in a water cooler discussion about last night's news report. More than just an upgraded version of letters to the editor, which creates an illusion of participatory journalism, user-generated comments embody a theoretically meaningful variable in explaining our news experience, inviting communication researchers to revisit some of the long-held assumptions about various communication processes and psychological effects, like HMP.

Table 1. Bivariate Correlations Among Key Variables
  1. Note.
  2. ap < .05
  3. bp < .01
  4. cp < .001.
1. Ego-Involvement.47b.37b.05.16−.09
2. Safe Discussion .70b−.01.21a−.09
3. Dangerous Discussion  −.05.06−.01
4. Congenial Public Opinion Perception   −.27b.11
5. HMP    −.38b
6. Perceived News Impartiality     


  • Eun-Ju Lee, Ph.D. ( is Associate Professor in Communication at Seoul National University. Her research focuses on social cognition and social influence in computer-mediated communication, which has appeared in Human Communication Research, Communication Research, and the Journal of Communication, among others. She has served as associate editor of Human Communication Research.

    Address: Gwanak-Ro 1, Gwanak-Gu, Seoul 151-742, Republic of Korea.