Preelection Selective Exposure

Abstract

The glut of media coverage prior to a presidential election requires individuals to selectively expose themselves to some messages and not others. The study involves a two-session online quasi-experiment with 205 participants that was conducted before the 2008 presidential election. Hypotheses on confirmation bias and information utility driving selective exposure prior to an election are tested. Results confirm that information utility can override a confirmation bias and motivate exposure if a government change is likely and the favored party is likely to lose the election. Moreover, participants with frequent habitual online news use do not exhibit a confirmation bias. However, participants whose favored party was likely to win the election and participants with infrequent online news consumption show a significant confirmation bias.

Keywords

selective exposure election confirmation bias informational utility online

It has long been argued that people selectively expose themselves to messages that are in line with preexisting attitudes while avoiding dissenting views (Lazarsfeld, Berelson, & Gaudet, 1944; for a recent meta-analysis, see Hart et al., 2009). Such preferences for like-minded political content may lead to a fragmented, unintelligent, and politically intolerant electorate (Mutz, 2002; Mutz & Martin, 2001; Sunstein, 2001). The new media environment, particularly the Internet, provides the audience with a platform to select or avoid content in great specificity, thus providing ample opportunity for confirmation-biased selective exposure. The present study examines whether such confirmation bias governs exposure to political messages right before a presidential election. When potentially fundamental political changes may soon occur, media users may be motivated to attend to information that will best help them adjust to the impending changes in the environment and would thus be driven by informational utility. Although earlier work on this question also examined preelection periods, the present investigation applies more precise exposure measures and also considers the role of Internet use. Hence, we will explore whether message preferences are more pronounced among those who use the Internet for news exposure more often, as this context facilitates greater selectivity. In the following, we will review theoretical frameworks and prior work before deriving hypotheses that will then be subject to empirical tests.

Confirmation Bias and Informational Utility

In 1957, Festinger proposed his theory of cognitive dissonance, which has basically defined the term of selective exposure in its traditional meaning ever since. The theory of cognitive dissonance is predicated on the idea that people strive for cognitive consistency or a drive to “produce consonant relations among cognitions” (Festinger, 1957, p. 9). Conversely, cognitive dissonance is a state of mental unbalance or unrest resulting from inconsistent relationships among cognitions. Festinger hypothesized that individuals would not only attempt to reduce dissonance but also actively avoid situations and information where dissonance is expected to occur. On the flip side, individuals are thought to prefer information in line with preexisting attitudes.

In an extensive line of research, Festinger’s theory of cognitive dissonance has not always been supported (for a recent review, see Donsbach, 2009). Empirical evidence has repeatedly shown that there are circumstances when attitude-discrepant messages are not necessarily avoided and at times actually sought out (Canon, 1964; Festinger, 1964; Frey, 1986). Citing the inconsistency of results, Sears and Freedman called for a more fruitful approach to interpreting the data. One factor in particular, informational utility, was suggested to systematically affect exposure behaviors.

In the most basic of terms, informational utility is the degree to which information can aid individuals in making future decisions, in political contexts and beyond. The more useful a person perceives information to be, the more likely he or she will be to engage with it, regardless of whether it is consistent with or discrepant from preexisting attitudes. Building off early work by Canon (1964), Freedman (1965), and Festinger (1964), Atkin (1973) synthesized a formal theoretical argument explaining the causal mechanisms driving informational utility exposure. Atkin postulated that the need for information was a result of uncertainty in how to respond to the environment, and thus, information functions as an aid for adaptation to the environment. The need for information serves four primary functions: surveillance (i.e., keeping cognizant of changes in the environment, monitoring threats), performance (i.e., how to do things), guidance (i.e., how to feel about things), and reinforcement (i.e. confirm attitudes; Atkin, 1973; Knobloch-Westerwick, 2008). More recently, information utility has been explained as follows:

information relating to individuals’ immediate and prospective encounter of threats or opportunities . . . [has] utility for these individuals, the degree of which increases with (1) the perceived magnitude of challenges or gratifications, (2) the perceived likelihood of their materialization, (3) their perceived proximity in time or immediacy, and (4) their perceived efficacy to influence the suggested events or consequences. . . . The increased utility of messages, in turn, fosters longer exposure to information. (Knobloch-Westerwick, 2008, p. 2273)

This perspective thus implies that, if information is expected to benefit individuals, they will be less concerned with whether or not it is attitude consistent and will be more willing to engage with attitude-discrepant messages. The utility motivation will then override the confirmation bias. This could be especially prominent among frequent Internet users, who typically report to not have strong loyalties to one particular news site over another (Purcell, Rainie, Mitchell, Rosenstiel, & Olmstead, 2010).

Confirmation Bias in the Internet Age

The “pull” nature of the Internet inherently affords individuals the opportunity to engage with whatever information they choose while ignoring the rest (Bimber & Davis, 2003). In this sense, the Internet can have a polarizing effect if users are motivated by a confirmation bias—it is possible for individuals to avoid dissonant messages altogether. This has led some to argue in favor of a “New Era of Minimal Effects” (Bennett & Iyengar, 2008) in line with the traditional views of media effects advanced by Lazarsfeld et al. (1944) and Klapper (1960). Their argument is predicated on the idea that, because media users can avoid dissenting views more easily online, they will. The Internet, therefore, has the potential to magnify confirmation-biased exposure patterns. However, not everyone is convinced that this is in fact happening (Gentzkow & Shapiro, 2010; Holbert, Garrett, & Gleason, 2010). The same characteristics of the online environment (i.e., personalization, ease of information selection as well as of information avoidance) that allow for increased confirmation bias in exposure can also provide the mechanism for information seeking of more diverse views when individuals are driven by motives other than a confirmation bias. Whereas traditional media use per television and newspaper may facilitate partisan-aligned exposure through habitual use of partisan channels and subscriptions to partisan newspapers, online news use may be different. A recent survey (Purcell et al., 2010) indicated that online news users report being less loyal to getting news from partisan websites. Thus, online exposure may be less driven by a confirmation bias, as suggested by some recent research (DiMaggio & Sato, 2003; Garrett, 2009; Stromer-Galley, 2003). In brief, the literature not only suggests that online information use makes individuals more effective in avoiding attitude-discrepant messages but also that online users exert such avoidance to a smaller extent. What specific mechanisms underlie these phenomena, however, remains unclear. Possibly, online news formats encourage different styles of engaging with news. Eye-tracking research (Poynter Institute, 2008) suggests that online news readers actually read more and in greater depth than print readers did, but print readers are still less likely to scan the news and also show greater avoidance of opinion columns/editorials.

Selective Exposure in the Election Context

It could be argued that, in American society, nothing garners more attention by both the media and the public than a presidential election. The time prior to an election is unique in that election outcomes are expected to have serious ramifications on the country’s economy, military, health regulations, education system, and so on, thus having a great effect on all citizens. The four subdimensions mentioned earlier that have been suggested to specify when informational utility will be high (Knobloch-Westerwick, 2008)—magnitude, likelihood, immediacy, and efficacy—are all present in election-related news prior to election day: Citizens can generally expect that a potential government change will have substantial consequences, that they are likely to be affected by these consequences, that the consequences will occur soon, and that they have the ability to influence whether this event of a government change will occur or not through their voting decision. Therefore, relative to selective exposure to political content, it is expected that, prior to an election, an information utility motivation will be especially pronounced among the electorate and may override a confirmation bias.

The present study examines the role of information utility on media exposure choices just prior to a presidential election. Exposure patterns before an election are often thought to be of great interest, as the following research review shows, because they could potentially affect and/or bolster voting intentions. A quasi-experimental study with behavioral measures of selective exposure, building on cleanly differentiated stimuli messages on several issues to define attitude-consistent and attitude-discrepant exposure coded per participants’ political preference voiced in separate first session, was conducted in the weeks leading up to the highly contentious 2008 presidential election. A brief review of prior related work, which mostly relied on self-reports and defined exposure mostly based on whether the source was the favored party or the competing party, will show how the present investigation extends this line of research.

In terms of selective exposure to political media messages in times before an election, Lazarsfeld, Berelson, and Gaudet’s landmark study, “The People’s Choice,” was the first to identify that individuals were selectively attending to particular political media messages while avoiding others. Even before Festinger’s (1957) theory of cognitive dissonance, these authors noted,

The fact that people select their exposure along the line of their political predispositions is only a special case of a more general law which pervades the whole field of communication research. Exposure is always selective; in other words, a positive relationship exists between people’s opinions and what they choose to listen to or to read. (Lazarsfeld et al., 1944, p. 164)

Yet Lazarsfeld and colleagues’ findings relied exclusively on self-report data. In contrast, some early work on selective exposure in election contexts built on measures of actually exerted message selections instead of self-reports. Freedman and Sears (1963, reported in Sears & Freedman, 1967) asked Californians about their candidate preferences in a gubernatorial election and offered them to choose from pamphlets about the election race. More than half of the individuals with a preference who picked a partisan brochure went for a pamphlet in line with their political preference. A parallel study with similar results was presented by Rhine (1967).

However, most prior selective exposure research in election contexts has relied on self-reports. Ziemke (1980) found some indication that selective exposure prior to a campaign depends on certainty of political preference. His findings were derived from survey data collected after the 1976 presidential election, which may suffer from impaired recall and respondent biases due to the outcome of the election. Chaffee and collaborators studied selective exposure in election contexts, mostly relying on survey measures of how much “attention” was paid to a campaign: Chaffee and McLeod (1973) as well as Chaffee and Miyo (1983) did not detect a confirmation bias among voters, whereas Chaffee, Saphir, Graf, Sandvig, and Hahn (2001) found that survey respondents reported a preference for messages consistent with their political leaning. Most recently, Stroud (2008) utilized the 2004 National Annenberg Election Survey and showed that respondents reported viewing media channels that aligned with their political leanings more often during the 5 months leading up to the 2004 presidential election than they normally would.

In contrast to studies relying on self-reports, research by Iyengar, Hahn, Krosnick, and Walker (2008) is comparable to the present work in that selective exposure to political messages was observed electronically right before a presidential election. Participants received a CD with information on political candidates in the 2000 presidential American election, which allowed tracking the information these respondents chose to view. The results showed that conservative partisans preferred the information provided by the Republican candidate. But liberal partisans were no more likely to choose messages on the Democrat candidate than his or her opponent. Iyengar et al. called for more research to examine this asymmetry in the confirmation bias. We will address this need for further investigation with our study on the 2008 presidential election context and by drawing on the informational utility framework. It is relevant to note that throughout the 2000 election campaigns, the Democrats mostly trailed the Republicans in the polls, even though some reversals occurred (Franklin, 2008). Hence, we suggest that Iyengar et al.’s findings reflected that Democrats were motivated by the informational utility of the messages on the Republican Party as likely election winner, which could then have compensated for the confirmation bias. After all, information about likely incoming governmental forces should hold great utility.

The 2008 American presidential election ended a tumultuous 8-year run by Republican President George Bush, who left the office with the second-lowest final approval rating in recorded history (34%; Saad, 2009). The year 2008 was the first election year in which more than half of the voting-age population used the Internet to follow the campaign and in which the Internet was utilized equally with newspapers as a source for election news (Smith, 2009).

The first signs of an ensuing governmental regime change occurred in 2005 when an increasing majority of Americans began to identify themselves as Democrat, or Democrat-leaning, in terms of party affiliation (Jones, 2009). During the midterm elections in 2006, voters put the Democratic Party in control of the House for the first time in 12 years. Pro-Democratic sentiment continued to grow through 2007, and by the first quarter of 2008, a 14% gap existed between Americans who identified as Democrat opposed to Republican. As election day drew near, it looked as if the 8-year Republican stronghold on the White House was going to be conceded to the Democrats. On November 2, just 2 days prior to election day Gallup’s final preelection poll based on data for the period October 31-Novemebr 2 showed Barrack Obama was 11 points ahead of John McCain, the Republican candidate, among likely voters (55% to 44%; Newport, Jones, & Saad, 2008). The aforementioned factors lead us to believe information utility in regards to political information in the weeks just prior to the election was at an absolute premium, as a governmental change was perceived as likely to occur.

Current Study and Hypotheses

Under the conditions of the 2008 election, it is plausible that the most “useful” information on political matters was information in line with the Democratic Party that was projected to win the election. Therefore, individuals whose preferred party is expected to win an election will be more inclined to choose messages in line with their preheld attitudes. On the other hand, individuals whose preferred party is not expected to win will favor messages that converge with views of the party anticipated to win the election, even though they diverge from preexisting attitudes. Thus, only individuals who favor the political party that is likely to succeed in the election will be motivated by both a confirmation bias and an informational utility that will lead them to consume messages aligned with stances of said party, whereas for other media users the confirmation bias may be overridden by utility motivations. In light of inconsistent views in the literature regarding the Internet’s impact, influences of habitual Internet news use will be addressed through competing hypotheses.

Hypothesis 1: Prior to an election, partisans of the party that is likely to succeed in the election prefer political messages consistent with their political leaning.

Hypothesis 2: Prior to an election, partisans of the party that is likely to lose the election, resulting in a government change, prefer political messages with the opponent, leading party’s views.

Hypothesis 3a: The confirmation bias suggested in Hypothesis 1 is more pronounced among media users with habitually low online news use.

Hypothesis 3b: The confirmation bias suggested in Hypothesis 1 is more pronounced among media users with habitually high online news use.

Hypothesis 3c: A difference as suggested in Hypotheses 3a and 3b results from different styles of interacting with an online site, such as different styles of choosing messages reflected in more time spent on an overview page or more selections in a given time.

To test these hypotheses, selective exposure will be measured both as exposure time to and frequency with which participants engaged with consonant or dissonant content on an experimentally manipulated news website. Both measures of issue attitudes and political partisanship were utilized to define an individual’s baseline attitudes, thus allowing for us to categorize exposure to consonant and/or dissonant messages. For the duration of this article, attitude-based selective exposure will be referred to as being either “attitude consistent” or “attitude discrepant,” whereas partisan-based selective exposure will be referred to as “partisan consistent” or “partisan discrepant.”

With the use of unobtrusive observation measures, the current study fits into the paradigm that has guided much of recent selective exposure research, going beyond political contexts. In this framework, selective exposure coins the phenomenon that media users do not allocate their media choices and time equally to the available cornucopia of media messages but rather demonstrate preference and avoidance patterns due to situational circumstances and personality factors (Zillmann & Bryant, 1985). The selective exposure paradigm postulates that the level of individual awareness regarding determinants of media choices is typically low, though some variation will exist (Knobloch-Westerwick, Hastall, Zillmann, & Callison, 2003; Zillmann, 1985; Zillmann & Bryant, 1985). As a result, the paradigm is associated with experimental research designs employing behavioral measures of media choices, in light of concerns regarding measures into motivation introspection and media use recall. These concerns apply to reported news exposure in particular, for which validity examinations have yielded devastating results (Prior, 2009). Along these lines, we propose that media users are not mindful of a confirmation bias or informational utility as driving their selective exposure to political messages. Because of these serious issues with self-report and introspection measures, the present investigation relies on unobtrusive observation of message exposure behaviors.

Moreover, the present study will employ political messages pertaining to several issues and will categorize exposure per issue-specific attitudes (attitude-consistent vs. attitude-discrepant messages) and partisanship (partisan-consistent and partisan-discrepant messages). The presented messages’ content will clearly differ regarding alignment with political party, whereas earlier related research often simply used political parties as sources to identify such alignment. Lastly, sensitizing participants to importance of partisanship or political attitudes will be avoided by measuring them in a prior session days before the main experiment and by embedding them in many distracter questions.

Method

Overview

A two-session online experiment was conducted with complete data from both sessions for 205 participants. In the first session, a web-based computerized questionnaire asked respondents about 12 political issue attitudes, partisanship, political interest, and news use habits. The second session, several days later, differed from the first in that participants were asked to interact with an online news magazine. The magazine featured 4 policy (target) issues chosen from the 12 political topics presented in the first session. A total of eight articles were displayed. Each of the four policy issues was represented by two articles featuring opposing topic perspectives. Selective exposure to specific news reports was unobtrusively logged by software. After the browsing, measures of perceived article quality were collected to provide closure.

Respondents

Participants were recruited through a two-step recruiting process. Students from an undergraduate communication class at a large Midwestern university were asked to recruit nonstudent participants between the ages of 30 and 65 and not enrolled at a university to participate in the study. Recruiting students received small increments of extra credit for each participant. Participants were rewarded with a US$20 check upon completion of both research sessions. The sample consisted of 205 complete data sets of participants who attended both sessions, with 53% women. The average age of participants was 40.57 (SD = 10.17), which did not differ by gender, t(202) = 0.314, p = .754. Furthermore, no significant differences for age and gender emerged when comparing participants with different partisanship (as explained below). In the sample, 75 respondents identified themselves as liberal, 76 as conservative, and 43 as nonpartisans. As many as 92 participants were from one Mid-Western state, and the others resided in 27 other states.

Procedure

The main experiment consisted of two sessions conducted online. The data collection was performed during the period September 28-November 8, 2008, and thus, in the final weeks prior to the 2008 presidential election. Students were contacted via email announcement from teachers, asking for their assistance in recruiting nonstudent participants between the ages of 30 and 65 for a study about “news outlets and judgments of news issues.” Instructions were provided leading students to access a website where they were able to create an account, from which they could start recruiting. Emails were sent to potential respondents, asking them to participate in a two-session online study regarding news outlets and judgments of news issues. At this point they were informed of the US$20 incentive check they would receive upon completion of both sessions. They were also informed of some technical requirements, as the online procedure was set up so that a Windows PC with an Internet browser (typically Internet Explorer) had to be used and a web player had to be downloaded to execute the application programmed in Authorware7. Participants were then asked to follow an embedded web link to access the online experiment site. The first screen page informed participants that the session required full attention and allowed them the option to quit if they wished to take the session at a later time. Questions in the first session pertained to attitudes, demographics, self-descriptions, including politically relevant adjectives, political interest, and news use habits.

Four days later, participants received another email invitation recruiting them to participate in the second online session. Again they were informed that full attention would be required, similar to the instructions above. Once participants agreed, further instructions were given as follows:

In the following, you will see a test version of a magazine. Please browse through to gain an impression of the articles. The scheduled time does not allow reading all articles. So please read what you find interesting, just as you normally would. There is no assigned number of articles that you should read, and you don’t have to read the articles as a whole. After the scheduled browsing time is over, a questionnaire will upload automatically, and you will be asked about your impressions of the articles.

The online magazine was then shown for 4 min, allowing participants to browse the 8 articles. At the end of the browsing period, a questionnaire was uploaded and asked participants how “credible/important/biased/interesting/well-written/relevant” the articles were; the items were rated on a 7-point scale, with response options ranging from not at all to extremely. These questions merely served to provide closure. Finally, participants were thanked and informed about receiving incentives within 10 days.

Experimental Internet Magazine

Display of available articles

The experimental web magazine was programmed with Authorware specifically for this study. This online news magazine had a similar look and feel to popular news magazines currently on the Internet. A masthead of the name and logo of the experimental platform, “American’s National Forum—Online Opinion,” was displayed across the top of the website. In addition, a navigation bar was placed on the left-hand side of the page. Although it was deactivated, the displayed navigation bar contained newspaper section headings such as economics, science, and so forth, which would be commonly found on a news site. The main frame initially contained an overview, which listed news leads for all available articles in two columns.

The overview page showed eight news leads that each contained a headline, a news lead, and a hyperlink to access the actual article. The positions of news leads on the overview page were randomized for each participant to prevent position effects. However, the two articles about the same issue were never displayed next to or above each other. The respondents made their reading selections by clicking on the hyperlinks to articles, scrolling through the selected articles and reading as much of them as they cared to, clicking to return to the overview, selecting other articles (or returning to the abandoned ones), and so forth, until the end of the reading period. Whenever a participant accessed or exited an article page via hyperlinks, the application logged the activity to accumulate selective exposure times.

News leads and article texts

The four policy issues and the eight headlines of the displayed articles were (a)“universal health care,” “personalized health coverage”; (b)“firearm threat,” “self-defense rights”; (c)“cruelty of prochoice,” “abortion is prolife”; and (d)“increase minimum wage,” “wage raising hurts.” The news leads were about the same length with 24 to 28 words. The eight employed articles were culled from partisan and lobbying websites such as heritage.org, nraila.org, and others. All articles were minimally edited, essentially shortened, to equalize their length to about 705 to 719 words (M = 716, SD = 5.2). The sans-serif typeface Verdana was used for all articles. Regular scrolling allowed access to the full text of the articles.

A pretest of news leads had been conducted (with participants from the same population as used for the current study) to select two opposing leads per topic that were perceived as different in political stance while being equally interesting. A separate stimuli test with participants from the same population had established that text pairs for all topics did not differ significantly for level of interest (Knobloch-Westerwick & Meng, 2009, for details).

Measures

Attitude (dichotomous)

Participants completed a practice trial to become familiar with the task and were instructed as follows:

The first task serves to show you how things work and to familiarize you with the procedure. In this task words will be presented for you to classify into groups—positive and negative adjectives. This task requires that you classify items as quickly as you can while making as few mistakes as possible. Going too slow or making too many mistakes will result in an uninterpretable score.

Press one of the two keys to categorize the displayed word.

The key with the symobol “Z” indicates NEGATIVE adjective.

The key with the symbol “/” indicates POSITIVE adjective.

It works best to keep one index finger above both these keys for quick responses.

Six adjectives (marvelous, superb, pleasure, painful, terrible, awful) showed up on each new screen in a random sequence. On each screen, the adjective appeared in the center with ‘Negative” and “Positive” at the left and right bottom respectively. Once a participant pressed “Z” or “/,” the procedure continued to the next screen.

For the actual dichotomous measurement of attitudes, eight political filler issues were presented first in randomized sequence and then the four target issues relevant for the later exposure measurement were displayed in randomized sequence, with three words for each topic. The instructions for this task were identical to those above except political policies were presented and respondents were asked to choose either “oppose” or “support” by pressing corresponding keys.

Attitude (Likert-type scale)

Respondents were asked to rate how strongly they oppose or support the policies on a 5-point Likert-type scale, with strongly support, somewhat support neither support nor oppose, somewhat oppose, and strongly oppose as response options.

Political interest

In the fist session, participants were asked how closely they followed news about government and public affairs on a 4-point scale, “very closely/not too closely/somewhat closely/not at all closely.”

Partisanship

Questions on the partisanship were embedded in general self-descriptions. The approach was in line with procedures by Markus (1977) and more recently, in communication research (Comello & Slater, 2009; Knobloch-Westerwick & Meng, 2011). The instructions were displayed over several screens as follows:

In the following, you will be asked about yourself based on characteristics that may or may not apply to you. We’d like to get your spontaneous responses, so please press the response key quickly to indicate whether a characteristic applies to you or not.

The key with the symbol “/” indicates that the characteristic applies to you; it equals “ME.” The key with the symbol “Z” indicates that the characteristic does not apply to you; it equals “NOT ME.”

It works best to keep one index finger above both these keys for quick responses.

The 24 adjectives concerning personal characteristics were displayed on separate screens, with the adjective displayed in the center, and “Me” and “Not me” shown at the right and left bottom of the screen, respectively. Once a participant pressed “/” or “Z,” a new screen was automatically uploaded. First, 16 filler items such as “moody” or “imaginative” were presented in a randomized sequence. Then the four target adjectives, “conservative,” “liberal,” “Republican,” and “Democrat,” embedded in four other politically relevant descriptions (e.g., “patriotic”), were shown in randomized order.

A condensed measure of partisanship was created based on the self-categorizations (not considering response times). First, two of the four-target self-categorizations were reverse coded so that lower scores indicated more liberal or Democrat leaning and higher scores more conservative or Republican leaning. A reliability test yielded a Cronbach’s alpha of .768. On the basis of this new measure, 76 respondents were categorized as conservatives/Republicans, 75 as liberals/Democrats, and 43 as nonpartisans.

News use frequency. Participants rated their news use frequency for online news, daily newspaper, and TV news, on a 6-point scale, with every day, several times a week, once a week, several times a month, once a month, and less often as response options. Separate news use variables were created for traditional news use frequency as well as Internet news use frequency. For traditional news use, television and newspaper use was collapsed into one average score (M = 2.95, SD = 1.30). A dichotomous variable for online news use was created with low (once a week or less frequently), which applied to 43% of the participants, versus high (every day or several times a week), which applied to 57%.

Selective exposure measures. The software application tracked participants’ exposure in seconds by logging every hyperlink use. Exposure was operationalized as article choice (clicked on hyperlink leading to article or not) and reading time in seconds, as earlier related research has yielded slightly different patterns for the two indicators (Knobloch-Westerwick & Meng, 2009). In addition to the selective exposure for the entire browsing time, reading behavior was also recorded for each 30-s interval.

For each political issue, the dichotomous measurement of attitudes from the first data collection session was employed to code articles and exposure to them as attitude consistent or attitude discrepant on an individual basis. The following measures were generated from hyperlink clicks for individual articles to test the hypotheses: (a) selection of attitude-consistent articles, (b) selection of attitude-discrepant articles, (c) exposure time in seconds dedicated to attitude-consistent articles, and (d) exposure time in seconds dedicated to attitude-discrepant articles. For a-d, measures were generated not only for each specific article but also for accumulated selections and exposure times, whereas only accumulated measures were used in later analyses. Furthermore, partisanship-based selective exposure measures were also created and used to assess partisan-consistent and partisan-discrepant exposure in article choices and reading times, resulting in four additional accumulated measures. Each issue was represented by both classic “liberal” as well as “conservative” viewpoints. Partisan-consistent and partisan-discrepant exposure was judged by comparing participants’ political partisanship with article selection either aligned with or divergent from said parties’ traditional issue stance.

Post Test

Validation of political partisanship measure

To validate our measure of partisanship, 68 undergraduate students completed the procedure of the first session in a research lab and furthermore responded to questions and filter questions taken from the American National Election Studies (ANES, n.d.) that were added to the computerized questionnaire: “Generally speaking, do you think of yourself as a Republican, a Democrat, an Independent, or what? Would you call yourself a strong Democrat/Republican or a not very strong Democrat/ Republican? Do you think of yourself as closer to the Republican Party or to the Democratic Party?” Based on their responses, a 7-point scale was constructed with 1 = strong Republican and 7 = strong Democrat. This scale was significantly correlated with our partisanship measure (r = .85, p < .001), corroborating the validity of our operationalization.

Validation of news leads’ and articles’ convergence with political parties’ stances

In the same posttest, participants were also presented with the news leads and rated them based on the following questions. “Does this news lead reflect more the views of the Democratic Party or of the Republican Party? Does this news lead reflect more the views of Barack Obama or of John McCain?” The 11-point rating scales were anchored with –5 = reflects the Republican Party and +5 = reflects the Republican Party for the first question and –5 = Reflects Barack Obama and +5 = Reflects John McCain.” Table 1 reports the analyses that demonstrated that the stimuli news leads were perceived as clearly aligned with either the Republican or the Democrat party and the respective candidates.

Table 1.

News Leads’ Alignment With Political Parties’ and Candidates’ Stances (M, SD in Parentheses)

Issue	Party stance		Candidate stance
News lead heading	Means (−5 to 5)	t Test (df = 67)	Means (−5 to 5)	t Test (df = 67)
Gun ownership		7.0*		7.83*
Firearm threat	−2.4 (2.5)		−2.0 (2.1)
Self-defense Rights	1.69 (3.3)		1.93 (2.8)
Abortion		−9.7*		−9.5*
Abortion is prolife	−2.44 (3.3)		−1.75 (2.8)
Cruelty of prochoice	3.06 (2.6)		2.69 (2.4)
Healthcare regulation		7.7*		8.3*
Universal health care	−2.76 (3)		−3.21 (2.8)
Personalize health coverage	1.71 (3.4)		1.31 (3.4)
Minimum wage		6.8*		6.8*
Increase minimum wage	−2.35 (2.7)		−2.13 (2.4)
Wage raising hurts	1.93 (2.8)		1.69 (2.6)

Note: Standard deviations reported in parentheses. t values with one asterisk (*) indicate a difference with p < .001. Stance alignment ratings were based on the scales ranging from −5 to +5: −5 = reflects the Republican Party and +5 = reflects the Republican Party; −5 = reflects Barack Obama and +5 = reflects John McCain.

The remainder of the posttest procedure was different for 4 groups of the 68 participants, as each group was presented with 2 of the 8 stimuli articles. The participants read each article broken down into three segments and answered the two questions mentioned below for each text segments, as well as for the entire text. As the analyses reported in Table 2 show, the stimuli articles were seen as clearly aligned with either the Republican or the Democrat party and the respective candidates.

Table 2.

Stimuli Articles’ Alignment With Political Parties’ and Candidates’ Stances (M, SD in Parentheses)

Issue	Party stance		Candidate stance
News lead heading	Mean across measures	F value, partial η² (df = 1, 30−33)	Mean across measures	F value, partial η² (df = 1, 30−33)
Gun ownership		31.8*, 0.491		26.2*, 0.442
Firearm threat	−1.88 (1.80)		−1.57 (1.66)
Self-defense rights	1.94 (2.20)		1.38 (1.75)
Abortion		51.9*, 0.611		41.7*, 0.558
Abortion is prolife	−2.04 (2.24)		−1.38 (1.87)
Cruelty of prochoice	2.96 (1.83)		2.51 (1.68)
Healthcare regulation		32.7*, 0.522		31.0*, 0.508
Universal health care	−2.10 (1.53)		−2.15 (1.54)
Personalized health coverage	2.12 (2.57)		1.85 (2.47)
Minimum wage		37.7*, 0.557		41.4*, 0.580
Increase minimum wage	−2.88 (1.76)		−2.62 (1.60)
Wage raising hurts	1.56 (2.26)		1.01 (1.59)

Note: Stance alignment was rated for article segments (three for each) and entire article on the scales ranging from −5 to +5: −5 = reflects the Democrat Party and +5 = reflects the Republican Party; −5 = reflects Barack Obama and +5 = reflects John McCain. Reported means are averaged across these repeated measures. Comparisons are based on ANOVAs with repeated measures, using articles with opposed stances as between-group factor.

F values with one asterisk (*) indicate a difference with p < .001.

Validation of focus on national issues

To examine whether the articles’ topics and content pertained to national politics, in line with our argument that participants in the main experiment considered them relevant in light of the upcoming presidential election, we asked the posttest respondents to rate the news leads as well as entire articles based on the question, “Does the topic of this article pertain more to a State/Local or a National political issue?,” with an 11-point scale ranging from –5 = pertains to a state/local political issue to +5 = pertains to a national political issue. For each of the 16 examined stimulus unit, leads, and articles, the ratings differed significantly (p < .031 in one-sample t tests) from scores indicating a state/local issue (–5 to –1). Thus the news leads and the articles were seen as pertaining to national political issues.

Results

Preliminary Analyses

Attitudes

Liberals/Democrats on average indicated significantly more support than conservatives/Republicans for each of the four issues as they were framed in this study (see Table 3). When interpreting these descriptive statistics, it needs to be considered that all issue cues were worded in line with liberal views. Thus, liberals/Democrats would most likely express “support” throughout, whereas conservatives/Republicans would most likely express that they “opposed” an approach issue. This setup of the questions explains why the conservatives/Republicans attitude ratings were not much above the scale midpoint. They only supported increasing minimum wage, but did so at a significantly lower level than nonpartisans and liberals/Democrats. Nonpartisans showed a preference for liberal views, supporting three of the four target issues while being relatively neutral toward the issue of legalizing abortion. These descriptives reflect that the four target issues were clearly aligned with political party ideologies.

Table 3.

Descriptive Statistics for Attitudes (M, SD in Parentheses)

	Political partisanship
Article issue	Liberal	Nonpartisan	Conservative
Attitude (1= support, 2 = oppose)
Increase minimum wage	1.05^a (0.23)	1.12^a (0.32)	1.29^b (0.46)
Legalize abortion	1.21^a (0.41)	1.44^b (0.50)	1.68^c (0.47)
Universal health care	1.12^a (0.33)	1.19^a (0.39)	1.61^b (0.49)
Stricter gun control	1.21^a (0.41)	1.30^ab (0.47)	1.47^b (0.50)
Attitude (1 = support, 5 = oppose)
Increase minimum wage	1.36^a (0.73)	1.53^a (0.94)	2.36^b (1.34)
Legalize abortion	1.80^a (1.34)	2.72^b (1.70)	3.78^c (1.61)
Universal health care	1.49^a (1.08)	1.84^a (1.36)	3.34^b (1.57)
Stricter gun control	1.65^a (1.12)	2.16^a (1.54)	2.89^b (1.63)

Note: Means in a row with different superscripts differ at p < .05 in Games–Howell post-hoc tests.

Convergence of attitudes with political party stances

For the examination of our hypotheses, the assumption that individuals’ topic attitudes converge with parties’ stances regarding these topics is of crucial importance. After all, it is possible that someone who thinks of himself or herself as a Republican favors prochoice abortion policies nonetheless. In that case, attitude-consistent exposure for this particular topic could actually undermine partisanship as political self-concept. However, individuals’ attitudes for the four target topics were generally highly consistent with the viewpoints supported by the favored party. This was demonstrated by an interitem reliability analysis that employed the dichotomous attitude responses for the four issues (which were all worded in proliberal terms) and the four target adjectives for political self (two of them recoded). Cronbach’s alpha was .76 for this analysis.

Selective exposure

Eleven individuals who clicked on one or less of the available articles or spent 3.5 min or more on the overview page were excluded from the analyses. Under these conditions individuals did not appear to have been engaged with actual news browsing and reading.

For the remaining sample of 194, the average time spent on the overview page was 59 s (SD = 41). The minimum time spent on the overview page was 8 s, the maximum was 209 s. Selective exposure measures showed that, of the four topics addressed in the 8 available articles, on average 2.75 (SD = 1.00) were accessed. Furthermore, 0.62 topics on average (SD = 0.73) were viewed only through attitude-discrepant articles and 0.90 topics (SD = 0.91) were viewed only through attitude-consistent articles, but most often (M = 1.23, SD = 1.30) participants attended to both views for a topic. Each of these three means differed significantly from the others in paired t tests (p < .020).

Effects of Partisanship on Selective Exposure

Four ANOVAs examined Hypotheses 1 and 2 with partisanship (conservative vs. liberals) and online news use frequency (low vs. high) as between-group factors and different selective exposure measures serving as within-group factor measures. Nonpartisans were excluded from these analyses. The first ANOVA utilized choices of attitude-consistent and attitude-discrepant messages, and the second drew on exposure times for attitude-consistent and attitude-discrepant messages. Along the same lines, the third ANOVA employed choices of partisan-consistent messages and partisan-discrepant messages, whereas the fourth ANOVA used exposure times for partisan-consistent messages and partisan-discrepant messages.

Table 4 reports the means and the effects pertaining to Hypothesis 1 and Hypothesis 2. Among liberals/Democrats, a preference for both attitude- and partisans-consistent exposure emerged, both for message choices and article times. However, this preference was absent among conservatives/Republicans, reflected in significant interactions between the between-group factor and the within-group factor (though one fell short of significance).

Table 4.

Selective Exposure to Political Messages as a Function of Partisanship (M, SD in Parentheses, ANOVAs With Repeated Measures)

Selective exposure measure	Liberals (n = 75)		Conservatives (n = 76)		Exposure type impact df (1, 149)			Exposure × Partisanship Impact df (1, 149)
	Consistent	Discrepant	Consistent	Discrepant	F	p	η²	F	p	η²
Exposure choices by attitudes	2.28 (1.11)	1.77* (1.25)	1.90 (1.23)	1.80 (1.22)	9.03	.003	0.058	4.46	.036	0.029
Exposure times (s) by attitudes	109 (58)	65* (48)	87 (59)	87 (60)	7.43	.007	0.048	7.51	.007	0.049
Exposure choices by partisanship	2.28 (1.15)	1.77* (1.24)	1.93 (1.14)	1.76 (1.23)	13.2	.001	0.082	3.54	.062	0.024
Exposure times(s) by partisanship	106 (61)	68* (52)	87 (58)	87 (68)	5.08	.026	0.033	5.01	.027	0.033

Note: Mean comparisons within groups of participants differentiated by online news use with an asterisk (*) differ at p < .05. Online news use habits served as additional between-group factor in these analysis, see impacts of this factor in Table 5. All main effects and all interactions were included in the model. Two-way interactions were not significant and are thus not reported.

Results pertaining to Hypotheses 3a and 3b are reported in Table 5. Online news use emerged as significant for two exposure measures and approached significance for the other two. Supporting Hypothesis 3a, participants who used online news only infrequently showed a confirmation bias, whereas frequent online news users did not. However, interactions between exposure type, partisanship, and online news use were not significant for any of the four exposure measures and are thus not reported in Table 4 and 5.

Table 5.

Selective Exposure to Political Messages as a Function of Habitual Online News Use (M, SD in Parentheses, ANOVAs with repeated measures)

	Low online news use (n = 68)		High online news use (n = 83)		Exposure type impact, df (1, 149)			Exposure x online news use, df (1, 149)
Selective exposure measure	Consistent	Discrepant	Consistent	Discrepant	F	p	η²	F	P	η ²
Exposure choices by attitudes	2.1 (1.2)	1.6* (1.2)	2.1 (1.2)	1.9 (1.2)	9.03	.003	0.058	2.16	.144	0.014
Exposure times(s) by attitudes	112 (59)	69* (56)	86 (57)	82 (55)	7.44	.007	0.048	6.84	.010	0.044
Exposure choices by partisanship	2.1 (1.1)	1.6* (1.2)	2.1 (1.2)	1.9 (1.2)	13.2	<.001	0.082	3.87	.051	0.026
Exposure time(s) by partisanship	108 (60)	74* (60)	88 (59)	81 (54)	5.08	.026	0.033	3.12	.080	0.021

Note: Mean comparisons within groups of participants differentiated by online news use with an asterisk differ at p < .05. Partisanship served as additional between-group factor in these analysis, see impacts of this factor in Table 4. All main effects and all interactions were included in the model. Two-way interactions were not significant and are thus not reported.

p < .05.

To examine the robustness of detected effects and to utilize news online use as a metric (instead of a dichotomized) variable, regression analyses for selective exposure measures as dependent variables were conducted. We utilized the difference of number of attitude- and partisan-consistent article choices and number of attitude- and partisan-discrepant article choices, as well as the percentage of exposure time for attitude- and partisan-consistent of the time spent on actual article pages as selective exposure indicators. In addition to partisanship and online news use habits, political interest, use habits for traditional news media, gender, income, age, and education served as additional predictors. The importance of partisanship was evident for three of the four exposure measures. The impact of online news use habits emerged specifically for the time-based exposure measures (see Table 6).

Table 6.

Impacts on Selective Exposure (Beta Weights)

	Number of attitude-consistent and attitude-discrepant articles	% of attitude-consistent exposure time	Number of partisan-consistent and partisan-discrepant articles	% of partisan-consistent exposure time
Political partisanship	—	0.18**	0.18**	0.20**
Online news use	—	−0.19**	—	−0.17**
Traditional news use	—	—	—	—
Political interest	—	—	0.18+*	0.18**
Education	—	—	—	—
Gender	—	—	—	—
Age	—	0.20**	—	0.17**
Income	−.31***	−0.25***	−0.22**	−0.16*
R²	.13**	0.15***	0.12**	0.13***

Note: Beta weights are significant at p values .01, .05, and .10.

p < .10. **p < .05. ***p < .01.

Exploratory Analyses on Origins of Online News Use Impacts

Various analyses were conducted to explore why online news use affected exposure patterns. Online news use did not affect evaluations of the presented news articles (ns). Participants who were frequent users of online news sites did not select more or less articles than others, even though they tended to spend slightly more time on actual article pages (as opposed to the overview page) than participants with low online news use did, but this only approached significance (p = .124). However, an analysis that considered how exposure patterns developed over time—with between attitude-consistent versus attitude-discrepant exposure as within-group factor, eight 30-s exposure intervals as additional within-group factor (time), and online news use and partisanship as between-group factors—yielded a significant interaction between the first three factors, F(7, 1029) = 2.12, p = .034, η² = 0.014. Figure 1 illustrates the related pattern. Participants who rarely accessed online news showed a consistent preference for attitude-consistent messages over attitude-discrepant messages, whereas frequent online news users initially also spent more time on attitude-consistent messages but then began to even favor attitude-discrepant messages later in the browsing period. On the other hand, impacts of political partisanship on selective exposure did not change significantly across time.

Figure 1.

Selective Exposure across Time as a Function of Habitual Online News Use

Discussion

The current work examined the role of information utility in politically motivated selective exposure prior to a presidential election based on unobtrusively observed exposure behaviors. Particular attention was paid to the role of habitual Internet news use in moderating selective exposure patterns. Findings from the current work not only corroborate an overarching confirmation bias but also support the notion that under certain conditions information utility does in fact override a confirmation bias and motivates selective exposure. In times when government change is likely imminent, information aligned with the stances of the party that is thought to take office soon carries high informational utility per subdimensions of this concept—magnitude, likelihood, immediacy, and efficacy. All voters can generally expect that this political change will imply considerable consequences, will likely affect them, and will do so soon, and that they may influence this event through their voting decision. However, only for individuals who favor the political party that is likely to succeed in the election, both a confirmation bias motivation and informational utility motivation will lead them to consume messages aligned with stances of this party. Individuals who do not prefer that party should be motivated to consume messages aligned with stances of this party to anticipate upcoming political decisions and circumstances—thus informational utility may then override the confirmation bias.

In accordance with Hypothesis 1, liberals/Democrats preferred articles consistent with their attitudes and partisanship by both time and frequency. In this case, it can be argued that these individuals were motivated toward such “like-minded” material by a confirmation bias or information utility or both, as all of these factors would have led to the same result. At the time the data were collected, it increasingly looked as though the Democratic Party would be taking control of the White House. Therefore, it was expected that the most useful messages were those in line with the Democratic Party as anticipated election winner. Conversely and as expected (Hypothesis 2), a bias toward messages consistent with preexisting attitudes and partisanship was not present among conservative/Republican-leaning participants. Hence, the confirmation bias may have been compensated for by an informational utility motivation—learning about the Democratic Party’s stances was more useful than attending to information about the preferred party’s perspectives. This interpretation also explains why Iyengar et al. (2008) found a confirmation bias only among Republicans, as their preferred party mostly led in the polls during the 2000 election campaigns, triggering informational utility as exposure motive among partisans of the Democratic Party.

Examination of competing hypotheses on whether habitual use of online use is linked to stronger or weaker preference for articles discrepant from preexisting attitudes and partisanship yielded support for Hypothesis 3a; the confirmation bias in selective exposure was more pronounced among media users with habitually low online news use. This finding calls arguments into question that were advanced by Sunstein (2001) and others, who believe the Internet is propagating political segregation. This view assumes a confirmation bias explains all selective exposure behaviors online. However, the present empirical evidence based on observed selective exposure behavior does not readily align with this position. Yet it should be noted that the present investigation considered impacts of habitual online use, whereas situation-based comparisons within the same individual across online versus traditional media use could still show that a greater confirmation bias exists in the online context; however, to our knowledge, no experimental study has been conducted to clarify this matter.

Unlike traditional media, which is “pushed” on consumers, the Internet facilitates seeking and “pulling” selected content. High levels of such selective habitual news seeking as well as information utility seem to have played a major role in selective exposure to messages that challenged existing views. A confirmation bias may be more pronounced among those with low online news use because media users in the traditional media context may habitually turn to highly partisan channels such as Fox News or subscribe to a newspaper with clear political leaning. Indeed, findings by Iyengar and Hahn (2009) and Stroud (2008) have shown that preferences for cable news sources strongly depend on political leanings. As a result, individuals with low online news use may be less at ease with messages that challenge their views than regular online news users. Therefore, these individuals were likely motivated by a confirmation bias and were not as concerned with the relative utility of information.

On the other hand, for individuals who frequently use online news, it was interesting to note that their exposure patterns evolved and changed across time. These developments did not reflect differences from participants who access online news only rarely, for instance, in clicking patterns or frequencies or more or less care in choosing available messages. Typically, frequent online news users’ early browsing favored messages consistent with political preferences and then later exposure turned to discrepant content. It appears that after some bolstering of the existing views these individuals felt equipped to explore alternative stances. Hence, it does not seem to be the case the frequent online users access more information in a given time or make their selections with less or greater care. But their exposure patterns reflect a strategy of counterbalancing exposure to attitude-challenging messages with prior exposure to attitude-bolstering messages. However, interpretations of the specific patterns remain somewhat speculative and call for further examinations in the future.

It is worth noting that both conservatives/Republicans and liberals/Democrats were not exclusively engaging with articles that bolstered their views. Although we found an overall preference for articles aligned with respondents’ attitudes or partisanship, the participants also engaged with considerable amounts of challenging material by both choice and time (see section “Preliminary Analyses”). This observation adds to the evidence that a general preference for messages that bolster preexisting views does not necessarily mean media users avoid messages that challenge existing opinions altogether (Chaffee et al., 2001; DiMaggio, & Sato, 2003; Garrett, 2009). Yet more often than not, such exposure to messages that challenged existing views was paired with exposure to messages that bolstered these views and pertained to the same political issue. Thus exposure to opinion-challenging messages only was outside the norm.

A limitation of the current study was the small sample of nonpartisans. Future research should aim to explore where nonpartisans fit into all of this and could also clarify how a bandwagon effect may play into exposure patterns. Exploratory results from this study showed a trend for nonpartisans to prefer articles in line with the leading party’s views; however, analyses (not reported) fell short of significance. Again, this is thought to be a result of the small number of nonpartisans in our sample. This work is also limited in that we do not have a measure of explicit voting intentions. Knowing whether an individual intended to vote as well as whom they intended to vote for could help further explain the role of information utility in selective exposure. Furthermore, future research should capture information consumers’ specific perceptions of political messages along the lines of informational utility—magnitude, immediacy, likelihood, and efficacy—in light of imminent political changes.

Further research in the domain of selective exposure to political messages should focus not only on the question of “what’s driving selective exposure” but also “what is the effect of selective exposure?” More work needs to examine how recipients actually respond to messages that challenge their views regarding affect, interest, evaluation, political participation, knowledge gains, and discussion. In terms of political elections, it is of the utmost importance to investigate effects that selective exposure has on voting intentions and behaviors.

Footnotes

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

The authors received no financial support for the research and/or authorship of this article.

References

ANES. (n.d.) Questions asked in ANES surveys. Retrieved from http://www.electionstudies.org/resources/questions/questions.htm

Atkin

C. K.

(1973). Instrumental utilities and information seeking. In Clark

(Ed.), New models of communication research (pp. 205-242). Newbury Park, CA: SAGE.

Bennett

W. L.

Iyengar

(2008). A new era of minimal effects? The changing foundations of political communication. Journal of Communication, 58(4),707-731.

Bimber

Davis

(2003). Campaigning online: The Internet in U.S. elections. New York, NY: Oxford University Press.

Canon

L. K.

(1964). Self-confidence and selective exposure to information. In Festinger

(Ed.), Conflict, decision, and dissonance (pp. 83-96). Stanford, CA: Stanford University Press.

Chaffee

S. H.

McLeod

J. M.

(1973). Individual vs. social predictors of information seeking. Journalism Quarterly, 50, 95-120.

Chaffee

S. H.

Miyo

(1983). Selective exposure and the reinforcement hypothesis: An intergenerational panel study of the 1980 presidential campaign. Communication Research, 10(1), 3-36.

Chaffee

Saphir

M. N.

Graf

Sandvig

Hahn

K. S.

(2001). Attention to counter-attitudinal messages in a state election campaign. Political Communication, 18(3), 247-272.

Comello

M. L. G.

Slater

M. D.

(2009, May). The effects of drug-prevention messages on the accessibility of identity-related constructs. Health Communication Division, 59th Annual International Communication Association Conference, Chicago.

10.

DiMaggio

Sato

(2003, August). Does the Internet balkanize political attention? A test of the Sunstein theory. Paper presented at the Annual meeting of the American Sociological Association, Atlanta, GA.

11.

Donsbach

(2009). Cognitive dissonance theory—Roller coaster career. How communication research adapted the theory of cognitive dissonance. In Hartmann

(Ed.), Media choice: A theoretical and empirical overview (pp. 128-149). London: Routledge.

12.

Festinger

(1957). A theory of cognitive dissonance. Stanford, CA: Stanford University Press.

13.

Festinger

(1964). Conflict, decision, and dissonance. Stanford, CA: Stanford University Press.

14.

Franklin

(2008). Polling trends in 2008 vs ‘04 and ‘00. Retrieved from http://www.pollster.com/blogs/polling_trends_in_2008_vs_04_a.php

15.

Freedman

J. L.

(1965). Confidence, utility, and selective exposure: A partial replication. Journal of Personality and Social Psychology, 2, 778-780.

16.

Frey

(1986). Recent research on selective exposure to information. In Berkowitz

(Ed.), Advances in Experimental Social Psychology,19, 41-80.

17.

Garrett

R. K.

(2009). Echo chambers online? Politically motivated selective exposure among Internet news users. Journal of Computer Mediated Communication,14(2), 265-285.

18.

Gentzkow

Shapiro

J. M.

(2010). Ideological segregation online and offline. Cambridge, MA: National Bureau of Economic Research.

19.

Hart

Albarracin

Eagly

A. H.

Brechan

Lindberg

M. J.

Merrill

(2009). Feeling validated versus being correct: A meta-analysis of selective exposure to information. Psychological Bulletin, 135, 555-588.

20.

Holbert

Garrett

Gleason

(2010). A new era of minimal effects? A response to Bennett and Iyengar. Journal of Communication, 60(1), 15-34.

21.

Iyengar

Hahn

K. S.

(2009). Red media, blue media: Evidence of ideological selectivity in media use. Journal of Communication, 59, 19-39.

22.

Iyengar

Hahn

K. S.

Krosnick

J. A.

Walker

(2008). Selective exposure to campaign communication: The role of anticipated agreement and issue public membership. Journal of Politics, 70(1), 186-200.

23.

Jones

J. M.

(2009). More independents lean GOP: Party gap smallest since ‘05. Retrieved from http://www.gallup.com/poll/123362/independents-lean-gop-party-gap-smallest-since-05.aspx

24.

Klapper

J. T.

(1960). The effects of mass communication. New York, NY: The Free Press.

25.

Knobloch-Westerwick

(2008). Informational utility. In Donsbach

(Ed.), International encyclopedia of communication (pp. 2273-2276). Oxford, UK: Basil Blackwell.

26.

Knobloch-Westerwick

Hastall

Zillmann

Callison

(2003). Imagery effects on the selective reading of Internet newsmagazines: A cross-cultural examination. Communication Research, 30, 3-29.

27.

Knobloch-Westerwick

Meng

(2009). Looking the other way: Selective exposure to attitude-consistent and counter-attitudinal political information. Communication Research, 36(3), 426-448.

28.

Knobloch-Westerwick

Meng

(2011). Reinforcement of the political self through selective exposure to political messages. Journal of Communication, 61(2), 349-368.

29.

Lazarsfeld

P. F.

Berelson

Gaudet

(1944). The people’s choice. New York, NY: Columbia University Press.

30.

Markus

(1977). Self-schemeta and processing information about the self. Journal of Personality and Social Psychology, 35, 63-78.

31.

Mutz

D. C.

(2002). The consequences of cross-cutting networks for political participation. American Journal of Political Science, 46(4), 838-855.

32.

Mutz

D. C.

Martin

P. S.

(2001). Facilitating communication across lines of political difference: The role of mass media. American Political Science Review, 95(1), 97-114.

33.

Newport

Jones

J. M.

Saad

(2008). Final presidential estimate: Obama 55%, McCain 44%: Independents break for Obama, boosting Obama’s broad democratic base. Retrieved from http://www.gallup.com/poll/111703/final-presidential-estimate-obama-55-mccain-44.aspx

34.

Poynter Institute. (2008). Eyetracking the news: A study of print & online news. Retrieved from http://eyetrack.poynter.org/index.html

35.

Prior

(2009). The immensely inflated news audience: Assessing bias in self-reported news exposure. Public Opinion Quarterly, 73(1), 130-143.

36.

Purcell

Rainie

Mitchell

Rosenstiel

Olmstead

(2010). Understanding the participatory news consumer: How internet and cell phone users have turned news into a social experience. Washington, DC: The Pew Internet & American Life Project.

37.

Rhine

R. J.

(1967). The 1964 presidential election and curves of information seeking and avoiding. Journal of Personality and Social Psychology, 5(4), 416-423.

38.

Saad

(2009). Bush presidency closes with 34% approval, 61% disapproval. Retrieved from http://www.gallup.com/poll/113770/bush-presidency-closes-34-approval-61-disapproval.aspx

39.

Sears

D. O.

Freedman

J. L.

(1967). Selective exposure to information: A critical review. Public Opinion Quarterly, 31(2), 194-213.

40.

Smith

(2009). The Internet’s role in campaign 2008. Washington, DC: The Pew Internet & American Life Project.

41.

Stromer-Galley

(2003). Diversity of political conversation on the Internet: Users’ perspectives. Journal of Computer Mediated Communication, 8(3).Retrieved from http://www.ascusc.org/jcmc/vol8/issue3/stromergalley.html

42.

Stroud

N. J.

(2008). Media use and political predispositions: Revisiting the concept of selective exposure. Political Behavior, 30(3), 341-366.

43.

Sunstein

C. R.

(2001). Republic.com. Princeton, NJ: Princeton University Press.

44.

Ziemke

D. A.

(1980). Selective exposure in a presidential campaign contingent on certainty and salience. In Nimmo

(Ed.), Communication yearbook 4 (pp. 497-511). New York, NY: Transaction Books.

45.

Zillmann

(1985). The experimental exploration of gratifications from media entertainment. In Rosengren

K. E.

Wenner

L. A.

Palmgreen

(Eds.), Media gratifications research: Current perspectives (pp. 225–305). Beverly Hills, CA: SAGE.

46.

Zillmann

Bryant

(1985). Selective exposure to communication. Hillsdale, NJ: Erlbaum.