Persuasive Essay

Impact of Popularity Indications on Readers’ Selective Exposure

to Online News

Silvia Knobloch-Westerwick, Nikhil Sharma, Derek L. Hansen, and Scott Alter

Selecting news online may differ from traditional news choices, as most formal importance indicators in traditional media do not convert di- rectly to online news. However, online portals feature news recom- mendations based on collaborative filtering. To investigate how recom- mendations affect information choices, 93 participants browsed online news that featured explicit (average rating) or implicit (times viewed) recommendations or no recommendations (control group) while news exposure was logged. Participants picked more articles if the portal fea- tured explicit recommendations, and stronger explicit recommenda- tions instigated longer exposure to associated articles. Implicit recom- mendations produced a curvilinear effect with longer exposure for low and high numbers.

The Internet has greatly contributed to the so-called information tide (Graber, 1984) that news consumers face. Using the World Wide Web to access news has become commonplace, with nearly two thirds of the people who “get news” using online sources at least some of the time (Fallows, 2004). Thus, communication scholars need to address new phenomena in news consumption that are unique to information re- trieval from the World Wide Web. Initial explorations into how news consumers per- ceive print and online news have revealed that the audience applies largely parallel criteria to both outlets (Sundar, 1999). However, with regard to issue perceptions and learning from news, differences between print and online news did emerge. For ex-

© 2005 Broadcast Education Association Journal of Broadcasting & Electronic Media 49(3), 2005, pp. 296–313

296

Journal of Broadcasting & Electronic Media/September 2005

Silvia Knobloch-Westerwick (Ph.D., University of Music & Drama, Hanover, Germany) is an Assistant Profes- sor at the School of Communication, Ohio State University. Her research addresses media effects and message selection in entertainment media and news.

Nikhil Sharma (B.A., Chandigarh College of Architecture, India) is a doctoral student at the School of Informa- tion, University of Michigan at Ann Arbor. His research interests include collaborative sensemaking and infor- mation reuse.

Derek L. Hansen (B.A., Brigham Young University) is a doctoral candidate at the School of Information, Uni- versity of Michigan at Ann Arbor. His research interests include health communication and computer-sup- ported cooperative work.

Scott Alter (B.S., University of Michigan, Ann Arbor). His undergraduate concentration was communication studies, and his research examines factors of selective exposure to information.

ample, Althaus and Tewksbury (2002) demonstrated reduced agenda-setting effects for online news readers compared to readers of the print news version. Furthermore, Eveland and Dunwoody (2002) showed that reading Web news produces smaller learning effects than print news. The authors of both studies attributed these differ- ences to increased selectivity in online news consumption. Hence, the well-estab- lished phenomenon of selectivity in media consumption (Klapper, 1960) also seems relevant for online media. This highlights the importance of determining which fac- tors influence selective exposure to online news.

Interestingly, online news platforms offer their users filtering techniques to access the abundant information selectively. Two broad classes of selection devices can be differentiated (Cosley, Lam, Albert, Konstan, & Riedl, 2003): content-based, in which the reader may enter keywords or section preferences to create customized newspaper versions (as in many so-called Daily Me projects in the 1990s); and col- laborative filtering, in which previous readers’ opinions on content are employed (see Lasica, 2002a, 2000b). The latter approach can be seen in various popular on- line portals from which many people retrieve news (e.g., Yahoo! News and Google News) and also in online versions of established news media (e.g., USAToday.com and CNN.com). Because news consumption is often considered important for sur- veying and interpreting the social environment (Lippmann, 1922; McCombs & Shaw, 1972; Neuman, Just, & Crigler, 1992; Noelle-Neumann, 1974, 1977), it is in- triguing that online news platforms now incorporate collaborative filtering cues, such as most e-mailed, number of page views, and average (avg.) rating associated with specific articles. The social functions of news may be even more salient when cues of collaborative filtering are present and indicate what other news readers ap- preciate and consume.

It is plausible that user-based recommendations of news have strong impacts on se- lective exposure to news. Various possible impacts of this kind are investigated in this study. We will review differences of print and online news that are relevant in this context, address how online news platforms offer new features to “tame the informa- tion tide” (Graber, 1984), and state research questions for the subsequent empirical investigation.

News Selections in Traditional Versus Online News Presentations

News editors apply a relatively standardized set of criteria when selecting informa- tion for publication or broadcast in news outlets (e.g., Galtung & Ruge, 1965; Gans, 1979; White, 1950) and, furthermore, emphasize certain news reports by prominent presentation, for example, on the front page, with illustration, or as the first item in a newscast. For traditional news platforms, such formal indicators of newsworthiness have been demonstrated to guide information selections of news consumers (e.g., Garcia & Stark, 1991; Graber, 1984; McCombs & Mauro, 1977; Wolf & Grotta, 1985;

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 297

Zillmann, Knobloch, & Yu, 2001) and also the level of importance attached to a re- ported issue (e.g., Kiosis, 2004; Wanta, 1988). Recently, though, more and more news consumers turn to the World Wide Web to retrieve information (e.g., Fallows, 2004), as it provides abundant and timely news coverage, mostly free of charge (Deuze, 2003).

It is entirely possible that the very same factors trigger information selection in both online and print news, although selecting might be more convenient in online settings and thus more important. In fact, studies by Zillmann et al. (2001) and Knobloch, Hastall, Zillmann, and Callison (2003) revealed that images in both print and online news attract news consumers in like fashion. Studies by Knobloch-Westerwick and collaborators (Knobloch, Dillman Carpentier, & Zillmann, 2003; Knobloch, Hastall, Grimmer, & Brück, 2004; Knobloch, Patzig, & Hastall, 2002; Knobloch, Zillmann, Gibson, & Karrh, 2002; Knobloch-Westerwick, Dillman Carpentier, Blumhoff, & Nickel, 2005) investigated the personal utility of news content for online and print set- tings and found the same factors to trigger selective exposure to media content. How- ever, all these studies focused on content features.

Yet many unique formal features of online news displays are likely to promote dif- ferent news selection behaviors. It has been noted that well-established, formal indi- cators of issue importance are no longer visible when news reports are retrieved on- line (e.g., Tewksbury & Althaus, 2000). By and large, online news outlets and portals list headlines that include hyperlinks to access the actual article. The length of an arti- cle, a strong determinant for selective reading of print news (e.g., McCombs & Mauro, 1977), cannot be previewed before clicking the hyperlink to access the article page for further scrutiny. Thus, the display on the entering page should have substantial im- pact on what online news consumers actually read and what they ignore. In addition, unlike in print newspapers, these headlines usually do not differ in terms of typeface size. Thumbnail-sized illustrations often accompany news headlines, but any empha- sis is lost as most headlines carry such images (e.g., Google News). In brief, news con- sumers receive hardly any indication regarding which of the accessible news reports has greater importance in the view of the news editors.

Recommender Systems in Online News

Instead of using the indications that traditional media employ, many online news displays carry other importance cues for their reports. Due to the interactive nature of the World Wide Web, it is easy for online news site designers to include the current number of times a report has been viewed, which news items have been the most popular in any recent time span, or how readers explicitly rated a news item. In gen- eral, such “recommender systems” can be found on many Web and e-commerce sites (Cosley et al., 2003). Media products are especially likely to be evaluated using these systems, such as music at music.yahoo.com, books at amazon.com, and movies at imdb.com. A great variety of approaches can be used; for example, recommendations

298 Journal of Broadcasting & Electronic Media/September 2005

can be derived from explicit ratings based on evaluation scales or from implicit appre- ciation inferred from observed selections (Resnick, Iacovou, Sushak, Bergstrom, & Riedl, 1994). Furthermore, there are a number of evaluation scale formats, including “thumbs up/thumbs down,” “one to five stars,” or scales ranging from 1 to 100 (Cosley et al., 2003). More sophisticated collaborative filtering algorithms detect sim- ilar preference patterns among different users to predict the liking of content and thus provide individualized recommendations based on shared tastes. This approach was pioneered by the GroupLens system (e.g., Resnick et al., 1994) and first implemented for Usenet Newsgroups, where Usenet participants provide news items.

A basic, yet common version of such recommender systems can be found in many online news displays that indicate how many readers have viewed a report (implicit recommendations) or how these readers evaluated an article (explicit recommenda- tions; see Claypool, Le, Waseda, & Brown, 2001). These indications of liking or im- portance assessments, based on audience behavior and judgment, are likely to influ- ence news selections. It has often been argued and demonstrated that consumers of traditional news derive social perceptions from the news coverage (McCombs & Shaw, 1993), employing a so-called quasi-statistical sense (Noelle-Neumann, 1974, 1977). Mutz (1992) looked more specifically at portrayals of mass opinions in the me- dia and contrasted such sources of “impersonal influence” from unknown others with personal influences. In her conceptualization, sheer numbers are often the basis of impersonal influence, whereas personal influence stems from trust in familiar others. Popularity indications of online news can be considered a source of impersonal influ- ence. Now, given that the online news media offer immediate and seemingly objec- tive numeric indicators on what news coverage other readers consume and appreci- ate, it appears likely that online news readers will orient their news consumption along these lines, as discussed in the following.

Theoretical Perspectives

At first glance, it seems logical that news consumers should simply follow the rec- ommendations they encounter in online news outlets—essentially a bandwagon ef- fect with a popular trend producing even greater attraction (Sundar & Nass, 2001). Chaiken (1987) argued that people use the heuristic that, if many think an opinion is valid, the opinion is probably correct. Accordingly, they could just follow the evalua- tions and choices made by others, thereby limiting their own cognitive selection ef- forts. Affiliation motives may be another reason for bandwagon effects, even when re- lating to strangers (e.g., Byrne, 1961). However, there are theoretical considerations that speak for alternatives to bandwagon effects. The following three approaches il- lustrate this for cognitive aspects, interindividual differences, and situational percep- tions as intervening variables.

First, given that bandwagon effects may result from heuristic and cursory process- ing, cognitive preconditions could determine whether individuals follow cues on

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 299

common selections or pursue their own distinct interests instead. Someone who is more cognitively engaged or more used to news processing will be able to invest more resources into individual news selection and thus not simply follow the cues of others’ choices. This thought has been advocated by Petty and Cacioppo’s (1979) dif- ferentiation of central versus peripheral processing. When more cognitive resources are applied to the news selection process (central processing), people are less likely to follow the heuristics of popularity indications. This is ultimately an argument of why the bandwagon effect may not apply. The next two theoretical ideas suggest that in some situations people actually avoid jumping on the bandwagon.

Second, individuals sometimes seek distinctiveness, either privately or publicly, in order to enhance their self-concept. For example, according to Brewer’s (1991) opti- mal distinctiveness theory, membership in large groups will activate a person’s need for differentiation, whereas membership in smaller minority groups satisfies the indi- vidual’s need for distinctiveness through intergroup comparisons while fostering as- similation patterns. Thus, news consumers who consider themselves members of a large group may be motivated to read different material than peers. Alternatively, news consumers who consider themselves members of a minority group may be more likely to read the same material as peers. The type of news site (e.g., general purpose, specialty news) may even affect the particular group membership that is evoked in the reader’s mind. Perceived group membership and perceived relative group size would then function as intervening factors, resulting in different news reading patterns. It is important to note that the resulting behavior does not have to be observable by others. Individuals might enhance their self-concepts through private behaviors such as news selections by basing inferences about their selves solely on their own behavior with- out others’ feedback, for instance, through social comparisons (Festinger, 1954; Wills, 1981, 1991).

Third, somewhat similar to Brewer (1991), Snyder and Fromkin’s (1980) theory on uniqueness-seeking argues that people pursue a sense of moderate self-distinctive- ness and try to avoid being overly similar or dissimilar to others because such states are experienced as unpleasant. Uniqueness-seeking can be applied to various do- mains, but most research has been conducted in the realm of consumer choices (Lynn & Snyder, 2002), of which news selections could be considered a specific case. Snyder and Fromkin suggested that individuals differ regarding uniqueness-seeking. Hence, those with a low need for uniqueness-seeking may make their news choices closely in line with others’ recommendations, whereas those with a high unique- ness-seeking need might just show opposite patterns.

These three approaches illustrate that cognitive circumstances (central vs. periph- eral processing), situational perceptions (group membership and size), or inter- individual differences (uniqueness-seeking) can account for the possibility that some news users let popularity indications guide their selections, whereas others make in- dependent decisions or even seek out news that others neglected. We cannot discuss all possible theoretical accounts for such behavior variations (see Mutz, 1998), but it should be clear that “following the crowd” is not the only option. Furthermore, differ- ent patterns might emerge for different popularity indicators.

300 Journal of Broadcasting & Electronic Media/September 2005

Related Empirical Work

Considerations on possible impacts of news recommender systems can also be de- rived from related empirical work. These investigations will be reviewed along estab- lished dimensions for print and online news (Sundar, 1999): (a) representativeness (for the genre of news), (b) credibility, (c) timeliness, and (d) liking.

If news consumers want reports to be representative, in the sense of broadly impor- tant and newsworthy (see Sundar, 1999), highly personalized filtering systems may not be particularly valuable in the realm of news. Audience members actually want to be informed of what is widely considered relevant to make sense of their social envi- ronment (e.g., Neuman et al., 1992). This need is reflected by the representativeness dimension and may explain why “Daily Me” formats of online newspapers, display- ing personalized selections of news based on content-oriented filtering, have re- mained a niche (Redelfs, 1996). Instead, recommendations on a collaborative basis appear especially helpful for this matter. In fact, a study by Sundar and Nass (2001) supports the idea that news consumers like information better when it appears to be selected and appreciated by many, which could be interpreted as representativeness. In the study, participants were led to believe that the news they read resulted from a traditional news arrangement from an editor, from a computerized news compilation, from other users’ selections, or from the individual user selections. These source asso- ciations significantly influenced news users’ perceptions. News reports selected by other users of the online news platform received significantly better evaluations on the dimensions of liking, representativeness, and quality than when the same reports appeared to be compiled by a news editor or by the users themselves. These findings imply that recommendations of online news will also affect news consumption. How- ever, Sundar and Nass looked at news evaluations instead of actual consumption and declared different source constellations instead of directly investigating the effects of recommendations. Hence, no definite conclusion on effects of news ratings can be drawn from their findings.

Furthermore, as prompt delivery of information is a key characteristic of online news platforms, the audience might not appreciate explicit recommendations be- cause gathering such evaluations on a broad basis requires the sacrifice of time, which inevitably causes news to become less timely. However, positive recommen- dations could make news appear more credible. Knobloch, Sundar, and Hastall (2005) investigated with American and German participants whether credibility and timeliness indications for news instigated longer exposure to the associated informa- tion. Yet the news display in their experimental procedure did not indicate credibility with recommendations from other readers. Instead, the news reports were associated with news sources for which either low or high credibility had been established in a pretest. The results indicate that credibility does affect selective exposure to news. The experimental manipulation of timeliness consisted of different publishing times for the specific news reports. These indications influenced only American news con- sumers; German participants did not vary at all in their news exposure with the timeli- ness manipulation. Also in the same study, U-shaped curvilinear impacts of newswor-

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 301

thiness, indicated by numbers of related articles, materialized in like fashion for participants in both countries. Given that higher credibility can increase selective ex- posure to news, user recommendations may have a parallel impact.

Selective exposure impacts of liking indications such as recommendations, to our knowledge, have only been studied in an earlier exploration that we conducted (Knobloch, 2004; Sharma et al., 2004), which consisted of an Internet-based experi- ment in a field setting. In this research, effects of implicit and explicit ratings on news selections were explored. Participants, recruited via e-mail, browsed through an ex- perimental online news magazine while their selective exposure times were unobtru- sively logged. The article leads on the front page were associated with five levels of recommendation indicators, two articles for each level, either for times viewed, avg. rating, or no numerical information at all (control group). To control for influence of article topics, the numbers for times viewed and avg. rating were reversed for two ad- ditional experimental cells, thus resulting in five experimental groups: times viewed, times viewed reversed, avg. rating, avg. rating reversed, and control group. The find- ings showed that both the level and the kind of numerical information associated with articles had a significant impact on selective reading. Online news readers spent more time on articles associated with low numbers for page visits, contrary to expec- tations. Reading times were also affected by quality ratings associated with articles, supposedly provided from other readers. However, these effects were less clear—pos- sibly because the meaning of those ratings may have been somewhat ambiguous to news readers.

As it stands, little is known about the impact of recommendations on selection of online news. Based on the considerations outlined previously, this study, while im- proving the research design that Sharma et al. (2004) used in an earlier exploration, investigates the following research questions:

RQ1: Do recommendations influence selective exposure to online news articles?

RQ2: Do explicit recommendations (i.e., avg. rating) and implicit recommendations (i.e., times viewed) influence the selective exposure to online news articles differently?

Method

Overview

In a Web-based experiment with three experimental groups, 93 participants browsed through an online news platform. The news stories were accessible in a format commonly found on the World Wide Web: the front page showed a news lead for each article, with a headline and the first words of the article text, allowing readers to access actual articles by clicking on hyperlinks. The displayed articles were associated with recommendations at varying levels that appeared to be de- rived from either explicit ratings (avg. rating) or implicit evaluation (times viewed).

302 Journal of Broadcasting & Electronic Media/September 2005

A control group browsed through the same articles without any recommendation indication. The kind of recommendation, or lack thereof, served as between-factor, whereas selective exposure to articles associated with various recommendation lev- els provided repeated measures. After the scheduled browsing period, participants completed a brief questionnaire.

Respondents

Participants were recruited from an introductory course in communication at a large California university and received extra credit. Of the 93 respondents, 68 were female. The average age was 21.2 years. The majority (53%) indicated that they read online news at least once a week or more frequently. The three experimental groups included 30, 31, and 32 participants, with equal gender distributions.1

Procedure

Participants convened for the scheduled session in a computer lab, up to 12 at a time. The experimenter greeted them and asked them to be seated at a computer, while leaving ample space between occupied seats to ensure privacy. All 30 comput- ers in the lab were identical, with 17-inch flat-screen monitors. The experimenter in- formed participants that they would go through a computerized research session in which all further instructions would be given via the screen. Then participants were asked to start the software by clicking a menu icon. The experimenter remained pres- ent during the session.

A consent form was displayed on screen, and all participants gave their consent by clicking a button. Then participants accessed the browsing instructions, stating that they would see a test version of an online newsmagazine. They were informed that time would not allow them to read everything so they should browse through the articles and read whatever they found interesting. It was indicated that a ques- tionnaire would eventually be uploaded to collect information on impressions of the newsmagazine. By clicking “continue,” participants accessed the online news- magazine and browsed through articles for 2.5 minutes. After this time span, a tran- sition page stated that the scheduled reading time was up. A questionnaire col- lected information on respondents’ perceptions and demographic information. Finally, the screen showed a debriefing page before participants were dismissed.

Experimental Internet Newsmagazine

The computerized procedure was created using the programming languages PHP, JavaScript, and HTML. Throughout the procedure, all of the participants’ actions were logged to a MySQL database server. This online news platform had a similar

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 303

look and feel to popular news magazines currently on the Internet; in fact, it mim- icked typeface, font size, and font colors of common news outlets such as Google News (see Figure 1). A masthead of the name and logo of the experimental plat- form—“World Wide News, U.S. Edition”—was displayed across the top of the Web site. In addition, a navigation bar was placed on the left-hand side of the page. Al- though it was deactivated, the displayed navigation bar contained newspaper sec- tion headings such as sports, world news, and so forth, that would be commonly found on a news site. The main frame initially contained an overview, which listed all available articles by headline and news lead in two columns. Except for the control group, which was not presented with recommendations, all articles were associated with numerical information to indicate recommendations for avg. rating or times viewed. In the lower right corner of the main frame, a graphic declared what the recommendations meant. For avg. rating, the statement read: “About News Ratings: Ratings averaged from Reader Recommendations; 1 = Not at all Recommended — 5 = Highly Recommended.” For times viewed, the statement read: “About ‘Times Viewed’: Data indicate how many readers viewed a story; Up- dated Every Hour” (see Figure 1). For the control group, a graph that just repeated information from the masthead (“World Wide News, U.S. Edition—Test Version”) populated this spot on the screen.

The headlines of the displayed articles were “Mickey drawing leads man on quest,” “High school principal ponders loss of job,” “Student overcomes learning disability,” “West Nile Virus: Bracing cautiously for another season of mosquitoes,” “Behcet’s dis- ease: ‘I diagnosed my rare condition’,” “Civilian rescues driver,” and “eBay quest for cookie jar leads to long-lost sibling.” All reports were edited to the same length of 450 words. The seven articles were chosen based on prior findings indicating that they all attracted similar levels of exposure (Knobloch, 2004; Sharma et al., 2004). The page placement of the seven news articles was systematically rotated across participants so that each article had an opportunity to be placed in every position.

Except for the control group, all displayed articles were associated with numerical information to indicate recommendations. The recommendation indications were systematically rotated across article topics. In the avg. rating condition, the specific numbers were 1.3, 1.5, 3.0, 3.1, 3.3, 4.4, and 4.6. The numbers did not have identical differences in order to make the ratings look credible. The times viewed condition employed numbers in close proportion (multiplied by 50) to recommendations in the avg. rating version: 64, 75, 152, 155, 165, 221, and 232. Recommendation levels were also rotated on the overview page, although independently from article posi- tions. This process assured that the various recommendations were assigned to differ- ent articles, and the position of the articles on the page varied for each participant. The position of articles was likewise rotated in the control group.

The respondents made their reading selections by clicking on the hyperlinks to articles (i.e., on the articles’ headlines), scrolling through the selected articles and reading as much of them as they cared to, clicking to return to the overview, select- ing other articles (or returning to the abandoned ones), and so forth, until the end of the reading period.

304 Journal of Broadcasting & Electronic Media/September 2005

Behavioral Measures of Selective Exposure

Unbeknownst to the participants, their article choices were recorded in a database monitoring the use of hyperlinks. In addition, scripts were written to measure, in sec- onds, the time that elapsed between entering and leaving a particular article. This was done for every article. In the case of repeated entries and exits, the duration of the multiple exposures was accumulated to a single measure of exposure time. Scripts

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 305

Figure 1 Screen Shots of Front Page of Online News Portal Featuring Explicit

Recommendations (Avg. Rating) or Implicit Recommendations (Times Viewed)

were also used to administer the initial random assignment of experimental condi- tions to respondents. In addition to the primary measure of selective exposure time, the decision to read or ignore an article served as a secondary measure of selective ex- posure. Any exposure time above 1 second defined article selection.

Questionnaire

At the end of the preset reading time, the application jumped to a page soliciting in- formation about the respondents’ impressions of the articles. These questions mostly served to gain an understanding of how participants interpreted the recommenda- tions associated with news articles on the front page and their own use of this informa- tion. In detail, the following questions and response options, to be selected via radio buttons, appeared on screen:

The articles in the online newspaper you just read were associated with individual numeric information. Please answer the following questions regarding this numeric information:

1. What did the numeric information indicate for each article? (a) how many read- ers viewed the news story, (b) how strongly readers would recommend the report to others, (c) how many people have e-mailed the story, (d) unsure.

2. Please select the rating information that you would associate with a stronger rec- ommendation: (a) avg. rating 2.0, (b) avg. rating 4.0, (c) unsure.

3. All news stories were posted … (a) at the same time, (b) at different times, (c) unsure.

4. The numeric information on the reports influenced my reading selections … (a) not at all, (b) somewhat, (c) a great deal.

These questions, thus the entire screen page, were skipped for the control group. On the next screen, participants indicated how often they consumed online news:

“daily,” “several times each week,” “about once a week,” “several times a month,” “about once a month,” “several times a year,” “about once a year,” or “never.” Finally, demographic information was collected before a debriefing page appeared.

Results

Perceptions of Recommendations

In the times viewed condition, 88% of participants responded that the numerical information they had seen was to indicate “number of times the story was viewed,” and 12% chose the response “how many people have e-mailed the story.” In the avg. rating condition, 73% of respondents selected the correct response option of “individ- uals would recommend the story to others,” whereas 10% chose the times viewed re- sponse, and 17% selected “how many people have e-mailed the story.” The differ-

306 Journal of Broadcasting & Electronic Media/September 2005

ences between the distributions in the times viewed and avg. rating conditions were significant, χ2(2, N = 62) = 42.2, p < .001. These percentages attest that the experi- mental treatment was clear while not overly obtrusive.

When asked which rating would indicate better quality (either a rating of 2.0 or a rating of 4.0), 77% of the participants in the avg. rating condition chose the response that was in line with how the ratings had been explained in the news portal. In com- parison 50% of participants in the times viewed condition selected the same re- sponse, which shows that using higher rating numbers to indicate recommendations that are more positive is not self-explanatory. As the percentages differed significantly, χ2(1, N = 62) = 4.71, p = .030, participants in the avg. rating group obviously paid at- tention to the graphic explaining the recommendations.

Interestingly, a significantly larger share of respondents in the times viewed condi- tion (36% vs. 13%), χ2(1, N = 62) = 4.72, p = .030, assumed that the articles had been posted at different times, although the news portal had displayed an upload time that was said to apply to all stories. Regarding the perceived influence of the recommen- dations on one’s own news selections, the two experimental conditions did not differ (p = .27). In both groups, respondents mostly thought they had not been influenced by the recommendations at all (60%) or, if so, only somewhat (37%).

Impacts of Article Position on Front Page on Selective Exposure to News

In a mixed-design analysis of variance (ANOVA), with experimental groups as be- tween-factor, exposure times for the seven news articles placed in different spots on the news portal front page were incorporated as within-factor with seven repeated measures. Only an effect for position emerged, F(6, 540) = 3.36, p = .003, η2 = .036. Articles positioned first and second in the left column produced average reading times of 23 and 21 seconds, whereas reading times for articles in other spots ranged between 9 and 15 seconds. No other significant effects emerged, which means that effects of article positioning were independent of recommendations. As a side note, the lack of an impact of the experimental group shows that the recommendations as such did not produce different levels of total news exposure, that is, the reading time accumulated for all articles.

Impacts of Article Topic on Selective Exposure to News

Articles had been selected based on the premise that they should not instigate dif- ferent levels of interest. The article selection was derived from a set of reports used in prior research (Knobloch, 2004; Sharma et al., 2004) in which they had produced similar levels of selective exposure. An ANOVA with the current data, again with kind of recommendation as between-factor and exposure times as repeated measures, for specific articles now instead of front-page positions, was conducted. The various arti-

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 307

cle topics yielded similar levels of exposure times that were not significantly different (F < 1). An analysis with selective exposure times for various articles as repeated mea- sures only for the control group showed again that the articles did not attract readers to different degrees. Significance was not even approached, despite the statistical power of repeated measures. Thus, the presented article topics as such had no differ- ent levels of appeal for the current sample as well.

Impacts of Recommendations on Selective Exposure to News

A mixed-design ANOVA employed the kind of recommendation as between-factor and the exposure to news articles associated with different recommendation levels as within-factor with seven repeated measures. This analysis did not incorporate the control group, as no recommendations had been presented to these participants. A significant interaction between the kind and level of recommendation emerged, F(6, 360) = 3.13, p = .008, η2 = .05. For the repeated measures in the avg. rating condition, a linear effect materialized at p = .049, whereas the impact of implicit recommenda- tions in the times viewed emerged as quadratic effect at p = .001. As shown in Figure 2, higher explicit recommendations (i.e., avg. rating) resulted in longer reading times. In the times viewed condition, a curvilinear effect occurred, as both very low and very high numbers for implicit recommendations resulted in longer reading times for the corresponding reports. No other effects reached significance in this ANOVA.2

Impacts of Recommendations on News Article Selections

Furthermore, an ANOVA examined selective exposure based on article selections instead of exposure time. The same mixed design as previous was applied while again

308 Journal of Broadcasting & Electronic Media/September 2005

Figure 2 Impacts of Kind and Level of Recommendations of Online News

on Selective Exposure to News (in Seconds)

excluding the control group. A main effect of the kind of recommendation material- ized, F(1, 60) = 4.3, p = .042, η2 = .067, because participants in the avg. rating condi- tion clicked on more articles in general (M = 2.7 vs. 2.2). As for exposure time, the analysis yielded an interaction between the repeated measures for the recommenda- tion levels and the kind of recommendation, F(6, 360) = 3.3, p = .006, η2 = .052, as depicted in Figure 3. Implicit recommendations fostered article selection for low and high numbers, resulting in a curvilinear effect, F(1, 31) = 7.6, p = .010, η2 = .196. Arti- cle selections were also influenced by explicit recommendations in curvilinear fash- ion but showed an inverted U-pattern, F(1, 29) = 4.8, p = .037, η2 = .141, although the linear effect fell short of significance (p = .076).

Discussion

As news consumers face an enormous amount of information to choose from on the World Wide Web, the current investigation examined how collaborative filtering rec- ommendations affect online news exposure. The results clearly show that both im- plicit (times viewed) and explicit recommendation (avg. ratings) affect selective expo- sure to online news. These recommendations influenced which articles the news consumers in our Web-based experiment selected and how much time they spent on the reports. Interestingly, the kind of recommendation did not affect how much time participants generally spent on reading news instead of scrutinizing the front page, but explicit recommendations fostered selections of more articles.

The two kinds of recommendations, explicit and implicit, also influenced selective exposure to news in different ways. Online readers preferred articles that had been viewed by either few or many other people (as indicated by the times viewed recommendation). On the other hand, articles with better explicit recom-

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 309

Figure 3 Impacts of Kind and Level of Recommendations of Online News

on Selection of News Articles

mendations (avg. rating) were read longer than articles with poor explicit recom- mendations.

The result that the least popular (i.e., lowest times viewed) articles were read most frequently replicated findings of an earlier exploration (Knobloch, 2004; Sharma et al., 2004). One potential explanation for this phenomenon is that participants be- lieved that the articles with fewer readers were more recently posted. Although in the present study, all news reports were explicitly associated with the same upload time, a significantly higher portion in this condition believed that the stories were posted at different times. Thus, it is still possible that some readers prefer news that merely a few others have read because they appear more recent.

Another possible explanation for the desire to read either very popular news or the least popular news articles is that some participants were looking for a “rare gem” be- cause these choices could add to news consumers’ self-uniqueness experience. By making less popular news selections, these media users may have felt more “special” while not risking any consequences of violating social norms. This could help explain our findings that articles with either high or low times viewed produced longer read- ings times and were selected more frequently. News readers may either want to read what others have favored and feel bonded or do the opposite to add to an experience of self-uniqueness (Snyder & Fromkin, 1980). Likewise, when applying the optimal distinctiveness theory (Brewer, 1991) to these findings, it appears that some news readers may have regarded themselves as members of a large group and were inclined to set themselves apart through their choices. Yet, interestingly, the explicit recommendations (avg. rating) did not trigger different selection patterns, as people followed these endorsements in fairly uniform fashion. Future research should investi- gate more specifically what factors result in these different selection patterns.

This study has some limitations, as it was conducted in a lab setting and employed a highly homogenous sample. However, the fact that the results reflected those of our previous study (Knobloch, 2004; Sharma et al., 2004), which included a far more di- verse population in a field setting, is suggestive that the findings presented here gener- ally apply. Furthermore, to remove confounding effects, the present study used arti- cles that were apparently equally interesting, as they yielded similar exposure times in the control group. It is important, therefore, to remember that the effect of recommen- dations may not be the same for articles that are extremely boring or eye-catching. Other characteristics of news stories may have a significant interaction with recom- mendations as well. For example, a high times viewed message may encourage more or less reading depending on if the article were popular entertainment versus health information. In this study, the fact that readers knew that they had limited time may have encouraged selective behavior. However, research has indicated that most ac- tual online news reading is, in fact, selective and of short duration (e.g., Harris Inter- active, 2004). Future work could certainly help clarify some of these important issues.

This study has focused on two specific types of recommendations because of their prevalence, but other types of both explicit and implicit ratings exist that could be ex- plored. For example, implicit ratings that measure times e-mailed, times printed, and

310 Journal of Broadcasting & Electronic Media/September 2005

times saved are all currently used on some news Web sites in addition to times viewed. These implicit ratings signal very different uses of an article and likely pro- mote different behavior when encountered. Likewise, different explicit ratings such as credible or interesting could be used in addition to recommended. A natural exten- sion of this work would be to measure the effect of some of these measures on selec- tive exposure. This would allow Web site designers to use better understood rating options that may encourage certain types of selective exposure.

The widely found recommendation systems in news platforms, in light of our re- sults, do affect which news reports are consumed and which are ignored. It appears, though, that interpretations of recommendation meanings are anything but equivo- cal. Although statistically significant patterns exist, they may not always match news Web site designers’ intuitions. The various forms of recommendations (e.g., most e-mailed, most viewed, highest rated) might instigate very diverse behaviors. More- over, news consumers may hold suspicions about the raters’ motivations because giv- ing recommendations for any item online might be driven by various intentions of the rater. Online users are in part aware of such factors and might distrust some recom- mendation systems for this reason (Cosley et al., 2003). Hence, the findings presented here may just be a starting point for investigating a highly complex process through which people choose public affairs information online.

Notes

1The data of the control group were collected a few months later than the data for the two other groups. The articles were not time sensitive, and the exact same setting and procedure were used.

2Additional analysis ruled out that these results were based on extreme cases.

References

Althaus, S. L., & Tewksbury, D. (2002). Agenda setting and the new news: Patterns of issue im- portance among readers of the paper and online versions of the New York Times. Communi- cation Research, 29, 180–207.

Brewer, M. B. (1991). The social self: On being the same and different at the same time. Person- ality and Social Psychology Bulletin, 17, 475–482.

Byrne, D. (1961). Interpersonal attraction as a function of affiliation need and attitude similarity. Human Relations, 14, 283–289.

Chaiken, S. (1987). The heuristic model of persuasion. In M. P. Zanna, J. M. Olson, & C. P. Hermann (Eds.), Social influence: The Ontario symposium (Vol. 5, pp. 3–39). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Claypool, M., Le, P., Waseda, M., & Brown, D. (2001). Implicit interest indicators. Proceedings of ACM Intelligent User Interfaces Conference (IUI). Santa Fe, NM. Retrieved March 3, 2004, from http://www.cs.wpi.edu/~claypool/papers/iii/

Cosley, D., Lam, S. K., Albert, I., Konstan, J., & Riedl, J. (2003). Is seeing believing? How recom- mender systems influence users’ opinions. Proceedings of CHI 2003 Conference on Human Factors in Computing Systems (pp. 585–592). Fort Lauderdale, FL. Retrieved March 3, 2004, from http://www.cs.umn.edu/Research/GroupLens/publications.html

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 311

Deuze, M. (2003). The Web and its journalisms: Considering the consequences of different types of news media online. New Media & Society, 5, 203–230.

Eveland, W. P., & Dunwoody, S. (2002). An investigation of elaboration and selective scanning as mediators of learning from the Web versus print. Journal of Broadcasting & Electronic Me- dia, 46, 34–53.

Fallows, D. (2004). The Internet and daily life: Many Americans use the Internet in everyday ac- tivities, but traditional offline habits still dominate. Pew Internet & American Life Project. Re- trieved October 19, 2004, from http://www.pewinternet.org

Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140. Galtung, J., & Ruge, M. H. (1965). The structure of foreign news: The presentation of the Congo,

Cuba and Cyprus crises in four Norwegian newspapers. Journal of Peace Research, 2, 64–91. Gans, H. (1979). Deciding what’s news. New York: Pantheon. Garcia, M., & Stark, P. (1991). Eyes on the news. St. Petersburg, FL: Poynters Institute. Graber, D. (1984). Processing the news: How people tame the information tide. New York:

Longman. Harris Interactive. (2004, May 19). Most Americans who are online use Internet for news, but

most say this does not reduce their use of other news media. The Harris poll #35. Retrieved October 18, 2004, from http://www.harrisinteractive.com/harris_poll/index.asp?PID=464

Kiosis, S. (2004). Explicating media salience: A factor analysis of New York Times issue coverage during the 2000 U.S. presidential election. Journal of Communication, 54(1), 71–87.

Klapper, J. T. (1960). The effects of mass communication. Oxford, England: Free Press of Glencoe.

Knobloch, S. (2004, May). Social determinants of selective exposure to media information: Re- cent Web-based experimental investigations. Panel presented at the International Communi- cation Association Conference, New Orleans, LA.

Knobloch, S., Dillman Carpentier, F., & Zillmann, D. (2003). Effects of salience dimensions of in- formational utility on selective exposure to online news. Journalism & Mass Communication Quarterly, 80, 91–108.

Knobloch, S., Hastall, M., Grimmer, D., & Brück, J. (2004, May). Predicting selective exposure to media information: The efficacy dimension in the informational-utility model. Paper pre- sented at the International Communication Association Conference, New Orleans, LA.

Knobloch, S., Hastall, M., Zillmann, D., & Callison, C. (2003). Imagery effects on the selective reading of Internet newsmagazines: A cross-cultural examination. Communication Re- search, 30, 3–29.

Knobloch, S., Patzig, G., & Hastall, M. (2002). Informational Utility: Einfluss von Nützlichkeit auf selektive Zuwendung zu negativen und positiven Online-Nachrichten [Informational utility: Impact of utility on selective exposure to negative and positive online news]. Medien- & Kommunikationswissenschaft, 50, 359–375.

Knobloch, S., Sundar, S. S., & Hastall, M. (2005, May). Clicking news: Impacts of newsworthi- ness, source credibility, and timeliness as online news features on news consumption. Paper presented at the International Communication Association Conference, New York.

Knobloch, S., Zillmann, D., Gibson, R., & Karrh, J. (2002). Effects of salient news items on infor- mation acquisition and issue perception. Zeitschrift für Medienpsychologie, 14, 14–22.

Knobloch-Westerwick, S., Dillman Carpentier, F., Blumhoff, A., & Nickel, N. (2005). Informa- tional utility effects on selective exposure to good and bad news: A cross-cultural investiga- tion. Journalism and Mass Communication Quarterly, 82, 181–195.

Lasica, J. D. (2002a, April 2). The promise of the Daily Me. Online Journalism Review. Retrieved April 1, 2004, from http://www.ojr.org/ojr/lasica/1017779142.php

Lasica, J. D. (2002b, April 2). The second coming of personalized news. Online Journalism Re- view. Retrieved April 1, 2004, from http://www.ojr.org/ojr/lasica/1017779142.php

Lippmann, W. (1922). Public opinion. New York: Harcourt, Brace. Lynn, M., & Snyder, C. R. (2002). Uniqueness seeking. In C. R. Snyder & S. J. Lopez (Eds.), Hand-

book of positive psychology (pp. 395–410). London: Oxford University Press.

312 Journal of Broadcasting & Electronic Media/September 2005

McCombs, M. E., & Mauro, J. B. (1977). Predicting newspaper readership from content charac- teristics. Journalism Quarterly, 54, 48–49.

McCombs, M. E., & Shaw, D. L. (1972). The agenda-setting function of mass media. Public Opinion Quarterly, 36, 176–187.

McCombs, M. E., & Shaw, D. L. (1993). The evolution of agenda-setting research: Twenty-five years in the marketplace of ideas. Journal of Communication, 43(2), 58–67.

Mutz, D. (1992). Impersonal influence: Effects of representations of public opinion on political attitudes. Political Behavior, 14, 89–122.

Mutz, D. (1998). Impersonal influence: How perceptions of mass collectives affect political atti- tudes. New York: Cambridge University Press.

Neuman, W., Just, M., & Crigler, A. (1992). Common knowledge: News and the construction of political meaning. Chicago: University of Chicago Press.

Noelle-Neumann, E. (1974). The spiral of silence: A theory of public opinion. Journal of Com- munication, 24(2), 43–51.

Noelle-Neumann, E. (1977). Turbulences in the climate of opinion: Methodological application of the spiral of silence theory. Public Opinion Quarterly, 41, 113–151.

Petty, R. E., & Cacioppo, J. T. (1979). Issue involvement can increase or decrease persuasion by enhancing message-relevant cognitive responses. Journal of Personality and Social Psychol- ogy, 37, 1915–1926.

Redelfs, M. (1996). Electronic publishing und computer-assisted reporting: Auswirkungen des Information Superhighway auf den Journalismus. In H. J. Kleinsteuber (Ed.), Der Information Superhighway: Amerikanische Visionen und Erfahrungen (pp. 257–276). Opladen, Ger- many: Westdeutscher Verlag.

Resnick, P., Iacovou, N., Sushak, M., Bergstrom, P., & Riedl, J. (1994). GroupLens: An open ar- chitecture for collaborative filtering of netnews. In Proceedings of 1994 Computer Supported Collaborative Work Conference (pp. 175–186). New York: ACM Press.

Sharma, N., Hansen, D., Buis, L., Carter, T. J. A., Ng, I., Alter, S., et al. (2004, May). It’s all in the ratings: Impact of ratings on selection of online news articles. Panel paper presented at the In- ternational Communication Association Conference, New Orleans, LA.

Snyder, C. R., & Fromkin, H. L. (1980). Uniqueness: The human pursuit of difference. New York: Plenum.

Sundar, S. (1999). Exploring receivers’ criteria for perception of print and online news. Journal- ism & Mass Communication Quarterly, 76, 373–386.

Sundar, S., & Nass, C. (2001). Conceptualizing sources in online news. Journal of Communica- tion, 51(1), 52–72.

Tewksbury, D., & Althaus, S. L. (2000). Differences in knowledge acquisition among readers of the paper and online versions of a national newspaper. Journalism and Mass Communication Quarterly, 77, 457–479.

Wanta, W. (1988). The effects of dominant photographs: An agenda-setting experiment. Journal- ism Quarterly, 65, 107–111.

White, D. M. (1950). The gatekeeper: A case study in the selection of news. Journalism Quar- terly, 41, 384–390.

Wills, T. A. (1981). Downward comparison principles in social psychology. Psychological Bulle- tin, 90, 245–271.

Wills, T. A. (1991). Similarity and self-esteem in downward comparison. In J. Suls & T. A. Wills (Eds.), Social comparison: Contemporary theory and research (pp. 51–78). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.

Wolf, R., & Grotta, G. (1985). Images: A question of readership. Newspaper Research Journal, 6(2), 30–36.

Zillmann, D., Knobloch, S., & Yu, H. S. (2001). Effects of photographs on the selective reading of news reports. Media Psychology, 3, 301–324.

Knobloch-Westerwick et al./POPULARITY INDICATIONS AND NEWS EXPOSURE 313