Selecting Science Information in Web 2.0: How Source Cues, Message Sidedness, and Need for Cognition Influence Users' Exposure to Blog Posts

Abstract

This research investigates how cues describing the authors of user-generated online science articles in blogs, and indications about whether the articles are 1-sided or 2-sided, affect others' decision about which content to read. It extended the elaboration likelihood model (ELM; Petty & Cacioppo, 1986), to predict whether better-quality arguments and individuals' need for cognition affected their content selections. In 2 experiments, 121 parents were asked to search for information on a blog concerning the effects of violent media. Results showed a general preference for texts composed by users with greater expertise and for 2-sided messages. Need for cognition magnified the effect of message sidedness, suggesting that the ELM is relevant for blogs and the selection of user-generated science stories.

The huge amount of content provided on the Web has led to renewed interest in the issues of credibility and quality of information (Metzger, 2007). On participatory websites such as blogs, forums, or wikis, one increasingly finds information that has been communicated by laypersons rather than experts or professional journalists. These user-generated statements may offer new insights and supplementary information to more fact-based articles in traditional media, but some of these sources may also be less reliable.

From the recipients' perspective, the Internet and especially Web 2.0 provide new chances to gather information on relevant topics (Flanagin & Metzger, 2008). Although this can be seen as a major advancement, a great degree of media literacy appears to be necessary to benefit from these opportunities. Preliminary challenges in this context are to find the most reliable or useful content amid the overwhelming amount of information, to assess credibility, and to decide what content to read more closely. Against this background, this article investigates several factors that influence readers' selection of user-generated content on participatory websites. Although previous research on information-seeking and selective exposure focused on topical relevance (Pirolli & Card, 1999) or attitude consistency (Knobloch-Westerwick & Meng, 2009), the present work adapts social psychological theories and research on persuasion. Specifically, our focus lies on the impact of source characteristics in terms of the expertise of the author and message sidedness of the information, which ranges from one-sided and unidimensional to two-sided and more balanced depictions of topics. In doing so, we analyze whether the mechanisms of the elaboration likelihood model (ELM; Petty & Cacioppo, 1986; Petty, Rucker, Bizer, & Cacioppo, 2004), originally describing persuasion and attitude change, can be applied to the selection of user-generated content.

The investigation employs readers' information selection from examples of science blog postings. In the field of science communication, debates typically involve conflicting viewpoints, and interested parties generally have to rely on others to reach an informed point of view (Friedman, Dunwoody, & Rogers, 1999). Science blogs (e.g., www.scienceblogs.com) offer content generated by different authors on topics such as health, climate change, or the influence of violence in the media on adolescent media consumers. The latter topic is used as an exemplary scenario for our experiments.

The ELM serves as a theoretical basis since it allows the interplay of source and message factors to be described in relation to readers' characteristics in order to predict certain outcomes. Within this framework, we will discuss the role of message sidedness and source information in the selection of content on participatory websites. A personality characteristic which is also likely to influence readers' content selections, the motivational disposition of need for cognition, is also taken into account. We present results of two experimental studies that analyzed selective exposure by 121 adult readers to blog posts on a science webpage.

Selective Exposure to Media Content

The question of which factors determine selective exposure has been aptly described as crucial, because a prerequisite for any media effect is that the recipients have to select a specific media contribution (Pease, Brannon, & Pilling, 2006). Based on the theory of cognitive dissonance (Festinger, 1957), research on selective exposure in traditional media has concentrated on the consistency between media messages and media users' attitudes. A recent meta-analysis confirmed a tendency to avoid counter-attitudinal information (Hart et al., 2009); however, this effect is relatively small and can be overridden by several factors, e.g., headlines that emphasize the importance of a topic or its high news value (Donsbach, 1991). These patterns have also been shown for the usage of online information (e.g., Johnson, Bichard, & Zhang, 2009; Knobloch-Westerwick & Meng, 2009). The presence of selective exposure with regard to the use of search engines has been approached through cognitive models and mainly focused on the topical relevance of documents in relation to readers' goals (Pirolli & Card, 1999). However, given that within Web 2.0 applications, “authority is no longer a prerequisite for content provision” (Metzger, 2007, p. 2078), it is likely that users' content selection also involves consideration of the credibility of those users who generated the science articles and other online information. Therefore, our research focuses on the role of information about the sources who contributed blog postings, and the apparent message sidedness of the content.

Application of the Elaboration Likelihood Model

In order to analyze the effects of message sidedness and source information, we utilize the propositions of the ELM (Petty & Cacioppo, 1986; Petty et al., 2004). The ELM originally pertained to processes of persuasion and attitude change. We argue that the application of this theory to the analysis of selection may help to explain the increasingly important process of how readers make decisions about what to read in Web 2.0, since the ELM offers a fruitful basis to describe the interplay of source, message, and audience characteristics. This research may contribute not only to providing a new theoretical basis for explaining content selection in Web 2.0 settings, but it may also expand the theoretical scope of the ELM.

With regard to persuasion, the ELM posits that, when presented with potentially persuasive information by a source, people who are motivated and able to engage in effortful information processing carefully examine the information. Such receivers are mainly influenced by characteristics of the message (quality of arguments), about which their thinking elaborates. This kind of information processing is referred to as the central route. Under low elaboration, that is, when receivers lack ability or motivation, they engage in superficial processing that is predominantly influenced by peripheral cues, that is, cues about the source of the information rather than the content of the information itself. Applied to the setting of science blogs, source information (i.e., the author of the specific blog post) can be regarded as a peripheral cue requiring little mental effort to assess, whereas the characteristics of the message constitute central processing and require more effort. If the processes that the ELM reports for attitude change are also relevant for information selection (when deciding which articles to read more closely), this would mean that users who are able and motivated to evaluate information will be more likely to focus on the characteristics of the message. They should be attracted to high-quality articles, one assessment of which may come from estimating whether the article will provide a two-sided and differentiated overview of the topic. Users with less ability and motivation to process information may primarily rely on source cues, preferring articles from credible sources, and pay less attention to the probable message characteristics. This is the adaptation of the ELM to content selection, based on different levels of the readers' motivation to engage in careful information processing, that drove the present research.

We will first review research concerning message sidedness and source information and the potential influence of these two factors on the selection of content. Then, with regard to the mechanisms of the ELM, we will take into account need for cognition, as a personality characteristic which influences elaboration likelihood.

The Influence of Message Sidedness

In the setting of science blogs, the one or two-sidedness of a content contribution may be visible in the headline or summary of a blog post. Headlines consist of three to seven words, and summaries (in the present study) contain between 32 and 44 words encapsulating the article to which these indicators link. Despite their brevity, these headings reflect important message characteristics which readers may equate with the (expected) qualities of the blog post they accompany: The sidedness of the associated content can be visible in the headline, e.g., “Dangerous effects of horror movies” (one-sided) vs. “TV for children - pros and cons” (two-sided).

According to Allen (1991), two-sided messages include evidence and arguments for both positions on a controversial issue, whereas one-sided messages only provide information supporting one particular position. The effects of message sidedness have been investigated mainly in the area of persuasion and in the context of advertisements. For instance, in some of the earliest studies by Hovland, Lumsdaine, and Sheffield (1949), two-sided messages were more persuasive than one-sided messages when the audience was highly educated and initially had a negative attitude towards the advocated position. However, meta-analyses on sidedness effects did not confirm effects of education and attitude as moderators (Allen, 1991; O'Keefe, 1999). Two-sided messages garner a general advantage in terms of credibility, since their speakers are likely to be perceived as more honest (Crowley & Hoyer, 1994; Eisend, 2007). Because they acknowledge the opposing side of an issue, they may inoculate receivers against counterpersuasion (see for review Insko, 1967). For these reasons two-sided messages are considered to be stronger than one-sided messages. One-sided messages are regarded as weaker since, among other reasons, they omit opposing arguments, rendering them more prone to counterpersuasion (Hale, Mongeau, & Thomas, 1991).

The strength or weakness of two-sided and one-sided messages, respectively, may apply beyond the persuasion context, and function as characteristics that affect one's decision about which messages to choose for further elaboration. A two-sided blog headline or summary which indicates that both positions on a controversial issue are being considered may appear more attractive to readers who are motivated to reach an informed position. This may be particularly true in the context of science-related debates, which are often characterized by conflicting positions, complex patterns of evidence, and degrees of uncertainty (Friedman et al., 1999). A two-sided overview of a scientific debate may offer greater value (Hagerty & Aaker, 1984) or more utility (Atkin, 1973) for a reader than a statement that advocates only one particular position. On the other hand, despite the advantages offered by two-sided messages, two-sided summaries may deter readers from inspecting their postings further since they require readers to weigh arguments on conflicting viewpoints (Hale et al., 1991), which may be difficult for laypersons (Bråten & Strømsø, 2006), and may challenge readers' pre-existing attitudes in opposing arguments.

The Influence of Source Cues

Whereas a strong, two-sided message is expected to appeal to readers with greater motivation to learn about a particular issue, readers who are less motivated may be attracted to user-generated blog postings based on other factors, such as the blog writer's source characteristics. The attributes of a message's source are relevant when assessing the credibility of a posting. For instance, if the author of a blog post has a profession that has a close connection to the topic, this pertains to the expertise dimension of credibility (Hovland, Janis, & Kelley, 1953). Persuasion research generally finds that messages presented by expert sources are more credible and lead to greater attitude changes in comparison to messages delivered by nonexpert sources (Wilson & Sherrell, 1993). Within the ELM framework, expertise has been shown to be particularly relevant to message recipients who are inclined to process advocacies in a peripheral manner. That is, expertise cues have their greatest effect on readers who are not interested (i.e., motivated) or able to engage in extensive information processing regarding a topic (e.g., Petty, Cacioppo, & Goldman, 1981).

In terms of this principle's application in the selection, rather than processing, of information, Sundar, Knobloch-Westerwick, and Hastall (2007) demonstrated that reputational characteristics of newspapers – which may be equivalent to expertise information – influence the likelihood of a reader selecting that newspaper's story within online news aggregators (such as Google News). Focusing on science blogs, Winter, Krämer, Appel, and Schielke (2010) showed that blog postings that are attached to expert authors (that is, authors who present themselves as having professions involving expertise on the discussion topic) are perceived as more credible and chosen more frequently for further reading. Aggregated positive or negative ratings from other users about authors had limited influence (the effects of which were restricted to nonexpert authors). The author's profession appears to be a more relevant marker of expertise affecting message selection than are ratings by others. Although expertise claims may be easy to falsify on the Internet (Donath, 1999), they nevertheless appear to drive message selection, at least by relatively less motivated readers.

Information Selection and Need for Cognition

The ELM, as discussed so far, allows for topic-by-topic differences in readers' ability and motivation to learn about a particular issue, one result of which is different levels of message elaboration an individual is likely to commit, and hypothetically, which messages (one- or two-sided) they are inclined to select for reading. The ELM has also admitted a stable individual difference characteristic that appears to affect readers' motivational disposition to engage in and enjoy analytical and effortful thinking: the need for cognition (NC; Cacioppo & Petty, 1982; Petty, Briñol, Loersch, & McCaslin, 2009). Within the ELM framework, research shows that people with greater NC more often tend to process information on the central route and are motivated to consider arguments thoroughly (Cacioppo, Petty, Feinstein, & Jarvis, 1996). Furthermore, individuals with greater NC paid more attention to and more often recognized the inconclusiveness of scientific evidence when reading a two-sided essay about a controversial issue, than did readers with lesser NC (Kardash & Scholes, 1996). With regard to selection of media content, the influence of NC and its orientation toward consideration of a message's strength, rather than the characteristics of a message's source, may also affect decisions about which user-generated texts to select for further processing.

Study 1

Objective and Hypotheses

The goal of the first study was to explore the effects of message sidedness on the selection of blog posts, in the absence of information about the source of the post. Based on the theoretical background described above predicting the appeal of two-sided messages, we hypothesized that:

H1: Two-sided blog articles are selected more often, read for longer, and selected earlier than one-sided articles.

With regard to the influence of need for cognition and its relation to message characteristics, we hypothesized that:

H2: Need for cognition influences the users' information selection with regard to message sidedness: People with greater NC show a stronger preference for two-sided over one-sided articles.

An experimental setting was created in which participants were asked to search for information on a science weblog. As an exemplary scenario, the blog dealt with the controversy regarding the effects of media depictions of violence on children and adolescents. The baseline motivation to attend to stronger messages, which is usually a matter of personal relevance when considering various topics, was elevated by choosing a group of participants for whom the topic of the experiment is important. This may enhance the study's external validity since it can be expected that these participants are initially interested in the topic, as would be most likely for actual users who read a science blog.

Method

Sample

Participants were 61 parents who had children between ages 2 and 18. For this group, the topic of violent media effects is likely to be personally relevant. Participants were recruited via newspaper ads and press releases in Duisburg, Germany. As an incentive to participate, they were offered 25 Euro to be in the experiment. Respondents (32 female, 29 male) were between ages 28 and 49 years (M = 40.72, SD = 4.55). Twenty-five percent of them had a university degree, 19.7 % had finished high school with a university entrance-level qualification, and 55.7 % had a lower level of education.

Design

A science weblog was created as stimulus material (see Figure 1). An overview page displayed headlines and summaries of eight texts dealing with media effects. Clicking on the respective link enabled participants to read the whole article. Author information was limited to nicknames, and the placement of presentation was rotated. No expertise cues about the authors were presented in Study 1, in order to ensure that message sidedness effects were not obscured by peripheral source characteristics.

The sidedness of information and the stance of the texts were systematically varied as within-subject variables. That is, half of the eight article headings, and the articles themselves, contained two-sided material (including limitations and counterarguments) and the other half contained only arguments for one position. With regard to the position that the article advocated, four articles suggested rather positive effects of media reception and four advocated rather negative effects. This arrangement resulted in four categories (with two texts in each category in order to minimize topic effects): one-sided information favoring of media usage by children and teenagers (e.g., the headline “Online games improve team skills”), one-sided information against media usage (e.g., “Dangerous effects of horror movies”), and two-sided information (e.g., “TV for children – pros and cons”), with a summary either expressing a predominance of negative media effects or with a summary emphasizing predominantly positive effects.

User characteristics

Participants' NC, that is, their tendency to enjoy effortful thinking was measured as a motivational disposition with the well-established NC scale developed by Cacioppo and Petty (1982), presented in its German version (Bless, Wänke, Bohner, Fellhauer, & Schwarz, 1994). It consists of 16 Likert-scale items such as “I really enjoy a task that involves coming up with new solutions to problems” and “I only think as hard as I have to” (reverse-coded), anchored from 1 to 5, and summed (Cronbach's α = .84). Participants had a mean NC score of 58.62 (SD = 8.85).

As a further moderating variable, we assessed the participants' attitude toward the topic of media usage. This was measured with seven original items¹ such as “Computer games have negative consequences on children and teenagers” (α = .74).

Pilot studies with the stimulus material

For the webpage, summaries and corresponding long versions of articles which differed in sidedness and stance but not with regard to comprehensibility and style were chosen based on the results of two online pilot studies. First, 16 headlines and summaries (four in each text category) were created. An additional 25 participants (age: M = 32.64, SD = 10.84; 13 female) rated the degree to which each headline and three-sentence article summary was one-sided to balanced (one semantic differential from −5 to 5), and which position with regard to media effects it advocated (positive vs. negative effects, two items, semantic differentials from -5 to 5, scores for every text were calculated: Cronbach's α between .71 and .98). Furthermore, the participants evaluated how comprehensible the text was (Likert scale from 1–7). Based on the results, eight summaries were selected which represented the categories in a complete array of one-sided/pro, one-sided/con, two-sided/mostly pro, and two-sided/mostly con (see Table 1). A repeated-measures analysis of variance (ANOVA) confirmed a significant effect of intended message sidedness on subjects' perceptions of one-/two-sidedness, F (1, 24) = 87.85, p < .001, η_p² = .79. There was also a significant effect of the direction of the arguments on subjects' perception of the text's stance, F (1, 24) = 134.09, p < .001, η_p² = .85. The texts did not differ significantly with regard to comprehensibility.

For the second pilot test, we developed blog articles on the chosen topics that resembled those that are typical in science blogs. Articles spanned 180 to 210 words. All texts consisted of seven information units: One-sided texts had seven arguments for the advocated position, and two-sided texts had four arguments for the predominant position and three arguments supporting the opposing position, and arguments were not directly refuted. Similar to the first pretest, 33 respondents (age: M = 27.61, SD = 7.98; 22 female) assessed perceived one- vs. two-sidedness and stance as well as comprehensibility of the articles. The results supported the intended manipulation: A significant effect of message type affected the assessment of one-/two-sidedness, F (1, 32) = 105.38, p < .001, η_p² = .77, as did a significant effect of the direction of the arguments affect perceptions of the text's stance, F (1, 32) = 185.93, p < .001, η_p² = .85. As intended, comprehensibility did not differ between the categories.

Procedure

The main experiment was conducted in a laboratory room of the University of Duisburg-Essen, Germany. First, the participants completed an online questionnaire in which their user characteristics were assessed. Following this, they were asked to search for information on the specific topic by reading the weblog. Reading time was limited to 3½ minutes in order to create a situation in which participants were forced to choose. Screen-recording software unobtrusively saved participants' behaviors for later analysis.

Dependent measures

Coding the screen recordings allowed identification of which texts were chosen (in which order) and for how long they were read, using the following measures. (1) Frequency of selection (number of clicks) from among the two-sided blog articles and the one-sided blog articles: These data generated a score for participants' preference for two-sided over one-sided blog articles by subtracting the number of one-sided text selections from the number of two-sided text selections. (2) Reading time for each article: The number of seconds between the time an article was clicked to open it and when it was closed generated these scores. (3) Order of article selection: If the text was chosen first, it was coded as 1, and increasing correspondingly for the other positions. If an article was not chosen, it was coded as 8 (last position). Furthermore, although not hypothesized, the coding generated the number of selected articles in favor of (pro) and against media effects (con).

Results

The participants chose an average of 3.48 articles (SD = 1.65) for further reading.

In order to investigate the effects of the text variables, ANOVA examined the effects of message sidedness and pro or con stance as within-subject factors (repeated measures) on article selection and reading time. With regard to frequency of selection as the dependent measure, a significant effect of message sidedness emerged, F (1, 60) = 6.66, p = .012, η_p² = .10. Scores indicated that, in the whole sample, participants clicked on two-sided blog posts (M = 0.50, SE = 0.04) more frequently than one-sided articles (M = 0.37, SE = 0.04). Furthermore, an interaction effect between sidedness and stance emerged, F (1, 60) = 4.95, p = .030, η_p² = .08; within the one-sided articles, texts describing positive media effects were selected slightly more often (one-sided pro articles: M = 0.41, SE = 0.05; con articles: M = 0.33, SE = 0.05), whereas in the group of two-sided articles, texts on detrimental media effects were more frequently selected (two-sided pro articles: M = 0.44, SE = 0.05; con articles: M = 0.56, SE = 0.04). The interaction was disordinal with regard to the effect of stance, but not with regard to the effect of message sidedness. That is, the effect of stance on article selection depended on whether the article promised to be one-sided or two-sided, whereas the main effect for the more frequent selection of two-sided articles over one-sided articles was not qualified by whether the article was pro or con.

The order in which readers selected which articles to read also reflected a preference for two-sided vs. one-sided blog postings. The data reflecting the order in which readers chose articles are comprised of ordinal ranks rather than interval-level data, and as such are amenable only to an analysis of main effects and not interactions. Data were analyzed using Friedman's nonparametric two-way analysis of variance for ranks (see Siegel, 1976). The analysis showed a significant difference, χ² (1) = 5.59; p = .018. Two-sided articles (mean rank = 1.35) were selected earlier than one-sided articles (mean rank = 1.65). The comparison between the groups of pro and con articles was not significant.

With regard to reading time, participants read two-sided blog posts longer (M = 22.39, SE = 1.56) than they read one-sided posts (M = 14.20, SE = 1.43), F (1, 60) = 9.70, p = .003, η_p² = .14. A main effect of stance also affected reading time, F (1, 60) = 4.49, p = .038, η_p² = .07; participants read texts about the negative effects of media usage longer (M = 20.54, SE = 1.38) than they read about the benefits of media (M = 16.05, SE = 1.16).

Based on the main effects of sidedness on information selection, order of selection, and reading time, the users generally preferred two-sided blog posts, supporting H1.

The second hypothesis predicted that need for cognition increases the preference for two-sided over one-sided texts. To investigate whether NC affected participants' blog selections, we performed a regression analysis testing the moderation effect of a continuous variable in a within-subject design (Judd, Kenny, & McClelland, 2001). For this purpose, the users' preference for two-sided over one-sided blog articles (i.e., the difference between the number of selected two-sided minus one-sided texts) was regressed on the NC scores. The regression included age and education (dummy coded: university entrance-level degree vs. lower education) in order to control for the influence of these two factors. In the regression model, R² = .131, need for cognition emerged as a significant predictor (β = 0.258, p = .048) and there was no significant influence of education and age. This means that greater need for cognition led to a stronger preference for two-sided over one-sided articles, supporting H2.

The total number of selected articles was not predicted by age, education or need for cognition. Furthermore, the prior attitude toward the topic did not influence the number of selected pro and con texts, suggesting that there was no significant confirmation bias.

In summary, the results suggest a tendency for readers who are somewhat involved with the topic to prefer two-sided blog articles when searching for information on a controversial scientific topic. Articles that were previewed with two-sided headers were generally clicked more frequently, selected earlier, and participants read these articles for longer. It is improbable that the reading time difference is due to participants needing more time to understand the particular texts since comprehensibility had been pretested and length of the texts was kept constant. This preference seems to be relevant in the whole sample and especially for users with greater NC.

Study 2

Objective and Hypotheses

The goal of the second study was to include the role of source cues when examining how readers select user-generated content on a science blog. Based on research on the persuasiveness of experts (e.g., Wilson & Sherrell, 1993), we predicted:

H3: Texts that are connected to expert authors are selected more often, read for longer, selected earlier, and evaluated more positively than texts by nonexpert authors.

Furthermore, we expected that the effect of message sidedness remains potent in a setting that includes information about contributors' source characteristics:

H4: Two-sided blog articles are selected more often, read for longer, selected earlier, and evaluated more positively than one-sided articles.

With regard to the interaction of source information and message sidedness and including readers' NC in line with the extension of the ELM described above, we hypothesized that:

H5: NC influences blog article selections such that people with greater NC are more likely to choose based on message sidedness (and prefer two-sided blog articles) whereas low-NC individuals choose based on source information (and prefer experts).

To test these hypotheses, the experimental setting of study 1 was replicated and extended by adding information on the blog authors' expertise.

Method

Sample

Sixty parents with children between the ages of 1 and 18 participated in the study. Again, they were recruited via newspaper ads and press releases and received an incentive of 25 euro. Participants (31 female, 29 male) were aged between 26 and 50 years (M = 38.95; SD = 7.03). Seventeen percent of them had a university degree, 25% had finished high school with a university entrance-level qualification, and 58% had a lower level of education.

Design and stimulus material

Similar to study 1, a science weblog dealing with the effects of violent media content was used as stimulus material. The overview page of the blog presented eight headlines and summaries that linked to the respective blog articles when clicked. Besides message sidedness (one-sided vs. two-sided) and stance of the articles (positive vs. negative effects), information about the authors' expertise was introduced as a new factor. This was operationalized in each author's self-reported profession which appeared after the authors' respective names. Greater expertise was indicated by an academic degree and a profession with a close connection to the topic (e.g., psychologist or media scholar), whereas authors with a lesser level of expertise appeared to hold a profession with no connection to the topic (e.g., banker or legal secretary). These qualifications appeared below each articles' headlines on the overview page. Thus, sidedness and stance of information, as well as expertise of the author, comprised within-subject factors, resulting in a 2x2x2 design. In each of the four text categories (one-sided/pro, one-sided/con, two-sided/predominantly pro, two-sided/predominantly con; see Method section of Study 1 for details), one text was associated with an expert author and one text with a low-expertise author. The combination of authors and texts within the categories was rotated across participants. Furthermore, texts appeared on the overview page in different orders in order to minimize position effects. Apart from these variations, the headers and texts were identical to those presented in Study 1.

Need for cognition

The users' NC (Cacioppo & Petty, 1982; Bless et al., 1994) was measured with 16 items (Cronbach's α = .81; M = 57.47; SD = 8.52).

Procedure

As in study 1, participants were asked to search for information on the topic by browsing through the weblog, and reading time was limited to 3.5 min. After this, participants rated the quality of the articles. To facilitate the rating, the questionnaire system displayed screenshots of the article summaries including author information. Participants assessed perceived text and author credibility as well as the quality of arguments on five original 7-interval Likert scale items, Cronbach's α from .92 to .97.

Dependent measures

For every article, coders extracted these variables: frequency of selection (number of clicks), reading time (in seconds), and order of selection (see Study 1). For every participant, the numbers yielded counts for the selections of two-sided texts, one-sided texts, expert texts, and nonexpert texts. These counts allowed the calculation of difference scores indicating individual preferences for two-sided over one-sided blog articles, and expert over nonexpert articles.

Results

The participants selected an average of 3.68 articles (SD = 1.66) for further reading. With regard to the effects of the text variables, repeated measures ANOVA proceeded with author expertise, sidedness, and stance of the article as within-subject factors. For the dependent measure of frequency of selection, significant main effects emerged for blog author's expertise, F (1, 59) = 7.19, p = .009, η_p² = .11; message sidedness, F (1, 59) = 6.35, p = .014, η_p² = .10; and stance, F (1, 59) = 6.23, p = .015, η_p² = .10. The mean values (see Table 2) showed that the readers preferred articles by expert authors, two-sided messages, as well as articles emphasizing the negative effects of media usage. A significant ordinal interaction of sidedness by stance emerged, F (1, 59) = 4.06, p = .049, η_p² = .06, with main effects unaffected. There were almost no differences between pro and con articles in the group of one-sided texts (one-sided pro articles: M = 0.40, SE = 0.05; con articles: M = 0.41, SE = 0.05), but among two-sided articles, readers more frequently selected texts with predominantly negative slant (two-sided pro articles: M = 0.42, SE = 0.05; con articles: M = 0.62, SE = 0.05).

The effect of two-sidedness impacted the order of selection. Friedman's analysis of variance for ranks again provided the comparison using the rank-ordered data describing selection order. Readers generally selected two-sided blog posts (mean rank = 1.33) earlier than one-sided blog posts (mean rank = 1.68), χ² (1) = 7.48; p = .006. There were no significant differences in the order of the selection of articles due to whether their authors were experts or nonexperts, however, or due to whether they took a pro or con position.

With regard to reading time, readers examined expert texts for longer than nonexpert texts, F(1, 56) = 4.53, p = .038, η_p² = .08. Similarly, two-sided blog articles were read longer than one-sided articles, F (1, 56) = 11.87, p = .001, η_p² = .18; and negative articles on media usage more than positive articles, F (1, 56) = 12.86, p = .001, η_p² = .19. These tendencies also appeared in the credibility and quality ratings of the articles: These scores differed as a function of sidedness, F (1, 59) = 51.18, p < .001, η_p² = .47; expertise, F (1, 59) = 10.87, p = .002, η_p² = .16; and stance, F (1, 59) = 19.37, p < .001, η_p² = .25 (see Table 2).

Hypotheses predicted that readers prefer expert texts (H3) and two-sided messages (H4) when searching for information within the blog. H3 is partially supported in that expert texts were selected more frequently, read for longer, and evaluated more positively; however, there was no effect of expertise on the order of selection. The main effects of sidedness on selection frequency, order of selection, reading time, and quality ratings support H4.

Regression techniques for the effect of a continuous characteristic on the influence of within-subject factors (Judd et al., 2001) tested the effect of need for cognition on the users' blog article selections concerning author expertise and message sidedness (H5). First, the difference score reflecting users' preference for two-sided over one-sided blog articles was regressed on age, education, and the NC scores. However, none of the variables was a significant predictor of preference for two-sided vs. one-sided texts. Second, the users' preference for expert over layperson texts was regressed on the same three variables. Here again, none of predictors reached a conventional level of significance, with the effect of need for cognition achieving (β = 0.251, p = .064, R² = .101). As in Study 1, the total number of selected articles was not predicted by age, education, or need for cognition.

In summary, the users preferred blog posts reflecting two-sided messages and those that are generated by expert authors when selecting blog texts on a controversial scientific issue. The main effect of message sidedness confirms the similar finding of Study 1. However, in this case, the preference was not magnified by need for cognition.

General Discussion

The fact that many websites are filled with participatory content has raised new questions about the way Internet users acquire information. This research investigated several factors that potentially influence the perception and selection of user-generated articles in blogs on scientific topics. Given the variations in information credibility and quality in Web 2.0, we tested whether and how the apparent source and the sidedness of messages affect the recipients' information selection. Transferring assumptions from the ELM from persuasion research (Petty & Cacioppo, 1986) to the domain of information selection, we examined whether users' general motivation to engage in effortful information processing (need for cognition) influenced their attention to and selection of blog posts. Because participants (parents) were likely to be generally interested in the topic of the blog (the effects of media on children), the situational motivation was presumed to be relatively strong in the whole sample.

In both the first and second study, the participants more often selected two-sided than one-sided blog articles (H1 and H4). This suggests that readers who want to gather information on a controversial issue prefer messages in which arguments and evidence for both viewpoints are expressed, the promise of which (as signaled in the message header) guides their decisions about which articles to choose for further processing. This may be due in part to expectations that two-sided messages provide greater quality and utility (Atkin, 1973; Hagerty & Aaker, 1984) with regard to the relevant debate. Users' individual differences in need for cognition magnified the preference for two-sided over one-sided blog articles (H2): Those who typically enjoy analytical thinking particularly preferred two-sided messages, which are likely to offer a more balanced but also possibly more complex picture of the debate. This is in line with research in the ELM framework, which showed that people with greater NC are more likely to follow the central route of processing and reflect upon arguments thoroughly (Cacioppo et al., 1996). Not only does this preference appear when individuals are digesting messages and adjusting attitudes (in ELM experimental settings involving forced exposure). Our results indicate that it also manifests in the earlier stage of information selection, when deciding which texts to choose for further reading.

The results of the second study also showed a main effect of author expertise: Readers preferred articles that were presented by authors whose profession has a close connection to the topic (H3). This demonstrates that effects of expertise that pertain in persuasion research (Wilson & Sherrell, 1993) transfer to information selection, insofar as the presumed credibility of the blog authors is concerned. Although the ELM predicts that source effects should be limited to those who have less interest in a topic, the importance of source factors may be particularly high in the setting of science communication: In a domain where laypersons have to rely strongly on others, the question of whom to believe might be more important than in domains where the merits of arguments are more apparent and can be inferred more easily by the readers themselves (Bromme, Kienhues, & Porsch, 2010). In this respect, it seems plausible that participants who are concerned about a topic—such as parents who may be attuned to issues about the effects of media on children—include source expertise in their information selection decisions. Further studies with different topics are needed to test this interpretation.

Another explanation for the importance of source characteristics could lie in the difference between the processes of selection and persuasion. In the selection phase, the presence of message factors (in this setting, in the headline and three-sentence summary) was relatively sparse compared to its abundance during the elaboration phase: When reading longer articles, users under high-elaboration conditions have more opportunities to evaluate the actual quality of arguments than when reading a blog headline and a summary.

With regard to NC, the pattern that emerged in Study 1 did not adhere in Study 2. In Study 2, where message sidedness, stance, and expertise cues varied, NC did not magnify readers' preference for two-sided over one-sided blog articles. Perhaps the impact of the three message properties are sufficiently potent to any reader, that those factors overcome readers' individual differences in NC. Perhaps the inclusion of research participants whose probable interest in the topic created a floor effect beyond which individual differences in NC were less relevant when three other factors—sidedness, expertise, and stance—were sufficient to guide article selections. With regard to theoretical modeling, our results are mixed regarding the importance of NC in the process of information selection. Consistent with predictions derived from the ELM, users with high NC tend to select more complex messages when confronting differences only in sidedness and stance, although there was no effect of NC when facing sidedness, stance, and source expertise. The individually based motivational mechanism of the ELM, originally pertaining to persuasion and attitude change, is also somewhat relevant for information selection processes, although greater specification of the boundaries of this effect deserves research attention in the context of Web 2.0 and beyond. The sample of parents with heterogeneous educational backgrounds (instead of a student sample) can be seen as a strength of the present studies, meaning that our results may generalize to people who do not possess a university degree and whose primary source of science-related information might be the Internet. Nevertheless, future research should test the effects of motivation and ability in different settings and in connection with other source and message factors in order to gain further insights into the information selection processes.

With regard to practical implications, the results provide a rather positive interpretation of the potential of the Web as a source of information for laypersons. The main effects with regard to message sidedness and source expertise indicate a general tendency to choose two-sided texts and articles that are attached to expert sources. These strategies are more pronounced among users with a high NC, but there was no “selection gap” in terms of participants with lower NC generally avoiding two-sided articles which may challenge pre-existing attitudes (Hale et al., 1991). From a normative point of view, it seems desirable that people (independent of their dispositions) prefer more balanced information on scientific topics, since scientific knowledge is tentative and there are conflicting views. The tendency to select articles presented by expert writers also appears to be reasonable, but there is cause for concern in that blog authors' claims about their expertise might be deceptive (see Donath, 1999). Therefore, it is important not to rely exclusively on self-reports, but to corroborate the background of a site or an author when possible (Metzger, 2007).

Although the science blogs on which these experiments focused are a useful context in which to investigate the influence of the message and source factors, we must acknowledge that actual Web 2.0 environments may offer readers even more choices. The content of the stimuli in these studies covered only a limited spectrum of the variety that can be found online. Furthermore, the task of surfing through a blog in a short time period may have affected responses. The procedures should be replicated using relaxed temporal constraints.

In summary, this research provides several contributions. First, it indicates that laypersons who search for information on science-related topics in Web 2.0 do not select randomly, but are guided by credibility and message quality cues when deciding which articles to read. Second, it extends theoretical scope and depth regarding selective exposure and the ELM's pertinence to online information selection processes. Third, the personality characteristic of need for cognition shows some influence in message selection, although more work is needed to clarify its role vis a vis the message characteristics that guide information selection. The findings generally offer a new basis for understanding how users decide what to read in Web 2.0. In order to gain a more elaborate picture of the usage of participatory websites, more research is needed on the interplay of source, message, and user characteristics in the selection, as well as also comprehension and processing, of user-generated content.

Acknowledgment

This research was funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) in the Special Priority Program 1409 “Science and the General Public” (Kr 2240/2). The authors would like to thank German Neubaum, Jana Appel, Kathrin Schielke, and Leonie Rösner for their assistance in data collection as well as the editor and two anonymous reviewers for their helpful suggestions.

Note

References

About the Authors

Stephan Winter, PhD. ([email protected]) is a researcher in the Social Psychology: Media and Communication team at the University of Duisburg-Essen, Germany. His research interests include social psychological aspects of Web 2.0, online journalism, and science communication.

Address: Forsthausweg 2, 47057 Duisburg, Germany.

Nicole C. Krämer, PhD. ([email protected]) is Professor in Social Psychology: Media and Communication at the University Duisburg-Essen, Germany. Her research interests are social psychological aspects of Web 2.0 applications and social effects of human-technology interaction.

Address: Forsthausweg 2, Rm LE 216, 47057 Duisburg, Germany.