In today’s high-choice media environment, it has become easier for citizens to only select news that is in line with their political leanings (
Stroud, 2010) or to avoid political news altogether (
Prior, 2007). In such a setting, incidental news exposure (INE) may become increasingly important to ensure that citizens encounter counterattitudinal information or at least remain connected to the political public sphere. Social network sites (SNS) have been accused of further limiting the chances of INE through algorithmic curation (
Thorson & Wells, 2015), but have also been hailed as a promising new source of INE (
Lee & Kim, 2017): Even Facebook users who have not intentionally subscribed to a news outlet or whose previous Facebook behavior has not marked them as a suitable target for sponsored posts by news organizations may encounter news articles in their newsfeed via social curation (
Thorson & Wells, 2015), that is, because someone in their personal network has shared an article. Although these personal news recommendations on Facebook are particularly important for citizens who do not follow news and politics on their own, we still know very little about the factors that turn incidental
contact with news on Facebook into incidental news
exposure.
Based on a multiple stimuli experimental study (
N = 507), the article addresses this research gap by investigating which personal news recommendations the recipients are more likely to follow. Drawing on the concept of informational utility (
Knobloch-Westerwick & Kleinman, 2012), we assume that incidental news contact is more likely to become incidental news
exposure if recipients attribute a higher degree of utility to the recommended content. As Facebook provides a social environment for news use, the recipient of personal news recommendation is likely to draw cues for informational utility from the recommender. Following the lead of
Turcotte, York, Irving, Scholl, and Pingree (2015) who investigated the impact of personal news recommendations by Facebook friends perceived as opinion leaders on
intentional news exposure, we assume that recipients use informational utility cues based on how politically knowledgeable or politically similar the recommender is (politico-social cues). However, we also look in more detail at informational cues based on the closeness of the relationship to the recommender (social cue) by contrasting close and weak ties. In addition, we investigate whether the media outlet as the source of the recommended article also provides relevant cues for its potential informational utility (media cues). And, finally, we uncover the underlying mechanisms that encourage people to read shared news articles on Facebook (e.g., social pressure, social monitoring, issue importance, or outlet curiosity).
Method
Design and Sample
We conducted a randomized 3 × 3 between-subject, multi-stimulus online experiment with two additional quasi-factors in Germany in March 2017. In all conditions, participants saw a mock-up Facebook stream containing a teaser for a political news article. First, we varied the type of centrist media as the source of the article (legacy newspaper vs. unknown digital-born outlet vs. tabloid newspaper). Second, tie strength manipulated whether the article was shared and recommended by a close friend of the participant (a strong tie), by a distant friend (a weak tie), or whether the article appeared without any recommendation in the Facebook stream. The recommender’s political knowledge and opinion similarity were both measured as non-manipulated quasi-factors, that is, as perceived by the participant receiving the recommendation.
Using an online access panel provided by the survey company Respondi, we recruited an interlocked quota sample representative in terms of age and gender of German Facebook users between 18 and 69 (following the official Facebook definition, we defined users as those using their Facebook profile at least once a month). We tried to reach a statistical power (1-β) of .95, which would require 560 finished interviews for medium effect sizes (d = .5) given our complex design. However, necessary data cleansing due to failed awareness checks left us with N = 507 finished interviews (completion rate: 31.05%), which equates to a power of .92 assuming no missing values on covariates or quasi-factors. The average age was M = 38.45 (SD = 12.49). Similar to the German Facebook population, there were slightly more men (n = 268, 52.9%) than women in the sample, and about half of the participants reported holding at least a high school degree (n = 281, 55.9%). Participants received vouchers for the completion of the survey.
Procedure and Manipulation
After receiving their consent, age, gender, education, residency, and frequency of Facebook use were measured. Then, participants answered questions on their interest in different topics. Next, we manipulated the tie strength, depending on the randomly assigned experimental group: We either prompted participants to think of a person to whom they feel close and with whom they have relatively frequent contact via Facebook or of a person from the participant’s distant network (weak tie condition):
Please think of a person who is close to you (you only know distantly), with whom you feel personally strongly connected (feel loosely connected) and with whom you have regular (rare) contact via Facebook, e.g., through seeing his or her Facebook posts or through private messages. This could be a friend, a relative or your partner (former school friend, distant acquaintance or a former workmate), for example.
To amplify the manipulation, participants wrote down the name of the person and rated how close they feel to the person and how often they had contact with them via Facebook. We presented the name of this person in each question and told respondents that they should keep them in mind before the stimulus was shown next.
The stimuli for all conditions consisted of three parts (see
Online Appendix B). The first part told participants to imagine that the following two posts originated from the chosen person, displaying the name again. Participants in the “no recommendation” condition saw the same reminder, but were told that only the first post originated from the imagined person. The second part was an apolitical post by the imagined recommender about vacation pictures. This part was constant in all conditions to increase the realism of the Facebook stream. The third part consisted of a posted news article showing the headline, a picture, and a short teaser. For tie strength, the weak and strong tie condition saw the news article as a shared post with a recommending comment as it typically appears on Facebook. The profile picture and the name of the recommender were pixelated so that participants could imagine the profile picture and the name of the chosen person. For participants in the no recommendation conditions, the news article appeared without a recommendation as an original post by the media outlet. For the media source cue, we varied the logo, name, and URL of the media outlet from which the news article was initially posted using their real Facebook pages.
To enhance our external validity, we used three multi-stimulus factors. First, we used two different but comparable media outlets per media type: Süddeutsche Zeitung (645,099 fans as of March 15, 2017) and Frankfurter Allgemeine Zeitung (458,649 fans) as legacy outlets; and Telepolis (20,574 fans) and Krautreporter (86,879 fans) as unknown digital-born media. The only exception is the tabloid condition, because Bild Zeitung (2,280,155 fans) is the only national tabloid. Second, the news article dealt with one of five comparable topics (regulation of monopolies, elections in Serbia, transparency of lawsuits, water protection, and broadband expansion), all of which were political topics, hard news, moderately relevant, and without issue ownership of a specific party. Third, we used two comparable sentences for the recommendation of the news article by the Facebook friend: “Finally a report that summarizes the whole issue. It’s worth it!” and “Finally an article that puts the topic straight. Readable!” Due to these multi-stimulus factors, our experiment used 135 different stimuli for the nine experimental conditions.
We measured our mediators and dependent variable after the participants had seen the stimulus. Next, we asked for quasi-factors and manipulation checks. Lastly, the participants answered some basic questions on their general media use before debriefing.
Measures
Dependent variable
The intention to read the news article was measured through 4 items using 6-point scales from 1 to 6 (e.g., “I would like to read what the news article contains”). The items showed high reliability (Cronbach’s α = .97) and were summarized to an index (M = 3.87, SD = 1.45).
Politico-social quasi-factors
Regarding the topic of the news article, we measured perceived opinion similarity between the participant and the recommender using a single item on a 7-point scale from 1 to 7 (“Please remember the topic of the news article. Would you say that you and [name] have rather the same or rather an opposite opinion on that issue?”, M = 5.11, SD = 1.32). With a second single item using a 7-point scale, the participant assessed the perceived knowledge of the recommender on this topic (“What do you think, how much knowledge has [name] about the topic of the news article?”, M = 3.80, SD = 1.64).
Politico-social cue mediators
The perceived importance of the topic of the news article was measured on three dimensions using 3 items on 6-point scales for each dimension: perceived importance of the topic for the society (e.g., “The topic is important for the whole society,” α = .94, M = 4.14, SD = 1.13), for own social environment (e.g., “The topic is relevant in my social environment,” α = .97, M = 3.27, SD = 1.35), and for oneself (e.g., “The topic is significant to me,” α = .97, M = 3.43, SD = 1.44).
Social cue mediators
Perceived social pressure to read the article was measured with 3 items (e.g., “I would have read the article because my social environment would expect it of me”) on 6-point scales and summarized to an index (α = .90, M = 1.71, SD = 1.03). Motivation for social monitoring was assessed with 3 items (e.g., “I would have read the article because I could better assess what opinion [name] has on this topic”), again using 6-point scales and computing an index (α = .97, M = 3.22, SD = 1.54).
Media cue mediators
The participants were asked to choose the one media outlet they had seen in the stimulus from a list of 12 names. The
perceived credibility of the outlet was measured with 5 items adapted from
Tsfati (2010), each of which used 6-point scales (e.g., “The media outlet is credible,” α = .98,
M = 3.08,
SD = 1.53).
Societal relevance of the outlet was asked with 4 items, again on 6-point scales (e.g., “Issues concerning the whole society are the focus of the media outlet,” α = .93,
M = 3.81,
SD = 1.30). In addition, we gauged participants’
outlet curiosity using 1 item (“I would have read the article because I did not know the outlet and was curious what kind of outlet it is,”
M = 2.25,
SD = 1.42). Only participants who correctly remembered the media outlet were asked their perceptions of the outlet to which they were exposed (
n = 254, 50.1%). However, our analyses still include those participants who could not remember the outlet because we assume that media cues could be effective even without a correct recall of the outlet. Participants with missing values on these mediators will only be excluded pairwise when testing the media cue mediators.
Control variables
We measured Facebook usage with 2 items allowing us to calculate the Facebook use per week in hours (M = 6.25, SD = 8.77). To measure the previous topic interest, we used a battery of 10 items on 7-point scales. Five of them asked for the participants’ interest in each of the five topics we used for the news article, and five were on other topics for distraction (M = 4.27, SD = 1.67). Lastly, we asked whether the participants are following any news site on Facebook: 59.8% (n = 303) reported that they did not follow any news site. We performed a simultaneous confirmatory factor analysis with all our measures showing that all items loaded on the expected factors (Root Mean Square Error Approximation [RMSEA] = .030, Comparative Fit Index [CFI] = .987, Standardized Root Mean Square Residual [SRMR] = .034).
Manipulation Check and Tests of Internal Validity
We asked the perceived social utility of information of the chosen recommender at the end of the questionnaire with 4 items on 6-point scales (e.g., “I like to be up to date regarding [name],” α = .97, M = 3.83, SD = 1.68). An independent t test confirmed that information from strong contacts (M = 5.14, SD = .84) had a significantly higher social utility than information from weak contacts, M = 2.43, SD = 1.15, t(370) = 26.03, p < .001, d = 2.69. We also asked about participants’ familiarity with different media outlets revealing that we succeeded in selecting media outlets with differing degrees of familiarity for the manipulation: 90% knew the screened legacy outlet, and 95.3% knew the tabloid, but only 10.2% were familiar with the digital-born outlet, χ2 (df = 2, n = 507) = 337.60, p < .001, Cramer’s V = .82.
Furthermore, we tested the internal validity of the experiment. First, we confirmed that the manipulation of the tie strength had no unintended spillover effects on users’ intention to read or on measured mediators of the control group. Second, we checked whether our multi-stimulus factors worked successfully. There were no differences in the dependent variable or any of the mediators between the two recommendation sentences nor between the summarized media outlets per media type. In addition, the multi-stimulus factor topic showed no main effect on the dependent variable nor on the social, and media cue, mediators nor interaction effects with our independent variables on these outcomes. That is, the results do not vary systematically by the five topics of the news articles. However, regarding our three politico-social mediators, the water protection issue was seen as significantly more important than the other topics. Therefore, we tested for metric invariance between this topic and all other topics displaying satisfying invariance (n = 507, RMSEA = .061, CFI = .957, SRMR = .059), and an alignment procedure showed no differences in the factor loadings. Thus, we summarized the topics as intended for the following data analyses but will control for the higher issue importance of the water protection topic in the mediation analysis.
Results
To test our proposed main and interaction effects (H1, H3, H4, RQ1, H6), we computed a factorial between-subject ANCOVA with the independent variables tie strength, knowledge of the recommender, opinion similarity, media type, and all possible interactions between them; the covariates previous topic interest, Facebook use per week, and the general following of any news site on Facebook; and reading intention as the dependent variable (see
Online Appendix A for descriptive data and
Table 1 for model coefficients). The recommender’s level of knowledge and opinion similarity were integrated by using median splits (
ndissimilar (1-5) = 151, 53.9%;
nsimilar (6-7) = 129, 46.1%;
nlow knowledge (1-3) = 118, 42.1%;
nhigh knowledge (4-7) = 162, 57.9%). Due to missing values on covariates and quasi-factors, our sample size decreased to
n = 382 for this analysis. That is, the statistical power is reduced to .79 to find effects with a medium size (
d = .5) but is still high enough to find strong effects (
d = .8) with 1-β = .99. The model explained 27% of the variance (
R2 = .27,
R2adjusted = .21) and was significant (
p < .001).
For tie strength (H1), we revealed a significant main effect (F(1, 352) = 4.20, p = .016, ηp2 = .02). Simple effects displayed that articles recommended by close friends led to a significantly (p = .024) higher reading intention (M = 4.21, SE = .11) than those recommended by distant contacts (M = 3.77, SE = .13) or not recommended articles (M = 3.79, SE = .13, p = .035; all simple effects analyses included the three covariates and used the Šidák correction, all reported p values are two-tailed). There were no differences between articles that were not recommended or recommended by distant contacts (p = .99). While we did not find a main effect of opinion similarity (H4, F(1, 352) = .41, p = .53), we detected a significant main effect for the recommender’s level of knowledge (H3, F(1, 352) = 9.41, p = .002, ηp2 = .03). Simple effects showed that recommendations from friends perceived to have high levels of knowledge about the topic led to a significantly (p = .007) higher intention to read an article (M = 4.24, SE = .11) than recommendations from friends perceived to have low levels of knowledge (M = 3.73, SE = .13) and articles that were not recommended (M = 3.79, SE = .13, p = .020); there was no difference between low levels of knowledge and no recommendation (p = .987).
For the media type (H6), we found a significant main effect (F(2, 352) = 4.20, p = .016, ηp2 = .02). Simple effects showed that the intention to read articles from legacy newspapers (M = 4.22, SE = .13) was significantly (p = .030) higher than for tabloids (M = 3.76, SE = .13), but there were no differences between legacy and digital-born outlets (M = 3.92, SE = .13, p = .28) or between tabloids and digital-born outlets (p = .76).
We found a significant interaction between opinion similarity and recommender’s level of knowledge (RQ1,
F(2, 352) = 12.30,
p = .001, η
p2 = .03). Simple effects showed that the aforementioned main effect of knowledge is mainly shaped by the condition of high opinion similarity (
Msimilar high knowledge = 4.58,
SEsimilar high knowledge = .15,
Msimilar low knowledge = 3.50,
SEsimilar low knowledge = .20,
p < .001) whereas knowledge does not have an effect when opinions are dissimilar (
Mdissimilar high knowledge = 3.90,
SEdissimilar high knowledge = .15,
Mdissimilar low knowledge = 3.97,
SEdissimilar low knowledge = .15,
p = .747). To secure our finding, we performed an additional moderation analysis with PROCESS using the metric, mean-centered values for recommender’s level of knowledge and opinion similarity, excluding the conditions with no recommendation from the analysis (new
n = 289) and controlling for the influence of tie strength. This analysis confirmed the significance of both the recommender’s level of knowledge (
b = .20,
p = .006) and the interaction between knowledge and opinion similarity (
b = .08,
p = .039). As
Figure 1 illustrates, the effect of knowledge depends on at least a moderate opinion similarity. The Johnson-Neyman significance value is –.77 for the mean-centered opinion similarity, equating to a value of 4.39 on the non-centered 7-point scale.
In addition, there were no interactions between media type and tie strength (F(2, 352) = .14, p = .87), media type and opinion similarity (F(2, 352) = .34, p = .717), or media type and recommender’s level of knowledge (F(2, 352) = .71, p = .49). That is, the recommender cues work the same for the three different media types.
To test for the proposed mediation processes (H2, RQ2, H7) and the experimental effect on issue importance (H5) beyond the already described main and interaction effects, we performed a structural equation model in R’s latent variable analysis (lavaan) package with the robust maximum likelihood (MLR) estimator using tie strength, opinion similarity, recommender’s level of knowledge, its interaction, and media source as exogenous variables; the social, politico-social, and media cue mediators as latent endogenous variables; and reading intention as the final endogenous variable. Again, we excluded the conditions without recommendation, controlled for the same three covariates, and used the metric mean-centered values for recommender’s level of knowledge, opinion similarity, and its interaction. We split the categorical independent variable media type into dummies using legacy outlets as the respective baseline. In addition, we controlled for the topic of water protection to account for its greater perceived issue importance (CFI = .977, RMSEA = .034, SRMR = .032).
The results (see
Figure 2 for the structural model and
Table 2 for statistical tests on indirect effects) show that tie strength influenced the two social mediators. Strong ties increased social pressure (
b = .48,
p < .001) and social monitoring (
b = .90,
p < .001), but only social monitoring is associated with reading intention (
b = .40,
p < .001). The indirect effect of tie strength via social monitoring (H2a) is significant (unstandardized coefficient = .36,
p < .001).
The recommender’s level of knowledge affected all three politico-social mediators (H5a issue importance society: b = .08, p = .035, H5b issue importance social environment: b = .19, p < .001, H5c issue importance self: b = .14, p = .001). Only issue importance for oneself in turn correlates with reading intention (b = .24, p = .018). However, the indirect effect of knowledge via issue importance for oneself was only marginally significant (unstandardized coefficient = .03, p = .056, RQ2c). Instead, the recommender’s knowledge showed an indirect effect on reading intention via social monitoring (unstandardized coefficient = .08, p < .001). Opinion similarity showed no effects, but the interaction between knowledge and similarity increased the perceived issue importance of the social environment (H5b, b = .08, p = .006) and oneself (H5c, b = .13, p < .001). However, only the indirect effect on reading intention via issue importance for oneself was significant (unstandardized coefficient = .03, p = .041, RQ2c).
Media type had no influence on the social or politico-social mediators; it only exerted effects on the media mediators. Tabloid articles led to a significantly lower perception of outlet credibility (b = −2.29, p < .001) and societal relevance of the outlet (b = −1.23, p < .001) than those from legacy media. Among these mediators, only the indirect effect on reading intention via the societal relevance of the outlet was significant (unstandardized coefficient = –.48, p = .015, H7b). Digital-born media were perceived as less credible (b = –.86, p = .003) and less societally relevant (b = –.65, p = .003) than legacy outlets, but only the indirect effect via societal relevance was at least marginally significant (unstandardized coefficient = –.25, p = .055, H7b).
Discussion
Facebook (and other SNS) have been seen as promising sources of INE to assure an informed citizenry (
Lee & Kim, 2017). However, the inherently social character of INE on Facebook may lead to changes in the heuristics used by citizens to decide which of the news articles that they incidentally stumble upon will actually be read.
According to our experiment, there are three paths that encourage news exposure based on INC among Facebook users. First, social cues of the recommender increase reading intention via social monitoring: If the recipient feels close to the recommending friend, he or she will read the recommended article to increase their knowledge of the recommender. Although respondents feel stronger social pressure to read articles recommended by close ties, this does not produce greater intentions to read. More importantly, recommendations from weak ties do not motivate recipients to read the recommended article compared with articles that are not recommended. In other words, though weak ties are expected to be more diverse and should, thus, have a higher potential to recommend counterattitudinal information, Facebook users are less likely to follow their recommendations, regardless of their political knowledge or the similarity of their opinions, or how respectable the recommended media source is. This phenomenon is further enhanced by the Facebook algorithm that assures that recommendations from weak ties as well as recommendations that do not match a user’s interests are less likely to appear in their news stream. Hence, INE on Facebook can only have a positive impact on exposure to political news, and in particular on exposure to politically diverse news, if users have close ties to people who are interested in politics and have different political outlooks.
The second path reflects the importance of opinion leaders for INE in a social setting such as Facebook (see also
Turcotte et al., 2015): Following recommendations by friends with high levels of topical knowledge and similar political opinions, participants believed the issue of the recommended article was more relevant to them personally and were, hence, more likely to read it. This, again, may facilitate exposure to like-minded news articles, but not cross-cutting exposure. Interestingly, political similarity alone does not predict reading intentions: According to our data, people are not motivated by a simple confirmation bias. They will not read just any article because they expect it to confirm their views; they will read articles by recommenders with similar opinions and high political knowledge, which are more likely to contain
useful arguments supporting their views. They are motivated by the
expected utility for opinion confirmation. Our results may, thus, represent a next step towards developing a theoretical understanding of how opinion leadership works in SNS.
The third path relates to traditional media source cues: Articles by legacy and (to a lesser degree) digital-born news outlets are perceived to have more
societal relevance than tabloid articles and are, thus, more likely to be read when recommended. Contrary to expectations, perceptions of credibility differ by outlet type but have no impact on reading intention. This finding contradicts a number of previous studies on the importance of credibility for news selection (
Choi et al., 2006), but might be explained by the particular usage situation on a social network. As news selection on Facebook is a social experience, respondents apparently place more importance on whether an article is likely to contain
relevant than
credible information. This would indicate that when using Facebook, users are more concerned with
what people are talking about than
whether it is true or not. This notion is corroborated by our finding that reading intentions, as well as perceptions of credibility and societal relevance, are higher for unknown digital-born outlets than for tabloids. Although legacy outlets are seen as more credible and societally relevant, Facebook users appear to give unknown sources considerable “benefit of the doubt” (at least compared with tabloids). This has important implications for the likelihood of mis- and disinformation by unknown sources being spread on Facebook. In addition, the non-significant interaction between tie strength and media cues indicates that recommender cues work the same for different types of media: Recommendations also have a substantive impact on exposure to tabloids, even though they are seen as significantly less credible than legacy media. In fact, a recommended tabloid article is just as likely to be read (
M = 4.03,
SE = .18) as an article without a recommendation from legacy media (
M = 4.02,
SE = .23).
Apart from the effects on reading intention, our study allowed a more detailed look at how agenda-setting effects may occur via agenda cueing on SNS. According to our study, SNS posts may not only affect perceptions of issue importance in general as found by
Stoycheff et al. (2018), their impact differs depending on the recommender’s socio-political characteristics and the scope of issue relevance: For evaluating societal issue importance, users followed the recommendations of highly knowledgeable, but not necessarily politically similar friends. In other words, they trusted perceived experts. However, indirect effects on reading intentions were limited to recommendations by highly knowledgeable, politically similar friends, which increased the perceived issue relevance for oneself. Thus, future studies on agenda setting (or agenda cueing) in SNS should include different social- and socio-political characteristics to further distinguish the conditions for changes in perceived issue importance of different scopes.
These interpretations have limitations. First, our manipulation of tie strength did not use the real profiles of participants’ Facebook friends but worked via imagination. However, the significant manipulation check and the main effect of tie strength showed that this manipulation worked. Still, one could argue that the entire experimental situation was rather artificial as the post itself did not appear in participants’ real Facebook stream but as a simple screenshot of a stream, and participants were asked to imagine it was their own news feed. Thus, our experimental setting reduces the ecological validity of our results compared with the study by
Turcotte et al. (2015) in which participants were deceived into thinking the friend actually posted the news article on Facebook. Second, our dependent variable only measured the
intention to read the article, rather than real usage. Third, the five topics we used for the news articles were non-polarized. More polarized topics may lead to stronger effects of opinion similarity regardless of the perceived knowledge of the recommender. Fourth, our mediation analyses should be interpreted carefully as the mediator outcomes are only partly explained by our experimental manipulations but also by personal predispositions. That is, the explained variance in news exposure by our mediation model originates only partly from our experimental factors. Lastly, we conducted our experiment in Europe, in a media system in which the audience is well aware of the differentiation between media types with respect to credibility and societal relevance. Thus, replications in different media systems are needed to confirm whether the media source still plays a role independent of personal recommendations.
Despite these limitations, this study showed that personal news recommendations motivate individuals to get involved with political news—if news is recommended by close friends or by contacts with high levels of knowledge and similar opinions on the topic. INE via Facebook may, thus, help keep people connected to the political public sphere, particularly those who are not connected via traditional media. However, our results also suggest that though news contact on Facebook may be “incidental” in the sense of “unintentional,” it does not occur “by chance,” that is, not all Facebook users will benefit in an equal manner. First, the likelihood of unintentionally encountering news (particularly diverse news) will depend for each user on their choice of Facebook friends (who may vary in their news-sharing habits) as well as their past reactions to shared news articles: The less they click on recommended articles, the less they will see news recommendations from friends in the future due to the news-feed algorithm. Second, people do not treat all “incidental” news recommendations the same. Their news exposure following incidental news contact depends on social and political preferences, they prefer to maintain existing strong relations with others and to engage with news they expect to confirm their existing beliefs (by preferring recommendations from like-minded, knowledgeable friends).
In sum, INE via Facebook has the potential to maintain a normatively desirable connection to the public sphere, but this is limited by the configuration of Facebook’s algorithm and the motivations of its users. As our study shows, people tend to follow INE recommendations by more homophile strong ties and knowledgeable, politically similar contacts, that is, INE on Facebook can easily be transformed into a form of self-reinforcing selective exposure. Thus, future research should continue to explore the potential of INE via Facebook and other social media platforms, but should pay close attention to its preconditions and circumstances because not all kinds of INE will contribute equally towards ensuring a well-informed public.