Taking Online Search Queries as an Indicator of the Public Agenda

Measuring the Public Agenda

The agenda-setting hypothesis is one of the best documented approaches of media effects research. The idea that the frequency of media coverage on political issues (media agenda) has an impact on citizens’ relevance ranking (public agenda) is plausible, because the public has hardly any other information available when they want to get a picture on socially relevant issues. Because already early agenda-setting studies found substantial media influences (e.g., Funkhouser, 1973; McCombs & Shaw, 1972), until today, hundreds of follow-up studies with different methodological designs, on different topics and in different social contexts have been carried out. The presumed effect of the media on the public agenda is generally recognized, even if it varies in its strength (for an overview, see McCombs, 2014; Wanta & Ghanem, 2007). In classical agenda-setting research, the media agenda is charged with media content analysis and the public agenda with more or less representative population surveys. The respondents are asked to mention one or more political issues that they find personally important. Issues can be measured on different levels (Yagade & Dozier, 1990): abstract issues (e.g., foreign policy), specific issues (e.g., the war in Iraq), or single events (e.g., the visit of the foreign minister in Iraq). Until some years ago, there was no alternative for this kind of measurement, although it is problematic for at least two reasons: First, survey research is subject to reactivity. In the context of agenda-setting research, respondents may simply be stimulated to think about important issues just because they are asked to mention them. Second, the availability of survey data is a significant problem for agenda-setting research. Similar to all media effect theories, the agenda-setting approach postulates a cause and effect relationship and can therefore be best tested with the help of longitudinal data. However, most longitudinal studies have to rely on survey data that are regularly charged by commercial surveys and provided for secondary analysis. Such data are rarely available on a daily basis. Instead, agenda-setting effects are often examined on a weekly, monthly, or even yearly basis, although one can assume that they occur in a shorter period of time.

Summarizing these considerations, it would be an ideal solution to measure the public agenda with a non-reactive method that generates continuous data on a daily basis. In recent years, some authors have therefore proposed to capture the public agenda via the use of online search engines. The online information behavior can be regarded as an indicator of the personal importance of an issue, because recipients are likely to search online for the issues that they consider to be particularly relevant (Granka, 2010; Scharkow & Vogelgesang, 2011; Weeks & Southwell, 2010). Those studies follow up early consideration on the role of behavior in the agenda-setting process (e.g., Becker, McCombs, & McLeod, 1975). Although most studies since then have regarded behavioral responses as a consequence of agenda setting (McCombs, 2014), some studies conceptualize behavioral responses also as an indicator of the public agenda. For example, some agenda-setting studies have captured the public agenda by asking which issues respondents discuss with other people (community issue salience; McLeod, Becker, & Byrnes, 1974). In this case, an issue is considered to be personally relevant in case it stimulates follow-up communication. Based on these considerations, Roberts, Wanta, and Dzwo (2002) examined the influence of media coverage on the issues discussed in political online discussion forums. They found that the frequency of discussing several issues in online forums followed the frequency of these issues in media coverage with a delay of 1 to 7 days.

The idea to capture the public agenda by behavioral responses has recently received a boost because freely available data about the use of online search engines are available via Google Trends and similar tools such as Yahoo Buzz Index and Experian Hitwise. For example, Google Trends aggregates the search requests done at the far most commonly used worldwide search engine Google. The data are presented as relative frequencies in comparison with other queries graphically or in tabular form in a certain period. For periods of analyses of a few weeks, the data are available on a daily, for longer periods, on a weekly or monthly basis. Therefore, they can easily be compared with the media agenda during those periods. Recent studies using that approach show considerable correlations: In the first study of that kind, Hester and Gibson (2007) compared the appearance of the term gay marriage in local and national U.S. newspapers and television news broadcasts with the Yahoo search term index for the same phrase. They found effects of the media agenda on search behavior in a period of about 2 weeks. In two more comprehensive studies, Granka analyzed the issue agendas of several American television networks, newspapers, and websites, and compared them with the corresponding Google searches (Granka, 2009, 2010). All in all, she found strong effects of the occurrence of the issues in the media on the search behavior in Google. One remarkable exception was the issue of unemployment where no effects were observed. Moreover, it was shown that the search volume decreased after a few days, although there was still media coverage on an issue, which seems to speak at first sight for very-short-term agenda-setting effects. In a similar study, Weeks and Southwell (2010) compared media coverage and online search volume based on a single event, rumors circulating during the 2008 U.S. presidential election saying Barack Obama was a Muslim. The study showed significant correlations between the reports of American television newscasts and newspapers and the frequency of the search term Obama Muslim on Google. The highest correlations were seen when the coverage was correlated with the search behavior on the same day. Most recently, Ragas and Tran (2013) and Ragas et al. (2014) used online search queries on several search engines pooled by Experian Hitwise as indicators of the public agenda concerning a person (Barack Obama) and a single event (an explosion at a BP oil rig). In both cases, the authors found evidence of agenda-setting effects as well as reverse agenda-setting effects indicating an effect of online search behavior on the media coverage about an issue. The effects persisted even after controlling for real-world indicators.

Recently, some studies also directly tested whether online search behavior is a valid indicator of the public agenda by comparing online search queries with survey data. On a daily basis, Scharkow and Vogelgesang (2011) examined the correlations between the naming of the German tax expert Paul Kirchhof as one of the three most important issues in the 2005 German national election campaign and the Google search Paul Kirchhof during the same period. Kirchhof was frequently criticized because of his tax concept and therefore became an issue in a broader sense. The authors show a moderate but clearly visible link between the two data sets, suggesting that the online search behavior reflects the agenda measured by classical surveys more or less well. Comparing weekly Google data on several political issues with monthly survey data from the United States, United Kingdom, and Spain, Mellon (2013, 2014) found that online search behavior can be used as an indicator of the public agenda for some issues but not for others. Because these findings result from an explorative statistical analysis, it remains unclear how those differences can be explained.

The Role of Public Uncertainty

Summing up the literature, we can conclude that online search behavior seems to be a valid indicator of the public agenda only under certain circumstances we do not know yet. This problem clearly results from a lack of research dealing with the role of variables moderating the relationship between online searches and the public agenda. In this article, we will bring forward the argument that recipients’ uncertainty is a crucial factor in this case. When theoretically reflecting the findings of former studies, it is evident that online search behavior differs from both the usual survey questions measuring the public agenda as well as the measurement of the public agenda by the frequency of issue specific follow-up communication, because it is a form of information seeking. Regarding online search behavior as a dependent variable in the agenda-setting process, the agenda-setting approach must therefore be associated with theories of information use. A link between the perceived relevance of an issue and information seeking is proposed by the concept of need for orientation (NFO). Following that concept, information seeking on an issue occurs in case an individual (a) perceives the issue as relevant and (b) feels uncertain about the issue (e.g., Weaver, 1980). In this case, uncertainty should not be misunderstood as an emotional state but can rather simply be defined as the absence of knowledge on an issue. It occurs when a situation is ambiguous, complex, or impossible to predict, when information is unavailable or inconsistent (Brashers, 2001). Consequently, it is perceived as an unpleasant condition that needs to be eliminated with the help of further information. Matthes (2006) shows that uncertainty clearly further increases NFO when an individual perceives an issue as relevant. However, NFO stays on a low level when someone is uncertain but perceives an issue as irrelevant. Therefore, uncertainty can be called a subsequent condition of NFO.

The role of uncertainty for information seeking is also highlighted by the information-utility approach (e.g., Atkin, 1973; Knobloch, Dillman Carpentier, & Zillmann, 2003; Knobloch-Westerwick & Kleinman, 2012). Following that approach, people compare the expected value of any information with the expenditures of obtaining and processing it. Thereby, they perceive information as valuable, in case it is able to reduce their uncertainty concerning an issue. Because online search queries take some time and cognitive effort (high expenditures), we can assume that the rising salience of an issue on the public agenda caused by increased media coverage is only transferred to online search queries in case there is public uncertainty regarding an issue. This would, for example, explain why the study by Granka (2009) showed no agenda-setting effects when the issue of unemployment was concerned: The increasing amount of media coverage on unemployment might make people believe that unemployment is an important issue. However, there is no need to change information behavior because there is no uncertainty that could be reduced with the help of online information. Most people might simply feel that they know enough about unemployment. If this consideration can be applied, the online search behavior would only be an appropriate indicator of the public agenda when one can assume a widespread public uncertainty about an issue. In an explorative study testing this, Holbach and Maurer (2014) found strong correlations between media coverage about the issue of EHEC, an epidemic threatening Germany in the summer of 2011, and online searches concerning the term EHEC. However, they quickly decreased after the first days of the outbreak of the epidemic. Authors suggest that this was the case because public uncertainty about the issue decreased simply because people learned enough about EHEC during the first days. Although this seems to make sense, to further examine the role of public uncertainty, the effects of media coverage on online search behavior concerning an issue causing uncertainty and an issue not causing uncertainty must be compared.

The EHEC Outbreak Versus Unemployment

Uncertainty probably leading to an increase in information seeking may primarily arise from surprisingly occurring events potentially affecting the entire population. This is often the case in the context of risk communication because there often exists uncertainty, because even the experts at the time of the occurrence of an event, for example, an environmental damage or a previously unknown disease, do not have reliable knowledge of its causes and consequences (external uncertainty; Kahnemann & Tversky, 1982). This assumption is, for example, supported by empirical studies showing that especially during serious epidemics people use information sources such as Google and Wikipedia to keep informed (e.g., Ginsberg et al., 2009; Tausczik, Faase, Pennebaker, & Petrie, 2012). Therefore, we stick to the EHEC issue as a good example for such an epidemic. In the middle of May, in northern Germany, there was an accumulation of reports of serious intestinal infections, the so-called HUS. The disease was caused by the strain of intestinal bacteria, called EHEC. Between May and July 2011, it affected several thousand people, especially in the Hamburg area, and 50 died. Immediately after the first deaths, it was initially suspected on May 26 that Spanish cucumbers where the source of the pathogen. In early June, the public was warned of sprout vegetables, a few days later the pathogen was detected in sprouts on an organic farm in a small town called Bienenbüttel. The population was asked, among other things, to dispense with the consumption of raw vegetables, resulting in significant negative consequences for the sales in agriculture. At the beginning of July, more than 1½ months after the outbreak, Egyptian fenugreek seeds were finally identified as the source of infection. By the end of July, the epidemic was declared over. Therefore, public uncertainty probably leading to online information seeking can be assumed for two reasons: First, EHEC has never been an issue in Germany before and, therefore, most people might have known nothing about that issue. Second, it took even experts several weeks to figure out the causes of the outbreak which can clearly be called a case of external uncertainty. Consequently, our hypothesis is as follows:

In case our theoretical considerations on the role of uncertainty as a necessary condition for information seeking are correct, we further can assume that the influence of media coverage on online search behavior will last only during the first days of media coverage. Later, public uncertainty about EHEC will be eliminated by recipients’ online searches. Consequently, further information seeking will decrease regardless of the amount of media coverage. Those short-term effects have, for example, been shown by Granka (2010) but have not been further explained. Therefore, our next hypothesis is as follows:

As an example for an issue without public uncertainty, we use the issue of unemployment. Unemployment belongs to the most important issues in Germany since decades and has been shown to be subject to agenda-setting effects in studies uncovering those effects by using traditional survey data (e.g., Brosius & Kepplinger, 1992). Being a constantly relevant issue not connected to any surprising event, we can assume that the public knows enough about unemployment and, therefore, uncertainty probably leading to information seeking does not occur when the amount of media coverage on unemployment rises. Therefore, we seem to have good reasons to pose H3 as follows:

Method

To determine the media agenda, we conducted a computer-assisted content analysis of 11 particularly wide-reaching online news media in German language. The selection includes the websites of the 9 most far-reaching German online news media: bild.de, spiegel-online.de, focus.de, welt.de, sueddeutsche.de, n-tv.de, zeit.de, stern.de, and faz.net. Because the German Wikipedia serves not only German users, but also visitors from Switzerland and Austria, the most far-reaching online news media from Switzerland (nzz.ch) and Austria (derstandard.at) were included. Concerning the EHEC issue, the period of investigation reached from May 15 (shortly before the outbreak of the epidemic) to July 31, 2011 (the official end of the epidemic). Articles of the online media containing the search term EHEC were recorded using a self-programmed crawler automatically searching for the term and counting the relevant articles. Overall, the 11 websites contained 2,393 articles on EHEC. Not to confound the results of the EHEC and the unemployment issue, the period of investigation concerning the unemployment issue reached from July 1 to September 15, 2012. This period corresponds to the investigation period of the EHEC issue concerning its length and more or less also concerning the amount of articles published (1,545) but did not include any surprising event.

Most previous studies have observed the online search behavior on search engines such as Google. This is problematic for several reasons: First, the Google user data are being criticized because its realization is relatively intransparent. Second, Google data are not available in absolute frequencies. Rather, relative frequencies are reported, where the queries are relative to a particular term in the searches of all terms in a given period. The choice of the period is limited. Third, Google Trends provides daily data only for relatively short periods. For longer periods, the data are aggregated on a weekly basis and can therefore not be used to uncover daily agenda-setting effects, which we mentioned as one of the major advantages of using online search data in agenda-setting research. We therefore decided to investigate the influence of media coverage on the EHEC outbreak and the issue of unemployment on information behavior in the free online encyclopedia, Wikipedia. Wikipedia is a multilingual online encyclopedia whose content is written collaboratively by a large number of volunteers and is licensed freely. The number of Wikipedia visitors in Germany is steadily rising. In 2011, 70% of the German online users visited Wikipedia at least occasionally; almost a third did regularly (Busemann & Gscheidle, 2011). To analyze Wikipedia search behavior, we use freely available aggregated-log files of the Wikimedia Foundation, which has at least two advantages compared with Google Trends: First, the data are available in absolute frequencies, which makes it easier to compare it with the frequencies of media coverage. Second, the data are available hourly and can therefore easily be aggregated on a daily basis. In contrast to regular surveys, log-file analysis is a non-reactive method because users are not aware of the fact that their search behavior is observed. In addition to the log files, their MD5 checksums are provided. Therefore, the completeness of downloads could be ensured by means of a local checksum comparison.

Both the media agenda and the online information behavior as an indicator of the public agenda are therefore available as time series on a daily basis. To calculate the influence of the media on the public agenda, we conduct a time-series analysis based on simple cross-correlations, which determines how changes in the media agenda influence changes in the public agenda. We take into account the contemporaneous correlations (the media agenda influences the public agenda on the same day) as well as lagged correlations up to 4 days. The 4-day interval was chosen because studies measuring short-term agenda-setting effects on a daily basis suggest that those effects occur in between 1 and 3 days (Krause & Gehrau, 2007). A problem arises from the fact that Wikipedia use—independent of certain issues—is much smaller on weekends than during the rest of the week. ¹ Because also German online news media publish less articles at weekends (e.g., Trappel, 2007), any existing correlation between the two time series could be at least partly due to this artifact. Therefore, we calculate the time-series analysis once with and once without the weekend days. Moreover, we interpret them carefully and with the aid of diagrams, in which it is clearly visible which effects are explained by the weekend data.

Findings

Figure 1 shows media coverage and online information behavior concerning the EHEC and the unemployment issue over time on a daily basis. As already mentioned, the 11 online news media published a total of 2,393 articles on the EHEC issue. In the same period, the German-language EHEC entry on Wikipedia was requested a little more than a million times. Figure 1 shows media coverage and online information behavior over time on a daily basis. Looking at the relationship of the two time series, the parallels are evident: The media coverage started on May 23, reaching its first climax at the end of the first week. The information use on Wikipedia increased during the first days much quicker than the media coverage and peaked on May 26. By the beginning of June, a number of largely parallel fluctuations in the time series can be identified, but these are, as expected, partly due to the independent decline of media coverage and Wikipedia use on weekends. Media coverage reached its peak on the day the pathogen was detected on a farm in Bienenbüttel (June 8). However, this did not lead to a further increase in the use of the EHEC entry on Wikipedia. Just the opposite was the case: After the first week of the epidemic, requests for the EHEC entry declined continuously. In mid-June also the media attention to the subject waned. Although the cause of the pathogen was not found and the number of infected continued to rise, online media were hardly interested in the issue. Even the slight increase in media stories in connection with the discovery of the actual pathogen in late June led to no further intensification of information use in Wikipedia. Although the online media under examination contained only slightly less articles on unemployment than in the case of EHEC (1,545), media coverage resulted in only a small amount of requests to the Wikipedia entry on unemployment (about 26,000). Nevertheless, Figure 1 shows parallels between the amount of media coverage and the Wikipedia requests, which we did not expect. During the whole period of examination, we find a substantial media coverage on unemployment, which is obviously rising when the monthly reports on unemployment statistics are published at the end of each month. Interestingly, that rise in media coverage does not lead to a remarkable increase in Wikipedia requests. Instead, there seems to be a parallel and regularly occurring up and down of media coverage and Wikipedia requests. A closer look at this parallels shows that they may be most likely caused by the artifact of lesser media coverage and lesser Wikipedia use on weekends, which we have discussed above.

OPEN IN VIEWER

Figure 1.

The relationship between the EHEC and the unemployment coverage in online news media and requests to the Wikipedia entry on EHEC and unemployment.

Concerning EHEC, the visual impression of a strong correlation between the two time series is confirmed by the statistical analysis: It results in a very strong correlation for lag 0 (.76, p < .01), which suggests an influence of the media agenda on the use of information on the same day. During the following days, the correlations decrease step by step but stay on a significant level. This is largely independent of whether the analyses are carried out with or without the weekend data, which indicates that the correlations found are not inflated artificially (table 1). Taken together, both the visual and the statistical analyses support H1 as well as H2. The correlations between media coverage and online search behavior would be actually larger one would only take the first three weeks of analysis into account. From then on, media coverage and online information use increasingly diverged, because the EHEC entry has hardly been requested. Regarding Wikipedia use as an indicator of the public agenda, one would therefore assume that the agenda-setting effects initially precipitated disproportionately (acceleration model; Brosius & Kepplinger, 1992) and then stayed out because the topic is no longer salient to the public. If one considers, however, that the online search behavior is a form of information use, which presumably requests uncertainty about an issue, it seems much more likely that the recipients simply had eliminated their uncertainty about EHEC after a few days, so that further use of the EHEC entry was no longer required. In this case, the data do not tell anything about how salient the EHEC issue was at this time.

OPEN IN VIEWER

Table 1.

Cross-Correlation Between the Occurrence of the Terms EHEC and Unemployment in Online News Media and Requests to the EHEC and Unemployment Entry on Wikipedia.

Time lag (days)	0	1	2	3	4
EHEC
Original time series a	.76	.63	.53	.44	.36
Excluding weekends b	.76	.64	.50	.36	—
Unemployment
Original time series a	.52	—	—	—	—
Excluding weekends b	.32	—	—	—	—

Note. Content analysis of 11 online news media (EHEC n = 2,393; unemployment n = 1,545); log-file analysis of Wikipedia use; all reported values are statistically significant (at least p < .05).

EHEC = Enterohaemorrhagic Escherichia coli.

a

Time series consists of 78 time points.

b

Time series consists of 55 time points.

Also concerning unemployment, Table 1 shows a significant influence of media coverage on Wikipedia use at the same day, but it is substantially lower than in the case of EHEC (.52, p < .05). Moreover, in contrast to the EHEC issue, correlations for other time lags are not significant. However, the most relevant difference to the EHEC issue is that the correlation almost completely disappears, when the weekends are excluded from the analysis (.32, p < .05). This suggests that most of the parallels between media coverage and Wikipedia use concerning unemployment are caused by an artifact. However, we still find a significant correlation even when the weekends are excluded. Therefore, H3 is only partly supported. In contrast to what we expected, a lack of uncertainty does not seem to completely eliminate the correlation between media coverage and Wikipedia use, but it clearly reduces it.

Discussion

In the present study, we assumed that an increase in media content about an issue influences online search behavior only when public uncertainty about the issue exists. Our main argument for assuming this was that only in this case further information seeking is likely. We tried to test this assumption by using the example of a typical case of public uncertainty, the EHEC outbreak in Germany in the summer of 2011, and compared it with the issue of unemployment, which we regard as an issue not causing uncertainty. For the empirical analysis, we carried out a content analysis of 11 online news media websites measuring the frequency of articles on both issues and compared them with a log-file analysis of requests to the corresponding Wikipedia entries on a daily basis. Our analyses concerning the EHEC issue seem to speak for strong agenda-setting effects. We find strong correlations between the occurrence of the term EHEC in the news media and the requests to the corresponding Wikipedia entry on the same day. This simultaneous connection is plausible, because one can assume that the recipients use Wikipedia more or less directly after reading EHEC articles in the online news media. Moreover, the graphical analysis shows that after a sharp increase in the use of Wikipedia in the first days of media coverage, the number of requests to the Wikipedia entries dropped sharply—no matter how the media coverage was evolving during the same period. Regarding online search behavior as an indicator of the public agenda, one would have to conclude that the EHEC issue was no longer salient to the public, even if the mass media still frequently covered it. This point of view ignores, however, that online search behavior is a form of information seeking. Following theories of media use, information seeking requires uncertainty about an issue. A decrease in the use of online information can therefore be caused not only by a decreasing relevance of an issue, but also by a decreasing uncertainty concerning that issue. Perceived issue relevance and perceived uncertainty may be accompanied or diverge. In the case of EHEC, they might have developed apart. After a few days of intense media coverage, the issue might still have been on top of the public agenda. However, the recipients felt well informed about EHEC that they had no reasons to search for further information.

Our analysis on the issue of unemployment, which we regarded as an issue not causing public uncertainty, generally confirms those considerations. Although the amount of media coverage in the period under examination—and on several single days during that period—was at about the same level as in the EHEC case, there were clearly fewer requests to the Wikipedia entry on unemployment. Moreover, the initial correlations between the amount of media coverage and the amount of Wikipedia requests were much lower. Most importantly, the correlations almost disappeared after excluding the weekend data, which suggest that they are mainly caused by an artifact: They appear because online media generally publish less articles on weekends, while Wikipedia also is used less frequently during those days. Obviously, no further information seeking occurs in the case of an issue that does not cause uncertainty.

These considerations can lead to two quite different conclusions: On one hand, one might argue that online search behavior is generally a poor indicator of the public agenda because perceived relevance of an issue and perceived uncertainty concerning an issue cannot empirically be disentangled when using that research design. Following that argument, it does not seem to be appropriate to consider our and similar analyses as agenda-setting studies, but as studies examining the equally relevant question of whether and under what conditions the media coverage of political issues leads to further information seeking. On the other hand, one might argue that agenda-setting theory, as well as other theories of media effects, has to be modified in the context of the online world. In this case, the role of uncertainty in agenda-setting research had to be rethought. In traditional agenda-setting research, NFO (including perceived relevance and uncertainty) is treated as an intervening variable enhancing information seeking that is enhancing agenda-setting effects. In recent agenda-setting studies using online search queries, information seeking is not treated as a cause of agenda-setting effects but rather as an indicator of the public agenda. The crucial question then is, whether this makes sense also in case of the absence of public uncertainty. Our study gives a preliminary answer to that question, but more research is needed. This especially includes studies systematically varying uncertainty and other possible moderators to be able to examine whether the findings presented here can be generalized.

However, our study design also shows some limitations. First of all, searching in Wikipedia may not exactly be the same procedure as using online search engines such as Google. Reading a Wikipedia entry indicates strong interest in the meaning of a certain term, whereas using a search engine rather indicates a general interest in the issue in a broader sense. Therefore, for most people, Google rather than Wikipedia might be the first choice for information seeking when feeling uncertain about an issue. Moreover, not everyone interested in information about unemployment will use exactly that search term when searching in Wikipedia. Instead, also entries such as unemployment data, employment, or even economy could be taken into account. Recently, some studies started to systematically identify valid search terms for different issues (Mellon, 2013, 2014). This research should be continued. Nevertheless, we are convinced that Wikipedia data should be preferred to Google because of its transparency and the fact that it is available on an hourly basis. Second, one can ask the question of representativeness. Still, not all Germans have an Internet connection and not all Internet users use Wikipedia. Consequently, the data collected here cannot be representative for the German population, of course. The question we are trying to answer here is, whether Wikipedia users especially use the online encyclopedia after they have received information on the two issues from the mass media. Because this is the case, we can probably assume that other people used other sources for further information. Of course, whether this is the case cannot be verified with our data. Finally, we use the concept of uncertainty to explain our findings but do not measure it. Obviously, public uncertainty cannot easily be measured in studies such as ours, which do not use survey data. However, we think that we have good reasons to assume uncertainty in the case of EHEC but not in the case of unemployment. As uncertainty is defined as the absence of knowledge, it should occur after surprising events such as the EHEC outbreak but not in the case of unemployment, which has been a regularly discussed issue in Germany since decades. Nevertheless, future studies should also try to measure the intensity of public uncertainty, for example, by analyzing social media such as Twitter or Facebook. Moreover, the causal relationship between online information behavior, uncertainty, and issue salience could be clarified by studies collecting survey data on the individual level. In those studies also the influence of other possible mediators such as fear could be tested to further clarify the role of public uncertainty. Finally, another direction for future research would be to deal with reverse agenda-setting effects, which might occur when mass media monitor online search behavior to fit their agenda to what the public is interested in.