Framing genetically modified mosquitoes in the online news and Twitter: Intermedia frame setting in the issue-attention cycle

Abstract

We investigate how the online news and Twitter framed the discussion about genetically modified mosquitoes, and the interplay between the two media platforms. The study is grounded in the theoretical frameworks of intermedia agenda setting, framing, and the issue-attention cycle and combines methods of manual and computational content analysis, and time series analysis. The findings show that the Twitter discussion was more benefit-oriented, while the news coverage was more balanced. Initially, Twitter played a leading role in framing the discussion about genetically modified mosquitoes. When the public learned about the issue, online news gained momentum and led the Twitter publics to discuss the risks of genetically modified mosquitoes. Based on the findings, we argue that the intermedia frame setting may change its direction over time, and different media outlets may be influential in leading different aspects of the conversation.

Keywords

computational social science framing issue-attention cycle genetically modified mosquitoes intermedia agenda setting intermedia frame setting time-series analysis

1. Introduction

According to the Pew Research Center (2017), more than half of the US population continues to obtain their science news from the general news media, including newspapers, TV news, and digital news. While the news media still exert a significant impact on the public about science information and understanding scientific issues, many people go to social media such as Twitter to obtain information about and discuss science-related issues (Brossard and Scheufele, 2013). This new landscape of science communication has challenged the conventional way of disseminating scientific knowledge. Rather than transmitting scientific knowledge from scientists to the public through professional journalism, science communication on social media encompasses two-way, one-to-many, and many-to-many communication (Büchi, 2017). Social media have also opened the gate to non-elite voices including individual and organizational activists, bloggers, and concerned citizens (Hopke and Simis, 2017; Newman, 2017).

Scholars have become increasingly interested in the role Twitter plays in science communication. At first, Twitter simply functioned as “a news information disseminator” (Shan et al., 2014: 924) and its coverage was generally consistent with news media (Veltri, 2012). More recently, studies have found significantly different topic contextualizations when comparing Twitter and news coverage (Büchi, 2017), suggesting that Twitter not only disseminates the information from news but may also be “a supplemental medium of science news” (Büchi, 2017: 954). Furthermore, journalists increasingly reference social media as an information source (Ahmad, 2010; Grzywińska and Borden, 2012). Non-elite users interact strategically with politicians and journalists to shape news coverage (Newman, 2017).

Although Twitter allows new voices in science communication and provides important perspectives, its social nature has intended and unintended consequences (Brossard and Scheufele, 2013). Because of its underlying algorithms, Twitter may present scientific information in a different way than the communicator intended. Cues embedded in the platform indicating accuracy, importance, or popularity may shape the information and influence flow. Furthermore, Twitter may facilitate selective exposure and contribute to polarization based on prior attitudes, beliefs, and values (Brossard and Scheufele, 2013). Still, like traditional media, Twitter plays an important role in the landscape of science communication. Exploring how the discussion of scientific issues on Twitter and traditional media converge or diverge and the impact the increased public discussion on Twitter may have on traditional media will enrich our understanding of the contributions Twitter makes to science communication.

Drawing upon the theoretical frameworks of intermedia agenda setting and framing (McCombs, 2004; Reese et al., 2001), we examined how online news and Twitter covered one specific scientific issue—genetically modified (GM) mosquitoes—and their lead-and-follow relationship in terms of issue framing between 21 October 2015 and 31 December 2016 in response to the Zika virus. Specifically, we traced the interaction between online news and Twitter throughout a full issue-attention cycle (Downs, 1972). Methodologically, we used manual and automated content analysis to annotate news articles and tweets, and conducted a time-series analysis to examine intermedia frame-setting effects. Results of the study shed light on how the discussion of a specific scientific innovation evolved in the emerging media landscape. Theoretically, the study contributes to intermedia agenda setting and framing in that it proposes a more nuanced understanding of intermedia influence in different stages of science communication, an area unexplored in previous literature.

2. Zika and GM mosquitoes

GM mosquitoes have been discussed as a tool to control mosquito-borne diseases such as dengue fever in the US Florida Keys. They were suggested again in response to the surge of the Zika virus. The World Health Organization (WHO) released its first update on the Zika outbreak on 21 October 2015 and declared Zika a public health emergency on 1 February 2016. Meanwhile, the US Centers for Disease Control and Prevention (CDC) confirmed the first local transmission case on 2 February 2016. Before these declarations, only a small group of people knew about the Zika virus in South America. Initially, discussion about the GM mosquitoes centered on controlling malaria and other mosquito-borne diseases, but not Zika. As the public awareness about Zika grew, more people joined the conversation about finding solutions to Zika, including recognizing the benefits and risks of using GM mosquitoes. Oxitec, the company that produces GM mosquitoes as an animal drug, has tried to license its product since 2011 by seeking approvals of field trials to test the safety and efficacy of the drug. In accordance with the National Environmental Policy Act of 1969, Oxitec prepared a draft environmental assessment, which was considered complete when Zika caught the public’s attention. The US Food and Drug Administration (FDA) made the environmental assessment and its preliminary finding of no significant impact available publicly and invited public comments for 60 days, as the proposed field trial was a case without precedent or “closely similar to, one which normally requires the preparation of an environmental impact statement” (Code of Federal Regulations Title (CFR) 21, 2017). In August 2016, the FDA approved the trial of using GM mosquitoes to fight mosquito-borne diseases in Key Haven, FL. However, it required Oxitec to obtain local approval before the field trial. Public discussion spiked during the commenting period and gradually declined once the FDA released its tentative approval until 12 November 2016 when Monroe Country, FL voted in favor of the trial, while Key Haven, the proposed site of the trial, voted against the release. As a result, Oxitec had to find a new site and seek public approval. After the conflicting votes, there was little discussion about GM mosquitoes.

3. Theoretical framework

Media framing

Erving Goffman (1974) defined frames as “schemata(s) of interpretation,” which allow individuals “to locate, perceive, identify, and label issues, events, and topics” (p. 21). Scholars have applied this concept of framing to communication research, and developed a variety of definitions and approaches in understanding media texts (e.g. Entman, 1993; Reese et al., 2001). We adopted Reese’s (2001) definition of media framing, which refers to an “organizing principle that is socially shared and persistent over time, that works symbolically to meaningfully structure the social world” (p. 11). Note that frames are not given or fixed, but are the outcome of negotiating shared meaning (Gamson, 1992). Pan and Kosicki (2001) suggested that framing is a strategic action in public deliberation in which various social actors including journalists, interest groups, and average citizens actively define and refine the discursive fields. Therefore, the process of framing can be highly contested.

Like other public debates, framing scientific discoveries also involves negotiation and contestation. Shwed and Bearman (2010) showed that the formation of consensus regarding various scientific issues takes different paths, with some stirring more controversy than others. They also suggested that this process is not limited to the scientific community and can be extended to public platforms such as the blogosphere. In today’s media environment, both traditional media and emerging social media can help gauge the development, generation, and elaboration of frames related to scientific issues in the public discourse. Twitter, in particular, conveys the voices of a wide variety of social actors who are very involved in science communication (Hopke and Simis, 2017; Newman, 2017).

Benefit- and risk-driven frames are common in news media when covering scientific issues. They help the audience understand complicated scientific issues and reveal the contestations of various social actors in interpreting scientific controversies. While GMOs have sparked heated debate since the 1970s and is still one of the scientific issues on which a sizable divergence exists between scientists and the public (Pew Research Center, 2016b), news coverage of GMOs in recent years has become more balanced in presenting benefits and risks (Bonfadelli et al., 2002; Ventura et al., 2017). In covering GM applications in biotechnology, subtle ways of framing with negative connotations are evident without open oppositions (Ventura et al., 2017).

The domain in which the benefit- and risk-driven frames are used is crucial to investigating the strategic framing effects. For example, personal benefits help reduce the likelihood of the moral rejection of GMOs, whereas environmental risks likely decrease public acceptance (Evensen et al., 2000; Marques et al., 2015). Debates in the United States and Europe on GM foods have outlined a recurring set of benefit- and risk-driven frames in the domains of scientific development, human health, environment, economy, and consumer choice (Augoustinos et al., 2010; Yu and Xu, 2016). The literature on GMOs and media framing is limited because most of the studies are about GM foods. Little is known about the media framing of other applications of biotechnology. Based on the previous literature, we asked:

RQ1: How did the news media and Twitter frame the benefits and risks of GM mosquitoes?

Issue-attention cycle and framing

Downs (1972) proposed five stages to describe how an issue usually emerges and then fades from the center of public attention. The first stage, the pre-problem stage, takes place when an undesirable situation occurs but has not yet caught the public’s attention. The second stage, labeled the stage of alarmed discovery and euphoric enthusiasm, entails a sudden public awareness of the problem. In the third stage, the stage of realizing the cost of significant progress, people begin to understand that solving the problem requires major societal restructuring and the sacrifice of someone’s interests. The hardship associated with the sacrifice and restructuring then leads to a gradual decline of intense public interest, the fourth stage. When public interest declines to a certain point, other issues emerge and replace the current one, which marks the fifth and final stage, the post-problem stage. Downs’ (1972) concept of the issue-attention cycle is helpful for understanding how the public gains and loses interest in a particular issue over time. However, a variety of factors can influence the development of public opinion. Therefore, it may not always follow these five stages (Nisbet and Huge, 2006).

Downs (1972) developed his framework when print media and television dominated people’s lives. Recent research postulated that we could also use the issue-attention cycle to understand attention span in the digital world (Jang et al., 2016). On one hand, the cycle might expand, because there are fewer limits on content production and dissemination on digital platforms. On the other hand, some evidence indicated that the digital attention span could be more volatile and shorter lived, and the discussion of social issues on social media could be unsystematic and irrelevant (David et al., 2016). To what extent the issue-attention cycle in this digital age deviates from the predigital era is worthy of scholarly attention.

Applying the model of the issue-attention cycle to framing research, scholars have found that the news media adopt different frames as the cycle progresses. For example, Shih et al. (2008) showed that the media coverage of both mad cow disease and West Nile virus emphasized uncertainty in the stages of heightened media attention. More evidence-based reporting occurred in the stages when the media and public attention decreased. Focusing on the media coverage of plant biotechnology, Nisbet and Huge (2006) found that technic frames such as policy and regulation were adopted more frequently in the early stages, while later on the media framed the issue from the perspectives of ethics and conflict. When applying the framework of issue-attention cycle to social media, most research focused only on differences in volume instead of framing as the cycle progressed. Specifically, when covering a social movement or disaster on social media platforms, there was a substantial amount of event-driven coverage in the early stages of the cycle (David et al., 2016; Grzywińska and Borden, 2012).

From intermedia agenda setting to intermedia frame setting

Intermedia agenda setting theory posits that media outlets influence each other in covering the news (McCombs, 2004). One reason for this effect is journalists’ co-orientation. Journalists often look to their peers to validate the newsworthiness of the stories. Earlier research showed that newspapers usually set the agenda of broadcast media, and elite newspapers such as the New York Times and Washington Post determine other media outlets’ reporting decisions (Golan, 2006). Recent research has demonstrated that emerging media such as political blogs and online partisan media have become the agenda setters in the new media landscape (Meraz, 2011). As more users come to social networking sites, especially Twitter for breaking news, how the news media and social media influence each other has become an important, new subject of intermedia agenda-setting research.

Just as journalists follow other news media to decide what stories to cover and how to cover them, they also consider Twitter an important source of information (Ahmad, 2010). Conversely, Twitter, where a large percentage of users are average citizens and groups, is also likely to follow the agenda of the news media (Vargo et al., 2014). There are mixed findings about the direction of intermedia agenda setting between the news media and social media. Vargo et al. (2014) showed that to a significant degree the news media led the public conversation on Twitter during the 2012 US presidential election. In contrast, Meraz (2016) demonstrated that the day-to-day political talk on Twitter influenced the agenda of the New York Times and Washington Post. Others argued that intermedia agenda setting is not unidirectional and established a reciprocal relationship between the news media and Twitter (Neuman et al., 2014).

In addition to the transfer of issue salience, intermedia agenda setting can also take place at the level of attribute, or frame. As McCombs and Ghanem (2001) maintained, we can consider the effect of framing the second level of agenda setting. The distinction between agenda setting and framing is not a focus of this study, which instead draws upon the commonality of the two theoretical traditions and examines the transfer of frame salience between media entities. We use the term intermedia frame setting to describe the media dynamics as previous studies have suggested (Scheufele, 1999; Zhou and Moy, 2007).

Research on intermedia frame setting between the news media and social media is rare. As an exception, Neuman et al. (2014) found a reciprocal relationship between the frames of the news media and those of social media. In other words, Twitter users are responsive to news coverage, and journalists also appear attentive to public responses to frames as reflected in Twitter spikes and trends. In the case of GM mosquitoes, Twitter enables the real-time communication of those involved in the debate, and the news media are important sources for the public to obtain information about the scientific innovation. Exploring how the news media and Twitter interacted with each other in covering GM mosquitoes would not only allow us to explicate the intermedia frame-setting effect, but also test the effect in a new context, that of science communication. Due to the scarcity of research on intermedia frame setting in science communication, we chose to pose research questions instead of hypotheses. We examine the transfer of issue salience first to establish a baseline and then focus on the transfer of the salience of this frame.

RQ2: Was the volume of news coverage of GM mosquitoes more likely to predict the volume of Twitter conversations, or the reverse (i.e. the volume of Twitter conversations was more likely to predict the volume of news coverage)?

RQ3: Was the news media’s framing of GM mosquitoes more likely to predict the framing in Twitter conversations, or the reverse?

More important to this study is determining under what circumstances intermedia frame setting occurs. Specifically, will the news media or Twitter take the lead in framing a controversial issue in different stages? Answering this question will help policymakers and health and science communication practitioners make better decisions about their public outreach strategies at various stages as an issue evolves. Zhou and Moy’s (2007) framing analysis of a controversial social event sheds light on this question. Focusing on the interaction between the news media and online bulletin boards, their study found that the new media platform decided how the news media framed the issue initially. Later, the two types of media influenced each other in their framing. Similarly, analyzing the media coverage of the ice bucket challenge, Jang et al. (2016) demonstrated that Twitter influenced how the news media framed the issue in the beginning, while a reverse relationship emerged toward the end of the campaign. These findings accord with Scheufele’s (1999) argument that journalists rely on public opinion for how to report relatively new issues when they have no established rules to follow (Grzywińska and Borden, 2012), while the news media appear to be more powerful in setting the agenda of issues for ongoing debate (Vargo et al., 2015). Therefore, it is logical to expect that Twitter influenced the media coverage when GM mosquitoes were still new to the general public. When more people became aware of the issue and joined the conversation, the news media might have started to shape the public debate about the innovation’s benefits and risks. We made two hypotheses. Again, we examine the media dynamics with respect to the transfer of issue salience first to establish a baseline and then turn to intermedia frame setting:

H1: The volume of tweets about GM mosquitoes was more likely to predict the volume of relevant news articles in the earlier stage of the issue-attention cycle, whereas the volume of news articles was more likely to predict the volume of tweets in the later stages.

H2: Twitter’s framing of GM mosquitoes was more likely to predict the news media’s framing in the earlier stage of the issue-attention cycle, whereas the news media’s framing was more likely to predict the framing in Twitter conversations in the later stages.

4. Method

Data collection

We examined the GM mosquito debate in response to the recent surge of the Zika virus. To capture the entire cycle of the discussion, we analyzed the coverage of GM mosquitoes in the online news and Twitter between 21 October 2015 and 31 December 2016. To locate Twitter’s coverage of GM mosquitoes, we used Crimson Hexagon’s ForSight social media analytics platform to retrieve all tweets that mentioned at least one of the following key words: “GM mosquito” OR “GMO mosquito” OR “genetically modified mosquito” OR “genetically engineered mosquito” OR “GMO aedes aegypti” OR “genetically engineered aedes aegypti” OR “genetically modified aedes aegypti.” The Crimson Hexagon program has access to the Twitter firehose - the unfiltered, full stream of tweets, thus allowing researchers to include all relevant tweets (N = 28,236) in the dataset.

Given that the digital intake of news is continuing to grow in the United States, we also used the Crimson Hexagon’s program to retrieve online news articles about GM mosquitoes (Pew Research Center, 2016a). The program archives fact-based articles from various online news sources around the world, including both elite news organizations, such as CNN and the New York Times, and emerging media such as yahoo.com and the Huffington Post (Crimson Hexagon, 2014). We selected top news organizations listed in the Pew Research Center’s (2015) report and all news sources that produced at least 10 articles about GM mosquitoes. The search terms about GM mosquitoes used to retrieve the tweets were also used to collect the news articles. Our final sample contained 464 articles from 55 media organizations.¹

Manual content analysis

We conducted a manual content analysis of all of the selected news articles (N = 464) and about 10% of the tweets (n = 2850). The Twitter sample was randomly selected from the data using a stratified sampling method to represent different time periods of the conversation.

To answer RQ1, we coded each tweet or news article for the presence of a list of benefit- and risk-driven frames related to the use of GM mosquitoes (1 = present, 0 = absent). Following prior research on GMOs (e.g. Augoustinos et al., 2010; Yu and Xu, 2016), we analyzed health and economic benefits, and health, environmental, and ethical risks. In addition to these well-documented benefits and risks, we added several new ones including cost-effectiveness benefits, cost-effectiveness risks, and experimental risks that were unique to GM mosquitoes and based on observations of media texts during the research design. Health benefits referred to the positive outcomes for human health such as curbing diseases and saving lives. Economic benefits referred to financial gains for a company, industry, or community. Cost-effectiveness benefits were coded as present if compared with other mosquito control approaches, the technology was considered more efficient, effective, and sustainable with regard to cost and time. Health risks were coded as present if the case mentioned potential harm to human health, such as the threat of certain diseases or the transfer of mutant genes to humans. Environmental risks referred to the potential harm to other species, biodiversity, and the ecological balance. Cost-effectiveness risks were coded as present if concerns about efficiency and effectiveness when targeting large geographical areas in the long run and at a reasonable price were mentioned. Ethical risks were coded as present when there was a moral debate. Experimental risks referred to concerns about experimental flaws and the possibility of unintended consequences. For both benefits and risks, there was an “other” category. In addition, we created a frame orientation index to determine whether a tweet or a news article was benefit- or risk-oriented. If a news article or tweet mentioned only benefits but not any risks, we recoded the item as benefit-oriented. If a news article or tweet discussed only risks but not any benefits, we recoded the item as risk-oriented. We categorized the news articles and tweets that discussed both benefits and risks and those that did not use any frames into the third group, “neutral/double-sided.”

To test inter-coder reliability, two communication researchers coded 283 tweets (10%) and 56 (14%) news articles, randomly selected from the sample of manual coding. For the news articles, Krippendorff’s alpha ranged from .71 to 1 with an average of .92. For the tweets, Krippendorff’s alpha ranged from .73 to 1 with an average of .91. The two coders discussed the discrepancies in the coding and then coded the rest of the items.

Issue-attention cycle

We divided the GM mosquito discussion into five distinct stages based on Downs’ (1972) model of issue-attention cycle. The first pre-problem stage (21 October 2015–30 January 2016) was between the time the issue emerged and the day before the WHO declared Zika a public health emergency. We defined the second and third stage as lasting until the week of 5 August 2016 before the FDA approved the trial of using GM mosquitoes to fight Zika and other mosquito-borne diseases in Florida. As Figure 1 illustrates, in these stages, there was heightened issue awareness, indicating alarmed discovery, and euphoric enthusiasm. Note that there was no key event marking the move from Stages 2 to 3, a challenge facing other recent research (Jang et al., 2016). Therefore, we made an arbitrary decision to divide the two stages by eyeballing the volume of conversation. Thus, we determined that Stage 2 lasted from 31 January to 30 April 2016, and Stage 3 from 1 May to 6 August 2016. We defined Stage 4 as the period from the FDA’s approval to the week when the public voted on the trial in Florida (7 August–12 November 2016) when public attention gradually declined. Stage 5 was marked by the conflicting voting results on the GM mosquito trial and the pause of the field trial until the end of 2016 (13 November–31 December 2016).

Figure 1.

The number of tweets and online news articles about GM mosquitoes from 21 October 2015 to 31 December 2016.

Supervised machine learning

RQ2–3 and H1-2 asked about the lead-and-follow relationship between the online news coverage of GM mosquitoes and the discussion on Twitter. Specifically, RQ3 and H2 focused on intermedia frame setting and how the use of benefit- or risk-oriented frames in one media would predict the use of the frames in the other. To investigate this issue, we included all tweets mentioning GM mosquitoes in this part of analysis (N = 28,236). Given the large volume of Twitter data, we used a supervised machine learning (SML) approach to identify the frame orientation of the tweets.

SML uses a set of prelabeled cases to build a concise model that can automatically assign labels to unknown cases (Kotsiantis, 2007). In text analysis, human coders label a sample of text documents (e.g. tweets), and the annotations are used to train an SML model. There are a number of algorithms available to perform SML analysis. We chose the Support Vector Machine (SVM) given its popularity and high performance in classifying texts. For example, Collingwood and Wilkerson (2012) compared the accuracy of four SML algorithms in analyzing Congressional bills and found that SVM outperformed the other three. Scholars in communication research have utilized the SVM to analyze tweets (Van Zoonen and Toni, 2016) and online news articles (Flaounas et al., 2013). We used the SVM to classify instances of benefit-oriented, risk-oriented, or neutral/double-sided tweets separately.

We used the initial sample of manually labeled tweets (n = 2850) to train the SVM models. The tweets were divided into the “training set” (70%) and “testing set” (30%) to assess the performance of each model. Given the poor performance of the initial assessment, we manually coded another random sample of 800 tweets and added them to the models. Given that all three SVM models achieved satisfactory performance (see Table 1), the models were used to predict the frames discussed in the rest of the tweets.

Table 1.

Machine learning model performance.

	Precision	Recall	F-score
Benefit-oriented	0.98	0.95	0.96
Risk-oriented	0.85	0.81	0.83
Neutral/double-sided	0.86	0.92	0.89

To evaluate the performance of supervised machine learning, we divided all the manually labeled tweets into a training set and a testing set. An SVM model was built using the tweets in the training set. The SVM model was then used to predict the sentiment of the tweets in the testing set. The SVM-predicted labels were compared with the manual labels to generate the precision and recall scores. Precision is the ratio of true positives to the total predicted positive observations. For example, predicting benefit-oriented tweets, the prediction score measures out of all the tweets predicted by the SVM model as benefit-oriented, how many of them were indeed benefit-oriented coded by humans. Recall is the ratio of true positives to all observations in the actual case. The recall score measures out of all the tweets coded as benefit-oriented by humans, how many of them were predicted by the SVM model. F-score is the weighted average of prediction and recall.

Data analysis

We addressed RQ1 through chi-square analysis, which is often used to test statistical differences when comparing media content. It analyzes the differences between observed frequencies in categorical data. We cross-tabulated the coded benefits, risks, and orientation with the type of media to answer RQ1.

To address RQ2–3 and H1–2, Granger-causality tests were performed to assess if a significant intermedia frame-setting relationship between the online news and Twitter was present in terms of the discussion about GM mosquitoes. Granger causality addresses autocorrelation (e.g. regressing past values of the outcome variable) and time lags (i.e. the time for intermedia agenda setting to happen), two necessary components to address time-ordered effects. Previous intermedia agenda-setting research considered time lags ranging from a few weeks to a day. In this 24/7 news environment, however, the salience transfer of topics or attributes can happen within hours. Harder et al. (2017) used a 6-hour time lag during the day—06:00–12:00, 12:00–18:00; 18:00–24:00—and then added news stories that were published during the night (24:00–06:00) to the morning lag. Harder et al. (2017: 283) considered a 6-hour time lag “a middle ground between daily or weekly time lags on one hand, and even shorter time lags on the other,” which captures the pace of online platforms and is meaningful. We adopted their approach in selecting time lags but acknowledge that the breakdown is arbitrary. Future research could consider using alternative aggregation methods or relying on information criteria to determine the optimal time lags.

We conducted Granger-causality tests on the number of articles/tweets, the number of benefit-oriented articles/tweets, and the number of risk-oriented articles/tweets, separately. We performed the analysis for the entire time period (RQ2–3) and then for each stage of the issue cycle respectively (H1–2). We used the Augmented Dickey-Fuller test to check whether each time series was stationary or not.

5. Results

News versus Twitter coverage of the GM mosquito debate

Descriptive statistics (see Table 2) show that the online news focused on the health benefits of GM mosquitoes (80.60%), followed by a great deal of deliberation about environmental risks (49.78%) and health risks (41.16%). The news coverage also framed GM mosquitoes from the perspectives of experimental risks (17.24%), cost-effectiveness benefits (14.44%), and cost-effectiveness risks (13.79%). In comparison, Twitter presented far fewer frames than the online news. Nevertheless, the top three frames in Twitter discussions were similar to those prominent in the online news. Twitter covered GM mosquitoes mostly from the angle of health benefits (47.44%), followed by discussions on environmental risks (6.70%) and health risks (4.00%).

Table 2.

Framing differences between online news and Twitter (manual content analysis).

		Online news(n = 464)	Twitter(n = 2850)	χ2	p	V
Benefits	Health	80.60%	47.44%	175.87	p < .001	0.23
	Cost-effectiveness	14.44%	0.00%	420.02	p < .001	0.36
	Economic	2.59%	0.04%	66.46	p < .001	0.14
Risks	Health	41.34%	4.00%	659.53	p < .001	0.45
	Environmental	50.00%	6.70%	666.52	p < .001	0.45
	Ethical	6.28%	0.91%	69.66	p < .001	0.15
	Experimental	17.24%	0.28%	444.07	p < .001	0.37
	Cost-effectiveness	13.79%	0.00%	400.85	p < .001	0.35
Orientation Index	Benefit-oriented	24.14%	46.18%	79.03	p < .001	0.15
	Risk-oriented	8.19%	10.39%	2.12		0.03
	Neutral/double-sided	67.67%	43.44%	94.12	p < .001	0.17

Overall, the discussion of GM mosquitoes on Twitter (46.18%) was significantly more benefit-oriented than that in the online news (24.14%), χ² (2, N = 3314) = 79.03, p < .001, V^* = .15 (Table 2). By contrast, the majority of online news articles were categorized as neutral or two-sided, presenting somewhat balanced coverage of the debate: news 67.67% versus tweets 43.44%, χ² (2, N = 3314) = 94.12, p < .001, V^* = .17.

Table 3.

The Granger causality modeling results.

Modeling	Online news to Twitter	Twitter to online news
Entire time period
Issue salience	13.43**	51.14**
Benefit-oriented frame salience	0.11	53.39**
Risk-oriented frame salience	4.88*	6.23*
Stage 1
Issue salience	0.03	14.3**
Benefit-oriented frame salience	2.51	46.64**
Risk-oriented frame salience	N/A	N/A
Stage 2
Issue salience	10.32**	16.27**
Benefit-oriented frame salience	0.46	0
Risk-oriented frame salience	0.03	2.78
Stage 3
Issue salience	20.84**	2.52
Benefit-oriented frame salience	0.31	0.37
Risk-oriented frame salience	7.57*	2.39
Stages 4 and 5
Issue salience	6.03*	1.82
Benefit-oriented frame salience	0.01	0.03
Risk-oriented frame salience	37.18**	2.88

The Granger causality parameter tests reported in the table are based on F distributions for significance.

N/A indicates that at least one of the time series tested was not stationary.

p < 0.05.

p < 0.01.

Who led whom?

RQ2 asked which media platform was more likely to lead in covering GM mosquitoes, the online news or Twitter, for the entire cycle. The results showed that the online news and Twitter had a reciprocal relationship in the coverage, setting, and responding to each other’s agenda. With respect to intermedia frame setting (RQ3), the online news and Twitter interacted with one another in covering the risks of GM mosquitoes. However, Twitter was more likely to predict the coverage of the online news in discussing the benefits of the scientific innovation.

H1–2 asked about the interaction between the online news and Twitter in different stages throughout the issue cycle. H1 expected that the volume of tweets about GM mosquitoes was more likely to predict the volume of relevant news articles in the earlier stage of the issue-attention cycle, whereas the volume of news articles was more likely to predict Twitter’s conversation in later stages. As the results demonstrated, in Stage 1, the volume of tweets successfully predicted the volume of online news coverage of GM mosquitoes. The reverse relationship—the online news to Twitter—was not found. In Stage 2, Twitter and the online news predicted each other’s agenda in terms of the amount of coverage. We found a reverse relationship in Stage 3 and Stages 4 and 5² when the online news began to set the agenda of Twitter in covering GM mosquitoes. Thus, we confirmed H1.

With respect to intermedia frame setting (H2), the results showed that in Stage 1, Twitter forecasted the online news in discussing the benefits of GM mosquitoes. However, we found no significant intermedia frame-setting relationship in Stage 2. Moving to Stage 3, the online news began to lead Twitter’s conversation in discussing the risks of GM mosquitoes and this pattern extended to Stages 4 and 5. Thus, we also confirmed H2.

6. Discussion

Using both manual and automated content analysis, we examined how the online news and Twitter framed the discussion about GM mosquitoes and the intermedia frame-setting relationship between the two. The news coverage of GM mosquitoes was more neutral and balanced, whereas users on Twitter were more likely than the news media to discuss the innovation’s benefits. Twitter played a leading role in framing the discussion about GM mosquitoes in the beginning of the debate, while the online news gained momentum when the public became aware of the issue.

From our results we can conclude that Twitter was more likely to take the lead in the pre-problem stage, especially in driving the news media to cover the benefits of GM mosquitoes. The frame-setting power of Twitter may have occurred because, as Scheufele (1999) argued, news reporters might not have had established rules to follow on how to frame GM mosquitoes in the beginning, and therefore Twitter appeared to be an important source for their coverage. Later, as the scientific discovery attracted increasing public awareness with euphoric enthusiasm (Stage 2), a reciprocal relationship between Twitter and the online news media emerged. Remarkably, the interaction took place only at the level of issue salience (i.e. the volume of coverage) but not frame salience. This result suggests that there is a point in science communication when the media and Twitter publics respond to each other’s coverage but they tend to insist on using their own frames to cover the issue. This pattern was not evident in earlier research and may indicate a process of frame contestation (Entman, 2008) during the early middle stage of the issue-attention cycle. Ambiguity and uncertainty often create a window for frame contestation. Finally, the news media won the competition. Online news emerged as a more influential player as the issue progressed and the conversation began to focus on the risks of GM mosquitoes. The pattern whereby the news media influenced Twitter’s agenda remained until the public interest in GM mosquitoes faded. The finding is consistent with previous research showing that Twitter is more likely to set the agenda in covering breaking news while the news media are more powerful in shaping the discourse of ongoing debates (e.g. Jang et al., 2016; Zhou and Moy, 2007).

The frame-setting power of different media entities in covering a scientific innovation as the issue evolves that we identified provides important insights for policymakers as well as public health and science communication practitioners for designing public outreach strategies. The Twitter publics often contribute to scientific discussions, and may even influence the news coverage in the beginning of the issue-attention cycle (Newman, 2017). In the initial stage of the debate, policymakers and public health practitioners may want to rely on the Twitter voices to gauge public opinion, particularly because the voices of the interest groups and “early adopters” are not influenced by the media agenda at this point. However, the news media appeared to regain their frame-setting influence as the debate evolved. This finding indicates that policymakers should use caution in interpreting Twitter’s “public opinion” in later stages of the issue-attention cycle because the elite discourse from the news media might be shaping the conversation. In the case of GM mosquitoes, the decision regarding a GM mosquito trial in Florida was not made until the last stage. To accurately assess the climate of opinion, policymakers may want to rely on other methods such as surveys to learn about public attitudes on this issue. On the other hand, the fact that the news media led the conversation on Twitter in discussing the risks of GM mosquitoes is reassuring. Well-rounded assessments of a scientific innovation are also crucial for the development of science and technology. Finally, we also found that there was a point at which frame contestation in covering GM mosquitoes occurred. This result indicates that various social actors and stakeholders can take advantage of certain points in the issue-attention cycle to push their frames in the public discourse.

Taken together, public outreach strategies should be developed at various stages of the issue-attention cycle for better science communication outcomes. To communicate their messages more effectively to the public, scientists should focus on engaging influential users on Twitter in the initial stage of an issue cycle, and supplying journalists with clear and accurate information in later stages. The best timing for science communication practitioners to exert a significant influence on issue framing might be the early middle stage when framing contestation occurs. At such a point, the situation is ambiguous and there is a need for information to fill the gap between current knowledge and sufficient information to make an informed decision.

Based on the findings, we argue that intermedia agenda setting may change its direction over time, and different media outlets may be influential in leading different aspects of the conversation. Thus, any conclusion about who leads and who follows in this emerging media landscape is premature. Nevertheless, our findings are based on one specific science communication case. Future research may consider applying the approach to examine intermedia agenda setting or frame setting in other sociocultural contexts.

As with any study, our research has several limitations. First, we obtained the sample of the online news for the manual content analysis from judgmental sampling that might not fully cover the sources of information an average Internet user encounters when searching or browsing online. Second, the comparisons of the benefits and risks frames between the online news and Twitter could be confounded by the significant difference in the length of these two genres of media due to the 140-character limit of Twitter. To account for this difference, we created an orientation index instead of focusing on the frequency of individual frames. Nevertheless, despite these limitations, which create new opportunities for further research, we maintain that our study adds to our understanding of how different media platforms frame scientific innovations in the emerging media landscape and intermedia frame setting as the issues develop.

Footnotes

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author biographies

Weirui Wang (PhD, Pennsylvania State University) is currently an Associate Professor at Department of Communication, Florida International University. Her research focuses on health and science communication for social changes, with an emphasis on understanding the impact of culture and social media.

Lei Guo (PhD, The University of Texas at Austin) is an Assistant Professor in the Emerging Media Studies Division at College of Communication, Boston University. Her research focuses on the development of media effects theories, computational social science methodologies, and emerging media and democracy in the United States and China.

References

Ahmad

(2010) Is Twitter a useful tool for journalists? Journal of Media Practice 11: 145–155.

Augoustinos

Crabb

Shepherd

(2010) Genetically modified food in the news: Media representations of the GM debate in the UK. Public Understanding of Science 19: 98–114.

Bonfadelli

Dahinden

Leonarz

(2002) Biotechnology in Switzerland: High on the public agenda, but only moderate support. Public Understanding of Science 11: 113–130.

Brossard

Scheufele

(2013) Science, new media, and the public. Science 339: 40–41.

Büchi

(2017) Microblogging as an extension of science reporting. Public Understanding of Science 26: 953–968.

Code of Federal Regulations Title (CFR) 21 (2017) Part 25 Environmental Impact Considerations. Available at: https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfcfr/CFRSearch.cfm?CFRPart=25

Collingwood

Wilkerson

(2012) Tradeoffs in accuracy and efficiency in supervised learning methods. Journal of Information Technology & Politics 9(3): 298–318.

Crimson Hexagon (2014) Content source: Advanced. 4 December. Available at: https://help.crimsonhexagon.com/hc/en-us/articles/203511215?flash_digest=1e82480f3341d90aa3516e7993164f6f87cbf65d

David

Ong

Legara

EFT

(2016) Tweeting Supertyphoon Haiyan: Evolving functions of Twitter during and after a disaster event. PLoS ONE 11(3): e0150190.

10.

Downs

(1972) Up and down with ecology – the “issue-attention cycle.” The Public Interest 28: 38–50.

11.

Entman

(1993) Framing: Toward clarification of a fractured paradigm. Journal of Communication 43: 51–58.

12.

Entman

(2008) Theorizing mediated public diplomacy: The U.S. case. The International Journal of Press/Politics 13: 87–102.

13.

Evensen

Hoban

Woodrum

(2000) Technology and morality: Influences on public attitudes toward biotechnology. Knowledge, Technology & Policy 13(1): 43–57.

14.

Flaounas

Ali

Lansdall-Welfare

Bie

Mosdell

Lewis

et al . (2013) Research methods in the age of digital journalism: Massive-scale automated analysis of news-content—topics, style and gender. Digital Journalism 1(1): 102–116.

15.

Gamson

(1992) Talking Politics. New York, NY: Cambridge University Press.

16.

Goffman

(1974) Frame Analysis: An Essay on the Organization of Experience. Cambridge, MA: Harvard University Press.

17.

Golan

(2006) Inter-media agenda setting and global news coverage: Assessing the influence of the New York Times on three network television evening news programs. Journalism Studies 7(2): 323–333.

18.

Grzywińska

Borden

(2012) The impact of social media on traditional media agenda setting theory: The case study of Occupy Wall Street Movement in USA. In: Dobek-Ostrowska

Lodzki

Wanta

(eds) Agenda Setting: Old and New Problems in the Old and New Media. Wrocław: University of Wroclaw Press, pp. 133–155.

19.

Harder

Sevenans

Van Aelst

(2017) Intermedia agenda setting in the social media age: How traditional players dominate the news agenda in election times. The International Journal of Press/Politics 22(3): 275–293.

20.

Hopke

Simis

(2017) Discourse over a contested technology on Twitter: A case study of hydraulic fracturing. Public Understanding of Science 26(1): 105–120.

21.

Jang

Park

Lee

(2016) Round-trip agenda setting: Tracking the intermedia process over time in the ice bucket challenge. Journalism 18(10): 1293–1308.

22.

Kotsiantis

(2007) Supervised machine learning: A review of classification techniques. Informatica 31: 249–268.

23.

McCombs

(2004) Setting the Agenda: The Mass Media and Public Opinion. Cambridge: Polity Press.

24.

McCombs

Ghanem

(2001) The convergence of agenda setting and framing. In: Reese

Gandy

Grant

(eds) Framing Public Life: Perspectives on Media and Our Understanding of the Social World. Mahwah, NJ: Lawrence Erlbaum Associates, pp. 67–81.

25.

Marques

Critchley

Walshe

(2015) Attitudes to genetically modified food over time: How trust in organizations and the media cycle predict support. Public Understanding of Science 24: 601–618.

26.

Meraz

(2011) Using time series analysis to measure intermedia agenda-setting influence in traditional media and political blog networks. Journalism & Mass Communication Quarterly 88: 176–194.

27.

Meraz

(2016) An expanded perspective on network agenda setting between traditional media and Twitter political discussion groups in ‘everyday political talk. In: Guo

McCombs

(eds) The Power of Information Networks: New Directions for Agenda Setting. New York, NY; London: Routledge, pp. 66–87.

28.

Neuman

Guggenheim

Jang

Mo Jang

Bae

(2014) The dynamics of public attention: Agenda-setting theory meets big data. Journal of Communication 64: 193–214.

29.

Newman

(2017) Tracking the release of IPCC AR5 on Twitter: Users, comments, and sources following the release of the Working Group I summary for policymakers. Public Understanding of Science 26: 815–825.

30.

Nisbet

Huge

(2006) Attention cycles and frames in the plant biotechnology debate: managing power and participation through the press/policy connection. The International Journal of Press/Politics 11(2): 3–40.

31.

Pan

Kosicki

(2001) Framing as strategic action in public deliberation. In: Reese

Gandy

Grant

(eds) Framing Public Life: Perspectives on Media and Our Understanding of the Social World. Mahwah, NJ: Lawrence Erlbaum Associates, pp. 33–66.

32.

Pew Research Center (2015) State of the news media 2015. Available at: http://www.journalism.org/2015/04/29/state-of-the-news-media-2015/pj_15-04-27_stateofthenewsmedia_summarygraphics_640x320-09/

33.

Pew Research Center (2016a) The modern news consumers. Available at: http://www.journalism.org/2016/07/07/pathways-to-news/

34.

Pew Research Center (2016b) The new food fights: US public divides over food science. Available at: http://www.pewinternet.org/2016/12/01/public-opinion-about-genetically-modified-foods-and-trust-in-scientists-connected-with-these-foods/

35.

Pew Research Center (2017) Science news and information today. Available at: http://www.journalism.org/2017/09/20/science-news-and-information-today/

36.

Reese

(2001) Prologue: Framing public life: A bridging model for media research. In: Reese

Gandy

Grant

(eds) Framing Public Life: Perspectives on Media and Our Understanding of the Social World. Mahwah, NJ: Lawrence Erlbaum Associates, pp. 7–31.

37.

Reese

Gandy

Grant

(2001) Framing Public Life: Perspectives on Media and Our Understanding of the Social World. Mahwah, NJ: Lawrence Erlbaum Associates.

38.

Scheufele

(1999) Framing as a theory of media effects. Journal of Communication 49: 103–122.

39.

Shan

Regan

De Brún

Barnett

van der Sanden

Wall

et al . (2014) Food crisis coverage by social and traditional media: A case study of the 2008 Irish dioxin crisis. Public Understanding of Science 23: 911–928.

40.

Shih

Wijaya

Brossard

(2008) Media coverage of public health epidemics: Linking framing and issue attention cycle toward an integrated theory of print news coverage of epidemics. Mass Communication and Society 11: 141–160.

41.

Shwed

Bearman

(2010) The temporal structure of scientific consensus formation. American Sociological Review 75: 817–840.

42.

Van Zoonen

Toni

(2016) Social media research: The application of supervised machine learning in organizational communication research. Computers in Human Behavior 63: 132–141.

43.

Vargo

Basilaia

Shaw

(2015) Event versus issue: Twitter reflections of major news: A case study. In: Cotton

Robinson

(eds) Communication and Information Technologies Annual, vol. 9 (Studies in Media and Communications). Bingley: Emerald Group Publishing, pp. 215–239.

44.

Vargo

Guo

McCombs

Shaw

(2014) Network issue agendas on Twitter during the 2012 U.S. presidential election. Journal of Communication 64(2): 296–316.

45.

Veltri

(2012) Microblogging and nanotweets: Nanotechnology on Twitter. Public Understanding of Science 22: 832–849.

46.

Ventura

Frisio

Ferrazzi

Siletti

(2017) How scary! An analysis of visual communication concerning genetically modified organisms in Italy. Public Understanding of Science 26: 547–563.

47.

(2016) Public discourse on genetically modified foods in the mobile sphere: Framing risks, opportunities, and responsibilities in mobile social media in China. In: Wei

(ed.) Mobile Media, Political Participation, and Civic Activism in Asia. Berlin: Springer, pp. 81–102.

48.

Zhou

Moy

(2007) Parsing framing processes: The interplay between online public opinion and media coverage. Journal of Communication 57: 79–98.