The problem stream and frame analysis

The MSF occupies an important place as a key theoretical approach in the study of public policy and the policy-making process (e.g. Howlett et al. Reference Howlett, McConnell and Perl2015; Cairney and Jones Reference Cairney and Jones2016; Jones et al. Reference Jones, Peterson, Pierce, Herweg, Bernal, Raney and Zahariadis2016; Rawat and Morris Reference Rawat and Morris2016).Footnote ¹ In Agendas, Alternatives and Public Policies, Kingdon (Reference Kingdon1997) argued that policymaking is made up of three separate and independent streams: the problem stream (problems that are perceived by the public as requiring government action), the policy stream (where experts and policy specialists examine problems and propose policy solutions) and the politics stream (which refers to changes in public opinion, executive or legislative turnover and advocacy campaigns). Kingdon argued that policy change occurs when a policy entrepreneur applies an idea from the policy stream to an issue in the problem stream at a time when this coupling is acceptable within the political stream. This coupling could be triggered by external focussing events (e.g. crises, or accidents), the presence or absence of policy entrepreneurs or institutionalised events such as periodic elections or budget deadlines (Birkland Reference Birkland1997, Reference Birkland1998). Thus, time and timing (along with ambiguity) are critical dimensions in the MSF because one’s influence over the policy process will always be temporally conditioned (Zahariadis Reference Zahariadis2003, Reference Zahariadis2007).

Our particular concern here is with the problem stream, which has been identified as an understudied component of the MSF (Knaggård Reference Knaggård2015). A concern with the problem stream is also broadly aligned with recent calls for greater engagement between the MSF and ideas (e.g. Béland Reference Béland2016; Winkel and Leipold Reference Winkel and Leipold2016), framing (e.g. Boscarino Reference Boscarino2016; Sager and Thomann Reference Sager and Thomann2017) and agency (e.g. Mukherjee and Howlett Reference Mukherjee and Howlett2015). At the general level, the MSF argues that each stream operates independently up until they cross and a policy window opens. The principle of stream independence has been contested (e.g. Robinson and Eller Reference Robinson and Eller2010,) but Zahariadis (Reference Zahariadis2007, 81) has argued that if it is understood as a “conceptual device” then scholars applying the MSF can assume that the streams act “as if” they were independent. To put it more succinctly, although streams are not empirically independent, when conducting research there is analytical value in approaching an issue from within one stream alone.

We can also analytically distinguish the problem stream because it is underpinned by a qualitatively different kind of logic. Although the politics stream may be influenced by constructions produced in the problem stream, it is fundamentally governed by an electoral calculus and involves coordinating the input of support or political opposition for an initiative (Sager and Thomann Reference Sager and Thomann2017). This is a different logic from that which is present in the problem stream, which is concerned with who uses what frames, and to what effect, in the public debate surrounding a contested policy problem. Issues in the problem stream are therefore addressed to a public audience and subject to public criteria of persuasion, rather than the technical criteria of the policy stream or the expedience criteria of the political stream.

The problem stream also exhibits some distinct characteristics at the actor level. For example, Knaggård has argued that “problem brokers” operate to “frame conditions as public problems and work to make policy makers accept these frames” (Reference Knaggård2015, 452). She argues that problem brokers perform a role that is distinct to the problem stream because many actors frame problems without making policy suggestions. This is consistent with Mukherjee and Howlett’s (Reference Mukherjee and Howlett2015, 67) “differentiated subsystem approach”, which analyses how “distinct subsets of actors” in each of the three streams undertake particular tasks, including the “articulation of problems”.

The problem stream also exhibits its own particular flow and temporality. The literature on issue attention cycles (e.g. Downs Reference Downs1972; Nisbet and Huge Reference Nisbet and Huge2006) has long shown that policy issues get problematised at different rates. In this article, we capture this variation by referring to “problem windows”, which we define as moments in time when a pressing problem emerges, or there is a perception that a pressing problem has emerged, within the problem stream. We describe this as a “window” to emphasise the punctuated nature of these moments as distinct from the overall flow of the problem stream. Problem windows appear when an issue in the problem stream becomes a subject of public contestation and debate. Most analysis has focussed on why problem windows open, rather than on why they close. However, understanding how problem streams “flow” requires a focus on both dimensions. As such, there can be both substantive and analytical reasons for distinguishing the problem stream from the policy and politics streams.

Contestation within the problem stream can also occur in multiple domains, although the broadcast media remains a particularly important venue for at least three reasons (Kingdon Reference Kingdon1997). First, broadcast media are a central site where actors on multiple sides of controversial issues are represented and policy actors compete to gain legitimacy and construct core meanings (Carragee and Roefs Reference Carragee and Roefs2004; Gamson Reference Gamson2004). Second, broadcast media outlets serve a legitimation role as an independent actor that selects certain issues, frames and actor representations over others (Gamson and Wolfsfeld Reference Gamson and Wolfsfeld1993; Carvalho Reference Carvalho2007; Shanahan et al. Reference Shanahan, McBeth and Hathaway2011; Evensen et al. Reference Evensen, Clarke and Stedman2014). Third, messages reach wider publics through the mass media (Kriesi Reference Kriesi1995; Boykoff and Boykoff Reference Boykoff and Boykoff2004). This is a conclusion supported by a US study on hydraulic fracturing, which found that most citizens continued to source their information from the broadcast media (Stedman et al. Reference Stedman, Jacquet, Filteau, Willits, Brasier and McLaughlin2012).

Although MSF scholars have examined the role played by framing in the problem stream (e.g. Sager and Thomann Reference Sager and Thomann2017), disagreements persist about what actually constitutes a frame, particularly in relation to the involvement of frames within cognitive processes (Schön and Rein Reference Schön and Rein1994; Stone Reference Stone1998; Chong and Druckman Reference Chong and Druckman2006; Bacchi Reference Bacchi2009; Cacciatore et al. Reference Cacciatore, Dietram and Iyengar2016; van Hulst and Yanow Reference Van Hulst and Yanow2016). However, we define frames with respect to properties of communications, treating frames as containing a set of meanings that organise objects and events in relation to a wider context of activity. They are external to the observer, directing one’s attention to particular attributes of an object while deflecting attention from other aspects, meanings and orderings of reality contained within communications (Burke Reference Burke1984, 111). Different observers may construe meanings in varied ways, but to the extent an observer can understand a language, frames obtain an autonomous coherence at least for a given time and place. This is particularly the case for broadcast media, which addresses a wide audience and is predicated on stabilities in meaning.

Although frames exist independent of particular advocates, Ferree and Merrill (Reference Ferree and Merrill2000, 456) have argued that linking frames with the agents who sponsor them is an important step towards providing a “more complex model of how people organise, use, and change their frames”. We use the term “frame sponsor” to refer to actors, including individual persons, organisations and government entities, who support particular frames. The inclusion of frame sponsors helps respond to those who have criticised the MSF for its lack of a detailed notion of agency (e.g. Mukherjee and Howlett Reference Mukherjee and Howlett2015). This is important because the decision by one frame sponsor to use a particular frame can place limits on the extent to which another frame sponsor can make use of the same frame on a contested policy issue; for example, if an industry actor concedes that their activities produce environmental harm then it may make it harder for others to deny that claim without first referring back to that industry actor’s framing of the issue. Certain actors, such as government officials or industry representatives, may also dominate domestic policy debates with a common set of priorities, preconceptions and frames, whereas the media also tend to “index” their coverage to the range of frames provided by political elites (Tuchman Reference Tuchman1978; Bennett Reference Bennett1990; Callaghan and Schnell Reference Callaghan and Schnell2001; Livingston and Bennett Reference Livingston and Bennett2003). In the case of controversies such as the CSG debate, the presence of frame sponsors therefore becomes an important empirical question. Although the broadcast media may legitimate claims of different policy actors, we anticipate that political authorities and their frames will prove particularly consequential in news reporting given that they are the ones who have the power to effect legislative and regulatory change.

Changes in the presence of frame sponsors and the frames that they advocate may also have an impact on a problem window, its duration and eventual closure. Frame adoption approaches presume that actors may use a variety of frames over the course of a debate. In a contested field, frames may change if the prevailing frame does not gain enough traction and even those frames that are successful are likely to be countered before long. For example, Boscarino’s (Reference Boscarino2009, 426) study of forestry policy found that interest groups “problem surf” by connecting “their solution of sustainable forestry to different policy problems” dependent on what is politically salient and likely to gain more traction. We expect that the introduction of new frames and new actors will prolong a problem window, whereas frame convergence between actors will close a problem window as convergence signals at least partial closure within a debate.

Finally, there is the question of what mechanisms cause problem windows to close. Kingdon (Reference Kingdon1997) argues that problem windows typically open in response to events. Events can disrupt existing patterns as they provide an opportunity for policy actors to “reframe” the narrative in their preferred direction (Tuchman Reference Tuchman1978; Gamson and Modigliani Reference Gamson and Modigliani1989; Iyengar Reference Iyengar1990; Benford and Snow Reference Benford and Snow2000; Greenberg Reference Greenberg2005). Frames can change when opponents attempt to capitalise on events that highlight a misfit between a frame and an unfolding event. However, elite convergence may also close a problem window. As the mediatisation of politics depends on the creation of a narrative of conflict (Blumler and Gurevitch Reference Blumler and Gurevitch1995; Strömbäck Reference Strömbäck2008), convergence in elite frames removes the driver behind the news coverage. Last, elections may also serve as a point of closure by legitimating an agenda through resolution by an independent arbiter: the voting public. In sum, we expect that problem windows will close when conflict subsides.

Research design

This research design develops a novel approach for identifying problem windows that combines algorithmic methods with qualitative hand-coding. Algorithmic methods, such as topic models, enable researchers to cover large amounts of text that would be too resource-intensive to code using manual methods alone (Grimmer and Stewart Reference Grimmer and King2013; Van Holt et al. Reference Van Holt, Johnson, Carley, Brinkley and Diesner2013). These methods have already generated interest among public policy scholars (e.g. Klüver et al. Reference Klüver, Mahoney and Opper2015), and the approach developed here adds a qualitative dimension to that prior work. Combining algorithmic methods with hand-coding has advantages: it provides contextual sensitivity in interpreting the data (Aipperspach et al. Reference Aipperspach, Rattenbury, Woodruff, Anderson, Canny and Aoki2006; Lewis et al. Reference Lewis, Zamith and Hermida2013; Hand and Hillyard Reference Hand and Hillyard2014) and, just as computer-based analyses of texts are prone to making errors, human coders are prone to nonsystematic biases even when they use highly established coding schemes (Mikhaylov et al. Reference Mikhaylov, Laver and Benoit2012; González-Bailón and Paltoglou Reference González-Bailón and Paltoglou2015).

Topic modelling has particular advantages for this study, as it can be used “diagnostically” to identify successive days of consistent news reporting on a particular topic.Footnote ² Although policy areas typically involve multiple dimensions (or “topics”), we are able to use topic models to construct problem windows within a single topic based on heightened and extended periods of topic prevalence within the news media (Nowlin Reference Nowlin2016). Topic prevalence is operationalised in terms of model “thetas”, which is a measure of the expected proportion of a topic in each document. Studies of framing have often used a simple article frequency count as a justification for “zooming in” on particular periods of time (e.g. Hurka and Nebel Reference Hurka and Nebel2013; Jaspal and Nerlich Reference Jaspal and Nerlich2014; Metze Reference Metze2014), but this does not always capture periods of heightened problematisation. For example, we found that brief stock market reports (several sentences long in the Finance pages) accounted for a high proportion of the total number of news articles published on CSG. These articles do not problematise the CSG issue but they still contribute to the article count.Footnote ³ Consequently, using heightened periods of topic prevalence on a policy-related topic gives greater certainty that selected problem windows relate to actual periods of problematisation, rather than the other types of extraneous reporting that might be captured using a simple article frequency count.

Once a policy-related topic with a high overall topic prevalence had been identified, the topic proportions across all articles for that topic were summed and aggregated to identify problem windows: successive days of high topic prevalence. We retrieved and hand-coded all the news articles that fell between those dates for the presence of frames and frame sponsors. In short, we used the topic modelling “diagnostically” to identify problem windows and the qualitative analysis “analytically” to investigate the reasons why problem windows close.

The research design was split into four steps:

• ∙ Step 1: we captured the relevant data and prepared it for the topic modelling.

• ∙ Step 2: we ran the topic modelling, selected a 15-topic model and one policy-related topic from within that model.

• ∙ Step 3: we identified four problem windows from within the selected topic.

• ∙ Step 4: we hand-coded for the presence of frames and frame sponsors within the news articles published during the four problem windows identified in Step 3.

In Step 1, 2,414 newspaper articles were retrieved using the database Lexis/Nexis and the search term “coal seam gas”. The search covered every news article (excluding “Letters to the Editor”) that matched the search term in four Australian newspapers: The Australian, The Australian Financial Review, The Courier Mail (Brisbane) and the Sydney Morning Herald, between 1 January 2008 and 2 August 2016. Once retrieved, the articles underwent text preprocessing so that they could be used in the topic modelling. Topic models use “model vocabularies”, which are the list of terms out of which patterns are identified to form topics. The topic models were unsupervised and prepared using Latent Dirichlet Allocation (LDA). LDA is a method commonly used in the study of journalism to analyse large volumes of news content and has been used by researchers studying contentious policy issues such as CSG (e.g. Blair et al. Reference Blair, Weible, Heikkila and Evensen2016; Günther and Quandt Reference Günther and Quandt2016; Jacobi et al. Reference Jacobi, van Atteveldt and Welbers2016; Bauer et al. Reference Bauer, Barberá, Ackermann and Venetz2017). It has also been shown to be effective at identifying topic distributions in newspapers when compared with hand-coding (Blair et al. Reference Blair, Weible, Heikkila and Evensen2016).Footnote ⁴ The model was unsupervised because we approached the data not knowing ahead of time the topics or their relative proportions.

Although LDA does not require any preanalysis, some text preprocessing is necessary to allow comparability of terms independent of their location within sentences. This step involves removing numbers, punctuation and standard stop-words (e.g. and, for, in, it), normalising text as lower case and reducing terms to their grammatical stems. Infrequently used terms and rarely used words are also removed to reduce unwanted noise and avoid creating topics that are only germane to one particular article. As a consequence, 625 very short articles were entirely eliminated and the topic modelling was conducted on the 1,789 reports that remained (see the Supplementary Information for further detail).

In Step 2, we ran the topic modelling (see Table 2 in the Supplementary Information for the results). It is a common feature of topic modelling to assume the number of topics before the modelling, and thus care needs to be taken when making this decision (Grimmer and King Reference Grimmer and Stewart2011). After running several tests (see below), a 15-topic model was selected within which several policy-related topics were identified (e.g. on the greenhouse effect, the mining tax and jobs and the economy). We then selected one policy-related topic related to “risk and regulation” as a basis for identifying the problem windows. We selected this topic for three reasons. First, the risk and regulation topic was the most prevalent topic among all of the topics – both those that concerned policy-related issues and those that did not (see Figure 1 in the Supplementary Information). From the 15 topics, risk and regulation was therefore identified as the most likely topic to exhibit the problem windows that were the particular motivation for this study and the research questions that informed it (Nowlin Reference Nowlin2016, 319).Footnote ⁵ Second, unlike other topics (e.g. the greenhouse effect), the risk and regulation topic is particular to the concerns surrounding CSG development. This makes the identification of discrete problem windows using algorithmic means more feasible as the topic contents are more likely be specific to the CSG controversy. Third, the risk and regulation topic provided a balance between having a topic that included enough problem windows to enable a thorough analysis and the time that it would take to hand-code the news articles that were selected through this process.

We interpreted the results from the topic modelling using the highest probability terms (those terms that are most frequent in the topic) and the “frex” terms (those terms that are most unique to the topic). When combined, these results define a topic according to both the salience of the terms that compose it and those terms that most distinguish one topic from the other topics present in the model. We gave each topic a name on this basis. The “highest probability terms” for our selected topic on risk and regulation were: said, gas, coal, seam, govern, nsw, water. “Coal”, “seam” and “gas” appear among the highest probability terms as these are the terms defining the policy area. In addition, “govern” and “water” reference both the governing process and concerns about water pollution, one of the regulatory concerns raised by CSG development. Last, “nsw” denotes New South Wales, one of the geographic parameters bracketing this policy controversy. The “frex terms” were: frack, windsor, pilliga, buckingham, okan, tara and epa. Focussing on these terms, and discarding “okan”, which seems to have multiple reference points, “frack” is a stemmed form of the term “fracking”, which is a means of extracting CSG. The local CSG industry maintains that it does not frack, but opponents either challenge that claim or maintain that it is a distinction without a difference. Windsor is a reference to Tony Windsor, an independent MP. Pilliga is a forest area where CSG development has been hotly contested in northern NSW. Tara is a location in Queensland where there is a CSG production facility. The EPA is the NSW Environmental Protection Agency, and Buckingham is a Green party MP. Although these data show the close association of particular actors with the topic, actors can frame the terms within a topic in a variety of ways that express either opposition or support for CSG.

There is no single diagnostic test to evaluate topic models, and thus we conducted several different tests to validate our choices. First, we measured and then compared the semantic coherence and exclusivity of models ranging from 10 to 20 topics (see Table 1 in the Supplementary Information). Semantic coherence measures whether the terms in a topic model appear with a high degree of terminological cooccurrence, whereas exclusivity measures the level of differentiation between the topics. The results from these tests show that a 15-topic model has the second highest semantic coherence (−85.10190) and a higher exclusivity (9.623557) than the 10-topic model. This suggests that a 15-topic model achieves an appropriate balance between the extremes of either mixing distinct issues together in the same topic, which would reduce topic exclusivity, or artificially parsing more topics than the underlying text would warrant. Second, we tested for the robustness of the risk and regulation topic by running models for 10, 15 and 20 topics. The risk and regulation topic was the most prevalent topic in all three models, which helps confirm that the appearance of this topic is not spurious. Finally, we also used the qualitative analysis to confirm the risk and regulation topic and its overall coherence as per the approach taken by others (e.g. Blair et al. Reference Blair, Weible, Heikkila and Evensen2016; Roberts et al. Reference Roberts, Stewart and Tingley2016). Our coders were not told about the topic, and the selected articles contained more than our single topic of interest; however, a thematically salient discussion and clear narrative concerning the risks associated with CSG was present in all four problem windows. Thus, we were able to combine a theoretical understanding of the object of analysis with small N comparison between the model results and the texts themselves.

In Step 3, four problem windows were constructed from within the risk and regulation topic. Problem windows are demarcated temporally, and thus it was necessary to develop a measure of topic prevalence as a function of time. The prevalence of the risk and regulation topic was summed on a daily basis, and a seven-day rolling mean of its daily topic prevalence was calculated to account for media cycles.Footnote ⁶ Figure 1 presents the results from this analysis where the y-axis shows the summated proportions across all documents, aggregated daily. As the underlying data reflect the proportions of each topic in each document, and there is no limit on the number of articles per day, this metric has no logical upper bound.

Figure 1

Risk and regulation over time.

The data in Figure 1 are quite “noisy”, necessitating criteria by which problem windows could be systematically identified and differentiated from other time points. We addressed this by summing the topic prevalence at the article level (model thetas) to create a measure of both the distribution of this topic across articles and the intensity of the focus on this topic in any one particular article. Hence, a topic featuring marginally in many articles might still have the same prevalence on a given day as a topic that features heavily in only two articles. We then ordered the results from the day with the highest topic prevalence through to the day with the lowest topic prevalence. We constructed the problem windows by looking within the top 100 days of topic prevalence for at least three days of successive media attention with no more than one day in-between. Where this existed, we took it as evidence of sustained reporting on the topic: that is both an initial increase in the topic’s prevalence and a follow-up in interest over successive days. The one-day gap was permitted to allow for other news outlets to pick up a story covered initially by a rival, which is sign of the continuing relevance of the issue. Table 1 presents the results from Step 3. The total topic prevalence in Table 1 refers to the summated topic prevalence across all articles between the date range specified for that window. The four problem windows have been named after the major news story reported in that window.

Table 1

Problem windows

We validated our selection of problem windows by running several tests. First, we attempted to construct a fifth problem window. This was only possible within the 200 top days of reporting. The window was also only three days long, and did not revolve around a concrete problem situation.Footnote ⁷ Second, the topic prevalence was substantially lower than for the other four windows. t-tests confirmed that the topic concentrations on a daily basis during our problem windows were substantially higher than during a hypothetical fifth window and that these differences were statistically significant (t=4.253, p<0.000). Finally, we charted the daily topic prevalence values in the four problem windows, which show a clear period of problem build-up and then decline in topic prevalence corresponding to the emergence, climax and closure of the problem window (see Figures 2a–2d in the Supplementary Information).

In Step 4, we hand-coded the selected articles (n=110) in Table 1 for the presence of frames and frame sponsors. We included all the news articles published within the date range covered by each problem window so that we could explore whether a problem window closes when a new issue intrudes. We also added a lead and lag of three days to capture the emergence and closure of a topic, including any transformation in the frames and frame sponsors. Frames were identified as coherent positions attached to an actor, which provide a description of a situation organised with respect to a normative position (for or against CSG) and a range of action. We also hand-coded for the presence of frame sponsors: entities who advocate a particular frame (e.g. individuals, organisations or position titles), as well as their abstract representations (e.g. “protestors” or “industry groups”) (Carragee and Roefs Reference Carragee and Roefs2004; Gamson Reference Gamson2004). We coded every frame to an actor whether they were quoted directly or paraphrased.Footnote ⁸ We then associated frame sponsors with particular social groupings. We used this coding to track the changing presence of frame sponsors, as well as similarities or differences in the frames that they were using. We checked for intercoder reliability on a random sample of 10% of the articles using the κ coefficient (Neuendorf Reference Neuendorf2002). The result from this test was 0.63, which is on the higher end of the fair to good agreement range (0.4–0.75) (Hayes and Krippendorff Reference Hayes and Krippendorff2007). These results match the findings from similar comparisons undertaken elsewhere (e.g. Klüver Reference Klüver2009; Klüver et al. Reference Klüver, Mahoney and Opper2015; Blair et al. Reference Blair, Weible, Heikkila and Evensen2016).

Problem windows and the CSG debate in Australia

We present our findings in two sections. First, we detail the aggregate results and then we discuss the four problem windows in turn. In total, the 110 selected news articles contained a total of 485 frames and 350 unique frame sponsor mentions (there are more frames than unique frame sponsor mentions because a frame sponsor may deploy more than one frame in a single mention). Nine categories of frame sponsors and nine categories of frames were identified during the analysis. The nine frame categories consisted of five anti-CSG frames [n=258 split between high risk (n=107), farmers and land (n=58), water (n=38), community (n=32) and business and politics (n=23)] and four pro-CSG frames [n=201 split between managed risk (n=103), jobs and growth (n=46), energy (n=38) and governance (n=14)]. We also included an “other” category (n=26) for any frames that did not match one of the nine major frame categories identified here. Although some of these frames draw upon overlapping premises, we differentiate them as they provide logically distinct grounds for supporting or opposing CSG development. Table 2 lists the nine categories of frames and their key features.

Table 2

Frame categories

CSG=coal seam gas.

Three frame sponsors consistently adopted anti-CSG frames [n=80 split between farmers (n=36), anti-CSG groups (n=25) and farming lobbyists (n=19)] and pro-CSG frames, respectively [n=82 split between energy companies (n=39), energy lobbyists (n=32) and trade unions (n=11)]. A further three frame sponsors [n=139 split between state politicians (n=71), federal politicians (n=61) and community members (n=7)] adopted a mix of anti- and pro-CSG frames. We also included an “other” category (n=49) for any frame sponsors who did not match one of the nine major frame sponsors. They also sponsored a mix of anti- and pro-CSG frames. Table 3 lists the nine major frame sponsors and their key features.

Table 3

Frame sponsors

CSG=coal seam gas.

The first problem window – industry development – contained 128 frames and 89 unique frame sponsors. Two news stories dominated this problem window during a 16-day period starting from Wednesday 13 October 2010. The first news story centred on a press release in which a CSG company announced the detection of Benzene, Toluene, Ethylbenzene and Xylene (BTEX) chemicals in its exploration wells (n=11 articles). This was followed two days later by a news story concerning the Federal Environment Minister’s approval of a LNG export facility on Curtis Island (n=16 articles). Five articles reported on these two stories in the same piece.

The major frame sponsors in this problem window included farmers (n=17), farming lobbyists (n=12), anti-CSG groups (n=10) and energy companies (n=13). Other frame sponsors were present (n=31) but they were mentioned less than ten times. Anti-CSG frames (n=83) dominated over pro-CSG frames (n=43). The three dominant anti-CSG frames were high risk (n=46), water (n=15) and farmers and land (n=13), whereas the two dominant pro-CSG frames were managed risk (n=28) and jobs and growth (n=14). Although the water frame was present across both news stories, the high risk and jobs and growth frames featured more heavily in the LNG story. Most frames emphasised the environmental and social risks of CSG and/or the need for tighter regulation. In short, this was a problem window dominated by anti-CSG frame sponsors (n=45) and anti-CSG frames (n=83). The problem window closes five days after the news story on the BTEX spill and three days after the government’s announcement approving the LNG export facility. The CSG industry’s managed risk frame appears to have been effective in closing the problem window, including its repeated reassurances that the risks associated with the spill were manageable and that the “community should have confidence in the processes we’ve put in place – as soon as there are issues we inform the regulator” (The Australian 2010). The bipartisan political support for the LNG export facility and its timed announcement also helps explain why this problem window closed.

The second problem window – land access – contained 145 frames and 111 unique frame sponsors. Two news stories dominated this problem window in a 19-day period from Wednesday 3 August 2011. The first news story concerned otherwise unconnected events such as the announcement of a parliamentary inquiry into CSG, a public meeting held by an energy lobby group and a group of anti-CSG protesters “gatecrashing” an industry conference (n=12 articles). The media used these protests to report more widely on community concerns towards CSG. This problem window was also extended by a second news story concerning comments made in a radio interview by the former Opposition Leader, Tony Abbott, in which he argued that farmers’ property rights needed greater protection (n=11 articles). The interview was interpreted as a departure from the bipartisan political support that had been shown towards the CSG industry, which helped spark a political debate about the role of the federal government in regulating CSG. Anti-CSG parties, such as the Greens, used Abbott’s interview to argue for more federal regulation and stronger land access requirements.

The major frame sponsors in this problem window were federal politicians (n=41) and state politicians (n=21) who both used a mix of pro- and anti-CSG frames. There was a fairly even split between anti-CSG (n=24) and pro-CSG frame sponsors (n=18), including agricultural voices (n=19) and the energy sector (n=18), but anti-CSG frames (n=81) were present more than pro-CSG frames (n=59). Abbott’s radio interview led to a significant rise in the farmers and land frame. There was a split between Green politicians who sponsored high-risk frames and the Government who tended to sponsor the managed risk and energy frames. Other prominent frame sponsors included state politicians, agricultural voices and energy lobbyists and companies. State politicians did not engage with the land and access frame but did sponsor the high risk, managed risk, jobs and growth and community frames. Agricultural voices consistently sponsored the water frame and the farmers and land frame, whereas energy sector voices sponsored the energy, managed risk and jobs and growth frames.

In terms of frame sponsors, the reporting notably shifts towards politicians (n=52 over a five-day period) in the week after Abbott’s radio interview. Other frame sponsors (n=22 over a five-day period), both pro- and anti-CSG, are marginalised during this period. The problem window closes once federal politicians converge around a governance frame, which places responsibility for CSG regulation with the states. Thus, elite frame convergence among federal politicians’ closes off debate and prevents other frame sponsors from extending the problem window’s duration.

The third problem window – buffer zone regulation – contains 60 frames and 43 unique frame sponsor mentions in a 13-day period from Saturday 16 February 2013. A government announcement that it would introduce a 2 km “no go zone” for CSG exploration adjacent to urban areas and prime horse/winery land dominated this problem window. Most of the coverage of this story took place in the two days before and after the announcement (n=9 articles). There was one other minor news story in this problem window (n=2 articles).

The major frame sponsors in this problem window included trade unions (n=9), state politicians (n=8), federal politicians (n=7) and the energy sector (n=8). The presence of trade unions is related to the minor news story, whereas the three other frame sponsors (n=23) all relate to the new buffer zone regulation. Anti-CSG (n=24) and pro-CSG frames (n=31) were fairly evenly distributed, but politicians and the energy sector heavily dominated the frame sponsorship. Although state politicians sponsored high risk and community frames to justify the new regulation, the energy sector sponsored frames that emphasised energy (security), jobs and growth. Pro-CSG frame sponsors used the energy frame to draw attention to NSW’s looming energy crisis and the jobs and growth frame to focus on the industry’s economic potential.

The final problem window – NSW state election – contains 152 frames and 107 unique frame sponsor mentions in a 15-day period from Friday 20 March 2015. The major news story in this problem window concerned the NSW state election, which took place on 28 March 2015 (n=12 articles). CSG exploration was a key election issue in several marginal seats where longstanding incumbents from the National party were seen as being under potential threat from anti-CSG candidates, including the Greens and other Independents. The NSW Labor party also took an anti-CSG position by announcing its intention to ban Santos’s controversial Narrabri Gas Project, one of only two serious CSG proposals in the state at the time. There were five other articles in this problem window.

The major frame sponsors in this problem window include state politicians (n=39), energy lobbyists (n=10), energy companies (n=13) and federal politicians (n=8). Notably, anti-CSG frame sponsors (n=8) are relatively marginalised compared with pro-CSG frame sponsors (n=27) despite anti-CSG frames (n=70) and pro-CSG frames (n=68) remaining fairly balanced. The “other” category (n=25) is relatively high and includes a range of actors from musicians to actors and radio talk-show hosts to ex-public servants. On the presence and distribution of frames, risk-based frames (n=72) are the most important. Managed risk frames (n=41) are mainly sponsored by the energy sector that presents CSG as a low-risk technology that has local community support. Opposition to CSG comes from “outsiders” who only want to “make trouble” by “politicising” the decision-making process. Conversely, high-risk and community frames are mainly sponsored by state politicians, particularly the Greens and Labor. The business and politics frame (n=10) – which focusses on distrust and collusion between industry and the government – also gains prominence in this problem window for the first time. New frame sponsors, such as Simone Marsh, an ex-public servant whistle-blower, and Dayne Pratzky, an activist/actor in the movie Frackman, sponsor this frame. The problem window closes three days after the election. Although the anti-CSG movement gains votes in several swing states, they fail to achieve an overall win and the problem window closes. There are two notable differences between this problem window and the others: first, a greater diversity in frame sponsors and, second, an even distribution between anti- and pro-CSG frames despite a larger presence of pro-CSG frame sponsors.

Discussion: why do problem windows close?

The results from this analysis suggest that problem windows generally close in response to institutionalised events and actions by formal political authorities. These events and actions lead to the closure of problem windows because they decisively reshape the context of problematisation, resolving or undermining the motivational grounds on which the problem window first emerged. Although the four problem windows detail distinct actions, they all function in the same way: the problem windows close once the provocation to continue the conflict has receded and there is no further interest in continuing to report on the original story that had provoked the initial contestation. These findings suggest that nonelite policy actors will find it challenging to open a problem window independently from either an action or intervention by a political authority or an institutionalised event.

We can also identify three specific mechanisms, which all point to how press coverage is primarily motivated by conflict. The first mechanism, elite frame convergence, occurs when an issue becomes noncontentious and a problem window closes because elites come to share a common framing of the issue. The second problem window provides an example of this mechanism at work. Abbott’s radio interview broke with the established bipartisan policy and led to frame divergence between the Government and Opposition. However, the Government and Opposition’s subsequent convergence around a governance frame effectively closed the problem window.

Problem windows also closed following public statements. In general, these problem windows are relatively short-lived because they are limited to the announcement itself. The introduction of a new buffer zone regulation in the third problem window provides an example of this mechanism. The new regulation was announced and state politicians were given an opportunity to justify the change. Without a new news story, the problem window closes. However, industries can also use public statements as an attempt to get ahead of a controversy and close off debate. In our analysis, this is illustrated by the BTEX press release, which successfully managed to close a problem window by promoting consistent and reassuring frames around managed risk. The company got “ahead of the story” by preempting public opposition and developing a response to it.

Our final mechanism is elections. Here, problem windows close because they provide democratic legitimation for particular issue framings in a space of contestation. The state election was branded as a “referendum” on the “future of CSG”. Although anti- and pro-CSG frames were fairly evenly distributed, anti-CSG groups were largely marginalised from the debate by state and federal politicians who sponsored anti-CSG frames. The problem window closed shortly after the election result when it was reported that the anti-CSG movement made important gains but not enough to win.

Our finding that the operation of problem windows is closely linked to the ability to control or prolong a controversy suggests opportunities for further research. First, a comparative examination of other policy domains could be used to further delimit the conditions under which problem windows persist and close. Second, the methodological approach outlined here could be used to compare problem windows in other venues such as social media, written submissions to parliamentary inquiries and government reports. The broadcast news media remains important, but social media is also a crucial communication platform and written submissions present frame sponsors with an opportunity to outline their position on an issue at length. Studying other venues may also present opportunities to analyse the norms within different media systems (Hallin and Mancini Reference Hallin and Mancini2004). This study found a more or less equal number of pro- and anti-CSG frames, but there were also times when new frames and new frame sponsors had more of an opportunity to enter into the debate (e.g. during elections). The news media’s tendency towards “balanced coverage” in which competing arguments are presented, the time and resource pressures of the news room and how journalists report issues more generally may both shape the conclusions of this study but also present an opportunity for comparison between how problem windows operate in the news media compared with other communication channels where similar norms and practices do not apply. Interviews may also present opportunities to explore problems that are being discussed out of public purview, as well as issues that the news media have not reported on.