1 Introduction

What explains the motivation of governments to take costly actions to address climate change? In many countries, especially democracies, the answer to that question depends heavily on public opinion. While many factors shape public opinion, a rich vein of scholarship has focused empirical attention on the media (Walgrave et al. 2008; Francke 1995; Yagade and Dozier 1990; Erbring et al. 1980; Tan and Weaver 2007; Commission on Accelerating Climate Action, Communication Working Group 2023; Sisco et al. 2021). Such a focus has been based on the idea that the media is both a shaper of public opinion and a good barometer of what the public thinks (Boykoff and Ravi Rajan 2007; Elias and Nelkin 1996; Programme 2007; Schoenfeld et al. 1979; Sampei and Aoyagi-Usui 2009; Wonneberger et al. 2020; Hart 2011; Sisco et al. 2021). Motivated by the proposition that media coverage is an important shaper and barometer, many studies have looked at patterns in news coverage about climate change and, in many cases, tried to explain those patterns. This paper contributes to that literature—less on the explanation for variation in coverage and more on the underlying data about media coverage that can be used for explanatory studies.

Before turning to patterns of media attention, we note that several lines of scholarship give confidence that the media is, indeed, a relevant quarry for studying political attention and policy pressure related to climate change. One line of thinking emphasizes that media outlets have finite resources and thus attention to one topic (e.g., climate change) is zero sum with others for which the public devote attention and apply pressure for social or policy change (Hase et al. 2021; Schmidt et al. 2013). In addition to revealing where public minds are (and are not) focused, other scholars also point to the role of the media as an organizer of complex information on a topic and thus a plausible central accelerant and focuser of public attention (Carvalho 2010). A substantial literature has emerged around how to communicate messages to the media on the logic that how the media interprets those messages and conveys them to the listening and reading public is material to public attitudes (American Academy of Arts and Sciences 2023; Somerville and Hassol 2011).

There are good reasons to believe that elite news coverage of climate change could be diverging from the rest of the country. As concerns about climate change have risen many of the elite news outlets have expanded their climate-related coverage. In the last five years, the New York Times, Associated Press, Washington Post, and the LA Times have all publicly expanded their climate news desks (New York Times 2017; Associated Press 2022; Buzbee 2022; Times 2023). Through wire services and syndication, some of this new coverage may diffuse outside those elite news outlets, but whether to republish such stories remains a local choice. Meanwhile, the last decade has seen a marked decline in the number and size of locals news outlets (Muse Abernathy 2023). This decline in local outlets has occurred in tandem with a rise in hyper-partisan (typically conservative) local outlets—often propped up by national conglomerates (Tow Center for Digital Journalism 2024). The hollowing out and shift to the political right may lead to big differences in the news consumed by Americans, although that pattern has not been investigated systematically—in part, perhaps, because empirical scholarship on news coverage about climate change has tended to focus narrowly on elite news outlets.

The literature on media attention is hard to summarize because it is so diverse. Following the approach in Schmidt et al. (2013), updated to include recently published studies, we offer one systematic look at the prior scholarship (Table 1). Our emphasis is on the empirical base for each study—media sources and search terms.

Table 1 Summary of sources that studied the trends in climate change-related coverage in U.S. news sources (based on the work of Schmidt et al.)

From that perspective, one pattern is clear: nearly all the existing research has concentrated an analytical spotlight on a small number of marquis sources of news, such as The New York Times. These studies reflect a combination of convenience (elite news sources are easy to spot and have long had fully searchable databases) and relevance. Elite news sources are key agenda-setters in policy circles and also frame how other news outlets cover complex topics like climate change. (Many sources simply reprint stories from elite sources; others engage in re-reporting—filing new stories that carry similar messages as the elite news outlet that broke the original story.). As such, many older studies that relied on researchers manually reading through each article individually, understandably, chose to focus their efforts on the most widely circulated, and read, papers of record. Of all the studies on Table 1, only three (Grundmann and Krishnamurthy 2010; Grundmann and Scott 2014; Bromley-Trujillo et al. 2023) look beyond an elite sample of news sources—in two of those studies, still, a restricted sample of news outlets covered by Lexis Nexis was used. The paper by Bromley-Trujillo and colleagues shifted the empirical focus away from elite papers by studying the top circulated paper in 49 out of 50 U.S. states. A few studies in countries other than the United States have attempted to look at news sources outside of this elite subset (Boykoff and Mansfield 2008; Boykoff 2008; Ghosh and Boykoff 2019).

In this paper we explore insights that could emerge from empirical research that looks at a much wider array of news sources. We use a new database, MediaCloud, which differs from nearly all earlier databases for its extremely broad coverage. As with earlier studies, we focus on the country that already has been the main subject of analysis—the United States. The expanded coverage of MediaCloud is particularly notable for its inclusion of news outlets based where most Americans live: over the full database examined in this study about 81% of all MediaCloud news articles covered are from the “State and Local” category (see table SI-1 in Supplemental Information “SI”). A significant portion of Americans still rely on local news as their primary source of information (though more look at national news). The Knight Foundation found that 18% of Americans pay a “great deal of attention” to local news, that figure was 32% of Americans for national news. Such data vary with a variety of factors, such as race. Black Americans, for instance, pay even more attention to local news (29%). More importantly, however, for research papers treating news outlets and agenda setters is the fact that on average, Americans place a higher trust in their local news outlets when compared with nation-wide elite sources (Knight Foundation 2023).

In this paper we are focused on two questions. First, does looking at a fuller sample of American news sources rather than a narrow slice of elite outlets lead to differences in the news coverage? The answer to that question, we will show, is that coverage is rising but highly variable with notable intense spikes. That leads to a second research question: are there significant differences in the events that appear to drive spikes in climate change-related coverage between elite news outlets and the broader array of heartland news sources? Some of the existing literature on media coverage of climate change has been probing what drives media coverage, such as links between extreme weather events or particular political events and coverage of climate change (e.g., (Saunders et al. 2018; Schäfer et al. 2014). The answers, we suggest, may depend on the type of media coverage.

2 Methods and Data

The centerpiece of this paper is a novel source of primary data: the news-article repository housed by MediaCloud. This is an open-source platform developed across numerous institutions, including the University of Massachusetts Amherst, Northeastern University, and the Berkman Klein Center for Internet & Society at Harvard University. We examine the entire MediaCloud repository from its origin (01/01/2011) through the end of the last full calendar year before we began our analysis (12/31/2022).

Our aim is to be as inclusive as possible. Thus, we built a dataset that included both MediaCloud’s “United States – National” and “United States – State & Local” datasets. In totality, this paper examined 170,477,176 news articles, from 9,000 unique sources, over 12 years (see SI-1). MediaCloud’s coverage has expanded over time (see Figure 1), with a large expansion in 2018 that roughly doubled the number of annual articles covered in the database. (Future research will need to pay close attention to these shifts in data coverage. For this paper, we note that the proportion of news types—local and national—shifts towards a preponderance of local articles as MediaCloud coverage becomes more inclusive.)

Fig. 1
figure 1

Total number of articles examined by year broken down by heartland and elite categories used in this paper. The dramatic rise in sources in 2018 is driven by MediaCloud adding a significant body of “State and Local” papers during that year. For detail see table SI-1, which shows the number of a articles by category by year and figure SI-1, which shows the total number of articles queried by day

Mindful of the first research question in this paper—differences between elite and non- elite news sources—we then created two distinct datasets. One is what we call “elite” news sources—the five most often the subject of prior empirical research (Table 1): The New York Times, The Wall Street Journal, The Washington Post, USA Today, and The Los Angeles Times. By focusing on these five media sources, we aim to use methods aligned with earlier studies rather than opine on whether any particular news source, such as USA Today, is elite in its orientation.

The other is what we call “heartland” news coverage—all the news sources included in MediaCloud’s “United States – National” and “United States – State & Local” datasets except for the five elite sources. In American parlance, the heartland is the broad swatch of territory in the middle of the country—geographically and intellectually distinct from more urban, more elite worlds of the coasts. Indeed, most of the coverage in MediaCloud—especially after the expansion in 2018—involves state and local sources that are not typically visible in elite spheres. All told, the narrower elite dataset accounts for only 2% of the stories of the combined datasets (see Figure 1 and also table SI-1).

We are mindful that it is hard to find a single term that well-describes the non-elite news sources covered in the MediaCloud database. Our choice of the term “Heartland” is not without flaws. It includes papers ranging from the Stamford Advocate to the Boston Globe. It also includes the Des Moines Register, the Bozeman Daily Chronicle, and the Fort Bliss Bugle. It’s a big category that, numerically, is dominated by papers designed for state and especially local coverage. By using the term “heartland” we are intending to convey the broadest geographical area centered on the whole of the country, not overweighted by elite news sources that are all edited from newsrooms on the coasts located in blue states. Our usage of this term is likely more expansive than most readers may assume from the term. But the key difference between elite and non-elite news sources is that the latter speaks disproportionately to state and local markets. And, as shown in figure SI-13, there are substantial and growing differences in climate-related coverage between “U.S. State and Local” source and “U.S. National” sources when compared to our “Elite 5” subset. National elite and national non-elite news sources show divergence (see Figure SI-13). But numerically the biggest source of divergence between elite and non-elite coverage will come from state and local news outlets. Before 2018 state and local news outlets accounted for two-thirds of the articles queried in our dataset; starting in 2018, when the yearly number of articles nearly doubled, state and local sources have accounted for about seven-eighths of the total articles we queried. Given that this paper is aiming to contrast a broad swath of U.S. papers to the elite subset commonly used for studies we favor the generic term “heartland” and invite future analysis to motivate a more refined array of categories.

As this paper relies heavily on the MediaCloud database, some additional detail on the data source is warranted. MediaCloud’s unit of analysis is a newspaper or online news source or magazine (e.g., The New York Times, Breitbart, The New Yorker, the Peoria Journal Star, Vox, and the Arizona Herald). From 2011 to the present, MediaCloud has a searchable database of all articles from its covered sources. The capacity to query the database is constrained in important ways by how MediaCloud already organizes its data. Notably, it has organized sources of news into categories: two of those being “United States – National” and “United States – State & Local”. The search algorithm allows for keyword searches across very broad geographical categories, such as “national” vs “state & local” but is not yet ripe for finer grained queries that many scholars who are interested in detailed patterns and explanations would want to perform. Without doing 9,000 independent queries, a researcher cannot answer questions such as: how many news sources in the non-elite news outlets (i.e., what we call “Heartland”) never mention climate change.

Analytically we then follow three steps for each of the two datasets. First, we identify all climate-related coverage, defined as any article that uses the terms “climate change” or “global warming” or “greenhouse gas” or greenhouse effect” in their article headline or body. While this approach may prove expansive (e.g., tagging an article about a climate change in Middle Eastern politics) we spot check the results to find these situations to be quite rare. We also expect that such hits will be unbiased with respect to time or clustering in time. Some scientists and federal agencies have stressed the differences between the terms “climate change” and “global warming” with the former being more encompassing and technical accurate for the broader phenomena in question (U.S. Environmental Protection Agency 2023). We find that the media are, indeed, shifting their vocabulary. In elite news sources in 2011, 52% of all news articles found using the search terms “climate change” or “global warming” used only the term “climate change.” By 2022 that share had risen to 82%. In heartland news sources the pattern is similar: a rise from 52% to 86%.

Nonetheless, using both terms in our search does not seem to bias our results (see figure SI-2). In addition, this multi-phrase approach is consistent with other studies (e.g. Brossard et al. 2004; Grundmann and Scott 2014; and Molder and Calice 2023). For completeness, we report in SI the main results for the paper had we taken a more restrictive search of just “climate change.” The differences are small and don’t materially alter any of the main results reported in this paper (see figure SI-3 and figure SI-7 to figure SI-12).

Second, we compute averages across the entire 11 years of coverage for two intervals. We calculate a six-month average over 22 discrete six-month periods for the full 11 years (six-month buckets were selected as a balanced choice between minimizing the day-to-day noise while still capturing the yearly and inter-year fluctuations). Then we calculate discrete three-day averages across the entire dataset (yielding 1,461 discrete three-day periods). These two intervals allow us to quantify spikes in coverage by calculating short-term (three day) deviations from the background (six-month) pattern of coverage. (For completeness, see equation SI-4 for a sample calculation.)

Having calculated deviations from background we then rank the top fifteen spikes in coverage—the episodes with the highest percent deviation from background six-month rates. For each spike we then use MediaCloud’s keyword analysis and Google News to identify the underlying events that are driving the news cycle for each spike.

3 Results

As every earlier study has found, attention to climate change is rising. In the full calendar year of 2011, 0.68% of all U.S. news articles (elite and heartland datasets combined) were related to climate change. In 2022 coverage had risen to 1.60%. This growth is notable (135% over 11 years, see Table 2) but extremely volatile as shown in Figure 2.

Table 2 Median, means, and standard deviations for climate coverage in the heartland and elite datasets
Fig. 2
figure 2

This figure shows two plots for both the heartland and the elite datasets. The first plot is the fainter, and more volatile, line. This plot is the, by day, percentage of articles that referenced climate change. The data is highly volatile, obscuring the long-term trend we aim to call out. As such, a 60-day moving average was added to the figure, the two darker lines. The moving averages start on the 60th day of the dataset and average the 59 days, plus the day the average is on, to produce the presented average

Four features stand out: the spike in coverage in late 2015 (the meetings in Paris that yielded the Paris Agreement), the spike in coverage in early 2017 (President Trump’s withdrawal from Paris), the decline in coverage in early 2020 (the shift in media attention to the global pandemic) and the spike in coverage in late 2021 (the 26th Conference of the Parties meetings in Glasgow).

Looking at the two datasets distinctly there are big differences in annual coverage. The heartland coverage saw a 144% increase between 2011 and 2022. The five elite sources saw roughly double that rise: a 299% increase in their volume of climate-related coverage. (Of those five, The New York Times accounts for more of the increase in coverage than any other.) Table 2, below, elaborates on this point and lays out the median, mean, and standard deviation for each dataset by year.

With such a noisy dataset there will be random days when heartland coverage of climate change exceeds coverage from the elite sources. If the medians of the two datasets are similar, then those random days will be frequent; as the medians diverge such occurrences will be rarer. Figure 3 shows those occurrences for the entire dataset (the full pie) and by year (slices). The driver of these patterns is seen in Table 2. In 2011 the median coverage of climate change in elite newspapers was only 50% higher than the median in heartland news sources. In 2022 the elite median was more than double the heartland median. As time passes the chance that heartland news sources, on any day, will have more to say about climate change becomes quite rare. That shift is evident, as well, in the average coverage levels. In 2011, the heartland sources discussed climate change proportionally more on 24% of the days – in 2022, it was 2.8% (Figure 3).

Fig. 3
figure 3

Chart that shows the number of days, per year, where the heartland dataset had a higher proportion of climate-related stories than the elite dataset. Over the entire 11-year period there were 640 days (out of a total of 4383 days in 11 years) when heartland coverage exceeded elite coverage. Three-quarters of those days occurred in the first six years of the dataset. Since 2019 the daily probability that heartland news sources will have more coverage of climate change when compared with the elite sources has been less than 5%

Figure 4 shows the six-month background rate of coverage for climate change along with three-day spikes. We then report the top 15 events for the heartland dataset (Table 3) and the elite dataset (Table 4).

Fig. 4
figure 4

The percent deviation and six-month averages were calculated in line with the equations discussed in the methodology section. The six-month averages are plotted on the right axis of the chart. The percent deviations are plotted on the left axis. Figure SI-6 shows the data from figure 4 with more fidelity – broken down by dataset

Table 3 Heartland dataset - top 15 three-day spikes in climate change-related coverage, sorted by percent deviation
Table 4 Elite dataset - top 15 three-day spikes in climate change-related coverage, sorted by percent deviation

4 Discussion

Both heartland and elite sources show that coverage of climate change is increasing. This finding is consistent with what many other analysts have already reported: the public is paying more attention to climate change, and so are the media (e.g., AAAS Commission). Heartland coverage more than doubled (increase in 144%) from 2011 and 2022. The five elite sources rose three-fold (299%) over that same period, when measured as the fraction of all news sources that addressed climate change. That finding of increased coverage is robust to many different ways of searching the MediaCloud database. Indeed, using just a narrower search on the term “climate change” yielded a more striking growth pattern: a rise of 227% for the heartland sources and 449% growth for the elite sources (Table SI-8).

While the daily data are volatile, even semi-yearly averages show some volatility. There are six periods for the elite dataset and seven periods for the heartland dataset when semi-yearly coverage declined (Figure SI-6). Those include:

  • early 2013 when coverage appears to shift to the early days of the second term of the Obama administration, a period with little attention to climate change.

  • early 2016 as the US national election began to attract attention.

  • early 2018 with an easing in coverage following the unusually high attention in 2017, the year when President Trump withdrew the United States from the Paris Agreement.

  • throughout the first half of 2020 when news coverage shifted, for a while, to the COVID-19 pandemic.

Of these declines, the largest for the heartland sources occurred in the latter half of 2013, where the six-month average dropped 39% from the previous six-month average. For the elite sources, the largest drop occurred in the first half of 2020 as Covid-19 began to spread in the United States and overall news coverage shifted to the pandemic. During the early period of the pandemic the six-month average in coverage of climate change dropped 41% from its previous six-month average. The covid crash was not long lived, and the upward trend resumed within a year. By contrast, the decline in coverage contemporaneous with the rise of the pandemic was much less pronounced in the heartland news sources—presumably because climate-related coverage was already so much lower when compared with the elite papers.

The single most striking finding in this study is the divergence between the two datasets over time. In the first four years of the study (2011-2015) on any given day there was a 20% probability that the fraction of total daily stories addressing climate in heartland news sources would exceed the fraction in elite sources. Elite news outlets have always given more attention to climate change than heartland outlets. But in the early 2010s those differences were small enough that one day in five climate change would play a more prominent role in the average American’s news feed than it did for the elite. In the last four years of our study the differences were much more extreme: only 5% of days saw more coverage in heartland sources than the elite sources. There are many potential reasons for this divergence in climate-related coverage. One that has been postulated is the well documented decline in local journalism. Many local papers have shut down between 2011 and 2022; moreover, the outlets that remain have become more reliant on the republication of national stories. Delving into such trends is outside the scope of this paper, but they seem unlikely to explain the patterns we observe. Shifting to mere republication should make national and local papers converge with the elite national papers—even though we observe the opposite. Moreover, state and local papers have choices about what to republish, and those choices reflect editors’ intuitions about local interests. On top of all that, while there has been a lot of death in state and local news coverage the declines in absolute volumes of news outlets should be compensated, at least to a degree, by the method we use—looking at fractions of total coverage, rather than total coverage.

The second difference illuminated by the analysis is a slight disparity in the events that drive extreme peaks in coverage of climate change between the two datasets. Overall, we find a lot of similarity across heartland and elite news sources. The top two spikes in both datasets were increases in coverage by a factor of four to five due to President Trump’s withdrawal from Paris in 2017 and the convening of the UN Climate Summit in September 2014, which set the final stage for the Paris conference a few months later (Tables 3 & 4). Both types of new outlets give extensive coverage to the Paris conference (2021) and the climate conference four years earlier, in Durban, that framed the diplomatic road that ended in Paris. (The Durban conference in 2011, known as COP17, was the first major reset after the disastrous COP15 in Copenhagen that ended in disarray with no agreement on the road ahead. By Durban the road was coming into focus.).

There are still some differences in the spikes in coverage. In the heartland dataset there is extensive coverage of a series of billboards erected by the heartland Institute (no relation) in 2012 opposing action on climate change. Heartland news sources also gave a lot of attention to two clusters of comments on climate change by Pope Francis in 2015 in the runup to the negotiation of the UN Climate summit and the issuance that year of the Pope’s encyclical Laudato Si. In the elite dataset there is heavy coverage of the US-China climate summit in 2014, which many analysts see as the key political agreement that made it possible to conclude, a year later in Paris, a similar agreement with nonbinding emission pledges that included all countries. The elite dataset also saw a spike in coverage when the Bulletin of the Atomic Scientists changed its Doomsday Clock to a mere one minute before midnight. Elites, it appears, care about the Bulletin; non-elites did not pay as much attention.

Overall, however, the spikes in coverage between the two sources show a general agreement. Our purpose in focusing on spikes is to help illustrate broad patterns with information about specific news cycles. But many caveats apply, of course. We look at only 15 spikes for each type of media. Future research might look at more, but such an approach must be attentive to the fact that news coverage is noisy: as one moves down the list of event spikes, the noise in the data becomes more prominent and true spikes are harder to detect. Moreover, spike detection requires attention to time periods when spikes are more common and the time interval over which a spike is measured. Most of the spikes identified in this paper happened before 2017—plausibly because our spike detection method looks at deviations in bursts of coverage compared with a longer-term average. As the longer-term averages have risen with an overall increase in climate coverage it may become harder to identify genuine spikes in coverage based on this statistical method.

Our findings parallel other divergences in American attitudes about climate change— between left and right, and between Washington and the rest of the country. While we focus on the incidence of coverage of climate change, not the content of the articles, there is a strong divergence over the period of study.

4.1 Future Research

Finally, this study points to several areas for fruitful additional research. Here we focus on three.

First, if MediaCloud is to be a richer source of data for research on climate change then more research is needed to understand underlying shifts in coverage and coding within the dataset. There are significant changes in total coverage in the dataset that could affect trends and other empirical patterns—even when those trends are computed (as we do) in percentage of total coverage. Among the empirical patterns that need more explanation—and accounting for in statistical assessments—are bursts of new coverage notably in 2013 and 2018 (see figure 1). In addition, there are brief unexplained drops in the number of articles available for query in the database. In particular, there was a three-week period in January of 2022 where the number of U.S. articles available for query dropped to just 5% of the number of articles available in the months preceding and proceeding the drop (figure SI-1).

There are clear structural questions that remain with the MediaCloud dataset. A lot of spadework is needed, and with such a huge dataset the spades will be hefty. Related, our search method—like nearly all other media pattern studies—is a dragnet approach with possibly lots of data bycatch. We are assuming that the errors are normally distributed and unbiased, but more work—with finer attention to search terms and their catch—is needed. Part of MediaCloud’s power, its breadth, is also a structural limitation. MediaCloud would better serve researchers in the future if they gave researchers a less restricted view of some of their background data. This would enable researchers to better answer questions arising from structural anomalies that appear in the dataset.

Second is the matter of explanation—both for the overall level of coverage and for possible differences in substance. The divergence in overall coverage is quite powerful but deserves more detailed explanation. As for substance: we looked at fifteen spikes in coverage in each of the two datasets and then used brute force to identify the story that was driving the cycle. That spike analysis was merely intended to help illustrate possible substantive differences. That approach, however, is quite limited in many ways—such as in what it can say, especially in the early years of the data set where a small number of articles over a three-day period can exceed the threshold that qualifies as a “spike” in our formulation. With automated text analysis it should be possible to do more granular analysis of spikes in news coverage and also, perhaps, begin to explain larger patterns in the stories that catch the attention of editors and readers. (Empirical research on letters to the editor, where those exist, could also help round out the picture by showing what motivated readers do in response).

While the purpose of the present study is not to explain, robustly, the diverging patterns in coverage we can already speculate on some explanations—informed, in part, by what earlier studies have explored. One plausible explanation lies with the elite news outlets, as suggested in the introduction. Most of these news outlets—with the partial exception of the Wall Street Journal—conspicuously expanded coverage of climate change (New York Times 2017; Buzbee 2022; Associated Press 2022; LA Times 2023). More reporters mean more stories. Of these news outlets, what happens at the New York Times—which perhaps most conspicuously expanded climate coverage—matters the most because it is the most heavily resourced of the elite papers thanks to its overwhelming advantage in subscribers (Watson 2024). The phenomenon of reporting begetting news is hardly contained to climate change, as the drama over news outlets hiring reports to cover Taylor Swift reveals (Silver 2023).

In parallel with changes at elite news sources there are many changes in incentives for coverage at heartland news outlets. The decline in local coverage may be a possible explanator, although our approach assesses coverage as a fraction of news output and thus, in principle, should not be affected by shifts in total volumes of coverage if they arise proportionally. More important may be the shift to the political right—often hyper-partisan right—of local outlets that are propped up by national conglomerates (Tow Center for Digital Journalism 2024). Studies that have emphasized partisanship effects in media coverage are consistent with this factor in the economics and organization of media coverage affecting coverage (e.g., Bolsen and Shapiro 2018). That shift could help explain a shift in both the supply of news material (from national parents) and selection of materials for inclusion by local editors.

Still other explanations may look to climate change itself as a topic that is more intrinsically national or global and thus more likely to attract editorial attention in national and especially elite news sources (Ejaz and Najam 2023). Research that seeks to find national factors explaining variations in national coverage is consistent with this idea that the “nation” is the right unit of analysis for explaining coverage (e.g., Vu et al. 2019). Similarly, other studies emphasize the global nature of climate change (e.g., Barkemeyer et al. 2017), suggesting that news outlets inclined to cover global stories might also be more inclined to cover climate disproportionately—and as more global events deal with climate so coverage should rise in such news outlets disproportionately. Most climate stories, at least historically, deal with topics that are more national in orientation (e.g., treaties, Congressional legislation, Presidential speeches). National outlets, especially those with large climate desks, may be well positioned to cover all climate-related disasters whereas local outlets may only cover the few disasters affecting their local community.

A wide array of studies has probed questions about how reporters cover climate change, such as the presence of debates that involve climate skeptics, the public interest in extreme events like extreme weather, and how climate change is “framed” in relation to other topics of public concern such as human health (e.g., Hase et al. 2021; Boykoff and Ravi Rajan 2007; Bolsen and Shapiro 2018). It could be valuable for future researchers to compile the wide array of explanations already offered for variations in news coverage and evaluate them with this larger data set (see for instance Saunders et al. 2018; Schäfer et al. 2014; Wetts 2020; Lochner et al. 2023).

Explanation may lie, as well, with the vagaries of the dataset itself. There are substantial changes as news sources—mainly state and local—being removed and added (mainly added) to the MediaCloud dataset. Although we normalized our indicator variables in an effort to remove these effects, bias in the patterns of coverage may explain some of the deviation between heartland and elite news sources.

Multivariate analysis of the entire data set can add to explanation but will require particularly careful attention to the construction of dependent variables given the large changes in coverage included in the MediaCloud data set. Explanation should also look at geographical focus of non-elite news outlets, political leanings, and other factors that plausibly could explain patterns. Better explanation will require looking more closely at the content of coverage, not merely incidence. Success will probably require a close collaboration with MediaCloud and not merely rely on public-facing open-access data queries as we have done for this article. More stability in the underlying data, allowing for replication, would also be useful.

Third is the need to disaggregate original stories from those that are simply reprinted. Wire services (like the Associated Press), in particular, generate a lot of reprinting through which the lines between elite and heartland might be blurred because news articles in multiple places have common origin. Reprinting should affect measures of incidence and may also be a fruitful way to explore where and how connections between elite and heartland news sources remain more durable. Such a reprinting effect would produce a tendency toward convergence between the elite and heartland datasets—suggesting that the divergence results we present here are robust.

The United States is early in a long process of developing and implementing a national policy strategy on climate change (American Academy of Arts and Sciences 2023). What it does at home will affect how it is seen abroad and its ability to get other nations to take a collective approach to climate change. The political viability of the American national policy depends, in large part, on what American think about climate change. For too long the efforts to understand what question have focused on elites and not the whole country.