The Two Theses of This Element

This Element examines data on media content and the public’s perceptions of that media coverage during the 2016 and 2020 presidential election campaigns. We present the following two fairly simple thesis statements. First, Donald Trump demonstrated how a presidential candidate in the social media era can manipulate the attention of the media system and the public to their benefit. From the moment he first gained local and national fame as a real estate developer, Trump had been adept at managing the news media and garnering attention for himself and his projects. As a political communicator, Trump is skilled at garnering attention for himself while also shifting the topics he regularly promotes. This reduces the time the media system spends covering the numerous negative stories about him during a campaign, and thus prevents much of the public from dwelling too long on any single dominant negative story.Footnote ¹ However, we don’t think Trump is the only person who can execute this strategy. Instead, it illustrates the opportunities available to other skilled politicians in the social media era. In this era, which stories voters pay attention to is more important than ever, and skilled candidates have more power than ever to influence the media system to shift attention onto their preferred stories.

On the other hand, if Donald Trump failed to get more attention than his opponents, despite having more experience than any other imaginable candidate getting media attention before capturing a major party nomination, that would be inconsistent with this thesis.

Yet our second thesis concerns the limits of attention-seeking. It is: The ability of candidates and other political actors to influence national attention is still limited by two factors: (1) the continued tendency of the media system to focus on the incumbent president’s performance and (2) the editorial judgments of journalists and news organizations about which stories are accurate and important. When you are the incumbent president, the media system will inevitably focus on your handling of major national problems. As we show in this Element, Trump’s ability to change the subject and distract from negative stories about him was substantially diminished in 2020. The COVID-19 pandemic was, in many ways, unique in American history. Still, we believe it illustrates a type of limit that other candidates will also face in the social media era. The campaign media system is not infinitely manipulable. The system still strongly tends to focus on current national conditions when covering a presidential campaign.

We also find that editorial judgment has consequences for which stories get attention, especially from high-quality media sources. The end of the 2020 campaign provided a clear illustration of this. As we document in Section 6, high-quality news organizations could have covered a story focused on potential corruption by the President’s son, Hunter Biden, in the last weeks of the campaign, but it seems that they made an editorial decision not to. This is an example of how, especially in the short term and in the crucial final weeks of the campaign, journalists still have the power to use their own judgment to decide which stories are worthy of attention.

If, instead of this, Trump were reliably able to change the subject away from negative stories about him and his record in both these campaigns, that would contradict this thesis. And if, when high-quality news outlets don’t cover a story because they doubt its accuracy or importance, that story still gets heavy attention from the public, that would contradict our thesis. This would be especially true at crucial times in the campaign, when prestigious journalists and editors are most careful about the consequences of their decisions.

Finally, the reader should be cautious about our conclusions because we arrived at these theses inductively, not deductively. In other words, we admit we had no set theses or hypotheses before beginning this project. These two theses were not the result of preregistered research questions. We came to them after a careful analysis of the data. As a result, some readers might fairly call our conclusions “exploratory,” and we would not object to that. Our claim is simply that these accurately summarize the data we have. We hope that future researchers will test these ideas with data in future elections.Footnote ²

Prominent Existing Theories Explaining Presidential Campaign Media Coverage

It’s helpful to consider our thesis statements in light of previous major political science theories that have attempted to explain variation in media coverage of American presidential campaigns. We don’t think our work in this Element is a definitive test that proves any of these theories correct or incorrect in the way advocated by Karl Popper (Magee Reference Magee1985), because we don’t think these theories are mutually exclusive with our theses or with each other. But we believe it is helpful to consider how our theses and findings suggest ways these theories might be updated, or our overall understanding of the causes of campaign coverage might be amended in the new social media era.

Three Eras of the American Media System

Key Differences in Context between the 2016 and 2020 Election Campaigns

Our datasets come from the 2016 and 2020 campaigns, which were different in several ways. For convenience and accessibility, we call these differences in the “context.” This overlaps with, but is not equivalent to, what political scientists call “the fundamentals” or “structural variables.” Political scientists have varying definitions of even these more technical terms,Footnote ¹² so it is necessary to provide our own definition to clarify what we mean by context.Footnote ¹³ We define the context of a presidential campaign as all the things that are true before the campaign begins, as well as events that occur during the campaign season that are beyond the control of the campaigns and the media. The campaigns and the media respond to this context during the campaign, but they can’t change it.

In our view, the three most critical contextual differences between these two campaigns were (1) the backgrounds of the Democratic candidates, (2) which party held the presidency, and (3) national conditions in the United States, especially the presence of a pandemic in 2020. It is worth considering how all three might have altered the effectiveness of Trump’s media tactics.

The first major difference in context between these campaigns was the identity of the Democratic candidate. Since becoming a national political figure when her husband was elected president in 1992, Hillary Clinton has been the frequent subject of false conspiracy theories and negative media coverage.Footnote ¹⁴ Her favorability ratings were relatively high during her 2008 presidential campaign, but took a hit in 2015 and onward (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020; Jeffrey Reference Jeffrey2017). In contrast, Joe Biden, who had served a remarkable forty-four previous years in elected federal office, was generally more well-liked. Clinton’s unfavorable ratings exceeded her favorable ratings in 2016, whereas the opposite was true for Biden in 2020 (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020; Jacobson Reference Jacobson2020). Coverage of Clinton and Biden, as well as voters’ reactions to it, could also have been influenced by gender. Research shows that voters react differently to candidates of different genders and some studies find that voters are less willing to vote for women (Badas and Stauffer Reference Badas and Stauffer2019; Sapiro Reference Sapiro1981).Footnote ¹⁵

The second and third contextual differences are connected. The second is the change in incumbency. And the third is the pandemic that struck while Trump was the incumbent in 2020. In 2016, Trump was an outsider and thus not responsible for national conditions. This made it much easier for him to influence media coverage by changing the stories he focused on in his social media feed and other public statements.

Many campaign observers noted how the 2016 campaign media agenda went from one outrageous tabloid story to another – from Trump calling a former Miss USA fat, to a tape surfacing where he bragged about grabbing women’s genitals, to a last-minute story about a laptop just discovered by the FBI and believed to contain hidden emails from Hillary Clinton (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020). In contrast, in 2020, Trump faced the longstanding journalistic tendency to focus heavily on the incumbent president’s record and national conditions more generally.

The most important national news story in 2020 was the COVID-19 pandemic. In March 2020, the World Health Organization officially declared COVID-19 a global pandemic, and the Centers for Disease Control declared a public health emergency (Centers for Disease Control and Prevention 2023; World Health Organization 2020). As of June 2024, over 775 million cases of the novel coronavirus would be tracked worldwide, with more than 103 million in the United States.Footnote ¹⁶ As we show in this Element, Trump struggled to influence media coverage in this environment. The administration held only sporadic official press briefings, forgoing one avenue for influencing reporters.Footnote ¹⁷ Yet Trump continued to post frequently on social media and make other outlandish statements – tactics that usually dominated media attention. Yet polls showed that the public remained overwhelmingly focused on the pandemic throughout 2020.Footnote ¹⁸

We want to highlight the limitations that come with looking closely at just two elections (four campaigns, two by the same person). These are essentially two case studies, and thus, we must be cautious about our conclusions. Yet, we believe our extensive data on these two elections sheds light on the dynamics of media coverage and attention during the social media era.Footnote ¹⁹

The Data

The Plan for This Element

In the sections that follow, we examine the information that was transmitted on Twitter, across high- and low-quality news outlets, and finally, which messages were received and remembered by the public. Examining patterns of attention (Section 2), the tone of coverage (Section 3), and the particularly important words and stories within these information streams (Section 4), we can gain insight into how the contemporary political campaigning environment puts information into citizens’ hands.

We then further interrogate the processes involved by directly examining some of the most prominent news stories in the two campaigns. In the latter half of Section 4, we compare two issues that consistently dominated the two campaigns: Hillary Clinton’s emails in 2016 and Donald Trump’s management of the COVID-19 pandemic in 2020. In Section 5, we compare events that could have triggered heavy coverage very late in the campaign and consider why they unfolded differently. Finally, in Section 6, we briefly revisit our two theses and previous theories of campaign content in light of the evidence.

Measuring the Volume of Media Coverage

To make sense of what Americans were hearing about the candidates in these two campaigns, we conducted textual analyses of mentions of the major party nominees across high- and low-quality conventional media sources, the social media platform Twitter, and responses in nationally representative surveys where we asked people what they had recently “seen, read, or heard” about each candidate, recording their answers verbatim (see Section 1 for more details on these datasets). We separate out the news recalled about the candidates among survey respondents who identify as Democrats or Republicans, including those who said they leaned toward one of these major parties.

First, we examine how the number of stories, tweets, or mentions on a given day compares to other days within the same media content stream. Second, we examine each topic as a share of all the coverage mentioning that candidate in each media stream. In combination, this allows us to see how attention varied over the campaign in each media content stream and the differences between coverage of the major party nominees.

A Caution about Differences in the Data Generating Process across Information Streams

Underlying the different types of data we track are four distinct data-generating processes. Editors and reporters usually produce the headlines in high-quality sources. There are also constraints on the overall quantity of content. The amount of content they produce is constrained by reporters’ working schedules and news judgment, as well as their organizations’ ability to disseminate their content. For instance, audiovisual news programs are typically of fixed duration, and both newspapers and online news remain relatively consistent in size from one day to the next. Reporters are often assigned to cover a certain candidate and are expected to produce regular stories about them (Dunaway and Graber Reference Dunaway and Graber2022; Gans Reference Gans2004). Thus, the norms of news production and the relatively steady supply of non-campaign news typically result in a relatively stable total amount of candidate news throughout the campaign. Increases in attention to a particular topic in traditional news are therefore likely to come at the cost of decreased attention to something else (Zhu Reference Zhu1992).

The decisions of high-quality outlets about which campaign stories to cover were traditionally fairly consistent and well understood: The decisions of larger, more prestigious outlets, such as The New York Times and The Wall Street Journal, shaped the coverage decisions of smaller news organizations (Gans Reference Gans2004). However, these norms have decayed in recent decades (Dunaway and Graber Reference Dunaway and Graber2022). More partisan and lower-quality news sources, especially on cable, the internet and social media, have increasingly been willing to adopt a partisan bent. While the impact of this trend is most prominent for lower-quality sources, it has been true of some high-quality outlets as well. Partisan outlets can release selective information or even misinformation to sway opinions.

In general, it is unclear whether the same factors influence lower-quality news sources as influence more traditional newsrooms. Sites designed to peddle misinformation or serve purely partisan goals may develop and broadcast their content strategically, releasing stories when they are poised to have the greatest virality or the largest potential impact on traditional media or the public. We therefore expect considerable variability over time in the quantity and content of information about candidates coming from low-quality sources.

Our social media data is somewhat similar to this. The number of tweets that can be generated about a particular candidate is essentially unlimited (Jang and Park Reference Jang and Jin Park2017; Jang and Pasek Reference Jang and Pasek2015). Because ordinary individuals are deciding whether and how much to tweet about a particular topic of their own volition, more tweets about the candidates do not imply fewer tweets about anything else. There is enormous potential for the volume of information on Twitter to fluctuate when external events prompt users to post heavily about one candidate or another. And many Twitter accounts may be motivated to post so that they achieve virality or partisan influence.

Finally, the public’s proclivity to talk about things they’ve “seen, read, or heard” recently about the candidates in open-ended survey responses has different attributes than our media datasets. Survey questions about media usage often overlook personal conversations with friends and family, where people can inadvertently hear media messages secondhand (Katz Reference 86Katz1957). Our survey questions, which simply ask what people have “seen, read or heard,” can capture campaign information from all sources that people have consumed and remembered. We see this as a key advantage.

On the other hand, it has limitations. Decades of research in survey methodology suggest that respondents are motivated to reply to survey prompts, even when they know little about the topic (Bishop, Oldendick, and Tuchfarber Reference Bishop, Oldendick and Tuchfarber1984; Kalton and Schuman Reference Kalton and Schuman1982; Zaller and Feldman Reference Zaller and Feldman1992). As a result, despite significant differences in the amount of political news people consume (Krupnikov and Ryan Reference Krupnikov and Barry Ryan2022; Zaller Reference Zaller1992), the format of the question leads most people to mention at least something about both candidates most of the time, although the length of their responses varies.

In conclusion, differences in the data-generating processes across these different datasets should caution readers against interpreting them identically. Tweets differ from open-ended survey responses, which in turn differ from news reports in high- and low-quality outlets.

Attention to the Candidates in the Media Information Streams

News Headlines in High-Quality Sources

Across the 2016 campaign, high-quality news outlets paid more attention to Donald Trump than to Hillary Clinton (Figure 1a). On any given day, the 1,048 high-quality sources that published stories in 2016 yielded between 612 and 4,561 headlines mentioning Donald Trump and between 230 and 2,092 stories mentioning Hillary Clinton. On average, Trump was mentioned 85% more often in headlines than Clinton, with Trump appearing in an average of 1,721 stories per day compared to 929 for his opponent.Footnote ²⁶ Trump stories outnumbered Clinton stories on 111 days, with mentions of Clinton only surpassing Trump mentions on nine days.

Figure 1

Trends in News Mentions of Candidates in 2016 (seven-day rolling averages)

(a)

Headline mentions of candidate in high-quality news sources in 2016;

(b)

Headline mentions of candidate in low-quality sources in 2016.

Clinton received more high-quality news coverage than Trump on a few notable occasions. These were: when primary opponent Bernie Sanders first endorsed Clinton (July 12th), around the Democratic National Convention (July 27th, 29th, and 30th), when Clinton fainted from pneumonia during a September 11th commemoration event (September 12th and 13th), and on the last three days of October, as the FBI resumed an investigation into whether classified information had been improperly shared in her emails.Footnote ²⁷ The tendency to cover Trump more in 2016, unless there was a significant event thrusting Clinton into the news, may reflect journalists’ sense that Trump’s campaign was more unusual and newsworthy than Clinton’s.

In 2020, the difference between Trump and Biden was even larger. Biden’s coverage in high-quality news outlets never approached that of Trump (Figure 2a). Even though we tracked a slightly smaller number of high-quality sources in 2020 (919 total outlets), Trump received a larger number of headline mentions than he had four years earlier, averaging 2,207 mentions per day. On the Democratic side, Biden received less coverage than Clinton had, averaging 792 mentions per day. Overall, Trump was mentioned more than two and a half times as frequently as Biden. Coverage of Trump peaked on the first few days of October when he was diagnosed with COVID after the first debate, reaching more than 5,000 daily headlines. He never fell below 1,412 daily headline mentions. Biden’s mentions only surpassed Trump’s lowest day ten times and reached its absolute maximum on October 23rd, following the final debate, with 1,945 headlines. On a relative basis, Biden’s coverage peaked at 82% of Trump’s mentions on August 12th, when he named Kamala Harris as his Vice-Presidential Nominee, and did not otherwise surpass three-quarters of the volume of Trump’s mentions.

Figure 2

Trends in News Mentions of Candidates in 2020 (seven-day rolling averages)

(a)

Headline mentions of candidate in high-quality news sources in 2020;

(b)

Headline mentions of candidate in low-quality sources in 2020.

Figure 2(b) Long description

The panel shows the same, but in low-quality sources. There are two lines showing trends, one for each candidate (Biden and Trump). Trump is mentioned more in high-quality sources, but there is very little candidate difference in low-quality sources.

Some of the gap in news attention in 2020 could be the result of Trump being a sitting president and thus attracting news coverage for presidential actions outside of the campaign. It may also reflect differences between Biden’s and Trump’s campaigns amid the Coronavirus pandemic. Biden’s COVID-cautious strategy meant that he limited travel and held relatively few campaign rallies, which likely decreased coverage. But when you consider 2016 and 2020 together, Trump consistently gets more attention from high-quality news sources than his opponents. Given that both Clinton and Biden led Trump in the polls during their campaigns (and thus were both favorites to win), this pattern contradicts Zaller’s (Reference Zaller1999) “rule of anticipated importance.”

In campaigns, we find a general increase in news coverage of the candidates over time, punctuated with spikes in attention around salient political events, such as the nominating conventions, debates, and immediately before election day. We also see bumps that appear to correspond with other developments, such as the release of the Access Hollywood tape in 2016, where Trump boasted about sexually assaulting women, and when Hillary Clinton fainted from pneumonia during a September 11th memorial event (see Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020) and when Trump contracted COVID in 2020.

Partisan Sources

Prior work suggests that liberal and mainstream outlets overlap considerably in their content and audiences, yet conservative content and audiences are distinct (Benkler, Faris, and Roberts Reference Benkler, Faris and Roberts2018; Levendusky et al. Reference 87Levendusky, Josh Pasek, Hardy, Kenski, Ophir and Renninger2023; Stroud Reference Stroud2011).Footnote ²⁹ Thus, we compare conservative-leaning outlets with those that are either liberal or mainstream. In 2020, the story was simple: Figure 4 shows that Trump received substantially more coverage in all four types of news sources, regardless of ideology or quality level. In 2016, the story was more complicated, as shown in Figure 3. Trump consistently received more coverage in the high-quality sources only. The low-quality sources focused relatively more on Clinton. The mainstream and liberal low-quality sources covered the two candidates roughly equally. In contrast, the conservative low-quality sources actually covered Clinton more than Trump from the end of August through election day. Clinton was mentioned 1.3 times as often as Trump in low-quality conservative sources in 2016.

Figure 3

Trends in Candidate Headlines by Outlet Quality and Orientation in 2016 (7-day rolling averages)

(a)

High-Quality Conservative Headline Mentions of Candidates in 2016

(b)

High-Quality Mainstream and Liberal Headline Mentions of Candidates in 2016

(c)

Low-Quality Conservative Headline Mentions of Candidates in 2016

(d)

Low-Quality Mainstream and Liberal Headline Mentions of Candidates in 2016.

Figure 4

Trends in Candidate Headlines by Outlet Quality and Orientation in 2020 (7-day rolling averages)

(a)

High-Quality Conservative Headline Mentions of Candidates in 2020

(b)

High-Quality Mainstream and Liberal Headline Mentions of Candidates in 2020

(c)

Low-Quality Conservative Headline Mentions of Candidates in 2020

(d)

Low-Quality Mainstream and Liberal Headline Mentions of Candidates in 2020

We can also look at how much these time series vary together over time. In 2016, variation over time in mentions of Trump were similar regardless of outlet quality or orientation (correlations ranged from .82 to .98 across categories; see trends shown with solid lines in Figure 3). In contrast, this was not true for Clinton coverage. Our lowest correlation between any of these headline streams was between Clinton headlines in low- and high-quality sources, which were correlated at only .61 (correlations ranged from .61 to .83). We did not find this in 2020, where patterns of mentioning each of the candidates were similar regardless of outlet quality and orientation (correlations >.92).

Taken altogether, Trump dominated attention in both elections. Overall, those producing news content were often twice as likely to mention Trump in a headline as his opponents. With this data alone, we cannot diagnose the reasons why news sources paid far more attention to Trump; however, this story resonates with claims that Trump was particularly effective in using tactics honed throughout his long career to attract news coverage (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020).

The only place where coverage of the Democratic candidates rivaled that of Mr. Trump was in conservative, low-quality outlets in 2016. This coverage was surely not favorable, as illustrated by its increase when Clinton fainted on September 11, the Comey letter was released, and in 2020 following the New York Post’s Hunter Biden laptop story.

Attention to the Candidates on Twitter

The party nominees were frequently discussed on Twitter in both years (Figure 5), with substantially more Tweets about the candidates in 2020 than in 2016. In 2016, over the 120 days preceding the election, approximately 366 million tweets mentioned Trump, and 160 million mentioned Clinton.Footnote ³⁰ Mentions of the candidates generally increased over that campaign, ranging from 656,000 to 14.6 million daily mentions of Trump and 344,000 to 5.5 million daily mentions of Clinton.

Figure 5

Attention to Candidates on Twitter

(a)

Twitter Mentions of Candidates in 2016

(b)

Twitter Mentions of Candidates in 2020

In 2020, we observed 1.5 billion tweets mentioning Trump (ranging from 1 million to 32 million per day) and 569 million mentioning Biden (ranging from 887,950 to 20 million daily). A typical day in 2020 yielded 5.4 times as many tweets about Trump as had been the case in 2016, and 4.3 times as many tweets about Biden as there were tweets about Clinton in 2016. This is surprising considering that the percentage of Americans using Twitter only grew from 20% to 23% between 2016 and 2020 (Auxier and Anderson Reference Auxier and Anderson2021).

Figure 5 shows that, as in other media, the campaigns became more prominent on social media around focusing events (including conventions and debates) and as election day drew nearer. Also, in other media, Donald Trump was far more likely than his opponents to receive attention in both 2016 and 2020. Overall, on Twitter, Trump received 69.6% of the attention allocated to the two major-party candidates in 2016 and 71.9% of the attention allocated to the major-party candidates in 2020. In 2016, Clinton was never mentioned more than Trump. Trump’s proportion of campaign tweets vacillated between 53% and 89%, reaching its highest point during the Republican convention. Clinton surpassed 40% of the candidates’ collective Twitter attention only during the Democratic convention and when she fainted on September 11th. In 2020, the gap between Biden and Trump was even larger. Biden’s attention on Twitter only very briefly surpassed Trump’s on July 28th, but the lines do not cross when we use weekly averages. Trump dominated social media attention in both years.

How Much Did People Remember Hearing and Seeing About the Candidates?

Attention to candidates on conventional or Twitter does not guarantee the public will notice (Pasek, McClain, et al. Reference Pasek, McClain, Newport and Marken2020). Social media’s relevance stems not from serving as a window into public opinion, but from its role as a source of information for journalists and some members of the public, and a reflection of the salience of that information. To see which stories broke through to the public, we asked people during the campaign what information they encountered recently and could recall about each candidate.

Figure 6 illustrates that most survey respondents could recall some information about the presidential candidates. In 2016, between 57% and 87% of respondents in our surveys recalled something about Trump in any given weekly period (averaging 73%), and between 55% and 87% recalled something about Clinton (averaging 70%). The pattern was similar but slightly more variable in 2020, with an average of 73% providing an answer to the Trump question (range: 52% to 90%) and 67% responding to the question about Biden (range: 43% to 89%).Footnote ³¹

Figure 6

Attention to Candidates in Survey Responses

(a)

Proportion of Survey Respondents Who Saw, Read, or Heard Something in 2016 (weighted)

(b)

Proportion of Survey Respondents Who Saw, Read, or Heard Something in 2020 (weighted)

People sometimes responded to the question, saying they had heard something about the candidates, but couldn’t remember exactly what it was. We included these responses. But we excluded people who responded only to say they “had not heard anything” or the equivalent. Some respondents provided a general evaluation of the candidate they were asked about rather than a direct answer to our prompt, which asked for recent news (e.g., “I don’t like him”). Given the somewhat subjective nature of interpreting these answers and the fact that even these answers might reflect something individuals had recently heard, we included responses like this. We excluded only answers that asserted a lack of knowledge or memory (e.g. “I don’t recall”).

People reported hearing more about the candidates over the length of the campaign, as the campaign became more salient in the media and paid campaign messages increased. While recalled information about Clinton in 2016 vacillated across events, it increased from just over 60% at the beginning of the survey period to consistently more than 70% throughout October, as shown in Figure 6a. In contrast, attention to Trump in 2016 was steadier. The source of this difference is not entirely clear. It may have something to do with his tendency throughout 2016 to manufacture controversy whenever attention to him dipped (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020), a strategy that was especially successful that year. Figure 6b shows that, in 2020, respondents’ comments increased following the conventions from a relatively steady 55% for Biden and 65% for Trump to more than 75% for each candidate in the last month of the campaign. In both years, people remembered more news about the candidates following their respective conventions, around debates, and following other notable events, such as when Clinton fainted from pneumonia in 2016 and when Trump announced that he had COVID in 2020. This reflects the patterns in media coverage documented above.

We saw few party differences in how much people recalled about the candidates (see Figure 7). Members of both parties recalled more about the candidates around major campaign events. And both parties remembered more news about Trump than about the Democratic nominee in both 2016 and 2020. Both parties paid more attention to Trump in both years, even though this gap was smaller than the coverage gap we found in the media content data. There were party differences in the reception of news about the Democratic nominees. Democrats were slightly less likely than Republicans to recall things about Clinton throughout the entire 2016 campaign. Yet in 2020, Republicans often had less to say about Biden than did Democrats, including during the events in Kenosha, Wisconsin, and around the time of the first debate.

Figure 7

Attention to Candidates in Survey Responses by Party

Figure 7 Long description

Panel (a) shows the prevalence of mentions of the two candidates in open-ended survey responses over the course of the 2016 campaign, with the numbers separated out by Democratic respondents and Republican respondents. Panel (b) shows the same, but over the course of the 2020 campaign. See the main text for an analysis of the results.

Associations between the Volume of Coverage in the Information Streams

Finally, we examine the correlations in the quantity of candidate news over time between different types of media content and the perceptions of survey respondents. The left side of Figure 8 shows the correlations across trends in the various information streams for both candidates in each election, using weekly averages. Most correlations are relatively high.Footnote ³² This partly reflects the fact that attention to candidates across all streams tends to increase as the election draws nearer.

Figure 8

Correlations Between Streams in Candidate Mentions

Figure 8 Long description

Panel (a) is a correlation matrix, which shows all the possible correlations over the course of the campaign between the different information streams (i.e., open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts) in both 2016 and 2020. Panel (b) shows the same information, except that the high and low-quality headlines are also separated into conservative sources versus mainstream and liberal sources. And the survey responses are separated into those from Republicans and those from Democrats. See the main text for an analysis of the results.

But a few other patterns are apparent. In most cases, the correlations across streams for different candidates were of similar strength. The media’s attention and the public’s attention move together. For instance, the correspondence between the surveys and the headlines in high-quality outlets in 2016 was .69 for Trump and .63 for Clinton. Survey content aligns more closely with low-quality headline mentions for Trump than for Clinton (.63 vs. .38), but these differences were generally modest. In contrast, mentions of both candidates on Twitter and in low-quality headlines were more distinct, implying that Twitter and low-quality news sources were likely not drawing large amounts of content from each other.

All these information streams were more correlated over time in 2020. The differences in headline volume between low- and high-quality sources were negligible, and trends in Twitter volume were far more likely to echo corresponding trends in headlines. In a more worrisome development, survey recall of the candidate news and Twitter candidate mentions in 2020 more closely mirrored candidate mentions in low-quality headlines than in high-quality headlines, indicating that low-quality sources may have become more consequential over time.

Breaking down these relationships by source orientation and survey respondents’ partisanship further illustrates the increasingly central role of low-quality news sources (Right side of Figure 8). Comparing across the diagonal reveals large increases between 2016 and 2020 in the correlation between low-quality conservative headlines and both Twitter and survey responses. Tweets about both candidates in 2020 were closely linked to the quantity of low-quality conservative media coverage. Even more significantly, survey respondents from both parties mentioned the candidates in 2020 at rates that tracked low-quality conservative headlines more closely than other media sources, even though they had been the least related to survey responses in 2016.

Making Sense of the Patterns in Attention

This section focused on patterns in the total quantity of candidate attention across these two campaigns: how much different types of media covered these candidates and how much the public noticed. In general, the media and the public’s attention to candidates increases over the course of the campaign, and is punctuated by notable spikes around key campaign events.

Another pattern is the changing role of low-quality conservative news outlets. In 2016, low-quality conservative news mentions of Hillary Clinton were an outlier, with the Democratic nominee receiving much more attention from these sources than elsewhere on several occasions. By 2020, attention patterns in low-quality conservative news were much more closely correlated with other media sources and with the information the public consumed.

However, the second and most notable pattern is a disproportionate focus on Donald Trump across various information streams and years. Trump received more attention than his opponent in both campaigns, as evidenced by headlines from high- and low-quality news sites, Twitter mentions, and open-ended survey responses. This imbalance in attention was even more pronounced in 2020 than it had been in 2016.

The theory that campaigns tend to be retrospective evaluations of the incumbent party might explain why the gap between Trump and the Democratic nominee was wider in 2020, but it doesn’t explain Trump’s consistent dominance of attention. Also, since Trump was behind in the polls in both election years, and his win was a surprise to many in 2016, we don’t think the rule of anticipated importance can explain this pattern either.

One can interpret this pattern as suggesting that coverage of Trump is more “resonant” with journalists (at high- and low-quality outlets) and the public because he is a longtime media celebrity, consistent with Entman’s (Reference Entman2004) ideas. Alternatively, you could interpret the pattern by suggesting that the media system treats Trump’s outlandish behavior compared to other politicians as warranting constant “fire alarm” mode coverage (Boydstun Reference Boydstun2013). We can’t rule this out.

However, we believe that Trump’s domination of the media system’s attention is such a significant pattern, so unusual in American politics, and has persisted for so long that an additional explanation is also needed. Campaign tactics matter. Trump appears to have used his media skills (conscious and instinctual, but all developed over his pre-political career) to exploit the opportunities of the new social media environment. In this environment, whether the candidate and their campaign are heavy social media users or not, the public’s attention is highly variable and uncertain. Candidates whose behavior stands out and who are good at generating compelling stories will get much more attention than others.

Of course, the volume of media attention to the candidates, while important, is only part of the story. One would expect positive or neutral attention to benefit its recipient, whereas negative attention may hurt public impressions of a candidate. In subsequent sections, we examine the tone of the information being transmitted (Section 3) and the specific stories garnering attention across these various information streams (Sections 4–5).

One Campaign Awash in Negativity, and One Less So

Media coverage of, and public opinion toward, the candidates were unusually negative during the 2016 campaign (see Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020). Views of opposition party members and political candidates have become increasingly negative for decades, a phenomenon that political scientists call “affective polarization” (e.g., Iyengar et al. Reference Iyengar, Lelkes, Levendusky, Malhotra and Westwood2019; Iyengar, Sood, and Lelkes Reference Iyengar, Sood and Lelkes2012) and “negative partisanship” (e.g., Abramowitz and Webster Reference Abramowitz and Webster2016; Lelkes Reference Lelkes2021). But even accounting for this, attitudes toward the major party nominees were anomalously negative in 2016. Trump and Clinton emerged from the 2016 campaign with an absolute majority of survey respondents expressing unfavorable views of each of them (Saad Reference Saad2016).

Was the public’s disapproval of the 2016 candidates an aberration or a new normal?

The answer from our 2020 data is ambiguous. Both candidates were still unpopular in 2020 by historical standards, but not as unpopular as Clinton and Trump were in 2016. Views of Trump remained net negative, although slightly better than in 2016. Biden was substantially more popular in 2020 than Clinton was in 2016, with about equal numbers of Americans viewing him favorably and unfavorably.

The Tone of Coverage in the Media Information Streams

This section tries to understand the tone of media coverage in these elections. We calculate positive-to-negative “sentiment” scores for each of our information streams. (We will use “sentiment” and “tone” interchangeably.) Like Section 2, the data comprise news headlines from high- and low-quality media sources that mention the candidates, tweets mentioning the candidates, and responses to open-ended survey questions asking what people had “seen, read, or heard” about each candidate in the past few days. Sentiment was calculated using the Lexicoder scores for the words in our data (Young and Soroka Reference Young and Soroka2012a, Reference Young and Soroka2012b).Footnote ³³ We compared the proportions of positive versus negative words in the units of each information stream, which were then averaged across a 7-day rolling period, with the contribution of each day weighted by the amount of data available. Scores below 0 indicate that negative words were used more frequently than positive words, whereas scores above 0 indicate that positive words appeared more often.

Figure 9 shows the tone over time of tweets about the candidates, headlines about the candidates in high- and low-quality news sources, and the open-ended survey responses. Information about both candidates was almost always negative in all information streams in 2016, which fits with previous research on public opinion and media content in 2016. For both Democratic candidates, coverage was more negative in low-quality sources, especially for Clinton.

Figure 9

Tone of 4 Information Streams Across the 2 Campaigns

Figure 9 Long description

There are four panels. Panel (a) shows the positive-to-negative tone of content focused on Clinton in four information streams over the course of the 2016 campaign. Each panel includes four lines, corresponding to surveys, Twitter posts, High-Quality News Headlines and Low-quality news headlines. The four information streams are open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts. Panel (b) shows the same thing, but for the tone of content focused on Trump in the 2016 campaign. Panel (c) shows the same thing, but for the tone of content focused on Biden in the 2020 campaign. Panel (d) shows the same thing, but for the tone of content focused on Trump in the 2020 campaign. See the main text for an analysis of the results.

For Trump, coverage was more negative in high-quality sources in both years. Trump’s coverage in high-quality sources was especially negative at the end of July 2016 after the Democratic Convention, when he asked Russia to hack into Clinton’s emails, and later in 2016 when the Access Hollywood tape became public, where he bragged about sexually assaulting women. But even in these dramatic cases, the tone of what people remembered about Trump in the survey responses recovered relatively quickly from these shocks. It returned to the baseline level (which was still usually net negative). The effect of these negative events dissipated rapidly.Footnote ³⁴ The pattern of reversion is present in all four information streams for all the candidates, but it is most dramatic for Trump in 2016. By the week before the election, coverage of Trump in high-quality sources had recovered from even the large negative shock of the Access Hollywood tape.

In 2020, Biden’s coverage in high-quality sources and on Twitter was more positive than in low-quality sources, a gap similar to what Clinton experienced in 2016, but not as large. In general, there were no campaign events that caused big declines in the tone of Biden news coverage that broke through to the public. Biden experienced a positive bump in coverage in high-quality outlets and on Twitter following the 2020 Democratic National Convention; still, it returned to prior levels by the end of the Republican National Convention.Footnote ³⁵ Coverage of Biden in low-quality sources was often negative, but unlike for Clinton, this was not reflected in other media streams or in 2020 survey responses. And significantly, as we will return to in later sections, there was a minimal drop in the tone of high-quality headlines, or in what people heard about Biden, after the “Hunter Biden laptop” story broke in the New York Post on October 14.

As in 2016, coverage of Trump in 2020 was more negative in high-quality sources than in other media streams, in contrast to the pattern for both Democrats. The biggest temporary drop in the tone of Trump’s coverage was when he visited Kenosha, Wisconsin, after the Republican Convention. Yet at no point in 2020 was there a drop in tone across all streams of Trump coverage comparable to what happened after the Access Hollywood tape was released in 2016. The tone of Trump coverage in both high- and low-quality sources was much more consistent in 2020 than it had been in 2016. But unfortunately for Trump, it was consistently negative.

The Tone of the News that People Heard and Remembered

While the information we tracked in all information streams was mostly negative in these two campaigns, the survey responses were the least negative. In 2016, however, the number of negative words mentioned by survey respondents outpaced the number of positive words on 108 of 114 days for Clinton and on 109 of 114 days for Trump. In 2020, the number of words mentioned when recalling news about Trump was negative for 67 of 102 days. Biden was the only candidate for whom people recalled mostly positive news – the words mentioned in recalled news were net positive for 87 of 102 days in 2020.

The tone of what people recalled about the candidates improved at a few key points during the campaigns. For example, news about Clinton was the most positive across all our information streams following the Democratic Convention and around the time of the first debate. But the survey responses were the most positive, rising all the way to neutral. In 2016, Trump experienced a positive bump in the media information streams and in the survey responses following the Republican Convention and during the window in September when Clinton was criticized because of health concerns and for a statement in which she referred to some Trump supporters as a “basket of deplorables” (Reilly Reference Reilly2016). Similar to Clinton, the tone of survey responses about Biden in 2020 increased temporarily after the Democratic Convention before reverting to its previous level.

The tone of what people reported hearing about Trump decreased dramatically when news coverage of him got more negative, especially in high-quality outlets: when he openly asked Russia to hack Clinton’s emails in 2016, following the release of the Access Hollywood tape in 2016, and when he visited Kenosha in 2020. However, perceptions of Trump were resilient. The tone of survey responses about Trump reverted to previous levels relatively quickly after these events.

When the “Hunter Biden laptop” story became public on October 14, 2020, the tone of news people reported hearing about Biden declined only modestly. It then recovered to its previous levels before the second presidential debate on October 22. That story had a small and short-lived effect on the tone of news people remembered about Biden.

Associations between the Tone of the Information Streams

Figure 10 shows correlations for the tone of information streams over time (using the same weekly aggregation strategy as in Section 2). Tone across the four information streams was closely related for Trump in 2016, but less so for Clinton. The net tone of information about Trump was correlated between surveys, tweets, and both types of headlines at or above .70. In contrast, although tweets and low-quality headlines yielded similar stories for Clinton, trends in the surveys were only moderately related to the other streams (correlations ranging between .41 and .63). As in our analysis of the volume of coverage in Section 2, we again see the importance of coverage of Clinton in 2016 in low-quality outlets. The tone of what people remembered hearing about Clinton was most closely correlated with headlines about her in low-quality outlets (.63), followed by the correlation with the tone of tweets about her (.54), and was least strongly correlated with the tone of headlines about her in high-quality outlets (.41).

Figure 10

Weekly Correlations in Tone Across Streams

Figure 10 Long description

This is a correlation matrix, which shows all the possible correlations over the course of the campaign between the positive-to-negative tone of content focused on the four nominees (Trump twice) in different information streams (i.e., open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts) in both 2016 and 2020. See the main text for an analysis of the results.

In our team’s previous book (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020) and in Section 4 below, we show that one story consistently dominated the public’s memory and media messages about Clinton in 2016: her “email” scandal. In contrast, a series of negative stories about Trump in 2016 faded quickly from media content and the public’s memory. The central role of low-quality news sources and social media in shaping perceptions of Clinton may have been linked to the prominence of this story.

Interestingly, the correlations across streams were generally lower in 2020, but, as in 2016, sentiment was more strongly correlated across streams for Trump than for his Democratic opponent. For example, Biden’s convention bump in sentiment was evident in high-quality headlines, survey responses, and tweets, but not in low-quality headlines. Overall, the tone of what people recalled about Biden in surveys was moderately correlated with his coverage in high-quality headlines (.44) and not at all with low-quality headlines (−0.08). In sharp contrast to Clinton, for Biden in 2020, the tone of what people heard and remembered about him had little relation to coverage in low-quality outlets.

The sentiment of what people recalled about Trump in surveys in 2020 was correlated with his coverage in high-quality headlines at 0.62 and with low-quality headlines at 0.55. The tone of information in high-quality outlets and survey responses about Trump was overall very negative. The nadir of the tone of survey responses about Trump following his visit to Kenosha was reflected in the high-quality headlines, but not in tweets or low-quality headlines.

Overall, while the volume of candidate coverage became more correlated across media streams from 2016 to 2020, the tone of the information became less correlated. The most likely explanation for this tone pattern was the increased coverage of Democratic candidates and the reduced importance of low-quality news outlets in 2020. The differences between Clinton and Biden were stark. In particular, what people heard about Biden was much less correlated with low-quality headlines than was the case for Clinton (-.08 vs. .63). In 2016, the tone of coverage of Clinton in low-quality sources was well correlated with the tone that people reported hearing about her. In 2020, the tone of coverage of Biden by low-quality sources was largely irrelevant to the broader public.

Interpreting the Patterns in Tone of Coverage

Despite Trump dominating the media’s attention in both 2016 and 2020, that did not lead to favorable coverage. In this section, we showed that people consistently encountered negative coverage of Trump in both these campaigns. In 2016, this was matched by negative coverage of his opponent. However, in 2020, when coverage of Biden was generally positive, this put Trump at a disadvantage.

As in the last section, the rule of anticipated importance does a poor job of explaining our findings. Both candidates led Trump in the polls for most of the election year, and Biden’s lead was larger. Biden was a probable future president, yet he did not receive the same negative scrutiny as Clinton.

The tendency to focus on the incumbent administration’s record during the presidential campaign could explain Trump’s negative coverage in 2000. While yearly real personal income growth after taxes and transfers was positive due to the U.S. government’s aggressive pandemic fiscal and monetary response,Footnote ³⁶ the country was suffering through a global pandemic that led to slow GDP growth and massive social and psychological hardships. And as we will see in the next section, much of this negative coverage of Trump focused on his handling of the pandemic. But this idea does not explain why Trump’s coverage was also negative in 2016.

Market incentives tend to favor negative coverage and more coverage of preexisting scandals and storylines (Searles and Banda Reference Searles and Banda2019). From this perspective, it is easier to get attention if you don’t mind much that the attention is negative. We should expect many attention-grabbing candidate strategies to produce negative coverage. Creating lots of positive coverage is much harder. Journalists may have seen Trump’s unorthodox background and behavior in all his campaigns as a “fire alarm” (Boydstun Reference Boydstun2013), drawing attention but the negative variety. It is also plausible that the media treated the entire coronavirus pandemic as a fire alarm during 2020.

The difference in the coverage of Clinton and Biden could be explained by culturally resonant frames (Entman Reference Entman2004). Clinton and her husband had a long, contentious relationship with the elite political press corps (Ladd Reference Ladd2015), and she was unpopular by the beginning of the 2016 campaign. While Biden had been a senator and vice president for many decades, he had never been a national cultural lightning rod or a symbol of the Democratic Party as Bill and Hillary Clinton had been. Negative news about Clinton might have resonated with the preexisting views of reporters and the public in ways that negative news about Biden did not.

As you can see, the patterns in tone are consistent with many existing theories. The major exception is the rule of anticipated importance. In the next section, we delve beyond the overall tone of media coverage to examine specifically which stories garnered attention.

Making Sense of Campaign Messages

We begin by examining the words that people mention in response to our open-ended survey questions.Footnote ⁴⁰ While examining word counts is a relatively simple approach, its simplicity lends it face validity, as will be clear when we examine the most frequent words in both campaigns. This straightforward method is particularly well-suited to these two campaigns because the major issues in them were widely described with the same one or two words. As you will see, words like “email,” “coronavirus,” and “women” play large roles in media content in these campaigns, and each straightforwardly connects to a major theme in campaign news.Footnote ⁴¹ While this simple method might not fit well with the dominant stories in every presidential campaign, in these two campaigns, we are lucky that the major stories could be described in a few words.Footnote ⁴²

The Words People Heard

To give you a sense of the stories that dominated these two campaigns, Figure 11 shows the words that appeared most frequently in our open-ended survey responses in 2016 and 2020.Footnote ⁴³ These words are arranged based on whether they were mentioned more frequently in relation to the Democratic nominee or Trump, and by whether they appeared more often in 2016 or 2020. The size of each term indicates the total number of times it was mentioned. Terms toward the left side of the graph were mentioned almost exclusively in 2016. Terms on the right were mentioned almost exclusively in 2020. For example, mentions of coronavirus and masking were unique to the 2020 election. Terms near the bottom of the graph were mentioned disproportionately in response to questions about the Democratic nominees. In contrast, terms near the top appeared more frequently in response to questions about Trump.

Figure 11

Proportions Comparing Mentions of Words by Candidate and Year

Figure 11 Long description

This figure is a crosstabulation where the data points are words. The words appear in the crosstab according to their value on the x- and y-axes. The x-axis is the proportion of all mentions of that word in our open-ended surveys that were in the 2020 surveys rather than the 2016 surveys. The y-axis represents the proportion of all mentions of that word in our open-ended surveys from both years that were in questions about Trump, rather than questions about the Democratic nominee. The words also vary in size. The size of each word indicates the total number of times it was mentioned. This is a correlation matrix, which shows all the possible correlations over the course of the campaign between the positive-to negative tone of content focused on the four nominees (Trump twice) in different information streams (i.e., open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts) in both 2016 and 2020. See the main text for an analysis of the results.

A large group of terms tended to appear with similar frequency across both candidates and years (located near the center of the figure). These were mainly words like debates, conventions, television, speech, and election, which reflect regular aspects of a political campaign without providing much information about specific details. Terms that appeared in one of the four corners of the figure reflect concepts that were associated with only one of the candidates in a given year. Of these, two stand out far more than the rest: mentions of email for Clinton in 2016 and mentions of coronavirus for Trump in 2020.

The dominance of these two issues is astounding. In 2016, 24.7% of those saying something about Hillary Clinton mentioned email, compared to under 7% of respondents mentioning debates or women in relation to Trump, which were his most common topics.Footnote ⁴⁴ Similarly, in 2020, 23.9% of those who mentioned Trump also mentioned the coronavirus, far outpacing “good,” the second most common word mentioned about him (8.1%), and “vice president,” which was the most common thing mentioned about Biden (9.3%). Voters perceived a clear dominant story in each campaign. In 2016, that was Clinton and her emails. In 2020, it was Trump and the coronavirus.

Stories in Different Information Streams in 2016

Figure 12a through d show the relative frequencies of the most common words across information streams. All three of our media content datasets in 2016 focused more heavily on terms related to specific election events and less on email than the survey responses did. With the average word in each figure given a value of 1, we can see how common different terms were relative to one another, as well as how their prevalence varied across streams.

(a)

Frequency of Top 25 Words Across Media Streams and Surveys in 2016.

Figure 12(a) Long description

This is a bar graph that shows the relative frequencies of the top 25 terms in 2016 across the four information streams (i.e., open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts). The length of the bar for each word and stream represents the proportion of total mentions for that word among all the mentions of the top 200 words. See the main text for an analysis of the results.

(b)

Frequency of Top 26–50 Words Across Media Streams and Surveys in 2016.

Figure 12(b) Long description

This is a bar graph that shows the relative frequencies of the top 26-50th most common terms in 2016 across the four information streams (i.e., open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts). The length of the bar for each word and stream represents the proportion of total mentions for that word among all the mentions of the top 200 words. See the main text for an analysis of the results.

(c)

Frequency of Top 25 Words Across Media Streams and Surveys in 2020.

Figure 12(c) Long description

This is a bar graph that shows the relative frequencies of the top 25 terms in 2020 across the four information streams (i.e., open-ended surveys, low-quality headlines, high-quality headlines, and Twitter posts). The length of the bar for each word and stream represents the proportion of total mentions for that word among all the mentions of the top 200 words. See the main text for an analysis of the results.

(d)

Frequency of Top 26–50 Words Across Media Streams and Surveys in 2020.

Among the 50 words most frequently mentioned in 2016 (Figures 18a and b), only a handful were relatively more common in survey responses than in the other streams. Among these, none approached the prevalence of email. Mentions of email accounted for more than 8% of the total mentions of the top 200 terms in the surveys in 2016. In the other streams, mentions of email comprised less than a quarter of that proportion. Survey respondents also mentioned “debate,” “speech,” “vice president,” “convention,” “scandal,” “immigration,” “lying,” and “television” at higher rates than these terms appeared in other streams. However, the difference was smaller than for email.

Not all scandals were mentioned more in the survey responses than they were mentioned in the media data. The “email” term was special. Even other incidents related to the email scandal but using different words were less common in survey responses than in media content. Mentions of the “FBI,” “Wikileaks,” and the “Clinton Foundation” – all connected in different ways to concerns about Clinton’s emailsFootnote ⁴⁵ – were found more frequently in news stories and on Twitter than in survey responses. Compared to their frequency in media content, people better recalled generic concepts in our open-ended responses. The word “email” became a catchall for this series of Clinton-related controversies, and other terms like “scandal” and “lie” were also more prevalent in the survey responses than specific aspects of the story because they better captured people’s feelings or impressions of that scandal.

Terms related to controversies springing from specific campaign events were more common on Twitter and in low-quality news headlines than elsewhere. This pattern was apparent for terms connected with Clinton like the “FBI,” the Clinton “Foundation,” and “Wikileaks,” and terms connected to Trump, like his treatment of “Women,” which was the strongest Trump-related theme in 2016, was about equally apparent in Twitter and in surveys.Footnote ⁴⁶ Finally, consistent with previous criticism of elite media campaign coverage (Patterson Reference Patterson2016), high-quality headlines in 2016 tended to prioritize “game schema” coverage more than the other streams, highlighting terms such as “campaign,” “poll,” “win,” and “election.”

Stories in Different Information Streams in 2020

Although Coronavirus/COVID-19 was among the most prominent words mentioned in every stream in 2020, yet again survey attention to the largest story heavily outpaced that story’s attention in other sources. As with the email story in 2016, only a few other top 50 terms received disproportionate attention in surveys versus the other streams (country, economy, speech, job, and television; Figure 18), and none by the margin of the coronavirus.

It is interesting to consider why the most prominent issues in both campaigns received a much higher proportion of attention in survey responses than in media content. It could be because popular attention is inherently different from media attention. Perhaps the way people remember information about candidates is simply different than the way stories appear in different parts of the media system. In contrast to most types of national media, people may only be able to remember a limited number of stories about a candidate at any one time.Footnote ⁴⁷ It could also be a function of measurement.

In Section 1, we discussed some differences in the data-generating process across these information streams, which make the datasets not perfectly comparable. In this case, people may feel limited in how long they want to talk and how many different stories they want to mention in their open-ended survey responses. Tweets may have different natural constraints. We collected all tweets from a 10% sample of the entire Twitter corpus (known as the “decahose”) that mentioned the candidates during the campaign. Social media users may naturally mention a wider range of things in connection with candidates than news outlets with limited space do, where editors make some attempt to employ editorial judgment about what topics are important and/or what people want to read. In summary, although we can only offer informed speculation about why, there is a recurring pattern in which the most prominent stories account for a much higher proportion of words in survey responses than in the media content datasets.

Nevertheless, the pattern that ordinary people viewing the campaign link certain prominent terms to certain candidates more often than to other term-candidate pairs indicates that there is something special about these term-candidate connections. The pattern of dominant-negative terms being associated with the losing candidate each year suggests that these stories may have cost those candidates votes. Given the outsized importance of these two issues, we dedicate the next section to a more in-depth examination of them.

As in 2016, words that reflect coverage of the campaign process and possibly the campaign in a “game schema” frame were prominent in the 2000 news and Twitter datasets: “campaign,” “debate,” “vote,” “election,” and “polling.” The vice-presidential candidates, the Supreme Court, taxes, and the economy also received attention (see Figure 12c and d). Most of these stories were more closely associated with Trump than with Biden. In fact, the sole issue that received notable attention in Biden coverage was his nomination of Kamala Harris as Vice President. Every other substantive issue was either candidate-neutral or mainly appeared in Trump coverage.

A Closer Look at the 2016 E-mail Story Versus the 2020 Coronavirus Story

The campaign media information environments in 2016 and 2020 were each dominated by one topic over all others. In 2016, the story that gathered the most attention across all sources was Hillary Clinton’s email scandal, while in 2020, it was the coronavirus pandemic.Footnote ⁴⁸ In this section, we dig deeper into both phenomena.

Both major stories had multiple components. Mentions of Clinton’s emails consisted of news about a private email server that Clinton used while Secretary of State, unrelated emails from Clinton’s campaign that were hacked and leaked online, an FBI investigation into the aforementioned server, and various other threads. All these might have been mixed together in the public’s thinking, but they were actually quite distinct. Similarly, the coronavirus discussion included information about the disease, questions about mask use, the virus’s economic and social impacts, and the development of a vaccine.

To better assess how coverage of these issues trended over time, we take a more inclusive measurement approach than in the previous subsection, which focused on individual words. Here, we combine mentions of a series of related words that encompass aspects of this topic as it was covered throughout the campaign. This allows us to capture a broader sense of what media outlets and ordinary Americans were attending to.

The topics of email and coronavirus have important similarities. Each was the most frequently mentioned campaign topic by Americans (as measured in our open-ended responses), yet received different levels of attention in various types of media. This made us especially curious about the coverage of these topics in low-quality outlets, given that previous sections found that they played a disproportionate role in media coverage in 2016 compared with 2020.

Table 1 presents the average percentages of all the information in each communication stream that was related to the email (in 2016) and coronavirus (in 2020) topics. The coronavirus was featured in more conventional media campaign coverage in 2020 than Clinton’s email scandal was in 2016. This holds for all media information streams, although the gap was smallest in low-quality right-wing outlets and on Twitter. The biggest difference between the two stories lies in high-quality left- and center-leaning news outlets, which paid more than three times the level of attention to COVID-19 in 2020 (19.3%) compared to the email scandal in 2016 (5.5%).

The difference in the attention the two stories received was smaller in the survey responses. The gap in attention in the surveys was primarily driven by Democrats’ recollections (9.3% mentioning email, 21.5% mentioning coronavirus), possibly reflecting the disparity in attention from high-quality left- and center-leaning outlets. Among Republican survey respondents, the percentages mentioning the two stories are essentially indistinguishable (14.4% vs. 14.9%). Even among Independent respondents, the gap in attention to these stories is not large. Eleven percent of Independents mentioned hearing about the email story in 2016, and 15.8% mentioned the coronavirus in 2020.

Overall, each story was covered and remembered more by the side of the political spectrum that benefited from it. In 2016, right-leaning news outlets (especially low-quality ones) covered the email scandal more than other media outlets, and Republicans discussed the story more than other people did. Analogously, in 2020, left-leaning and centrist media outlets covered the coronavirus pandemic more than other outlets, and Democrats discussed the story more than others did.

Coronavirus coverage was generally higher in low-quality news sources than in high-quality ones, while low-quality conservative outlets primarily drove coverage of the email story. This is consistent with Section 3’s finding that low-quality conservative outlets led the overall negative coverage of Clinton. Opponents of Clinton drove the email story. Yet no such pattern is present for the coronavirus in 2020, in which both high- and low-quality liberal outlets gave it comparably heavy coverage.

The Leading Stories Over Time

Next, we look at changes over the course of the campaigns. Erikson and Wlezien (Reference Erikson and Wlezien2012, 44–46) make a very useful distinction between two different ways an event can affect the state of a presidential race: a “bounce” versus a “bump.” A bounce is a temporary change to the state of the race which entirely fades, while bump produces a change in the race that never wholly dissipates. In their words: “As a series of short-term bounces, the campaign-induced shocks to the electorate’s collective choice over the campaign have little net impact on the final outcome…. If campaign effects are bumps, conversely, they persist and affect the outcome” (45–46). Erikson and Wlezien’s book examined horse race polls. Our focus in this Element is different: media content and information reception. But these concepts are still valuable for us. It is illuminating to determine whether events affecting attention to these stories are bounces or bumps.Footnote ⁴⁹

Tables 2 and 3 list some major events related to these stories during the campaigns.Footnote ⁵⁰ Figures 13 and 14 show that the prevalence of the email and coronavirus stories varied over time, with these events highlighted. Real-world campaign events seemed to increase news attention to both stories, but these spikes look like temporary “bounces” rather than “bumps.” The overall level of attention and the height of the peaks are greater for the coronavirus. Both stories had their highest peak close to Election Day, but the peak for email was much closer to the end of the campaign because of James Comey’s announcement and letter (see Section 6 here, and Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020, Chapter 7). As a result, both topics were likely top of mind for voters as they headed to the polls, especially the email story.Footnote ⁵¹

Figure 13

Email Story Prevalence over Time across Different Online News Producers

Figure 13 Long description

This is a line graph that shows relative attention to the email story (on the y-axis) across time (on the x-axis) during the 2016 campaign. There are separate lines for high-quality mainstream and liberal sources, high-quality conservative sources, low-quality mainstream and liberal sources, and low-quality conservative sources. See the main text for an analysis of the results.

Figure 14

Coronavirus Story Prevalence over Time across Different Online News Producers

Figure 14 Long description

This is a line graph that shows relative attention to the coronavirus story (on the y-axis) across time (on the x-axis) during the 2020 campaign. There are separate lines for high-quality mainstream and liberal sources, high-quality conservative sources, low-quality mainstream and liberal sources, and low-quality conservative sources. See the main text for an analysis of the results.

How Did the Public Receive These Two Dominant Stories?

How much did absorbing these major news stories depend on people’s partisanship? As noted above and in Table 1, there were partisan differences. In 2016, the email story was prominent in low-quality, right-leaning outlets and was mentioned most frequently by Republicans. And in 2020, the pandemic was covered most in high-quality left-leaning and centrist outlets and mentioned most by Democrats.

Figures 15 and 16 illustrate the prevalence of these two stories in survey responses over time, alongside other information streams. All the information streams are generally more correlated with each other in 2016 (correlation coefficients ranging from 0.67 to 0.98) than they are in 2020 (ranging from 0.37 to 0.95). However, there are many similarities. The highest peaks of these major stories (the Comey letter in 2016 and Trump getting COVID in 2020) move all streams in similar magnitudes. For instance, at its peak after the Comey letter, the email topic accounted for roughly 33% of all Republican survey responses. This number is lower among Democrats but remains at an impressive level of approximately 29%. The difference between the parties after the WikiLeaks Podesta emails revelation earlier in the campaign is much larger – 25% vs 14%. In other words, events with mid-level prominence seem to drive divergence and weakened correlations. There are also differences in how the time series reverts to normal levels after events. While we see a reverse u-shape after events like the Wikileaks Podesta emails, some even seem to change attention levels more permanently. For instance, the increase in attention after the Wikileaks Sanders emails is not followed by a reversion back to low levels for low-credibility right-leaning news and Republican responses. They continue to focus on this story longer than other information streams do.

Figure 15

Email Story Prevalence Over Time Across Information Streams

Figure 15 Long description

This is a line graph that shows the relative attention to the email story (on the y-axis) across time (on the x-axis) during the 2016 campaign in eight separate information streams, each represented by a line on the graph. The streams are high-quality mainstream and liberal news sources, high-quality conservative news sources, low-quality mainstream and liberal news sources, low-quality conservative news sources, survey responses from Democrats, survey responses from independents, survey responses from Republicans, and Twitter posts. See the main text for an analysis of the results.

Figure 16

Coronavirus Story Prevalence Over Time Across Information Streams

Figure 16 Long description

This is a line graph that shows the relative attention to the coronavirus story (on the y-axis) over time (on the x axis) during the 2020 campaign across eight separate information streams, each represented by a line on the graph. The streams are high-quality mainstream and liberal news sources, high-quality conservative news sources, low-quality mainstream and liberal news sources, low-quality conservative news sources, survey responses from Democrats, survey responses from independents, survey responses from Republicans, and Twitter posts. See the main text for an analysis of the results.

In summary, a closer look at these two stories indicates that, while they were very different events in the real world, they had some similarities in how they dominated attention during the campaign. They both constitute a remarkably large percentage of news during the campaign. Attention to both spiked after major campaign events connected to them. All these spikes were temporary “bounces,” not permanent “bumps” in attention, although the time it took for the bounces to fade out differed across types of media. And for both, there was some evidence of partisan selective coverage and perception. Stories that hurt your party were covered less by sympathetic partisan media and were mentioned less by partisans in surveys.

However, there were two big differences. The coronavirus pandemic was covered more extensively overall, although the gap in attention was larger in the news headlines than in public perceptions (as measured by survey responses). Also, among the news sources, coverage of the email story was heaviest in the low-quality outlets, while coverage of the coronavirus pandemic was heaviest in the high-quality outlets.

Why Do Certain Stories Dominate Coverage of Certain Candidates? Some Possible Explanations

Donald Trump’s ability to capture media attention and shift the communication environment’s focus away from specific negative stories about him encountered a major limitation in 2020. In the two campaigns we examined, a few issues stuck to their associated candidates, while others had less staying power in the media system or with the public. Two issue-candidate pairs stand out: email and Clinton in 2016, and the coronavirus and Trump in 2020. One could argue that the coronavirus story was unique. It was so prominent in media coverage and in what people recalled hearing about Trump because he was the incumbent during a once-in-a-generation pandemic that affected almost all aspects of Americans’ lives. But given that the email story was also so prominent in Clinton coverage in 2016, despite it not being tied to a global catastrophe, it raises questions about what types of issues stick to candidates in this way. This phenomenon may be a recurring feature of the social media era.

How do these two major stories fit with the campaign media coverage theories we have utilized in this Element? Taken together, they did not provide good coverage to facilitate retrospective voting. The email and coronavirus stories were indeed both scandals of the nominee of the incumbent party. However, the email scandal had little to do with the incumbent administration’s overall record and essentially nothing to do with national conditions in 2016. The law of anticipated importance is a better fit here than with the evidence in previous sections, but still imperfect at best. Clinton was favored to win in 2016, yet Biden was ahead in the polls and favored to win throughout 2020, and he did not receive consistent campaign coverage of any similarly negative story.

You could see the coronavirus as an example of the “alarm/patrol hybrid model,” in which the beginning of the pandemic served as an “alarm,” and “aftershock” events all year led coverage to persist.Footnote ⁵² However, the prominence of the email scandal in Clinton coverage is more puzzling from this perspective. This story also received “alarm mode” levels of attention throughout the campaign, despite its lack of a triggering event of the seriousness of a pandemic outbreak that temporarily shut down much of the economy and social activity.

Other theories more plausibly fit this evidence. Hillary Clinton’s long history of negative press coverage over decades and her connection in the public mind to elite Democratic politics and liberal causes may have made the email scandal more “culturally resonant” with journalists and the public. It also could be that sexism led a story that depicted a woman as unlikable or untrustworthy to be more resonant with these groups. Coverage of both the email and coronavirus stories is consistent with market incentives, which lead to a preference for repeatedly covering scandals or other negative stories that have previously received heavy coverage, as well as scandals involving less popular politicians. All of these conditions seem to fit both the email and coronavirus stories.

Whatever the ultimate reason, in the social media era, an issue can be connected to a candidate early in the campaign cycle, and that connection then becomes self-reinforcing. Past coverage of the story and its prominence in the public’s thinking lead the media system to continue covering it.

Donald Trump avoided this dynamic in 2016. Despite many negative stories that could have received sustained coverage throughout the campaign, none did (Bode et al. Reference Bode, Budak, Ladd, Newport, Pasek, Singh, Soroka and Traugott2020). One Trump campaign story after another faded in attention, replaced by another (also negative) story. But in 2020, he could not escape the global pandemic. His role as the incumbent connected him to that story. He could not divert the media system’s attention to something else.

Public Responses to the October Surprises

How did such coverage percolate to the public? We investigate this by comparing survey responses and Twitter content with the news information streams. Figure 19 provides the time series of the degree to which the Comey topic was mentioned across different streams. The peaks in public attention and news attention are generally aligned, although the increase in news attention is more pronounced. Among the public, the highest emphasis is among Republican survey respondents, followed by Independents and Democrats. This is particularly clear at the peak, right after the first Comey letter was released. Public attention is generally lower than news attention, which is encouraging, given that the event had minimal to no policy implications. Yet this might be a measurement difference. If this were the fourth or fifth most important issue in the public’s mind, it might be hardly mentioned in survey responses because people do not want to provide a very long answer. Yet, it could be seen in a long list of headlines if it were the fourth or fifth most-covered story on the national news agenda.

Figure 19

The Degree to Which Different Information Streams Focus on the Comey Story over Time

Figure 19 Long description

This is a line graph that shows the relative attention to the “Comey letters” topic (on the y-axis) across time (on the x-axis) during the 2016 campaign in eight separate information streams, each represented by a line on the graph. The streams are high-quality mainstream and liberal news sources, high quality conservative news sources, low-quality mainstream and liberal news sources, low-quality conservative news sources, survey responses from Democrats, survey responses from independents, survey responses from Republicans, and Twitter posts. See the main text for an analysis of the results.

Figure 20 shows that the Hunter Biden story in 2020 had some similarities with the Comey story. The attention from survey respondents mimics the 2016 patterns in important ways. Much like 2016, Republican respondents were more likely to recall the Hunter Biden topic and mention it in their responses, followed by Independents, and then by Democrats, who recalled it the least.

Figure 20

The Degree to Which Different Information Streams Focus on the Hunter Biden Topic over Time

Figure 20 Long description

This is a line graph that shows the relative attention to the “Hunter Biden” topic (on the y-axis) over time (on the x-axis) during the 2020 campaign across eight separate information streams, each represented by a line on the graph. The streams are high-quality mainstream and liberal news sources, high-quality conservative news sources, low-quality mainstream and liberal news sources, low-quality conservative news sources, survey responses from Democrats, survey responses from independents, survey responses from Republicans, and Twitter posts. See the main text for an analysis of the results.

Unlike the reaction to the Comey letter, the differences in attention between the public and the media are stark here. Low-quality right-leaning domains are intensely focused on the Hunter Biden story. This focus is somewhat matched by attention from right-leaning high-quality outlets. But unlike in 2016, left and center-leaning outlets cover this last-minute story much less than their right-leaning counterparts. And public attention never comes close. In fact, public attention among Republicans appears to be more aligned with that of left-leaning outlets.

Our analyses thus far focus on the alignment between the public and the media’s timing of attention. Next, we turn to the language. We aggregate all headlines and survey responses that are related to these stories and compute a KL Divergence measure across streams. KL Divergence ranges from 0 to 1, and is commonly used to measure similarities and differences in language. The lower the value of this measure, the more similar the two distributions representing text are. Figure 21 summarizes our findings.

Figure 21

KL divergence between Comey (in 2016) and Hunter (in 2020) Related Headlines and Survey Responses

Figure 21 Long description

This figure has two panels. Each of these panels is a crosstabulation that shows the KL Divergence between different information streams. The cells are shaded to indicate the level of KL divergence for each pair of information streams. KL Divergence varies between 0 and 1 and is commonly used to measure similarities and differences in language. The lower the value of this measure, the more similar the two distributions representing text are. Panel (a) shows the KL divergences for all possible combinations of six information streams in coverage of the “Comey letters” topic in the 2016 campaign. Panel (b) shows the KL divergences for all possible combinations of six information streams in coverage of the “Hunter Biden” topic in the 2020 campaign. The information streams used in both Panels (a) and (b) are: high-quality mainstream and liberal news sources, high quality conservative news sources, low-quality conservative news sources, survey responses from Republicans, survey responses from Democrats, and survey responses from independents. See the main text for an analysis of the results.

Our primary concern is the differences between the two years.Footnote ⁵⁴ First, coverage of Comey by high-credibility left- and right-leaning outlets was more similar to each other. In 2020, high-quality right-leaning outlets were more similar to low-credibility right-leaning outlets than to their left-leaning high-credibility counterparts. A similar pattern appears for high-quality left-leaning outlets. While they are closer to high-quality right-leaning outlets in 2016, their alignment in 2020 is closer to low-credibility outlets with similar ideological leanings.

A similar pattern of alignment appears across survey groups. Independents look more similar to Republicans in both years compared to Democrats. Furthermore, the divergence between Democrats and Republicans in 2020 is greater than in 2016 (as indicated by the lighter color in the corresponding cells).

Next, we look at the alignment between the public and the media. In 2016, Democrats are most aligned with all the media streams, except for low-credibility right-leaning outlets (signified by the darker colors). This is not the case in 2020, when they are more closely aligned with high-credibility left-leaning outlets, but not the others. The divergence between Republicans and Democrats also grew by 2020.

Editorial Decisions by News Organizations Are Still Important

These two “October Surprises” had commonalities. Both could have served as “triggering events” that led the media system into a heightened “alarm mode” (Boydstun Reference Boydstun2013). Because both in some ways reflected and amplified preexisting rumors about the candidate, they might have been easier for news organizations to cover (Searles and Banda Reference Searles and Banda2019) and “resonant” frames that spread through a media system and public that was already primed to accept them (Entman Reference Entman2004). For these reasons, both events could have led to negative coverage of the Democratic candidate’s character through election day. The differences in how they were treated in 2016 versus 2020 suggest limits in how much external events can grab the attention of the campaign media system.

These two case studies demonstrate that the editorial decisions of high-quality news outlets matter, even in the social media era. In 2016, when traditional media reacted strongly to James Comey’s letter about Clinton’s emails, focusing on it heavily, the public followed. In 2020, low-quality media behaved similarly in covering the Hunter Biden story. If low-quality media had strong agenda-setting power, we would expect a similar public reaction. We do not see that.

This is a good reminder of the enduring agenda-setting power of traditional high-quality media outlets (Chen et al. Reference Chen, Liu, Singh and Budak2024). These results align with recent scholarship arguing that social media is not the primary mechanism creating political harm and spreading misinformation (e.g., Guess et al. Reference Guess, Malhotra, Pan, Barberá, Allcott, Brown and Crespo-Tenorio2023).Footnote ⁵⁵ This is also a good reminder that we should hold traditional media sources accountable and demand policy-relevant, informative coverage, which remains crucial to the health of American democracy.

The Power and Limitations of Grabbing Attention

Consistent with prior studies (Carlson, Robinson, and Lewis Reference Carlson, Robinson and Lewis2021, chapter 2; Patterson Reference Patterson2016), we find that Trump dominated attention in both the 2016 and 2020 campaigns, as measured by media content and the public’s perceptions. However, this high volume of coverage was not positive. In both campaigns, media coverage of Trump that reached the public was mostly negative.

In 2016, this led to him winning the Electoral College and becoming president. A major reason was that none of the specific negative stories about Trump were covered for an extended period of time or at the very end of the campaign. Trump was able to change the subject frequently, from one story to the next, even though most of them reflected negatively on him. The week before the election, the dominant story in America was James Comey’s letters to Congress and his reopening of the FBI investigation of Hillary Clinton.

Trump’s 2016 opponent received less media attention, and the coverage she received was predominantly negative, similar to Trump’s. The negative information about Clinton in high-quality news sources and survey responses largely mirrored her coverage in low-quality conservative news sources. She was also dogged throughout the year by various iterations of the email scandal, which the Trump campaign could highlight in its campaign messages, and which the Comey letters reinvigorated in the final days before the election. Overall, this evidence was consistent with our first thesis: “Donald Trump demonstrated how a presidential candidate in the social media era can manipulate the attention of the media system and the public to their benefit.”

Yet in 2020, Trump’s ability to dominate campaign attention led him to less success for two main reasons. First, the media system still has a bias toward focusing on the incumbent’s record and national conditions generally – the sort of coverage that facilitates retrospective voting. This pattern, long highlighted in the academic literature on presidential campaigns, persists in the social media era. In 2020, the coronavirus pandemic was the dominant story of the campaign by a wide margin. Much of this constituted negative coverage of Trump. As the incumbent, it was much more difficult for Trump to use his attention-grabbing tactics to change the subject.

Second, high-quality media outlets seem to have exerted more editorial judgment in 2020 in a few crucial ways. Their coverage diverged more from what was covered in low-quality outlets than it did in 2016. And when a major event occurred several weeks before the election that could have triggered large quantities of negative coverage of the Democratic candidate, high-quality outlets decided it wasn’t accurate and/or newsworthy, giving it little attention. Likely as a result, coverage of that topic (Hunter Biden) did not dominate the public’s thinking in the last week of the campaign. Thus, the evidence is also consistent with the second thesis that we introduced in Section 1: “The ability of candidates and other political actors to influence national attention is still limited by two factors: 1) the continued tendency of the media system to focus on the incumbent president’s performance and 2) the editorial judgments of journalists and news organizations about which stories are accurate and important.”

Lessons for the Political Communication Literature in the Social Media Era

Our findings indicate that the “rule of anticipated importance” (Zaller Reference Zaller1999) does poorly in explaining campaign attention in the social media era. Trump consistently trailed in the polls in both campaigns. Yet he still received the majority of media attention.

However, other patterns can be seen as illustrations of existing theories of political media coverage. As noted above, there remains a tendency for the media system to focus on the incumbent president’s record (Hetherington Reference Hetherington1996; Sides, Tesler, and Vavreck Reference Sides, Tesler and Vavreck2019; Vavreck Reference Vavreck2009). One could argue that some stories proliferated in the media because they were more “culturally resonant” (Entman Reference Entman2004). Hillary Clinton’s email scandal could fit this pattern. Was it the dominant story of 2016 because Clinton had already been depicted as untrustworthy and worse by the press for decades, or because this story fit preexisting negative stereotypes about powerful women? We can’t prove that either way, but it is certainly possible. Broadly, the patterns we found are also consistent with a general media system preference for negative news, and for more coverage of scandals that are preexisting, aren’t currently competing for attention with other major stories, and involve less popular participants (Nyhan Reference Nyhan2015; Searles and Banda Reference Searles and Banda2019).

“Triggering events” can cause the entire media system to shift its focus to a given topic, temporarily dominating media and public attention (Boydstun Reference Boydstun2013). This remains an essential dynamic in the social media era. If anything, it is enhanced by social media algorithms and the attention economy. One could view Trump’s outrageous statements and actions as attempts to create triggering events that distract from other recent stories while holding the media’s attention on himself. You could also see the Comey letters to Congress at the end of the 2016 campaign as some of the most important media-triggering events in American history.

However, the New York Post’s 2020 story about the Hunter Biden laptop shows the limits of this phenomenon. Coverage in low-quality right-leaning news outlets spiked after the initial Post story. However, high-quality outlets (even to some degree right-leaning high-quality outlets) were wary of the story, giving it little attention. While we cannot be sure of their motivations, high-quality outlets may have made the journalistic judgment that the convoluted tale of how the laptop hard drive found its way to journalists cast doubt on the story’s accuracy. Alternatively, they questioned the importance of this story in relation to other news in the weeks leading up to the election. Or, it was some combination of both. As a result of this editorial caution, unlike the email story following the Comey letters, the Hunter Biden story had low salience for most people on election day.

Lessons for Politicians and Journalists

The primary purpose of this Element is to advance the academic literature on campaign communication, the media system, and the attention economy, rather than making recommendations. However, in these final few paragraphs, we conclude by offering a few lessons for politicians and journalists.

Politicians and those designing presidential campaign strategy should know that using unusual and sometimes outrageous behavior and rhetoric to dominate media attention can lead to political success, but does not guarantee it. It is more likely to be successful if neither candidate is an incumbent and there is no national crisis during the campaign. When an incumbent is running, the media system tends naturally to focus more on their record, especially if that record includes a major calamity or crisis. In this circumstance, no campaign tactic will divert attention from these issues.

On the other hand, if you are running in a race where neither candidate is the incumbent or closely tied to them (such as the sitting Vice President), and the nation is not facing a major national problem like a pandemic, economic downturn, national security crises, or overseas war with casualties, the situation is different. In the absence of these things, the media system will be more open to focusing on whatever stories the competing candidates can draw its attention to, as it was in 2016. In an election year like this, campaigns must be skilled at diverting attention from negative stories about their candidate and highlighting those that harm their opponent.

Journalists, editors, and management at high-quality news outlets should know that they have more power and autonomy than they might think, especially at the very end of presidential campaigns. The editorial decisions made by those at high-quality outlets still matter. And with this power should come responsibility. Throughout the campaign, but especially in its final weeks, those trying to influence the election will inevitably attempt to draw the public’s attention to stories that benefit their favored candidate or harm the opposing candidate. However, as we saw in 2020, high-quality news organizations can decide that they cannot confirm a story’s accuracy and/or doubt its importance, and therefore choose not to focus on it. If they do, social media and the public will follow their lead.

There is no guarantee that journalists and news organizations will always make the right decisions. There is never a guarantee that any human institution will. There have always been major debates among journalists about what the best moral standards are for making editorial judgments (e.g., Carlson, Robinson, and Lewis Reference Carlson, Robinson and Lewis2021, chapter 5). Yet at their best, high-quality news organizations don’t just produce whatever content generates the most profit, but also follow a set of professional norms that include providing the electorate with the verified information it needs to make good decisions (Schudson Reference Schudson1978; Stensaas Reference Stensaas, Hoyer and Pottker2005; Streckfuss Reference Streckfuss1990; Tuchman Reference Tuchman1972). High-quality news organizations together constitute a key institution in a democratic system (Cook Reference Cook1998; Ladd Reference Ladd2012; Schudson Reference Schudson2002). It is hard to have a healthy democracy without also having a healthy news media system, where influential decision-makers prioritize spreading accurate and useful political information.

Even in the social media era, journalism still has the power to draw more attention to accurate and essential stories, and less to others. If they choose to, journalists and news organizations can improve the quality of voters’ decision-making. We think they should.