The Dynamics of Issue Attention During the Democratic Nomination: Twitter, Candidates, and the News Media
Our first issue tracker followed the ebb and flow of social media discussion about 25 policy issues referenced in Tweets about the Democratic nomination. You can read more the methodology we use to assign Tweets to issue categories in our previous blog posts. Here we zoom in on several of these policy areas to assess the temporal dynamics of issue attention on social media among candidates, news media, and the broader public during the Democratic primaries (January-April 2020). Did candidates or journalists respond to what they saw on Twitter in a bottom-up fashion, in which a spike in Tweets about an issue was followed by increased media and candidate attention to that issue? Or did the interactions follow a more top-down model, in which candidates and the news cycle drove social media discussion of policy issues? What role did the media play? Did they help amplify particular signals from the Twittersphere, or did they help set the agenda by directing social media attention to an issue in the first place?
In recent years, scholars have taken advantage of Twitter to examine such agenda-setting dynamics. Members of Congress, for example, appear to respond to what they see on social media, increasing their attention to issues after their party supporters Tweet about them. There is also evidence that traditional media outlets pay attention to social media. Twitter is now a routine part of news production, particularly political reporting. In 2016, journalists used online sentiments, trending hashtags, and other social media metrics to “take the pulse” of public opinion throughout the presidential election.
To unpack such dynamics in the context of the Democratic nomination, we focus on four issues that polling data suggest are highly salient for Democrats heading into the 2020 presidential election and that, in some cases, differentiated the positions of the crowded field of contenders who vied for the nomination: civil rights and discrimination, economic inequality, healthcare, and immigration. Previously, we examined how much attention these issues garnered in Tweets that referenced any of the Democratic candidates or debates, and also considered how often the Democratic candidates Tweeted about these same issues. Now, we bring in the amount of attention these issues received in Tweets from national newspapers (New York Times, Washington Post, USA Today, Wall Street Journal) and in Tweets from cable news outlets (MSNBC, CNN, Fox).
Our question is simple: when do these groups (Twitter, candidates, newspapers, cable news) lead social media discussion about these four issues and when do they join the debate?
In the figure below, we plot the daily average percentage of Tweets devoted to each issue between January 1 and April 8, 2020. For example, we categorized about 14 percent of daily posts about the Democratic nomination or debates on Twitter between January and April as civil rights and discrimination. Civil rights and discrimination was also frequently mentioned in candidates’ Tweets: just under 1 out of 10 Tweets authored by one of the Democratic candidates, on average, referenced racial equality, gender equality and/or LGBTQ rights. In contrast, the average daily attention to this issue was much lower in Tweets emanating from traditional and cable news outlets, largely because these groups Tweeted less frequently about the Democratic nomination on any given day.
Issue Attention Across Groups, January 1-April 8, 2020
The figure above gives us some sense of the correspondence in issue prioritization across these groups but says nothing about temporal dynamics. Who leads whom? To get to the heart of this question, we employ a vector autoregression (VAR) approach to explore the extent that each group’s expressed issue priorities (Twitter, candidates, newspapers, cable news outlets) respond following a change in issue attention from the three other groups. VAR models treat all variables as endogenous by regressing each of the variables in the system on the past lags of those variables and on the past lags of all the remaining variables. We selected lag length in the VAR model by comparing several information criterion procedures; all four models use a 7-day lag. We then used Granger causality tests to examine whether increased issue attention from one group (Twitter, candidates, newspapers, or cable news outlets) predicted subsequent changes in issue attention among the other groups. We estimate separate models to assess potential differences in dynamics across issues.
First, consider civil rights and discrimination. In the figure below, we plot the share of all issue-related Tweets about the Democratic nomination that we had previously categorized as civil rights and discrimination. This issue received sustained attention from the general public and candidates on Twitter during the first four months of the year. We also see spurts of attention in Tweets from national newspapers and cable news outlets.
Granger causality tests show that increased issue attention from candidates and newspapers were more likely to follow changes in issue attention on Twitter. Conversely, we find no evidence that Twitter responded to changes in attention from either the candidates or in news coverage. In other words, Twitter led most other groups in directing attention to concerns about civil rights and discrimination (racial equality, gender equality, LGBTQ rights). The agenda-setting dynamics in this case are strictly bottom-up: Twitter led but did not follow.
Compare these dynamics to what we see for economic inequality. Here we find a much more interactive dynamic between Twitter and the candidates. When there was an increase in issue attention to economic inequality on Twitter, candidates followed suit by devoting more attention to the issue on social media. But we also observed the reverse: when candidates Tweeted about economic inequality, there was also a subsequent increase in attention to this issue on Twitter. Candidates also appeared to respond to changes in news coverage about economic inequality. When the national newspapers Tweeted about inequality, candidates’ Tweets soon followed.
Healthcare followed a similar dynamic between candidates and Twitter. Candidates responded to increased social media discussion about healthcare on Twitter, and vice versa. Whereas Tweets from the national newspapers helped drive attention to economic inequality, we find a larger role for cable news outlets on social media discussions about healthcare. Both candidates and Twitter increased their attention to healthcare following spikes in attention to the issue from cable news outlets.
Finally, we consider the agenda-setting dynamics regarding social media discussion of immigration and the Democratic nomination. Unlike the other issues, increased Twitter attention to immigration was unrelated to any shifts in attention among the other three groups. Nor do we find any evidence that Twitter responded to increases in immigration-related Tweets from the candidates or news sources. We observe no leaders and no followers. The ebbs and flows of Tweets about immigration from the general public, candidates, national newspapers, and cable news outlets are all independent of each other.
In sum, we show that issue attention dynamics do not look the same across these four issues. For civil rights and discrimination, Twitter predicted subsequent social media attention from candidates and the media. These results point to an important way that social media may be broadening the public sphere, drawing the attentions of candidates and journalists to issues and debates not already dominating the news cycle and campaign trail.
Social media discussion of economic inequality and healthcare, on the other hand, revealed both bottom-up and top-down dynamics, with Twitter and candidates responding to each other, and news media also having some ability to set the agenda on these two issues.
Finally, immigration stands out as the one issue considered here where no group leads any others. Why might that be? Some possibilities include the field of candidates and the idiosyncrasies of the Democratic debates. In 2019, the Democratic candidates spent, on average, about 10 percent of the debates talking about immigration policy. Yet, in the last four debates held in 2020, they spent, on average, about 2 percent of the debates on immigration. No doubt the field of candidates helps explain this shift. Candidates who prioritized immigration, including Julián Castro, Beto O’Rourke and Cory Booker, withdrew in 2019; those who remained staked their campaigns on different issues. The debate moderators likely played a role too. Indeed, not a single moderator brought up immigration policy in three of the final four debates. Whether these patterns of agenda-setting dynamics hold as we shift into the general election campaigns remains to be seen.
It is also worth noting that other issues might reveal even different dynamics. Take social media discussion about the economy and the Democratic nomination. Applying the same approach to Tweets about the economy from all four groups, we find a more media-dominated dynamic. We estimate that Twitter and candidates both responded to increased news coverage about the economy, but not the other way around. Nor did we observe any responsiveness from candidates to increased attention to the economy from the general public on Twitter. In this case, the ebbs and flows of social media attention to the economy over the course of the Democratic nomination appeared to be purely set by the rhythms of the news cycle. That said, we have seen how the coronavirus outbreak has upended the issue landscape, with the economy and healthcare now dominating social media discussions about the 2020 elections. And so, we will revisit these dynamics again and expand our analysis to a broader set of issues as we assess shifts in expressed issue priorities from now until November.
Student team members, Savannah Charles, Kaley Gilbert, and Emma Hazeltine, contributed to this report.