How did concerns about Hillary Clinton dominate voters’ concerns in 2016, despite the scandal- & gaffe-prone campaign of Donald Trump? Did media coverage and social sharing doom Clinton? And can we expect a similar pattern in 2020? Jonathan Ladd finds that citizens heard a succession of negative things about Trump, but remembered the one big scandal about Clinton: her emails. Hal Roberts finds that conservative media produced and shared harmful stories about Clinton, while Democrats mostly followed mainstream and less partisan news sources. They both see important implications for the 2020 campaign ahead.

Interviews: Hal Roberts, Harvard University, and Jonathan Ladd, Georgetown University

Studies: Words That Matter and Network Propaganda.


Matt Grossmann: This week on The Science of Politics, how news coverage and social media shape American voters. For The Niskanen Center, I’m Matt Grossmann.

To gear up for 2020, we’ll look back at the dynamics of 2016: how did the scandal prone campaign of Donald Trump escape with an investigation into Hillary Clinton’s emails leading the headlines and priming voters? Can we expect a similar pattern in 2020 with conservatives better able to consolidate and share negative messages about the democratic nominee.

Today, I talked to Jonathan Ladd [of] Georgetown University about his new book, Words that Matter: How the News and Social Media Shaped the 2016 Presidential Campaign. Written with Leticia Bode, Ceren Budak, Frank Newport, Josh Pasek, Lisa Singh, Stuart Soroka and Michael Traugott.

They find that citizens heard a succession of negative things about Donald Trump in 2016 but remembered one big scandal about Hillary Clinton, her emails. I also talked to Hal Roberts of Harvard University about his book Network Propaganda, Manipulation, Disinformation, and Radicalization in American Politics, written with Yochai Benkler and Robert Faris. They find that the conservative media ecosystem produced and shared harmful stories about Hillary Clinton. While Democrats mostly followed mainstream and less partisan news stories that stopped untrue anti-Trump stories from spreading. They both see important implications for the 2020 campaign ahead. Roberts says lots of media stories aren’t sharing data point to asymmetric polarization.

Hal Roberts: And we try to understand the media ecosystem around the national political presidential election in 2016 that elected Donald Trump. The book uses a lot of quantitative data that we get from our platform called media cloud, where we collect about a million stories a day from all around the world mostly about politics but about really any topic. For the election, we ended up capturing about two and a half million stories about the election by crawling the web using our platform. Our large findings were that the defining feature of the US national political media ecosystem is of asymmetric polarization and that’s a fancy word for saying that the media system is very polarized but there’s, instead of the traditional polarization that we expect to see between the right and the left, what we say as the right versus the rest.

One of the reason we see a strongly segregated, an insulated right wing media ecosystem on one side and then we see certainly a liberal ecosystem on the other side but the liberal ecosystem is closely intertwined with the big mainstream media outlets. The way that we measure this is not by subjective measures of our own, instead what we’re looking at is audience metrics and attention. So we’re looking at what sites people tend to tweet and that tells us whether a site is conservative or liberal. So sites that tend to be tweeted only by conservative Twitter users, we labeled as conservative and liberal Twitter users would be labeled as left. What we find in the book is that this is a problem that’s baked into the architecture of the American political ecosystem. So this is not a feature.

Main problem with the election to the degree is the election featured its post-truth moment in which it becomes difficult to tell what’s true or not true or who cares about that. The cause of this issue is not a sort of fancy, more technical causes that have been brought up such as Facebook or filter bubbles of people only listening to their own group or a Cambridge Analytica brainwashing people through looking at their Facebook data or even really Russia hacking into our systems. The fundamental problem is that are you at least since 1950s but for sure since the 1980s black conservative movement has been building a media ecosystem with express purpose of separating that ecosystem from large mainstream journalistic institutions that have some large set of processes and norms for truth telling, which are far from perfect but exist.

Matt Grossmann: Ladd says they also attract what people heard in 2016 comparing it to the news coverage and online sharing.

Jonathan Ladd: We wanted to in this book understand the information environment in 2016, so sometimes in political science people write campaign books that are just the story of the campaign and everything that happens in the campaign. And those books are often extremely well done and that’s not our goal here. Our goal is a bit more specific to get at what was in the news media and what of the information the news media did people really notice and really remember. So it’s that specific question but there’s a lot to answering that question because that information is in a lot of different places and measuring what people actually picked up is tricky. So we did that with a number of different datasets and I’ll just mention the major ones that we use in the book.

We wanted to collect information on what was on social media and to do that, we collected data throughout the campaign from July to election day on a random sample of what was on Twitter that mentioned Donald Trump or Hillary Clinton. We were taking out a random sample throughout the whole campaign from the beginning of July until election day. We also wanted to measure just what journalists were talking about. So we went back to Twitter and just measured what political journalists were saying on Twitter about Hillary Clinton and Donald Trump. We got a long list of politically active journalists and monitored what they were saying throughout the whole campaign as a measure of what journalists were thinking about. We didn’t have a measure of conventional media, which is a list of major newspapers across the United States. And we track the full text of all articles in those major newspapers in that same time period throughout the whole fall campaign and tracked what topics were in those stories about Donald Trump and Hillary Clinton in major newspapers.

And then finally we didn’t just track the media directly but we also had a survey measure that measured what from the media environment people remembered. So this is our partnership with Gallup and this is the last major data set I’ll talk about at the outset. We had those on social media, there was a random sample just out there. We had what journalists were saying on social media, we had what’s in conventional media in terms of newspaper text. And then we had this role in cross section survey to see what people actually absorbed. And this was our partnership with Gallup so we had a role in cross section which if you’re not fluent in survey research terminology that means you have a new sample every day. And here we had a random sample of 500 new adults every day from the beginning of July until election day.

And we’re entering 500 people every day and the advantage of doing that, which servers you should call roll and cross section is any, for example, three day window throughout the whole campaign you have 1500 people, a fairly large survey sample in any three day window. Gallup was asking different things on these surveys but we for our research project put two open ended questions on there for the entire campaign, 500 people every day. An open ended question that said quote, what specifically do you recall reading, hearing or seeing about Hillary Clinton last several days? And that was open ended and then we recorded the words people said because they were phone interviews. And then we had a separate question SAS, same thing about Donald Trump. What specifically do you recall reading, hearing or seeing about Donald Trump in the past several days?

We have exactly what they responded, exactly the words they respond to those questions. So we know what in our pollster through our role in cross sections throughout the campaign, people say they can recall recently hearing from the media environment about each candidate and we can look separately about what they’re hearing. Those are the major data sets because we want to see what information was out there and what information people actually absorbed and remember.

Matt Grossmann: People remembered only emails about Hillary Clinton but not due to the media alone.

Jonathan Ladd: One thing we thought is that it’s interesting that in the Gallup open-ended polls where we measure what people actually remember and can report back to a pollster, people consistently mentioned remembering the email story about Hillary Clinton and when there wasn’t any other big story in the news they went back to mentioning that over and over and it dominated what people could remember about Hillary Clinton. The fascinating thing though is that in our measures of news content we couldn’t find media content that’s skewed towards email, so there wasn’t a similarly large percentage of email mentions in the random sample of all Twitter users.

There wasn’t a similar skew towards the email story dominating among what journalists were writing on Twitter. Some people might think that that was the case, but not in our data. And there wasn’t a skew towards email in newspaper coverage but it wasn’t what people remembered. That’s fascinating to me and I can think of several possible explanations but in our book raises more questions than it answers. And we say that in the conclusion that this raises a lot of questions that hope we hope future work continue to answer. One possibility is that people remember the email story about Clinton because it was a story that had been around longer or just suspicions about Clinton rightly or wrongly had been around in their head longer, possibly for decades.

And they were more suspicious of her based on whatever predispositions they had about her. And that stuck around even though it wasn’t as omnipresent in the actual news environment as it was in their minds. So that’s one possibility. Something about the story made it easier to remember or make people’s minds go back to it apart from what messages they were getting. A second possibility, which is fairly different is that they were hearing about the story in media information streams we didn’t measure, so it’s possible that… So we didn’t, we didn’t measure people’s Facebook feeds. We don’t have access to that, we have access to Twitter because it’s public. It could be that the email story was dominating Facebook feeds that we don’t measure. We in this book have a lot of media content measures but we don’t have a measure of conservative talk radio or just Fox news content.

So it could be that that story was covered so much there and that explains the entirety of why it was remembered so so much. We think either way it actually what either explanation, it does illustrate that it’s easier for news organizations to repeat and people to remember one story and then get them to remember then multiple stories. That doesn’t mean it’s necessarily has a bigger effect on people’s votes but it does mean it’s easier to remember and it’s easier to remember than a different story about a candidate every week.

Matt Grossmann: Robert says Trump took advantage of the unique American infrastructure that long predated him.

Hal Roberts: When we look at data from 2012, we see a probably similar architecture. So there is a slight growth in degree of the installation of the right, but the Bay, if we go back to 2012 we see the same basic architectures. So our reading of what happened in 2016 is not that we had a healthy ecosystem and then Trump came along and broke it or populism came along and broke it but instead again, we really trace it back to the 1990s with the creation of Fox news popularization of a fresh slime ball for a variety of reasons that those were sort of the stars next to the Drudge report as well online. Those were the start of this sort of really strong segregation of right wing in the US and we see Trump not as sort of causing this change in how we operate news but rather just taking advantage of this architecture of media that was existing when he showed up in 2016 internationally.

The question is much, much harder. The first hard thing about studying and understanding other international cases is that the vast majority of countries don’t have a two party system. So when we talk about polarization it’s easier to study and it’s queer or what’s happening in the US because we just have two major parties. Ours in most other countries, they have multiple parties in their multiple polls and that means that it’s harder to study and also just means as academics a lot of our tools that we use to study the US don’t work with use different and more sophisticated tools to try to understand these multipolar systems.

We have looked at some countries and what we found is that different countries work differently. There was some very good work done by my colleague, even Zuckerberg and our partners at the CN Sport Science that used our tools and some of their tools to look at yellow sharp movement and France. And what they found is that the French media ecosystem is much more strongly centered and anchored still in the bid mainstream institutions of journalism and that has provided them with more protection from these kinds of populous movements. Every scene coming online, you can the US and other countries.

Matt Grossmann: Ladd says Clinton faced skepticism going in enabling Republicans to raise objections.

Jonathan Ladd: Both candidates have problems with their longterm image. Both candidates were more disliked than any other previous presidential candidates. And so you would think Hillary Clinton being around so long and being secretary of state, maybe it’s impossible for a candidate to being around so long to maintain a positive image when you’re in the fray of politics. There’s an older literature on presidential approval that it used to always say, almost always presidents get less popular over time. In more recent decades it’s not clear that’s always true that’s what political scientists used to think. But either way people had long… I think it’s fair to say people have longstanding skepticism of Hillary Clinton despite… And I think it’s fair to say that that is despite her temporarily high personal approval ratings when she was in a less political job of secretary of state or one way or the other way to put his people had long, long standing skepticism of the Clintons.

And if you wanted to find political figures where it was easiest to prime partisanship and get Republican voters to come home where they might be hesitant to support unorthodox candidate like Donald Trump. The end it seems like the Clinton’s as an opponent was the easiest to a canvas to do that, to activate negative partisanship, to activate longterm partisan attitudes. And as a result of the defections of Republicans away from their presidential candidate in 16 even though it was a very strange candidate in Donald Trump, there were very few defections.

Matt Grossmann: They both see traditional media playing more of a role despite concerns about fake stories, here’s Robert’s.

Hal Roberts: Immediately after the election there was a big spike in stories about sort of exotic causes of what happened to the media ecosystem to some degree separate from the results it was not just a panic we had the surprising result of who got elected but as much as sense that something really different was happening in the kinds of conversations that were happening where it seemed like built up this sort of expectation and norm for objective journalism over the last 100 years and there was something quite different happening where it didn’t apply anymore. We actually have quantitative results showing that there’s enormous dimension a amount of attention paid to Russian hacking of our media system. There’s enormous amount of sort of Facebook, if you remember after the election we brought Mark Zuckerberg up in front of the Congress to testify about why he ruined our media system about a company called Cambridge Analytica that was claiming to have used data from Facebook to do, what it called Psychographic profiling.

So basically claiming to be able to learn so much about you that it could brainwash you with it’s sort of secret sauce methods of Facebook advertising. And we did find in our data for sure the stories that tended to get a lot of attention on Facebook and not a lot of attention on the open web through hyperlinks or even actually through Twitter shares, who’s sort of Facebook heavy attention stories tended to be much less high quality than stories that were getting a lot of attention say on the web or on Twitter. This certainly to some degree confirmed the reason that we brought Mark Zuckerberg blame him for what had happened. But when we look at our data, what we find is we do find some stories among the most influential stories that were clearly either just pure fabrications.

So imagine Estonian teenager sitting in his bedroom just making stuff off the top of his head and then publishing it and then getting a quarter million Facebook shares for that story. So we found those stories but when we looked at our data we found maybe… I forget the exact numbers now but out of the top 100 stories that we found in over the course of a whole election by Facebook shares, I think four of them were these sort of fake use out of whole cloth type stories. So that’s concerning that the elections are really big story and it’s really hard to get attention at that level. So that the fact that four of these stories were just, someone just literally sat there and made it up. It’s not good but it does not warrant the level of attention they got immediately after the election.

Matt Grossmann: Ladd says fake news was shared less often but it was remembered by Republicans.

Jonathan Ladd: We examined fake news stories versus more high quality news stories and how they’re shared on Twitter. And this is really nice especially in knowledge this is a really nice our lead on doing this was Sharon Boondock who’s a computer scientist at university of Michigan and really did a great job taking the lead on putting together this analysis. First of all we found that the high quality non fake stories do get shared more so everything isn’t horrible. Everything on social media isn’t horrible and so they do get shared more and then send to circulate more. And so the social media leads social media or at least Twitter does a pretty good job sharing legitimate stories from legitimate news outlets.

On top of that, we did find that in our surveys, Republican voters were a little bit more likely to remember the fake news stories about Clinton than the stories that were in traditional news coverage, like our newspaper coverage. So there was a bit of a tendency to believe stories you want to believe which we did find but we did not find that Twitter was a big vector of fake news stories. I should be clear that because people use fake news to me and all sorts of things these days. Our definition in the book is stories that are not just have a bias you don’t like are slanted but are like made up fabricated from scratch.

Matt Grossmann: He says you can’t completely blame the mainstream media.

Jonathan Ladd: Did the public really get their perceptions from media content in this campaign and our story is that not entirely, it seems like certain stories resonated more than others despite what was in the media environment. You can imagine some Democrats thinking, well people only remember the email story about Hillary because the media talked about it incessantly. We don’t find data that would easily confirm that belief, it could be that real right wing media did repeat it incessantly. But if you’re looking for like the media more broadly mainstream media, it did it.

So certain stories resonate and is not because the mainstream media is repeating them. Our data actually pretty much complicates the traditional priming or agenda setting stories where these classic and which were cutting edge at the time and a great studies where you just track what topics are heavily covered in the New York times you see if that primes voters by just tracking a few major newspapers and in a previous media environment, I think that those studies were correct. The stories in national newspapers were the stories people remembered because they were also reflected in local TV news and other things like that. We find a lot of complications to that. We don’t find that the stories that seem to be where the media set the agenda were straightforwardly because they were in major conventional media.

Matt Grossmann: And Roberts thinks offline, Fox News influence may have been even more important.

Hal Roberts: One reason we study stuff online is because it’s online, we have some data about it. Nobody really has the perfect data that we would want for online. So we have various things that are kind of give attention but what every researcher would really like is they would just like to know how many times each story on the web was read for lots of reasons we don’t have that data so the data that we use, the main metrics we use are how many times a story was linked to by someone else in the web, how many times a story was tweeted and how many times it was shared on Facebook. So all of those are really interesting, great pieces of information that tell us a lot about the world, but they all kind of circle around this central question of how many times did someone read something.

But that’s certainly much better and more detailed data than we have about consumption of media offline. Another piece of data we have from Pew is just again, a survey data and not by us just asks people what their primary source of the news are. And what we find there is overwhelmingly Fox news is the most popular single source for the 2016 election. The numbers for something like 20% of all voters in Fox news as their primary news source, about 40% of Republicans. So it is 20% are almost all Republicans and there is no other source that comes even close. The next closest source of something we take person and virtually all of that, our understanding is virtually all of those users are not reporting going to the Fox websites.

It’s just reporting watching Fox on cable TV. That alone cuts strong against this idea that the main problem is Facebook shared stories because when you actually ask people, they’re saying 40% of them saying, no, I just Fox on cable TV. We have a lot of data also that shows that most of those viewers are much older than average. So again, this idea that this election was primarily driven online by Facebook or by other, by Russian hackers or whatever is sort of belied by really basic data that most of enormous amount of their really influential content was being circulated through 1997 technology of cable TV.

Matt Grossmann: Online right wing sites do share more stories of the kind that gets shut down by mainstream stories on the left.

Additional portion of transcript coming soon.