A group of Duke researchers found that some Twitter users are indirectly revealing their political affiliations without necessarily tweeting about politics. The three researchers, graduate students in political science, analyzed the accounts politicians, celebrities and institutions follow on Twitter as well as the users that follow those accounts. This process allowed the researchers to place users on a spectrum ranging from the far left to the far right. The Chronicle’s Julia Ni spoke to David Sparks, one of the researchers and a doctoral candidate, about the findings of this study and its implications.
The Chronicle: Why did you choose to explore this issue?
David Sparks: Our paper is actually about ideology in primary elections and whether relatively extreme ideology is an advantage or disadvantage to candidates in primary elections. So the problem is, it’s not easy to estimate ideology for people who have never formally cast a vote in Congress. We use Twitter as a source of information on which we could base our estimates and look at whether the more extreme ideology was actually advantageous. After doing that, we added a bunch of entities, political figures and interest groups to see if their estimates matched with what we would expect.
TC: Could you explain the methodology of your study?
DS: Essentially, we collected a list of political and public figures with Twitter accounts and identified each individual following and being followed by each of them. We made this into a huge matrix, several million rows long, of 1s and 0s and found the correlation between each of the individuals we were interested in. Then, we used a technique called non-metric multidimensional scaling to essentially reduce the amount of information required to represent the patterns driving following preferences, finding that something very like ideology was a factor. As it turns out, our estimates correlate very well with existing measures of ideology, which are based on roll-call voting in Congress.
TC: How did you decide which organizations and individuals to analyze?
DS: For political figures, we tried to find every incumbent Senate and House [of Representatives] member on Twitter. At the time we collected the data—which was in Fall 2010—we think we had a complete list of all House and Senate members on Twitter at that time. And then, also, we think we had a complete list of every Senate primary challenger on Twitter. So we tried to come up with individuals from both sides of the spectrum. We tried to be complete. As for non-political figures, mostly, we just sat down and tried to decide who would be interesting... media members who covered the election, some celebrities.
TC: What were the most surprising or unexpected results?
DS: I think the most surprising thing about the estimates was that we got a very, very nice discrimination between Republicans and Democrats. If you look at our estimates—which go from about negative 1 to 1.5—at the 0 point, it’s almost exclusively Democrats to the left and Republicans to the right. So it discriminates almost perfectly between parties, which was pretty cool. We were glad to see that. So in this study, we did find that relatively extreme ideology was an advantage to helping people win primary elections. There were a lot of interesting estimates. If you looked at the estimates we did for Senate challengers, the far right is inhabited almost exclusively by candidates affiliated in some way with the Tea Party, which sort of makes sense and is a confirmation that our method was doing what we were thinking it would do. And then, some interesting specific things like Kelsey Grammer coming out on the right and Jane Fonda did come out on the left. Jesse Jackson is a little bit to the left. Basically, a lot of the things that we would expect to happen did happen. Every time we found [a] Michelle Malkin [or] Sean Hannity to the right, that was an interesting finding because it shows the mechanism does seem to work.
TC: Do you think the results you ended up with accurately depict the political affiliations of the people and organizations you studied? How so?
DS: Well, I think it’s tricky in that it does not necessarily say something about their own personal ideology. I think we need to be really careful because I think what we’re actually measuring is the perception of their ideology or partisanship, not necessarily that which they themselves hold. And in some cases, what we measure doesn’t represent their personal or perceived ideologies. So, for example, Katie Couric is quite far to the left and that was totally surprising. I don’t think she is as liberal as it comes out necessarily, but I think rather that during the 2008 election, her interview with Sarah Palin may have generated large numbers of liberal-leaning Twitter users who also follow liberal Senators and Democrats. And that is why she is being correlated with those individuals. So, this needs to be taken with a grain of salt, especially for those working outside the realm of politics. And again, I think it’s important to emphasize that this is perception and not necessarily true reality of how these individuals behave.
TC: What opportunities are there for future research in other fields or using other social networking sites?
DS: There are many ways we could use Twitter behavior to study political phenomena. One in which we are particularly interested is the use of “retweets” and hashtags to trace the diffusion of ideas throughout the political online network. Do ideas or messages tend to emanate from a few central party leaders, or are they more grassroots-driven, with ideas disseminating from non-elites, throughout the network and eventually appropriated by more visible political elites? The vast amount of data being generated daily on sites like Twitter and Facebook presents many opportunities for social scientists.
Get The Chronicle straight to your inbox
Signup for our weekly newsletter. Cancel at any time.