Monday, April 4, 2011


I saw some very good papers at MPSA this year, but one that struck me as simultaneously novel, fun, and useful was this one by Duke grad students Aaron King, Francis Orlando, and David Sparks. The authors are interested in figuring out just how much it helps to be ideologically extreme in a primary contest. Unfortunately, we don't have very good measures of candidate ideology, unless the candidates are incumbents (in which case we can approximate their ideology from their roll call voting records). We're mostly left with guessing at candidates' ideological positions from their speeches, donors, endorsees, etc.

King et al decided to look at candidates' Twitter accounts to see who was following them. In theory, the decision to follow or not follow someone on Twitter is in some ways analogous to the decision to vote/not vote for them or to donate/not donate to their campaign. (No, it's not exactly the same -- I follow Sarah Palin but wouldn't give her money or vote for her. But there are certainly similarities.)

The authors use social networks techniques to boil down the literally millions of connections between hundreds of candidates and other political elites to come up with something akin to ideal points for each person. You can see some lists of these ideal points for House and Senate incumbents and their primary opponents here. The authors also scaled some media figures just to see if the results would be credible. (It turns out that Brendan Nyhan is just to the right of Regis & Kelly but just to the left of Toby Keith.) It doesn't appear on the charts, but they scaled me, as well. I have an ideal point of -.077, putting me to Nyhan's left but to the right of Jesse Jackson. (This is believable.)

This is hardly a perfect measure of candidate ideology, as the decision to follow someone is often made for non-ideological reasons. But it works surprisingly well, confirming other studies' findings about ideology and primary elections. Their ideal points almost perfectly predict party for candidates, with the exception of Mickey Kaus. (Frankly, I'd have distrusted their method more if it had gotten Kaus correct.) Within party, the measures are a bit noisy. Interestingly, they have Andrew Romanoff to the right of Michael Bennet, although not by much. It would probably help if they could bootstrap some sort of standard error equivalent in there just so we could tell whether these differences matter.

One other praiseworthy item: the authors distributed a full-color handout during their presentation which contained a QR code, linking to the project's website. Very cool.

Anyway, this is a project worth watching.


dsparks said...

I just updated the summary PDF (see here) to include smotus, in the "Other" category at right, between the Brookings Institution and the American Association for Justice.

Seth Masket said...

Thanks, David. Most helpful.

Rick said...

An interesting pattern in the Media/Online and Other categories is the sharp break about a third of the way down on each graph, between (what I would guess to be) 'movement' conservatives and commentators who are a bit right-leaning but not 'movement.'