At the upcoming meeting of the Midwest Political Science Association, Aaron King, Frank Orlando, and I will be presenting a paper that investigates the determinants of success in Senate primary elections. We are primarily interested in whether voters are best modeled as voting by ideological proximity, or whether primary electorates strategically select candidates who offer a better chance of victory in the general election. Essentially, we are trying to identify whether ideological extremity is an advantage or a hindrance to primary electoral success.
Unfortunately, estimating the ideology of many of these candidates can be problematic, given that many, for example, have not cast a roll-call vote which could be used in a NOMINATE-like scaling. Absent a more explicitly political record, we turn to the social networking/microblogging site Twitter, and collect data on the connections between elected officials and the mass public of Twitter users.
We use a nonmetric multidimensional scaling algorithm to estimate a space which represents users’ Twitter behavior, and find that the second dimension of that space correlates very well with Poole and Rosenthal’s NOMINATE scores for Senators and Representatives. Our main results can be seen in the figure below, and the paper is now available for download here.