My recent article, “Federal Election Commission Divided: Measuring Conflict in Commission Votes Since 1990“, published at the Election Law Journal, looks at polarization in voting at the Federal Election Commission (FEC) between 1990 and 2018. The FEC is the federal agency tasked with enforcing and administering federal election laws. It has six commissioners, with no more than three of them from one party. The FEC is often accused of being weak at enforcement and riddled with partisan fighting among commissioners. My paper explores the partisanship in voting among the six commissioners. One of the assumptions of the modeling strategy in the paper is that the FEC votes on a relatively consistent set of topics over time. Otherwise, any change in voting could be either a consequence of changes in commissioner approaches to enforcement (i.e., polarization in commissioner ideology) or a result of changes in the types of controversies coming before the commission (i.e., harder topics on which to find consensus). My assumptions were focused on inferences with respect to the former, meaning I need evidence against the latter.

When the FEC investigates a possible violation of election law, it assigns it to a case number as a Matter Under Review (MUR). Each enforcement case lists a set of topics on the FEC site. For example, see here. The purpose codes assigned to MUR #7063 is “Disclaimer; reporting.” This means the potential violation involves proper and mandated reporting around an electioneering expense, including the inclusion of a disclaimer on the communication that properly identifies the sponsor. With the help of Nick Foley and Morgan Edwards, both Bowdoin Class of 2022, we downloaded the listed purpose codes and did some preliminary topic modeling. (For the record, I’m new to topic modeling approaches and am using these data as my crash course in learning the methodology.)

Below is a word-cloud of the most common purpose tags. Contribution-related controversies dominate, as do reporting problems. Below the cloud is a list of word frequencies, which show the clear ranking of topics.

Finally, here is the topic modeling results, which is both a pretty boring graph but also some confirmation of agenda stability in FEC enforcement cases. The modeling approach looks for clusters of common words. I asked the model to find clusters of 10 common “topics” and plot the proportion that these “topics” appeared in each year. As is evident, there is consistency over time (good for inferences about commissioner-induced changes in voting patterns). Still, the purpose codes assigned by the FEC are often perhaps too blunt. “Contributions” could be about fund-raising by candidates under “hard money” rules or problems in the fund-raising of parties or outside groups around “soft money.” In practice these are very different enforcement controversies.

I’m working with Nick and Morgan to do something with the 1977 to 1990 cases reviewed by the FEC. More on those results when we have them.