Congressional Networks: Co-sponsorship Ties

Wil Mackey

Introduction:

According to political analysts, the US House of Representatives is becoming increasingly polarized. Fewer bills are being passed, and fewer still have bipartisan sponsors.[i]

Why is that? Obviously, ideology plays a part. But what other factors—such as re-election timelines—can explain who cosponsors what legislation.  

Research Questions:

In the US House of Representatives, whose bills get the most cosponsors? When there are cosponsors, what attributes do they share with the person drafting the legislation? And over the years, have various cliques developed between sponsors and cosponsors—for instance, is there a clique among Freedom Caucus members when it comes to bill co-sponsorship? (Just to clarify, sponsors are those who originally propose a bill. Cosponsors sign on, as a signal of their support, later on.)

Hypotheses:

In the House, I would expect junior members to co-sponsor more bills. They will be more eager to get publicity, and will, most likely, want to start forming relationships with the older members. Those older members, on the other hand, will probably sponsor more bills, but be less likely to cosponsor. Also, representatives’ election timelines will be a factor. If members are up for re-election, they will be more prone to co-sponsor legislation, so they can prove to their constituents that they are “fighting” for their district. Lastly, I bet that members from lower socio-economic backgrounds will tend to cosponsor more legislation, because they will be more concerned about generating campaign donations later on. (Presumably, worse-off members will have to rely more heavily on outside interest groups for campaign donations. They will, as a result, want to cosponsor more legislation to show that they share an industry’s interests, such as the energy sector’s.  Meanwhile, better-off members will be able to rely on past family and work connections for fundraising—or simply fund themselves if they have the means to do so.)

Data collection:

Previously, scholars have used social-network-analysis techniques to examine the legislative process, and there is a lot of publicly available data. The Library of Congress, for example, keeps a record of all legislative activity—including sponsorships and co-sponsorships—and that information could be used to create a one-mode network of all representatives and their sponsorship ties. The ties would be directional, indicating who cosponsors who, and the data would cover the legislative sessions from 2006 until 2014. (The session starting in 2010 would be especially interesting. It was the year when the Tea Party “wave” swept into Congress.)

Afterwards, attribute information would be added to the network, including the members’ party affiliations and their socio-economic backgrounds. Admittedly, finding information on the members’ socio-economic levels would be difficult. Members of Congress are not obligated to release their tax returns.  Also, I’d include information about the safety of members’ congressional districts. If a district was split—meaning that the district voted for one party in a congressional election and another party in the following presidential election—it would receive a score of 1. If it was not split, it would get a score of 0. Perhaps, not surprisingly, there wouldn’t be many 1s. Only 20 districts out of 435 were swing districts in 2012, according to The Economist. Additionally, I’d enter in attribute information about when members are up for re-election—which would be one or two years. I’d also create attribute information about members’ states and regions. For instance, if a member is from New England, he or she would get a 1. If they are from the South, he or she would get a 2 and so on. I’d also add in information about how many years each member has served in Congress, and lastly I’d include information on the members’ involvement with various caucuses on the Hill.[ii]

Social Network Analysis:

At first, I’d look at the whole network, and I’d compare the density levels from the different congressional sessions. It would be interesting, in particular, to see if the density levels drop over time, indicating fewer overall connections among House members. Next, I would use a strong-clique analysis to see if there are any subgroups with particularly close ties, and then I’d see if those cliques share any common attributes, such as caucus membership. Also, using clique analysis, I’d identify if there are any nodes acting as bridges between cliques—in other words members who are part of two or more cliques. Perhaps, if those cliques are affiliated with two different parties, that bridging member could be viewed as a potential bipartisan dealmaker. Lastly, I’d do directional- and non-directional-centrality measures. I’d focus on the members’ InDegree scores as an indication of their institutional clout—the highest-scoring nodes would be those who sponsor the most legislation in each session and get the most cosponsors.

In each step listed above, I’d compare the results with the information contained in the attribute table. For example, if there are nodes acting as bridges between cliques affiliated with different parties, are there common attributes those nodes share, such as being from split districts?  In addition, I would use the E-I Index on the wider network to examine what factors, if any, explain the connections between House members. (For instance, I’d expect members of similar caucuses to have homophilic relationships in terms of bill sponsorship, so their E-I Index scores, based on that caucus attribute, would be close to negative 1.)

At a subgroup and individual level, I’d examine the Egonets of the members with the highest InDegree scores in each legislative session. Then, I’d see if those members remained the most influential in the following years. Also, it would be interesting to see which nodes have highly partisan Egonets and which ones do not. Moreover, by using eigenvector scores, it would be possible to see which InDegree leaders are not affiliated with their party’s leadership—perhaps an indication that they may lead intra-party insurrections in the future. (But to draw that conclusion, the assumption that the parties’ leaderships are well connected must hold.) Additionally, I’d examine the density measures of the various cliques, thereby indicating which of the subgroups have the strongest networks.

Conclusion:

Needless to say, it is hard to draw any conclusions from an uncompleted study. But it would be interesting to compare this analysis with those done previously.[iii] If I had to guess, I’d imagine that the types of networks have changed quite markedly from 2006 to 2014—from networks based on seniority to ones based on caucus affiliation. I’d also bet that this analysis would show that cliques have become smaller, yet denser, over the years—an indication that the House doesn’t have two distinct parties, but in fact contains many sub-parties that should be dealt with independently when legislative horse-trading is required.

---

[i] Laurel Halbridge. Is Bipartisanship Dead? (New York: Cambridge University Press, 2015), 62.

[ii] For qualitative attributes, such as names and caucuses, I’d give each variable a distinct ID, so I could easily match the attribute and network information.

[iii] Fowler, James and Wendy Tam Choo. “Legislative Success in a Small World.” The Journal of Politics. 72:1 (2010): 124-135; Fowler, James. “Connecting the Congress.” Political Analysis. 14 (2006) 456-487; Faust, Katherine and Jon Skovertz. “Comparing Networks Across Space and Time, Size and Species.” Sociological Methodology. 23:1 (2002): 267-299.