Coevolution, interconnections and communities of social and political networks in the United States CongressNetworks, or graphs, provide a powerful tool for representing and analyzing complex systems of inter¬acting agents. Individuals and organizations are embedded in numerous types of networks whose structure plays an important role in explaining their behavior. Connections between agents in a network can thus be represented by many different quantities that each yield different estimates for the strength of the tie between two agents. For example, individuals communicate through multiple media (in person, by telephone, online, etc.) and scientists collaborate via both coauthored papers and social interactions at academic conferences. The goal of this project is to apply this "multinetwork" perspective to legislators in the United States Congress, who are connected both politically and socially through membership of the same committees and subcommittees, common voting behavior, cosponsorship of legislation, repre¬sentation of geographically proximate constituencies, and so on. The presence of many different types of connections (which are inter-related and that evolve in time), in this and other networks, raises the question that our research will address: How do the various social and political connections of legislators in the United States Congress influence each other and change dynamically in time as a result of their interconnections?
Although the quantitative study of real-world networks has a long history in the social sciences, such investigations experienced a major expansion in popularity in the late 1990s, in part because of interest in the internet and online networks. Together with the improved availability of large-scale network data, high-performance computers, and better algorithms, this has led to a rapid development in the tools for studying social, biological, and technological networks. Such research has generated important insights into the effects of network topology on individuals behavior, including collaborations, community forma¬tion, and hierarchical and modular organization. For example, studies of interorganizational networks have yielded new insights into how alliances and other ties are formed, how they affect organizational performance, and how various organizational practices spread in such networks.
Despite these advances, there are few results on the dynamics and coevolution of networks. Most studies of social networks are modeled mathematically using a single graph, which is represented by an adjacency matrix whose entries consist of the ensemble of unique connection strengths between each pair of individuals. This implicitly assumes that there is only one type of connection in the social network and does not consider different ways of measuring the ties between individuals. This simplification leaves out a lot of information, but it has been warranted because it greatly simplifies the analysis. Our goal is to use our wealth of Congressional data to take the important step of explicitly considering coevolution; we will exploit the fact that nodes are typically embedded in multiple types of networks (with multiple ties, of differing strengths, between individuals), and that the structure of the networks and the behavior of the actors embedded within them are often interdependent.
Over the past few years, we have led a nascent effort to apply network theory to the United States Congress and other political groups and organizations. Political networks include a tremendous amount of information that has remained untapped by conventional qualitative, game-theoretic, and statistical approaches; and new advances in network theory promise to uncover the ways in which social relationships shape political outcomes. As the first steps in this process, we have considered, on an individual basis, Congressional networks composed of ties based on committee and subcommittee assignments, legislation cosponsorship, and roll call votes.
By computing the hierarchical and modular structures of communities in the House of Representatives and combining this information with an analysis of the ideologies of their constituent legislators, we have investigated correlations between the political and organizational structure of House committees and subcommittees. For example, we revealed close ties between the House Rules Committee and the Select Committee on Homeland Security in the 107th (2001-02) and 108th (2003-04) Congresses. We have also identified structural changes (e.g. an increased presence of modules in the community structure) in the House of Representatives that resulted from the 1994 elections, in which the Republican party earned majority status in the House for the first time in more than forty years.
Mapping the network of cosponsorships can help us uncover the "hidden" social networks between politicians, as legislators who work closely together on pieces of legislation are likely to have friendly (or at least cordial) relations. That is, although they do not broadcast lists of their friends and foes, their bill cosponsorships provide a paper trail for these types of social relationships. We examined these hidden networks by analyzing 150 million cosponsorship decisions made from 1973 to 2004 and the results were astonishing. The top 20 legislators at the center of this network read like a who's who list of American politics, including the likes of Bob Dole [R-KA], John McCain [R-AZ], and Ted Kennedy [D-MA]. We showed that these individuals are consistently successfully in lobbying for amendments to legislation (previously the best measure of legislative influence), and that they were personally influential in swaying votes during roll calls. They were also more likely to seek higher office in the future (including John McCain, Hillary Clinton [D-NY], and Ron Paul [R-TX]) or to succumb to scandal. Finally, the networks clearly identified close personal relationships that had no basis in shared committee memberships or geographic affiliation. For example, they showed John McCain and Phil Gramm [R-TX] had one of the strongest relationships ever recorded, which is unsurprising given that McCain ran Gramms campaign for the presidency.
Recently we conducted a longitudinal analysis of the modularity of the above legislation cosponsorship network and another network constructed from the full history of Congressional roll call votes. Graph modularity indicates how well a given network partition reflects the actual groups in a network, and our analysis revealed patterns that deviate (in some cases drastically) from prior studies of Congressional polarization. These deviations indicate that parties are not the most significant communities in Congress for certain periods of history. This finding also allowed us to use techniques from graph theory to study historical party realignments and suggests that the 1994 party-changing elections were a manifest consequence of a rise in partisan polarization rather than an event that caused abrupt polarization in the country.
We are now ready to build on our past research of individual networks in Congress with a multinetwork approach that simultaneously considers multiple ties between different people and/or organizations. We will take advantage of the enormous amount of information that has now become available because it contains evidence of the inner workings of important political institutions. The quantitative study of ideology in the Congress has greatly improved understanding of what politicians want, but scientists are still uncertain about the precise social and political mechanisms they use to get it. By studying the complex interactions that underlie millions of decisions about elections, campaigns, and legislation, we hope to use a multinetworks approach to better understand the role of information flows and personal influence in political outcomes.
A multinetworks approach adds a significant level of sophistication to our previous investigations. It is well-known that personal relationships greatly affect group decisions and outcomes, but these relation¬ships are difficult to observe and quantify in large groups such as the U.S. Congress. It is common practice in network studies (including our own) to look at one type of connection at a time and then to make some simple comparisons between the separate networks after the fact. However, our current data (in-eluding information on committee assignments, legislation cosponsorship, roll call voting, and campaign finance contributions) yields multiple families of Congressional networks (multinetworks), which allow a direct comparison between different social networks that arise from the same sets of individuals. By investigating these different networks simultaneously, we will analyze the interrelations of political and social communities among legislators and obtain a better understanding of the U.S. legislative process, much as network theory has already yielded a better understanding of corporate practices. With our longitudinal data, we will also be able to investigate the dynamical evolution of microscopic (individual legislators), macroscopic (collective behavior), and mesoscopic (communities of various sizes) structures in Congressional networks with political party reorganizations, elections, and so on.
With the tremendous impact that U.S. policy has on the world, it is crucial that we better understand how Congressional networks influence everything from deficit spending to foreign policy. Our study also promises to be amenable for generalization far beyond the U.S. Congress. Because the abstract entities of networks-nodes and links-are independent of application domain, our analytical techniques and multinetworks perspective will be directly applicable to a variety of domains. Our study will thus produce a contextual roadmap, as well as new mathematical tools, for how to perform similar analyses in many other contexts including international trade and military alliances, food webs and ecological niches, drug use and sexual contacts, the social organization of university life, and more. The time is ripe for new tools and perspectives that exploit the sudden explosion of data in all these areas.
