Cosponsorship in the U.S. Senate: A Multilevel Approach to Detecting the Subtle Influence of Social Relational Factors on Legislative Behavior

Why do members of the United States Congress choose to cosponsor legislation proposed by their colleagues and what can we learn from their patterns of cosponsorship? These are questions that have attracted increasing interest among legislative scholars over the past several years, and are, fundamentally, questions about relationships. Unfortunately, most methods of statistical inference with which re- searchers, even methodologists, are likely to be familiar tend to be ill-suited for the analysis of relational data, in which observations are typically interdependent. Previous empirical research on cosponsor- ship in the House and Senate has suffered from two principal limitations. First, it has used statistical tools that ignore the systematic clustering of observations, leading to incorrect inferences. Second, too much emphasis has been placed on large-scale influences such as party and state or region at the expense of more subtle social factors that operate at a lower order of magnitude. In the current paper, we show how carefully chosen random effects might be included within a generalized linear model, in order to better handle network-type patterns of dependence. By explicitly modeling the multilevel structure of the data, we can more confidently investigate whether various dyad-specific properties predict senators' ten- dency to support one another's proposals. In addition to being better suited to a network setting, this approach allows us enormous flexibility in the choice of covariates we may incorporate and the degree to which we pool observations from different levels of analysis. To illustrate, we examine whether a number of potential social factors, capturing homophily, proximity, and role are associated with varying odds of cosponsorship among senators.