| Much recent empirical work on legislative decision-making, in the United States Congress in particular, has taken a spatial approach, driven by improved methods of estimating legislator ideal points and locations of bills within a shared latent Euclidean space. However, for research questions that involve relationships and interactions among members of Congress, models relying heavily on an assumption of independence among voter decisions will not be suitable, at least in isolation. In the thesis, I consider research questions that are relational in nature the observations of primary interest are taken on connections among United States senators, rather than on the senators themselves. In asking what sorts of situations and social factors raise the propensity of a senator to cosponsor a colleague's proposed legislation, I utilize data on the senators themselves, but the key insights pertain to links in a directed social network. In modeling a cue-taking process by which late deciders rely on better informed early deciders in casting roll call votes, here too the phenomenon under study manifests itself not in the behavior of individual political actors, but rather in the space between, in what the declared preferences of each reveals about the likely preferences of the others. The models and statistical methods employed must thus be chosen carefully to handle the sorts of dependence that most widely used techniques cannot. In the essay on decisions to cosponsor, I employ a form of generalized linear mixed effects model that alleviates major sources of bias and provides more believable standard errors on estimates than would otherwise be the case. In the chapters on cue-taking in roll-call voting, I consider the process by which late-deciding senators make up their minds as analogous to the operation of recommender systems in the marketing context. The cue-taking models employed all begin with the assumption that legislators under heavy constraints on time and information must rely on simple heuristics, yet will approach the task in the spirit of repetitive problem-solving rather than in a haphazard manner impervious to the researcher's scrutiny. I demonstrate the usefulness of the relational approach by applying it to two questions: (1) Who were the mavericks of the 108th Senate and (2) Are multiple cues that include one's political opponents among sources more informative than those that originate with only allies?... |
