Examining Networks of Influence: Using Semantic Similarity Clustering and Affiliation Network Analysis to Reveal Lobbying Dynamics

Abstract

Scholars of lobbying have been limited in their ability to measure organizational lobbying agendas, positions and coalitions. They are either forced to rely on time-consuming interviews or overly-broad Lobbying Disclosure Act-mandated issue codes. We propose a new approach. We use a hierarchical agglomerative clustering (HAC) algorithm to group bills within LDA issue codes based on their similarity, calculated using latent semantic analysis of a corpus we constructed from summary text provided by the Congressional Research Service (CRS). This technique allows us to disaggregate within the existing categories and label individual bills with higher resolution than was previously possible. We then use the clustering groups to label bills within a weighted affiliation network based on the volume of lobbying by a given industry on a given bill. The topic labels of bills within the network provide more detailed insight into the specific policies or provisions different industries have supported, and might be likely to support in the future. As a test, we apply this approach to lobbying on immigration legislation during the 109th-112th Congresses. Prepared for the 6th annual meeting of the Political Networks Section of the American Political Science Association (PolNet), June 28th 2013.