Measuring Social Ties from Roll Call Votes: A Fused Latent Factor and Social Network Approach

Congressional literature suggests that the motivations behind roll call votes are complex, spanning the legislator's ideology, party strategies, and social influences. In terms of methodology, latent factor models have dominated roll call analysis, where the estimated "ideal points" are interpreted as the legislators' partisan-ideological positions, but these models do not account for partisan or social motivations behind the votes. On the other hand, some researchers have explored the social influence behind these votes using network models, but this approach often overlooks the role of ideology or parties. We address this gap by integrating the partisan-ideological and social approaches through a fused latent factor and social network model. This model decomposes the effects of partisan-ideology and social connections on roll call votes while giving priority to the former. Additionally, our method provides a direct measurement of social ties from roll call votes, rather than relying on proxies such as cosponsorship to first estimate the social effect and later make connections to political outcomes. We apply our model to the 101st Senate and find that the model successfully decomposes ideology and partisanship from social ties. The estimated social network captures notable friendships and geographical communities. We also demonstrate that cosponsorship and shared committee membership, commonly viewed as indicators of social connections, are either closely aligned with the legislator's revealed partisan-ideological preferences or have minimal legislative impact.