Recovering Collective Structure in Congressional Voting: A Fused Latent Factor and Network Approach

Congressional voting reflects collective behavior structured by chamber-wide patterns such as partisanship and relational patterns among particular subsets of legislators, including factions and coalitions. Yet latent factor frameworks that dominate roll call analysis assume conditional independence across legislators and represent their ideal positions in a linear space. These restrictions can obscure relational patterns that do not align with chamber-wide latent dimensions. To model these behaviors, we adopt a fused latent factor and network model that captures chamber-wide patterns with the latent factors and attributes residual structure to the network. Applying this fused model to the 101st (1989--90) and 114th (2015--16) Senates, we first validate that the latent factors recover partisan voting patterns and identify two systematic failures of the latent factor: senators with moderate latent positions are assigned uncertain probabilities on party-line votes, and intra-party split votes exhibit systematic co-defection patterns. The network component closes the misprediction gap entirely for party-line votes but only partially for intra-party splits, capturing stable co-defection patterns while leaving more transient coalitions less predictable. Next, we apply a degree-corrected stochastic block model to the fitted network and recover interpretable voting blocs in both Senates. The 101st Senate features within-party factions that match historically documented coalitions and are driven by shared issue interests despite overlapping positions in the latent space. In the 114th Senate, the blocks take on an entirely different interpretation, uncovering two institutionally opposed coalitions: a cross-party governing bloc, including members of the Republican party leadership, and a Republican insurgent bloc, both driven by opposition of the other. Our findings demonstrate that roll call data alone contains recoverable signals of collective legislative behavior beyond partisanship that remain after accounting for chamber-wide partisan structure, and the organization of this collective behavior shifts from issue-based coalitions toward institutional conflict across decades of polarization.