Research

Rhetorical organization is politically consequential but difficult to measure at scale. I introduce rhetorical structure embeddings (RSEs), a representation of paragraph-level rhetorical organization derived from adjacent layer transformations in BERT. Combined with an unsupervised dimensionality-reduction and clustering pipeline, the approach discovers recurring organizational patterns in political text. In both a three-source register-recovery task and a content-controlled variant, the RSE outperforms conventional readability, complexity, and grammatical benchmarks, as well as the pooled final-layer output of the BERT model. I then apply the method to U.S. Democratic and Republican national party platforms from 1980 to 2024, identifying six recurrent patterns of rhetorical organization and partisan differences in their use. Democrats rely more heavily on a Problem-Solution style, while Republicans more often use a Purpose-Driven style. The method provides a scalable way to discover and analyze rhetorical organization without predefined categories or collapsing rhetorical form into a single dimension such as complexity.

Political scientists often rely on proxy networks to study relational political processes, yet theoretically important relationships such as patronage, trust, and antagonism are often latent and multiplex. Existing approaches risk conflating the target latent relationship with its proxies and introducing non-classical measurement errors. We propose MuLTI (Multi-indicator Latent Tie Inference), a framework that recovers latent ties from rich but noisy observational data by combining relational traces, dyadic characteristics, and features of the documentation process. We establish nonparametric identification using conditionally independent, tie-specific anchors, and estimate the model by maximum likelihood with stratified stochastic optimization. Simulations show that MuLTI consistently outperforms proxy-based benchmarks, with the largest gains when indicators are noisy and multiple indicators are available. We apply the framework to biographical records of 1,500 revolutionary elites from the Chinese Communist Revolution, estimating positive and negative tie probabilities from twelve extracted observational traces. The application shows that MuLTI recovers nuanced, multidimensional relationships that conventional binary codings would miss.

How do rebel organizations transform wartime networks into post-victory ruling hierarchies? Existing accounts of rebel victory and postwar state-building emphasize ideology, battlefield performance, and formal rank, but pay less attention to the relational foundations of elite incorporation. We argue that wartime social capital becomes a politically valuable asset after victory, but only when it survives the transition from insurgency to regime formation. Using MuLTI, a multi-indicator latent tie inference framework, we estimate positive, negative, and residual wartime ties among 7,798 Chinese Communist ex-rebels. We distinguish between remaining social capital, ties to militants who survived until the founding of the People's Republic of China, and lost social capital, ties to militants who died before victory. The results show that remaining social capital consistently increases the likelihood of attaining central, provincial, or senior military positions. Conversely, lost social capital has a consistent negative impact on postwar elite incorporation outcomes. Key mechanism suggests that this penalty reflects not only the disconnection of individual contacts but also a severe community-level shock: deaths within close-knit, in-group networks weaken the local network infrastructure through which endorsement, information, and sponsorship flow, actively depressing the political utility of am ex-rebel's surviving ties. The findings show that revolutionary victory does not simply reward wartime service; it converts surviving relational capital into state power while leaving those embedded in shattered wartime communities at a disadvantage.