Programme 2026

Socialization through institutional engagement is a key mechanism through which states develop communities of common interest and internalize shared norms. While intergovernmental organizations (IGOs) are the classic example, we explore how socialization mechanisms interact across institutional domains. This paper investigates the co-evolutionary relationship between two major socializing structures: IGOs and defense cooperation agreements (DCAs). These institutions do not operate in isolation. Rather, socialization in one domain affects propensity for engagement in another, creating an interconnected system of state socialization. Using a relational operationalization of socialization, we measure integration into communities of states rather than monadic attributes like ideology. This approach better captures the fundamental mechanism of socialization: routinized interaction. To test this interdependence, I develop a stochastic actor-oriented model (SAOM) estimating both the DCA and IGO networks simultaneously, incorporating multilevel terms that capture cross-network effects. This model reveals a bidirectional relationship between these networks: firstly, states with similar IGO portfolios are significantly more likely to establish defense cooperation agreements. Notably, this similarity effect is substantially stronger than simple counts of shared memberships, suggesting that community structure drives DCA formation. Secondly, bilateral defense cooperation agreements increase the likelihood that states subsequently join the same IGOs. These effects indicate that the defense cooperation and IGO networks form a cohesive co-evolutionary system. These results demonstrate that focusing upon a on single institutional context misses dynamics of state socialization through institutional engagement. The interdependence between DCAs and IGOs reveals that security interdependence is partially reflected in diplomatic community structures, and vice versa.

Why do informal intergovernmental organisations (IIGOs) such as the G7, G20 and BRICS select certain multilateral development banks (MDBs) as intermediaries in orchestration arrangements, while others are consistently bypassed? Existing orchestration scholarship has emphasised the functional attributes of candidate intermediaries, including resources, expertise, and mandate fit, but has paid less attention to how an MDB's relational position within the broader inter-organisational field shapes its likelihood of being selected. This paper asks whether the structural embeddedness in the MDB network exerts an independent effect on intermediary selection.

The analysis draws on an original panel covering 3 IIGOs and 28 MDBs from 1999 to 2018. The data comprise two linked networks. The first is a bipartite IIGO–MDB orchestration network, in which ties record whether an IIGO selects a given MDB as an intermediary in a given year. The second is a multiplex MDB–MDB network, in which ties capture inter-organisational relations across three layers: co-financing, information-sharing, and co-donor membership. Together, these two networks allow the analysis to link an MDB's position within the inter-organisational field to its likelihood of being selected as an orchestration intermediary.

The paper proceeds in two methodological stages. A Firth penalised logistic regression provides a dyadic baseline appropriate for rare-events data. The core analysis then specifies a bootstrapped Temporal Exponential Random Graph Model (TERGM) that simultaneously captures endogenous network dependencies. Preliminary results suggest that functional focality is a consistent predictor of intermediary selection, while several network-level terms point to additional selection logics that are not reducible to functional attributes. Some covariates display sensitivity across the two specifications, indicating that dyadic and network models may recover different aspects of the orchestration process.

The paper contributes to political network scholarship by treating intermediary selection as a bipartite network outcome, and to IIGO research by showing that informal governance arrangements are shaped by both functional and relational considerations.

IR literature treats strategic autonomy as a zero-sum game of alignment. Yet, this fails to explain the “Network Paradox” of Central Asia: deep structural integration into competing global orbits without formal alignment. This paper identifies a gap in “multiplex brokerage” and introduces the “Structural Bypass” – a theoretically grounded mechanism enabling middle powers to navigate zero-sum constraints by decoupling their network layers.

This study analyzes a multiplex network of 40 sovereign states, stratified by institutional affiliation (EU-27, SCO, observers). Three distinct layers are operationalized:

1.#Legal Layer: Regulatory alignment depth, measured by EPCA implementation rates.

2.#Diplomatic Layer: Fine-tuned LLM-based sentiment extraction (RoBERTa-large) from joint communiques (2015–Q1 2026).

3.#Connectivity Layer: Edge-level covariate based on Middle Corridor (TITR) cargo throughput (TEUs) and investment flows.

Methodologically, I employ a Generalized Exponential Random Graph Model (GERGM). To ensure model convergence at $N=40$, I utilize MCMC diagnostics and sensitivity checks on cross-layer parameters. This identifies an architecture of “Network Decoupling”, where legal integration is structurally decoupled from diplomatic signaling to preserve maneuverability.

This paper contributes by:

1.#Reframing hedging theories through the "Structural Bypass" mechanism to explain middle power survival.

2.#Applying GERGMs for the first time to Central Asian geopolitics using an up-to-date (Q1 2026) dataset.

3.#Mapping regional network dynamics following the post-2022 fragmentation.

Findings uncover a multiplex mismatch between normative integration and diplomatic signaling, posing structural constraints to external transformative power. Ultimately, strategic autonomy is a function of multiplex structural position, where power is derived from the ability to maintain non-overlapping institutional dependencies.

The Advocacy Coalition Framework (ACF) argues that policy subsystems are structured by collaboration among actors who share policy-relevant beliefs. Empirical research has typically operationalized collaboration using binary or undifferentiated collaboration ties, implicitly assuming that collaboration relationships are symmetric. This paper challenges that assumption by distinguishing between limited and extensive collaboration and examining how these forms of collaboration are structured asymmetrically across different types of organizations. Using original climate policy network survey data collected in Spain in 2024–2025, we introduce a measurement strategy that distinguishes limited collaboration — characterized by episodic, routine contact — from extensive collaboration — characterized by sustained interaction, joint activities, and coordinated action.

We show that powerful governmental actors, such as central ministries, tend to report numerous limited collaboration ties, reflecting their central position and broad reach within the subsystem. In contrast, non-governmental organizations often report extensive collaboration with ministries, indicating relationships that are more consequential for them than for their more powerful counterparts. These asymmetric relations reveal how collaboration ties embody power differences and unequal dependencies within policy subsystems. Policy networks thus reflect not only shared beliefs and strategic alignment, but also power asymmetries and competing organizational interests involved in policy processes.

The major assumption of the Advocacy Coalition Framework (ACF) is that advocacy coalitions are belief homophilous and tend to persist over time. In other words, policy actors tend to collaborate with like-minded others, which may result in the formation and maintenance of such coalitions. However, they may be susceptible to alteration in their beliefs through the process of policy learning under certain conditions, such as exposure to novel information or various external and internal events. This research utilizes organizational survey data collected during two governmental periods (2013-2017 and 2021-2025) in the Czech Republic, which are divided by the adoption of the European Green Deal in 2019. To disentangle belief-driven selection (belief homophily) from network-based belief alteration (policy learning), the study employs stochastic actor-oriented modelling (SAOM), jointly examining changes in information ties, collaboration ties, and policy beliefs over time. Our expectation is to identify two complementary, co-evolutionary tendencies. First, we expect that the selection via belief homophily is present. Second, we expect policy learning to be differentiated, depending on the dominant information sources. Thus, we explore whether the belief homophily shifts to operate on different sets of policy core beliefs and policy instrument beliefs – for instance, moving from the nature of climate change to beliefs about specific climate policy instruments, such as the ban on combustion-engine cars.

Drawing from the Advocacy Coalition Framework (ACF), this paper argues that research on policy learning faces two critical barriers: the systematic measurement of belief systems, and the systematic measurement of policy networks. Both of these phenomena (beliefs and networks) are integral to a theory of learning, however both are characterized by conceptual vagueness and extreme measurement challenges.

This paper demonstrates a new approach to face these challenges. Using invasive species management in the Southwestern United States as a policy context, we use natural language processing (NLP) and large language models (LLMs) to implement a theoretically-grounded ontology of networks and beliefs. The ontology is then applied to the machine-automated measurement of various types of beliefs and networks from text archives of invasive species management activities. We thus demonstrate the possibility of LLMs to measure beliefs and networks on a large scale and over time.

These data are used to test core hypotheses of learning, particularly that policy actors tend to adopt the policy beliefs of their collaborators (social influence) and that actors actively adjust information environments based on belief similarity (homophily). While these hypotheses have been tested in some prior research, this research has often been limited to static observations at one point in time, or limited to only a single type of belief and/or form of policy-relevant coordination. We show that social influence and homophily, as key drivers of learning, operate differently depending on the type of actor, type of belief, and type of coordination. Overall this paper simultaneously provides a proof of concept for a new approach to the measurement of learning inputs, and the use of these measurements to advance the theoretical treatment of learning within the ACF.

Understandings and conceptualizations of network data to represent real world systems have meant that social network scholars frequently move beyond the binary one-mode networks that have previously served as the main social networks data structure. Here, we reformulate the network autocorrelation model to adapt to these new types of data. Specifically, we show how the model can be applied to predict tie weight in both one- and multi-modal systems. These models do not predict whether or not a tie exists, but instead, given that the tie exists, is there credible social influence motifs that affect the strength of any given tie? We apply this model to some classical datasets which have dense but variably weighted ties as well as networks where the ties represent ratings of policy performance. In this case, we hypothesize that how respondents perceive policy effectiveness depends on the ratings that their collaborators give, and more importantly collaborates with whom they share additional partners.

Exponential random graph models (ERGMs) are widely used to study political networks, but standard implementations assume homogeneity across actors once covariates are included. This assumption rarely holds in political systems, where networks are routinely embedded in administrative, partisan, or territorial hierarchies. Existing solutions face a fundamental trade-off. Fixed-effects specifications absorb group-level heterogeneity but discard cross-group information, produce unstable estimates for small groups, and distort structural network parameters. Pooled ERGMs recover efficiency but impose a homogeneity assumption that is often implausible and require structural zeros between groups, rendering them invalid whenever ties cross group boundaries. Bayesian hierarchical ERGMs achieve partial pooling but—besides prohibitively expensive for large networks—assume multiple independent networks, precluding their application to a single connected network with embedded group structure. In this paper, I propose a mixed ERGM framework that incorporates group-level random effects as a computationally tractable alternative that addresses these limitations. Building on the iterative estimation procedure of Kevork and Kauermann (2022), we generalise their framework to group-level intercepts, combining simulation-based maximum likelihood for structural parameters with penalised pseudolikelihood for group effects. This yields partial pooling across groups (i.e., shrinking group-specific propensities toward a common mean while preserving cross-group ties) at computational costs feasible for the large sparse networks common in political research, and without the bias in structural parameter estimation that afflicts pure pseudolikelihood approaches. We introduce a pseudo-intraclass correlation coefficient to quantify the degree of group-level clustering and assess whether a hierarchical specification is warranted. Monte Carlo experiments demonstrate that fixed-effects and pooled alternatives produce biased estimates under heterogeneous data-generating processes, and that the proposed estimator outperforms both, particularly when groups vary in size. We illustrate the method using canonical policy network data and a novel dataset on intergovernmental environmental collaboration in Colombia.

Exponential random graph models (ERGMs) provide an efficient framework for modelling local mechanisms that generate global network structure, but their estimation remains challenging due to their computational intensity. This study proposes combining three heuristic approaches, bootstrapped Maximum Pseudolikelihood Estimation (BMPLE), Equilibrium Expectation (EE), and an egonetwork-based approximation, with LASSO regularised ERGM variable selection. Using a collection of synthetic networks with known generative processes, as well as a large-scale empirical collaborative network of rock musicians, the study evaluates the ability of each method to rank term importance and reduce computational costs. Results show that the EE-based LASSO method provides the best balance between reliability and efficiency, consistently giving greater importance to true generative terms, and running several times faster than alternatives. Application to a large empirical collaboration network of rock musicians further confirms EE’s ability to preserve theoretically meaningful effects while discarding randomised attributes. These results highlight the potential of heuristic ERGM-LASSO methods for automated, scalable model specification selection.

While "Stop Asian Hate" originated as a North American racial justice movement during COVID-19, its global expansion catalysed localised solidarity building. In the UK, this mobilisation took on a distinct trajectory, witnessing the rapid emergence of grassroots organisations from 2020 and the articulation of a new panethnic political identity: ESEA (East and Southeast Asian). Adopting a mixed-methods social network analysis that combines surveys, ethnography, interviews, and social media data, this research aims at investigating the consolidation of ESEA identity and the network structures of these social movement organisations.

Central to this study is the duality of network and identity, positing that movement identities are not pre-existing essence but shaped and reshaped through relational configurations, and networks are embedded with identity process. While the research is still under data collection stage, the preliminary finding reveals that although ESEA-centric networks existed prior to 2020, the institutionalisation of the non-profit organisation besea.n (British East and Southeast Asian Network, launched in September 2020) provided the critical semantic foundation and network position for this identity process. Subsequently, besea.n acts as a pivotal broker who occupied a structural hole within the UK’s racial justice field, bridging fragmented ethnic-specific clusters and mainstream activism. Therefore, ESEA identity is argued to be a relational outcome for “social footing” (White, 2008) by the social actors who are historically marginalised and politically silenced in the UK’s racial regime.

The first "Great Migration" saw the greatest redistribution of Black American populations across the nation. Motivated by the expansion and preservation of Jim Crow subjugation, widespread racial violence, and declining economic conditions, Black people sought to lay down roots in new places outside of the South. Now, amidst a resurgence of racial policy threats and new economic crises, Black Americans find themselves in a familiar dilemma: should they stay in their current place of residence or seek refuge elsewhere? In this paper, we explore the logics motivating Black peoples' migration considerations in a time when many feel under assault by state-level policy shifts. Using surveys of nearly 2,500 Black Americans, I construct networks that map (1) which states Black people are inclined to leave and (2) which states are perceived as ideal landing spots. From there, I explore the various rationales behind migration considerations and use ERGM models to align expressed concerns (i.e., to find states that offer better racial, economic, and political conditions) with states' characteristics. This process leaves us with a more comprehensive understanding of what is at stake for Black Americans as they navigate the current political climate.

Despite the extensive theoretical work on political representation within parliamentary democracy, its empirical application often remains underdeveloped. Current parliamentary and legislative studies frequently employ reductive metrics, such as counting Prime Minister's Questions appearances or simple voting patterns, to measure this phenomenon. This paper contends that while the theoretical literature has largely evolved past the Principle-Agent framework, the methodological approaches used by parliamentary scholars have lagged behind.

A significant oversight in existing methods is the failure to capture the dynamic interplay among various actors, including MPs, Lords, NGOs, CSOs, and Parliamentary staff, and how these connections fundamentally shape the process of representation. To address this gap, this study applies Social Network Analysis to the UK Parliament's refugee policy domain during the 2019–2024 period. This paper uses the refugee policy arena as its case, due to refugees acute need of high quality representation. 

Adopting a time-series perspective, the research tracks shifts in refugee policy network dynamics across successive ministerial reshuffles and three different premierships. By using ego-level centrality measures, the paper identifies the key individuals driving policy debates around refugees and the flow of information crucial to producing representative outcomes. It further distinguishes between clusters of core advocates and those actors peripheral to the network. The findings reveal a structural weakness: refugee representation relies on an active but vulnerable network that is highly dependent on specific brokerage roles. Disruption to these roles, often caused by administrative turnover, significantly impairs the quality of representative output. However, the data also highlights the presence of robust, cohesive clusters that ensure policy continuity despite executive volatility.

In conclusion, this paper argues that refugee representation is not simply the result of individual MP action, but rather the product of an engaged and active policy network. Ultimately, the study seeks to move the parliamentary representation literature beyond a reductive Principle-Agent model toward a more accurate, dynamic, and network-based understanding.

The international system can be represented as a multiplex network of state-to-state relations across diplomatic, economic, security, and interaction domains. To understand relationships of dependence among the layers, I apply the nonnegative Tucker decomposition (NNTuck) to a high-dimensional representation of the international system consisting of twelve relational layers. The results point to a layer‑dependent structure consisting of three layer communities, with reference layers for these layer communities being alliances, arms transfers, and bilateral trade. Together, these three reference layers can reproduce the bulk of the multiplex structure, with other layers expressible as linear combinations of these references. Further examination of the layer dependence matrices indicate a major partition between longstanding institutional layers and security-interaction layers, which indicates a divide between networks representing sticky, normative layers and practical, security-oriented layers. Methodologically, the NNTuck offers a scalable, interpretable framework for identifying structurally informative layers and uncovering layer interdependence structures in richly multiplex networks like the international system.

This paper develops a multi-layer network approach to analyzing governance systems by demonstrating how different types of interaction produce distinct structural patterns. Rather than representing state relationships through a single network, the study constructs separate layers capturing key governance functions: signaling through United Nations voting, institutional coordination through shared memberships and agreements, and operational collaboration through joint activities and partnerships. Each layer is independently defined, allowing comparison across interaction types while holding the set of actors constant.

The empirical analysis draws on two domains, maritime and space governance, to illustrate how this framework can be applied across issue areas. These domains provide variation in institutional design, operational activity, and patterns of state engagement, making them useful for examining whether network structure is consistent or context-dependent. Across both cases, the analysis compares density, community structure, and patterns of connectivity within and across layers.

Preliminary results indicate that network structure varies systematically with the type of interaction being modeled. Differences emerge in how communities form, how central actors are positioned, and how coordination is distributed across the network. Importantly, these structural patterns are not consistent across layers, suggesting that governance relationships are function-specific rather than unified across domains. The same set of actors may exhibit cohesive alignment in one layer while appearing fragmented or hierarchical in another.

The paper contributes to political network analysis by emphasizing the importance of edge construction and multi-layer design in shaping observed outcomes. By demonstrating how structural divergence emerges across governance functions and domains, the study provides a framework for analyzing complex systems in which coordination, hierarchy, and alignment are distributed unevenly across different forms of interaction.

How do bilateral diplomatic ties shape normative alignment in multilateral climate politics? This research develops a dyadic network approach to the study of UNFCCC party statements, arguing that normative alignment is conditioned not only by coalition membership or structural position in the wider diplomatic system, but also by patterns of bilateral engagement between states. Building on a broader dissertation on diplomatic interaction and climate discourse (Ali & Voinov Vladich, 2026), the project introduces a dyad-year dataset that links pairs of UNFCCC parties across successive COPs to measures of diplomatic contact and normative diffusion. The explanatory side combines bilateral engagement indicators derived from diplomatic interaction data, including leader travel (Moyer et al., 2025) and related dyadic covariates, while the outcome side captures alignment in public statements delivered in UNFCCC plenary settings.

To improve conceptual precision, the research also incorporates a representation of climate discourse based on norm clusters (Winston, 2018). Rather than treating statements as undifferentiated texts, the analysis distinguishes among clusters of problems, values, and mechanisms, allowing discursive alignment to be measured with greater substantive specificity. This approach, combined with techniques such as discourse network analysis (Leifeld, 2017, 2020) makes it possible to ask whether bilateral interaction is associated with convergence across climate discourse in general, or whether it is more strongly related to particular normative components.

Methodologically, the project combines computational text analysis (Grimmer & Stewart, 2017) with dyadic network modelling (Kenny et al., 2006). The research presents the architecture of the dyadic dataset, the operationalization of bilateral engagement, and preliminary evidence on how interaction structures may condition convergence in climate diplomacy. In doing so, it contributes to political network research by showing how bilateral ties embedded in multilateral arenas can structure discursive outcomes, and to climate politics by offering a relational account of norm alignment within the UNFCCC.

Local planning policy is an area of policy-making with direct impacts on people's living conditions. Yet planning policies are understudied in terms of their diffusion. They are often both necessarily hyper-local and translated to specific contexts, but also influenced by broader discourse. We suggest that only by including a diffusion network perspective, research can disentangle these factors.

We theoretically propose that diffusion networks operate at three levels. Local planning policies spread through 1) spatial proximity, through 2) hierarchical polity networks (especially in federal systems) and through 3) broader discourse networks.

We empirically test the relative influence of these three network processes by tracking how specifically speed reduction policy for motorized individual traffic (so-called Tempo 30 proposals) have diffused in Swiss municipal, cantonal and federal parliaments. To do so, we utilize a newly available data source on Swiss parliamentary affairs, OpenParlData and analyze time-stamped occurrences of Tempo 30 affairs, which we identify using automated text classification. We fit a model to predict the occurrence of Tempo 30 initiatives based on spatial network diffusion among municipalities and hierarchical network diffusion across federal levels, adjusting for larger effects of rising prominence in broader discourse, language barriers and party composition in parliaments.

Migrating from one social network platform to another can reconfigure communication networks, visibility, and influence in political and scientific online spaces. Such migration may be driven by several distinct mechanisms at once, including exogenous temporal shocks, organizational coordination, and peer effects. Identifying which mechanisms are at work is a challenge because adoption unfolds over time on an underlying network. From a contagion perspective, one actor's adoption of a new platform changes the future network neighborhood of others, inducing dependence among platform-adoption events. Modeling this dependence at the network level, and distinguishing it from coordination or common temporal shocks is therefore not trivial, since the network connectivity between actors violates the typical assumption that event observations are independent. Using the pre-adoption Twitter interaction network of more than 16000 climate policy actors and the users' observed Bluesky adoption times, we develop a network event-history model with a joint likelihood over interdependent adoption events. We study how the adoption of Bluesky can depend on local network exposure while controlling for exogenous and environmental factors such as the toxicity and disagreement an actor experiences. We moreover compare alternative simple and complex contagion mechanisms in their ability to explain observed adoption events.

How does scientific knowledge travel through the policy ecosystem, from its initial production to its ultimate use in policymaking? Existing research has largely examined this process in fragmented terms, focusing either on how policymakers selectively use scientific evidence (demand side) or on how intermediary institutions such as think tanks produce and disseminate science-based policy knowledge (supply side). This project integrates these perspectives by modeling the full, dynamic system through which scientific knowledge diffuses across policy institutions.

Building on large-scale data linking approximately two million policy documents to nearly one million scientific publications, the study treats citations to scientific research in policy documents as observable traces of diffusion. When multiple institutions cite the same scientific work over time, these sequences form diffusion cascades that reveal underlying pathways of knowledge transmission. Leveraging probabilistic network inference methods, the project reconstructs a latent, time-varying “science diffusion network” connecting policy institutions across the United States.

The analysis proceeds in three steps. First, time-stamped citation cascades are constructed from integrated bibliometric and policy data. Second, dynamic network inference models are used to estimate directed diffusion ties among institutions. Third, the structure and determinants of the resulting network are analyzed, identifying early adopters, bridging actors, and clustered communities, as well as the political and institutional factors shaping diffusion pathways.

This approach generates the first system-level map of how scientific knowledge enters, circulates within, and shapes policymaking. Substantively, it sheds light on whether scientific knowledge flows broadly across institutional and ideological boundaries or remains confined within polarized enclaves. Methodologically, it advances the integration of large-scale text data, bibliometrics, and dynamic network modeling within computational social science. More broadly, the project contributes to understanding the informational foundations of democratic governance in an era of polarization and contested expertise.

Despite a well-known overlap and common origins between spatial econometrics and statistical network modelling, the two literatures have developed separately such that core concerns in one tend to be ignored in the other. While spatial econometrics focuses on node features, taking the network as exogenous, statistical network modelling focuses on tie features and network structure. This study focuses on the importance of network structure for the performance of estimators for the spatial autoregressive and the spatial error model.

In Monte Carlo studies of the relative performance of different estimators in spatial econometrics (e.g. Calabrese and Elkink, 2014), the spatial contiguity matrix is typically based on highly regular network structures, with low variation in degree between nodes and high connectivity, for example by using a random geometric network (Dall and Christensen, 2002). I argue that the structure of the network affects the amount of information available in the estimation of the spatial coefficient.

Using simulated networks (Morris, Handcock and Hunter, 2008), I construct random networks that vary in terms of levels of clustering, connectedness, density of ties, degree distribution, and other common characteristic statistics on network structures, and generate data assuming varying levels of spatial clustering in the node features. I then evaluate through simulation the performance of standard linear and binary spatial econometric models in terms of the estimation of spatial autoregressive and spatial error coefficients, as well as coefficients on covariates.

Spatial time series (STS) data are central to many scientific domains, yet standard analytical approaches often blur the distinction between temporal co‑evolution and genuine spatial dependence. This conflation obscures subtle but consequential spatial anomalies. We present a Random Matrix Theory (RMT)–based framework that extracts core spatial association by selectively trimming dominant temporal signals while preserving meaningful spatial structure. The pipeline integrates a Hilbert space–filling curve for spatial ordering, the spatial Bergsma measure of nonlinear dependence, and generalised SVD to robustly analyse high‑dimensional STS data.

We demonstrate the method using climate and conflict datasets spanning both lattice and non‑lattice spatial designs. In climate applications, the framework reveals how regional variability emerges from interactions between physical geography and anthropogenic influences, and uncovers hidden spatial behaviour in key variables. For conflict data, we show its effectiveness for univariate event series and outline extensions to multivariate observations collected at each spatial location. Overall, the approach provides a transferable, model‑agnostic foundation for isolating spatial dependence in complex spatio‑temporal systems.

This talk analyses the percentage of votes for the Republican nominee in the US Presidential Elections across different states from 1976 to 2020. We are interested in building a model for how the percentage vote share for a given state for a given election depends on (i) the percentages of that state from previous elections and (ii) the percentages from neighbouring states from previous elections. The neighbours of a state are those states that share a border. We also consider influences from neighbours of neighbours and so on. We model these influences using a recently developed generalised network autoregressive (GNAR) model, which can obtain parameter estimates for the influence due to red, blue or swing states. Our new Dynamic GNAR (DyGNAR) model also allows for the modelling of a time and seasonal trend, fixed network effects and interactions between communities driven by a GNAR stochastic model. As part of the modelling we will show both Corbit and R-Corbit plots that quantify the internal network autocorrelations within the series and how these can be broken out into different types of state. We will discuss several interesting conclusions that can be drawn from the analysis, such as how swing states appear to have weaker influence on red or blue states compared to the influence of red or blue states on swing states or that blue states vote as strongly for the Democrat nominee as red states voted for the Republican nominee eight years prior (whereas red states react half as strongly).

In the literature about protest mobilization many authors have  emphasized the central role of interpersonal networks and recruitment processes in shaping political participation. This line of research consistently demonstrates that social ties are a key mechanism through which individuals become mobilized. In particular, individuals are significantly more likely to participate in protest when they are directly invited or encouraged by others, especially by those with whom they share strong social ties. Empirical studies show that recruitment through friends, family members, and close acquaintances is among the most powerful predictors of participation, often outweighing individual-level factors such as political attitudes or resources.

But at the same time, existing research often overlooks the relational structures, focusing instead on the individual characteristics of those who are recruited or on the attributes of those who recruit others. Thus, the relational structure through which recruitment occurs remains comparatively underexplored. The mobilization tie itself, that is, the interaction between recruiter and recruit, has rarely been treated as the central unit of analysis. 

Our contribution addresses this gap. We develop a relational typology of protest recruitment by identifying distinct classes of mobilization ties that vary in terms of tie strength. In doing so, we move beyond the question of whether networks matter to examine the structure of recruitment networks, as well as how these patterns relate to the social backgrounds of participants.

Empirically, we draw on a large dataset of protest surveys encompassing almost 20000 participants across 120 protest events in twelve European countries. This dataset enables us to reconstruct mobilization ties and classify them using latent class analysis, thereby identifying recurring patterns of recruitment. In a subsequent step, we apply multinomial regression models to examine how social characteristics differ across the identified mobilization classes.

Our findings have important implications for the study of political participation and collective action. By uncovering distinct types of mobilization structures, we provide a more nuanced understanding of how protest participation diffuses through social networks.

Social movements manifest multiple and overlapping temporalities from short-lived coordination around contentious events to longer protest cycles and structural shifts catalysed by eventful protests. Relational approaches to collective action maintain that a certain degree of stability in the informal networks of movement actors is required to sustain social movements beyond single initiatives, protest events, and temporary coalitions. Online movement networks, usually associated with sudden bursts of activity around key protests, volatile engagement, and supposedly ephemeral networks of interactions, have been dismissed by some scholars for not being able to build durable relations. However, online interactions are not only confined to contentious moments, but they also signal and nurture more stable relations of alliance between individuals, movement groups, and organizations. Research on online movements has tended to conflate these different network temporalities, limiting our understanding of how they differ and intersect in longer processes of movement development.

Going back to the distinction between interactions, relations, and structures, recently proposed to study stability and change in collective action fields (Diani, in press, “The temporal dimension in the network analysis of collective action fields”), this contribution engages with different temporalities of online collective action networks. The analysis draws on online interactions expressed in the Instagram posts of 962 social movement organizations in the transnational socio-environmental field between 2019 and 2025, distinguishing between transient interactions, stable relations captured through recurrent interactions, and network structures as persistent relational patterns. Applying longitudinal social network analysis on network motifs and communities, we disentangle and compare the temporalities characterising interactions, relations, and structures. Further, we explore how they intersect with key transnational events within or beyond the movement (global climate strikes, Covid19 pandemic, COPs and counter-summits). Lastly, the contribution reflects on different strategies of temporal networks segmentation and their analytical consequences for capturing the persistence and change in movement networks.

How does the position of migrant voluntary associations within organisational networks shape their political action? This paper addresses this question through a comparative analysis of migrant organisational networks in five European cities—Zurich, Budapest, Barcelona, Madrid, and Athens—focusing on the relationship between network structure and political engagement. Drawing on theories of collective action, social capital, and network embeddedness, the study examines whether node-level network properties, degree centrality and cutpoint status, influence how associations engage in political action.

Political action is operationalised as a count index of ten political activities undertaken during the two years preceding the survey, including demonstrations, petitions, lobbying, public campaigning, and institutional engagement. Because the dependent variable is a count measure and exhibits overdispersion across city samples, the analysis employs negative binomial regression models. This modelling choice is suited to capturing variation in political activity while accounting for differences in association-level characteristics such as age, membership size, type, and regional origin.

At the same time, the paper addresses a methodological challenge in network-based quantitative research: the violation of the independence assumption. Since associations are embedded within shared organisational networks, node-level observations are structurally interdependent, raising questions about conventional regression estimates. To strengthen the robustness of the findings, the negative binomial results are cross-checked using node-level permutation regression in UCINET, presented in the methodological appendix.

The findings show that associations occupying central positions in the network tend to engage in higher levels of political action, particularly in Budapest, Barcelona, and Madrid. Moreover, organisations acting as cutpoints—bridging otherwise disconnected parts of the network—display significantly greater political engagement in several city contexts. By combining substantive findings with methodological reflection, the paper contributes to current debates on migrant associational politics and offers a discussion of how network effects can be modelled in comparative organisational research.

Global development is at a critical juncture following the 2025 dismantling of USAID, which disrupted aid systems and exposed the pre-existing structural vulnerabilities around the developing world. This paper introduces the “Bridging Link Fallacy” to explain how externally driven coordination structures can produce network fragility over resilience in development governance.

Applying social capital and network theory, particularly weak ties (Granovetter, 1973) and structural holes (Burt, 1995), the paper challenges the prevailing assumption that bridging actors inherently strengthen networks. While these approaches highlight the structural benefits of bridging ties and brokerage, this paper also engages Bourdieu (1986) to emphasize that social capital is embedded in relations of power and inequality. From this perspective, bridging effects are conditioned by asymmetries in resources and influence.

Donors such as USAID function as dominant bridging nodes, connecting governments, NGOs, contractors, and civil society actors. This centrality facilitates coordination, but it can also suppress lateral ties, limit network redundancy, and concentrate dependency within a single node. When donor presence is disrupted or withdrawn, networks fragment due to the absence of locally embedded coordination. This paper demonstrates the Bridging Link Fallacy empirically through qualitative fieldwork with USAID and development policy actors in Pakistan and Washington, DC, complemented by comparative qualitative analysis across Afghanistan, Sudan, Gaza, Lebanon, Syria, and Egypt. Methodologically, the paper combines qualitative network mapping with conceptual network analysis to examine how donor centrality impacts coordination.

This study has crucial implications for the conversations surrounding localization, sustainability, and post-aid governance. The paper concludes that resilient development systems require distributed and localized bridging roles, stronger horizontal ties, and locally embedded coordination mechanisms capable of surviving donor exit. The analysis highlights the significance of power asymmetries and coordination ties in aid networks, especially in moments of systemic disruption, and how these dynamics influence network resilience in development interventions.

Russia’s war of aggression against Ukraine has triggered a rapid expansion of municipal partnerships across Europe, particularly between German and Ukrainian local governments. While traditionally framed as symbolic and depoliticised, these partnerships increasingly function as sites of coordination for humanitarian aid, logistics, and recovery. This raises a central question: how can such evolving forms of municipal cooperation be conceptualised as political networks under conditions of crisis?

The contribution presents a relational and process-oriented research approach to examine the wartime transformation of municipal paradiplomacy. It conceptualises partnerships as translocal crisis governance networks and investigates variation in their functional expansion, with particular attention to different network configurations (direct, brokered, and trilateral ties). In particular, we investigate existing dyads and triads of town twinnings and contrast them with discourse networks of paradiplomacy in Germany and Ukraine.

Methodologically, the project advances a qualitative and reconstructive approach to social network analysis. It combines network mapping with document analysis and semi-structured interviews to identify actor constellations, brokerage structures, meaning and coordination patterns. Particular emphasis is placed on analysing informality, partial visibility, and dynamic change as constitutive features of translocal wartime networks.

Initial evidence suggests that some municipal partnerships evolve into dense, hybrid networks that facilitate rapid coordination and resource mobilisation, while others remain largely symbolic. Brokerage – often involving intermediary actors or third-country municipalities – emerges as a key mechanism shaping network connectivity and access. Furthermore, network ties appear unevenly institutionalised, with significant reliance on trust-based and informal relations.

These initial findings also indicate that wartime municipal networks can develop infrastructural qualities, stabilising expectations of support and enabling decentralised crisis governance. Conceptually, this strengthens the analytical focus on networks as emergent relational infrastructures. The study thus contributes to political network analysis by extending its scope to contexts of conflict, uncertainty, and rapid institutional transformation.

The turnout in the Romanian elections in 2024 showed that diaspora voting is of strategic importance. As migration continues to reshape the demographic landscape of the modern world, the political voice of diaspora communities has emerged as an increasingly consequential force in electoral politics. Furthermore, the presence of social networks in political campaigns unveils a more complex phenomenon of voter behaviour that cannot be fully captured by traditional surveys. Thus, studying such phenomena involves both epistemological innovation (adequate semantic data description) and the application of appropriate methods for network analysis, as well as alignment with current standards.  

In this talk, we present a series of strategies we are undertaking for research method design, data collection and analysis of voter behaviour of the Romanian diaspora residing in Catalonia. The focus is on a specific network node phenomenon: the link between religious practice, institutions and voter behaviour. Even though research starts with a traditional survey, building the data set requires some decisions: preserving privacy, ensuring data replicability (for use by other diaspora communities), and ensuring data discoverability (aligned with current AI-ready strategies). These strategies mean adopting existing technical frameworks developed by the wider open science communities as the ML Commons, or aligning with data stewardship practices such as CARE and FAIR. However, our innovation focuses on the data semantic descriptions, bridging quantitative and qualitative design methods. The results not only represent a contribution to studies in behavioural sciences and methods for political network analysis, but also to the ongoing efforts to build cross-domain data standards.

Global trade is a primary vector for emerging pests and pathogens (EPPs). Despite this, it remains unclear how well sanitary and phytosanitary (SPS) measures in trade agreements align with trade patterns and ecological risks. We assess alignment of SPS policy with crop and livestock trade flows filtered by ecological similarity of trading partners to identify systemic gaps in EPP governance. Using normalized CEPII-BACI data, we constructed global networks for livestock and crop trade flows filtered by ecological similarity and compared them to a network mapping bilateral SPS measures in preferential trade agreements. The SPS policy network is sparser and more modular than trade networks, reflecting geographic and political ties rather than trade volume or ecological risk. Crop and livestock trade networks differ structurally: livestock trade is largely intra-regional and lower volume, while crop trade is frequently inter-regional and high volume. Crucially, many countries share high ecological similarity with distant partners with whom they lack robust SPS agreements. Current SPS frameworks are misaligned with ecological reality, particularly for inter-regional crop trade, leaving countries ill-equipped to manage risks from geographically distant but ecologically similar trade partners. Policy effectiveness could be improved by decoupling livestock and crop risks to reflect their distinct trade architectures and increasing SPS harmonization for high-risk trade ties.

Typical approaches to governance network research that rely on survey data have a common challenge: how to understand the forces behind a lack of a tie or the breaking of a tie. How do you study the actors that aren’t part of the network and the connections that don’t exist? Although tools like Temporal Exponential Random Graph Models enable studies of dissolution, dissolution is often understood as a consequence of shifts in desired connections over time, such as if the network undergoes shifts in stages of development. However, a fundamental tension in organizational dynamics is that collaboration choices stem not only from institutional level predictors and network level features, but also from influential individuals within these organizations. When influential individuals enter, alter, and leave a system, this can cause cascading effects throughout the network. Although network studies in natural sciences and engineering study these cascading processes, the concept of cascading failures appears in governance literature only as a theoretical construct. There has not yet been a quantitative network analysis treating collaboration costs as load flows that interact with actor capacities to dynamically influence the system, particularly in the case of node failure. Informed by empirical data from environmental governance networks in California, we introduce an agent based model simulating the vulnerability of governance systems to disturbances of individual nodes, applying a model for cascading failures in complex networks. In this research design, the load on each organization, represented by its number of connections, is weighed against its capacity, as determined endogenously through the security of its connections and defined exogenously by its number of staff and available resources. This study represents both a methodological contribution and theoretical advancement: by simulating the resilience of governance systems to destabilizing events, we progress closer to understanding dynamic socio-ecological systems in the real world.

Organic and natural farming are expanding rapidly in India, yet the domestic market for these products remains limited, leaving many farmers struggling to identify buyers. While low domestic consumer demand contributes to this challenge, the reported low availability and affordability of organic and naturally farmed foods and beverages point to broader structural and governance issues in the market. There has been limited research on this sector, including who the key stakeholders are, how they are connected, and how these networked relationships shape market governance and policy coordination. This study addresses this gap through a mixed methods social network analysis of 471 stakeholders, including national- and state-level policymakers, certification bodies, civil society organisations, and private sector actors. It maps and analyses networks of (1) collaboration, (2) governance, (3) funding, and (4) market-related information for organic and naturally farmed foods and beverages in India. Across the four networks, we found low density and concentration of power, with some highly-connected stakeholders who nonetheless did not exhibit substantially greater structural influence or stronger links to government actors. The networks showed fragmentation, particularly based on differing certification standards, and information flows were highly limited. Stakeholders identified growing domestic demand as a key market strength, while key challenges were the small and fragmented nature of the market, the lack of differentiated market infrastructure, and difficulties in establishing trust. Overall, our study highlights the need for a more coordinated, multi-stakeholder strategy to strengthen the market for organic and naturally farmed foods and beverages in India, which will require greater regulatory coordination and policy convergence.

Despite being labeled domestic conflicts, civil wars are usually highly internationalized with multiple actors militarily supporting different warring parties. While there is a meanwhile extensive body of research on third-party intervention, the majority of those contributions still implicitly assumes the independence of support ties neglecting their inherent network structure. In contrast, this paper perceives military interventions not as isolated events, but as inherently interconnected phenomena. Outside actors form a complex network echoing their relationship towards conflict scenarios as well as other third parties within the international arena. My research attempts to disentangle this network and aims to trace trends and patterns in respect to the support network as a whole as well as regarding single actors within. By merging the three most widely used datasets on external support in civil wars – the UCDP External Support Dataset (ESD), the Non-state Actors in Armed Conflict (NSA) dataset, and the State-Nonstate Armed Group Cooperation (NAG) dataset – I structurally analyze intervention patterns between 1975 and 2010, the shared coverage of these datasets. Using social network analysis and temporal community detection algorithms I find major disruptions in the international support network after the end of the Cold War as well as immediately after the events of 9/11. The number of homogenously intervening communities doubled after the end of the Cold War, indicating a more fragmented international system, the 9/11 terrorist attacks, however, reversed this. In terms of competitiveness, the end of the Cold War marked a substantive drop in opposed interventions, however, this trend reversed shortly after with new central providers of support entering the international arena, including transnational terrorist groups. In general, the international support network became increasingly imbalanced, indicating fragile and frequently shifting patterns of amity and enmity among major and middle powers.

Dynamic social networks can be observed as sequences of dyadic interactions between actors over time. The relational event model has been the workhorse to analyse such interaction sequences in empirical social network research. When addressing possible unobserved heterogeneity in the interaction mechanisms, standard approaches, such as the stochastic block model, aim to cluster the variability at the actor level. Though useful, the implied latent structure of the adjacency matrix is restrictive which may lead to biased interpretations and insights. To address this shortcoming, we introduce a more flexible dyadic latent class relational event model (DLC-REM) that captures the unobserved heterogeneity at the dyadic level generalizing stochastic block structures. The applicability of the model is shown for relational event data of militarized interstate conflicts between countries.

Why do some civil wars produce inclusive governance structures while others generate arrangements that fuel renewed violence? The quality of rebel governance during civil war remains one of the most consequential and least understood dimensions of armed conflict, shaping civilian welfare, the legitimacy of armed actors, and the relational foundations on which any eventual political settlement must be built. Yet existing research has focused predominantly on the organizational characteristics of individual rebel groups or movements (Weinstein 2007; Mampilly 2011; Staniland 2014; Krause 2014), the ideological foundations of insurgent behavior (Huang 2016; Stewart 2018), or the strategic calculations of elites at the negotiating table (Walter 2002; Mattes and Savun 2009; Wennmann 2009), neglecting the broader relational architecture within which these actors are embedded. This study asks why some rebel coalitions produce inclusive governance structures while others generate arrangements that consolidate authority or fuel fragmentation. I argue that the answer lies in the composition of wartime alliances, specifically in how identity homogeneity and power distribution among allied groups shape the credibility of institutional commitments. Classifying coalitions along these two dimensions, I derive predicted governance trajectories and test them through two complementary empirical strategies: a large-N quantitative analysis of rebel governance outcomes across post-Cold War conflicts using the Quasi-State Institutions dataset (Albert 2022), covering 167 rebel groups across 2,387 group-years from 1989 to 2012, and a Syrian network dataset of cooperation and rivalry ties among armed groups from 2011 to 2024 combined with process tracing of three cases of HTS, the SDF, and ISIS that controlled overlapping territories yet produced radically different political orders.

In the aftermath of the January 6th U.S. Capitol Riot, Reddit's CTO argued that the platform's policies had successfully curbed toxic discourse that fueled the riot on competing platforms. In this article, we take that claim seriously as a methodological challenge: how can researchers compare behavior across platforms with different architectures and governance structures?

We contrast two methodological approaches, labeled 'top-down' and 'bottom-up,' motivated by the 1903 debate between sociologists Émile Durkheim and Gabriel Tarde, as well as more contemporary theories of collective and perceived social norms from the communications literature. Top-down (Durkheimian) approaches start at the collective level to analyze the whole network and make distinctions between subgroups. Bottom-up (Tardean) approaches start at the individual level, constructing each actor's local, dynamic social environment to analyze their ego networks.

We apply both approaches to compare norms surrounding toxic language use in discussion threads on Reddit and Twitter in the 24 hours surrounding January 6th. We analyze both descriptive norms (what people do) and injunctive norms (how behaviors are rewarded and penalized). The top-down approach shows that Reddit has higher rates of toxic language use than Twitter and that toxicity is more likely to be rewarded on Reddit. However, the bottom-up approach reveals a cycle of positive reinforcement on Twitter that is largely absent on Reddit: users rewarded for toxicity continue deploying it, producing toxicity cascades deep into threads through a kind of 'playground fight' dynamic not visible at the collective level.

The two approaches yield different, though complementary, conclusions about how anti-social norms vary across platforms. Together, they illustrate how a researcher’s choice of analytical lens can shape their substantive findings. We argue that digital traces open up new investigative directions in comparative methodology, with theoretical roots dating back to sociology's founding debates about how to best understand 'the social.'

Culture emerges from continuous contestation over competing visions of the world. Dominant groups routinely seek to impose institutional cultures on others, while reactions of local groups lie not only in total compliance or outright resistance, but often between these poles. Since such asymmetric interactions stimulate complex adaptive behavior, influence models, based on binary adoption/non-adoption of an otherwise static recipient, are insufficient. This paper introduces a model of change under directed influence attempts, incorporating also creative autonomy of local culture subjected to institutional pressure. To empirically test our model, we consider four distinct cases of flood risk knowledge management across Europe, in each of which a group of experts under government mandate sought to impose its institutional culture on a weaker and dependent local flood action group. Operationally, we model the change in local semantic networks exposed to expert semantic networks in each case pair using a semantic extension Longitudinal Exponential Random Graph Models (LERGMs). We find substantial evidence that local groups do not simply adopt the institutional culture of dominant groups but systematically react to it in the ways that preserve their cultural autonomy. That is, they not only align or non-align, but also align-while-adjusting parts of institutional culture to their own needs and misalign—likely depending on the variations in power and authority across the cases. The model proposed here has broad implications for understanding social interaction, culture, influence, and diffusion more generally.

This study examines how the digital public sphere responds to rapidly evolving, unfamiliar, and often conflicting information. In such contexts, narratives are recombined from existing elements, making the way themes are combined as important as which themes appear. Using news shared on social media during the early COVID-19 pandemic, we conceptualize atypicality as the extent to which themes are linked in ways that deviate from established co-occurrence patterns. For example, an atypical combination may link vaccine rollout to blood-clot risks or college mandates, whereas more typical combinations connect vaccines to case counts, public health guidance, or routine distribution. Using a dynamic topic network of pandemic discourse, we show that atypical ties are consequential. More atypical combinations are associated with greater future salience: atypical ties tend to more persist, they connect to topics that later become more prevalent in the network, and nodes embedded in many atypical ties tend to have higher future centrality. While these ties do not fully converge to typical patterns, they exhibit a gradual shift toward more typical configurations over time.

We deepen this observation with three questions. First, which attributes make atypical ties persist? We find atypical ties associated with blame directed at governments, policy actors, or elites persist more than others, whereas ties with high levels of conspiracy content are surprisingly comparatively short-lived. Second, when do central topics help sustain atypical ties? Anchoring an atypical tie in more central topics improves its persistence selectively: blame-related ties benefit from central anchors, whereas highly conspiratorial ties do not. Third, how do atypical ties evolve in terms of their content? Over time, atypical ties become increasingly associated with evidentiary uncertainty, while conspiracy and blame show no significant change. Taken together, the digital public sphere adapts atypical narratives widely, but not those combining conspiracies.

Traditional public opinion research suggests that average citizens are non-ideological or ideologically moderate, with only a sophisticated minority exhibiting constrained political beliefs. Challenging this view, we propose focusing on citizens’ meta-cognitive beliefs about ideological consistency among other people — a phenomenon we refer to as first-order, or ‘projected’ belief constraint. Leveraging belief network analysis on experimental data, we show that citizens overestimate the degree of ideological alignment in strangers, perceiving them, on average, as more streamlined than they are based on their actual, self-held beliefs. Our preliminary results further show that ideological overprojection cannot solely be explained by individuals’ own levels of issue constraint. The tendency to over-project ideological alignment, we argue, may have significant implications for interpersonal behavior, potentially contributing to prejudiced inter-group interactions and affective polarization. In this registered report manuscript, we present the results of an initial pilot study measuring the degree of ideological over-projection and outline an experimental design to test the downstream behavioral consequences of ideological projection. Our study aims to enhance scholarly understanding of ideological thinking and to device strategies to reduce issue-based hostility in daily interactions between members of different partisan-ideological groups.

For Congressional mediation to function effectively, the process of social influence among members must facilitate behavioral coordination. Contrary to pervasive skeptical views regarding congressional rhetoric, we argue that members use their public speeches on the congressional floor to participate in the informal coordination of their legislative behavior. Analyzing the effects of SIA on voting agreement among members served in the House from the 43rd to 111th Congresses, we establish that issue attention serves as an informal mechanism to promote coordination in members’ legislative voting. The results of our analysis show that SIA exerts a robust positive effect on legislative voting agreement across the period analyzed (4.1 percentage points per 1 standard deviation increase in SIA, translating to coordination for approximately 47.6 additional votes in a modern Congress). Crucially, the effect size is more than twice as large between members of different parties, while it is marginal and unreliable among members of the same party. Subsequent analysis shows that the bipartisan coordination effect of SIA is particularly pronounced in the presence of party pressure, suggesting that SIA functions to counterbalance party pressure. These coordination effects are estimated using a doubly robust approach and dynamic identification, and inferred while modeling the error structure of dyadic data. This study reveals that floor speeches are not merely instruments for individual or partisan branding but serve as effective informal channels for bipartisan coordination, highlighting the deliberative potential of Congress.

Despite considerable scholarly attention on the institutional and normative aspects of development cooperation, its longitudinal dynamics unfolding at the global level have rarely been investigated. Focusing on aid, we examine the evolving global structure of development cooperation induced by aid flows in its entirety. Representing annual aid flows between donors and recipients from 1970 to 2013 as a series of networks, we apply hierarchical stochastic block models to extensive aid-flow data that cover not only the aid behavior of the major OECD donors but also that of other emerging donors, including China. Despite a considerable degree of external expansion and internal diversification of aid relations over the years, the analysis has uncovered a temporally persistent structure of aid networks. The latter comprises, on the one hand, a limited number of major donors with far-reaching resources and, on the other hand, a large number of mostly poor but globally well-connected recipients. The results cast doubt on the efficacy of recurrent efforts for “aid reform” in substantially changing the global aid flow pattern.

Does the seat neighbor in parliament affect MP's productivity? This paper examines whether proximity to certain colleagues in parliament shapes MPs’ legislative productivity. We focus in particular on the gendered effects of seating arrangements. Existing research shows that female MPs tend to support and reinforce one another, but can they also indirectly enhance the productivity of their male colleagues?

To address this question, we compile original data on seating arrangements and seat changes in the Swiss parliament and construct a temporal network of seating proximity spanning four legislative periods. We then analyze how the seating network relates to multiple dimensions of legislative productivity, including bill sponsorship, network collaboration patterns, vote attendance and even speaking patterns.

We complement this analysis with a causal design based on a staggered difference-in-differences approach. Specifically, we compare MPs whose seating arrangements changed (for example, male MPs who moved from sitting next to a male colleague to sitting next to a female colleague) with matched MPs whose seating environment remained unchanged. This allows us to estimate how productivity evolves following changes in seat neighbors.

Our findings suggest that male MPs who sit next to female MPs become more active collaborators: they develop stronger collaborative ties with other MPs, participate more frequently in legislative debates, and are even more likely to deviate from party lines.

Social networks have been found to impact outcomes ranging from health to the labour market. This paper seeks to build on these findings by analysing how the social networks of British MPs impact their career outcomes. We conduct our analysis by creating a multiplex network using multiple data points, including university alumni networks, election-year cohorts, and co-sponsoring Early Day Motions. Using a multiplex network allows us to account for the various ways in which MPs may form social networks. We use the multiplex network to calculate the centrality of each MP both within their party and within the entire House of Commons for the 2015 Parliament, and estimate the effect of centrality on the vote share each MP received during the 2017 election.

This paper examines how voter-level polarization shapes elite legislative cooperation in a fragmented parliamentary democracy, using the Israeli Knesset from 1988 to 2022 as a case study. Combining a complete record of private-member bill co-sponsorships with survey data from the Israeli National Election Studies, we construct monthly networks of dyadic co-sponsorship and link them to party-level measures of ideological distance and affective polarization between electorates. We develop the concept of “protective representation” to theorize how, under high affective polarization and institutional stalemate, legislators may come to represent voters less by producing shared policy and more by visibly refusing cross-camp cooperation. Empirically, we estimate Knesset-specific bootstrapped temporal exponential random graph models (BTERGMs) across multiple tie-intensity thresholds, incorporating structural network terms and institutional controls. The results show that ideological polarization robustly depresses cross-party cooperation; affective polarization is more contingent, coexisting with dense cross-party collaboration in earlier periods but becoming more tightly associated with non-cooperation during the post-2019 cycle of electoral deadlock. These patterns suggest that protective representation is not a constant equilibrium but a conditional regime that emerges when affective polarization intersects with stalemated institutions and identity-defining cleavages, transforming obstruction itself into a credible and electorally rewarded mode of representation.

There is widespread concern about the present age of misinformation. Using computer simulations of communities of rational agents, the current paper explores the effects of social network structures, as well as information-processing strategies, on the spread and uptake of true belief. It finds that poor informational environments lead to delays in discovering the truth - especially when agents employ sceptical information-processing strategies, and/or when social networks have realistic structural features - and that disinformation in particular can result in opinion polarisation. The paper also extends the scope of the antecedently known Zollman effect, showing that in the presence of misinformation, networks of larger size show a trade-off between accuracy and efficiency, with better connected networks faster but less reliable in arriving at the truth.

This paper studies how communication networks shape the ability of individuals to coordinate collective actions such as strikes, political revolts, or military coups. We develop a theoretical model in which heterogeneous agents decide whether to participate in a risky collective action whose success depends on aggregate participation. Communication occurs over an exogenously given network that determines which agents can securely exchange information.

A key contribution of the paper is to characterize equilibrium communication and participation behavior in such environments. Despite allowing for multiple rounds of communication and rich network structures, we show that equilibrium information transmission is severely limited. Agents can credibly convey only coarse information—specifically, whether they are broadly in favor of or against the action—rather than the intensity of their preferences. This result highlights a fundamental constraint on information aggregation in strategic environments with complementarities.

Building on this insight, we derive equilibrium participation thresholds that depend on agents’ beliefs about others’ actions, which in turn are shaped by the network structure. We provide explicit characterizations for several canonical networks, including circles, lines, and star networks, and compare their implications for the ex-ante probability of successful coordination. Our analysis suggests that network topology plays a crucial role: more decentralized structures may facilitate coordination relative to highly centralized ones, due to differences in how information propagates locally.

Large Language Models (LLMs) are increasingly used as both conversation partners and autonomous agents online. This has implications for political discourse, depending on how expressed LLM opinions shift when their expressed opinions are challenged or reinforced. This research aims to uncover the LLM opinion dynamics which emerge in a network setting, beginning with the smallest network units of dyads and triads.

We begin by constructing dyads of LLMs with fixed initial opinions on a political statement (Agree (A), Disagree (D) or Neutral (N)) and evaluate the extent to which these opinions shift after discussion rounds with other concurring or disagreeing LLMs. This can then be extended to both open and closed triads of LLMs with a range of opinion configurations e.g. A-N-D or A-D-A. After each discussion round, the LLMs select the extent to which they agree on a Likert scale, and indicate their opinion of the other LLMs.

Statements are based on a wide range of questions from the World Value Survey and experiments use a range of open-source LLMs with additional survey robustness measures.

From the results, we will evaluate the influence of both group composition and structure, allowing for differences by model or statement topic. Possible mechanisms include a tendency towards consensus [1] or polarisation [2], and sycophancy [3]. These results will further provide a foundation for extending similar experiments to more complex network structures.

[1] Jenness, A. (1932) ‘The role of discussion in changing opinion regarding a matter of fact’. The Journal of Abnormal and Social Psychology.

[2] Pattie and Johnston. (2016) ‘Talking with one voice? Conversation networks and political polarisation’. The British Journal of Politics and International Relations.

[3] Perez et al. (2023) ‘Discovering Language Model Behaviors with Model-Written Evaluations’. ACL.

We introduce ResIN, a package for the R programming language designed to estimate, visualize, and analyze Response-Item Networks (ResIN). ResIN is an increasingly popular method for modeling complex socio-political attitude systems as sparse, spatially interpretable networks. The package simplifies and unifies the underlying workflow while offering a range of convenience features, including network plotting, community detection, psychometric score extraction, bootstrap-based uncertainty estimation, and the export of ResIN objects to other statistical software environments. We provide an overview of the ResIN method, its computational implementation in R, and demonstrate the utility of key package features through an applied replication tutorial based on a recent study by Lüders et al., 2024.

Accounting for measurement error is a well-documented challenge in survey research; yet, its consequences for scholars conducting network analysis of political attitudes remain poorly understood and are often ignored altogether (at least one of the present authors included). The dominant network approach for analysing heterogeneous survey items uses mixed graphical models (MGMs), for which measurement error in any variable can induce bias in the estimated edges, inflating some associations, attenuating others, and generating spurious ties between nodes. This problem is compounded in exploratory analyses where networks encompass many variables simultaneously, increasing the risk that spurious edges will be mistaken for substantively meaningful relationships. We demonstrate, for the first time, how the MultiTrait MultiError (MTME) framework, which uses within-person experimental designs embedded in panel surveys to decompose response variance into trait, acquiescence, method, and random error components, can be integrated with MGMs to produce measurement-error-corrected network structures. Using data from the 2023 Citizens' Opinion Parliamentary Election Study, a probability-sampled Finnish online panel, we apply MTME to a battery of political value and attitudinal items, including the traditional left-right placement, as well as GAL-TAN orientations common European parliamentary democracies, and subsequently estimate corrected and uncorrected MGMs. Results from network comparison tests reveal that the uncorrected network systematically underestimates edge weights primarily due to random error, which is consistent with classical attenuation, with the corrected network explaining substantially more variance in node structure. Notably, differenced networks reveal predominantly negative corrections, suggesting that a subset of ties reflect acquiescence or method artifacts rather than genuine inter-attitude associations. Heterogeneity analyses further show that measurement error is not uniform: respondents with lower political sophistication exhibit higher random error, magnifying network misspecification in precisely those subgroups most consequential for research on democratic representation. These findings have broad implications for the growing use of network analysis to study political belief systems and value structures, and we offer practical guidance for researchers seeking to incorporate measurement error correction into network analytic pipelines.

Understanding the relationship between issue-specific attitudes and ideological identity is important for understanding how issue polarization manifests in a comparative context. By representing political survey data as a bipartite network, we show how network-based Bayesian inference methods can be used to uncover the high-dimensional group structure of both survey respondents and survey questions. We define an attitude group as a set of individuals sharing coherent socio-political attitudes such that, from an information-theoretic standpoint, it is more efficient to describe them collectively – with responses drawn from shared probability distributions – than to store each individual's responses separately. Groups are defined analogously for survey questions: questions in the same group elicit statistically similar response patterns, identifying clusters of aligned or redundant survey items.

We operationalise these ideas using a layered, hierarchical stochastic block model as a generative mechanism for observed survey data, applying Bayesian inference to identify the most likely group structure based on the available evidence. The response distributions defining each attitude group may be loosely interpreted as different ideological worldviews, constructed organically from respondents' attitude positions rather than imposed through conventional markers such as "left-right" or "protectionist-globalist."

We demonstrate the method using World Values Survey data from 54 countries, analysing responses to attitudinal questions spanning diverse socio-political issues. The method partitions both questions and respondents into hierarchically organised groups, revealing that social issues carry substantially greater discriminatory power than economic issues in determining primary respondent divisions. To validate these structures, we test the relationship between inferred attitude groups and self-reported left-right ideological positions by computing the statistical distance between ideologically-stratified respondent group distributions. Results show that attitude group membership aligns strongly with self-reported ideology in high-income countries, particularly Anglophone countries (Spearman’s correlation >0.8), but diverges across the Global South, highlighting important cross-national variation in the relationship between attitudes and ideology.

This paper presents results from a survey experiment conducted in Bogotá, Colombia, with a sample of 575 adolescents (M_age = 14.8; SD_age = 0.9; 51.0% identified as female, 97% Colombian) recruited from eight charter and traditional public schools serving families from low socioeconomic backgrounds across five localities of the city. Participants were randomly assigned to one of two name generators that differed in the subject of discussion. In Form A, participants were asked to name individuals with whom they discuss social issues they care about, whereas in Form B they were asked to name individuals with whom they discuss politics.

The results suggest that issue-based and politics-based name generators recover potentially overlapping yet distinguishable segments of adolescents’ discussion networks. Participants assigned to the issue-based form report, on average, 1.27 additional discussants (p < .01) relative to those assigned to the politics-based form, corresponding to an increase of 64.3% over the mean network size observed under the politics-based form (M = 1.97). Mothers, cousins, and friends are more likely to be named at least once under the issue-based form than under the politics-based form, with estimated differences in probabilities of .25, .09, and .19, respectively (all p < .05). The proportion of female discussants is also higher in networks elicited through the issue-based form (delta = .12, p < .01). By contrast, I fail to reject the null hypothesis of no differences between survey forms in mean frequency of conversation, perceived discussant knowledge, similarity of opinions, or network density.

Overall, this paper provides novel evidence on the effects of name generator design in an understudied context and age group, with implications for how adolescent political discussion networks are measured and, more broadly, for research on political socialization and political discussion networks.

Having a public that is exposed to challenging information, perspectives, and norms in everyday life is widely considered essential to the practice of democracy; concerns over backsliding and discussions of polarization underscore the importance of understanding the extent to which Americans are subject to political disagreement. Using name generators embedded in surveys, some scholarship has emphasized the mostly agreeable nature of Americans’ core social networks. In this paper we build on recent work to consider the extent to which data gathered via these techniques produce accurate representations of the public’s social experiences. Drawing on multiple nationally-representative studies, we: 1) replicate standard political name generator prompts, 2) randomly assign respondents to (additional) generators that explicitly ask about disagreement, and 3) examine patterns of broader exposure to disagreement via aggregate relational data (ARD) batteries. We find that Americans are exposed to more interpersonal political disagreement than is commonly suggested --- chronic underestimates and accompanying narratives are partially a function of data collected via name generators, but also because (name generators miss, but ARD methods reveal that) people who interact with more politically agreeable contacts also interact with more politically disagreeable contacts. We close by noting that name generators and ARD batteries have distinct strengths and weaknesses, and that design decisions should balance such considerations. Including both types of measures in studies potentially offers researchers the most complete information on patterns of social interaction in mass publics.

In competitive authoritarian regimes, incumbent control over traditional mass media forces opposition actors onto social media platforms like YouTube to mobilize. This study examines how incumbents counter this migration through strategic infiltration, a friction-based censorship tactic designed to degrade political discourse and raise the informational costs of preference revelation. Drawing on a corpus of approximately 28 million user comments from Turkish political YouTube channels, this research employs zero-shot Large Language Model stance classification to map the discursive strategies of competing political blocs. The analysis reveals a distinct resource asymmetry: while opposition actors dedicate their limited capacity to in-group agenda-setting, resource-rich incumbents systematically pivot toward out-group disruption. Furthermore, longitudinal event-study analyses of pre-election windows demonstrate that this interference is temporally concentrated. Rather than maintaining a constant baseline of friction, incumbents deliberately surge their infiltration efforts during the final week before an election to maximize coordination costs when they matter most. Beyond these semantic and temporal dynamics, this study emphasizes the profound structural consequences of digital disruption through network disintegration. High-intensity infiltration acts as a solvent that fractures the opposition's digital topology, effectively driving moderate users into silence while radicalizing the remaining participants. To formally measure this schism effect and the breakdown in community cohesion, this research proposes a dynamic temporal network analysis utilizing Temporal Graph Neural Networks. By integrating high-dimensional latent space embeddings, this advanced methodological approach moves beyond static snapshots to explicitly model how the geometric boundaries of political discussion warp and network modularity collapses under systemic friction. Ultimately, this framework reveals how autocrats neutralize the opposition's capacity for collective action by strategically shattering digital solidarity.

Although deliberative practice is essential for the legitimacy of institutions, increasing studies portray the U.S. Congress as an institution paralyzed by polarization, where members’ speeches serve as strategic branding rather than deliberative exchange. This article challenges that view by advancing the concept of Shared Issue Attention (SIA)—the alignment of public signals weighted by topical relevance. Analyzing policy speeches in the U.S. House of Representatives between 1873 and 2010 (43rd–111th Congresses), we construct the congressional social network to uncover the structural patterns and mechanisms of legislative discourse formation. Our analysis yields two critical insights. First, the structure of the congressional SIA network exhibits a small-world architecture—characterized by both high clustering and short distances— suggesting sustained efficiency in information transmission. Despite a modern rise in party assortativity, the network remains integrated through dominant inter-party triads, indicating that subcommunities are permeable rather than segregated by party boundaries. Second, the analysis of the causal speech diffusion reveals that social influence among members is robust, dynamic, and reciprocal. Speech diffusion between members occurs almost without exception in the analyzed period. Contrary to the narrative of partisan echo chambers, inter-party and intra-party diffusion occur at nearly identical magnitudes in all Congresses. Furthermore, we observe significant minority-party influence, challenging theories of unilateral majority agenda control. These findings indicate that Congress operates on an incidental model of deliberation, retaining a resilient capacity for deliberative policy-making that persists beneath the surface of partisan conflict. Beyond this legislative context, our causal diffusion framework provides a scalable method for analyzing power and interactive alignment across diverse multi-agent systems.

I propose to present the penultimate chapter of a 10-chapter book manuscript on political capital. The broader argument of the book is that political capital accrued by members of the US Congress can be understood as emergent properties of legislative networks. I measure political capital across five observable networks (campaign finance, caucus memberships, roll call agreement, committee assignments, and cosponsorship) as four distinct types of political capital: connection capital (degree), status capital (eigenvector centrality), cooperation capital (triadic closure), and brokerage capital (inverse constraint or betweenness). In the penultimate chapter, I offer a meta-analysis of political capital by treating the observed networks as a multiplex. I aggregate across networks and forms of political capital to generate a composite index of political capital for each member of Congress in the 103th (1992) through 116th (2020) Congresses. In addition, I perform sensitivity analyses on the index, including a principal components analysis on the four component parts, which shows political capital neatly summarizes as two dimensions. I label these embedded capital and brokerage capital. I demonstrate that members of House who are high in embedded capital are more likely to be effective legislators while those high in brokerage capital are more likely to become chamber leaders. This chapter serves as an analytical culmination of a much longer book manuscript on the topic of understanding power among members of the US Congress as the byproduct of relationships among its members.

Since the Congressional Black Caucus' founding, Black women have been essential to the development and advocacy of the primary legislative agenda for Black advancement in Congress. However, for many of those years, Black women were few in number and lacked a great deal of power in the institution. As time progressed, Black women increased both their critical mass and power within the chamber. We ask: how has advocacy for Black interests changed as Black women become better incorporated into the institution? To answer this larger question, we examine the discourse networks captured in CBC Special Order Hour Speeches - round robin style floor speeches delivered by an organization tasked with bringing about racial policy change. We argue that, as more Black women enter the CBC, their floor speeches will become increasingly intersectional. However, we also expect much of the labor in pursuing race-gendered issues to fall on the shoulders of the women members of the Caucus.

The employment of conceptual networks to map and analyse ideas and concepts is becoming increasingly popular for studying issues central to politics. Despite their growing use, no systematic approach has been established to assess the structural properties and potential biases of such networks. This study introduces a method for systematically detecting and monitoring structural features and potential biases in conceptual networks. The method is empirically tested on a Causal Loop Diagram (CLD), an increasingly popular conceptual network technique. CLDs are manually constructed and represent factors as nodes connected by positive or negative causal relationships. These relationships form feedback loops—balancing or reinforcing—that can be interpreted substantively or used for system dynamics simulations. A CLD can be considered a finite, information-rich dataset, yet no standard method exists to assess the quality of the information it conveys. This study applies Exponential Random Graph Models (ERGMs) to a CLD developed to analyse digitalization in the Dutch construction sector, in order to test: (1) whether the structure exhibits the typical features of a CLD; (2) whether contributions from different stakeholder groups introduce systematic differences in the network; (3) whether qualitative attributes of factors influence connectivity patterns. The results show that the CLD displays the expected features of this modeling approach. Furthermore, the findings reveal that the five stakeholder groups involved in the data collection process hold markedly different understandings of the problem and construct portions of the diagram that are only loosely connected to one another. Finally, qualitative attributes of factors do not systematically drive connectivity, suggesting that thematic categories are integrated across feedback loops and that there is limited evidence of thematic clustering. The proposed method is generalizable to other conceptual networks with appropriate adaptations and is recommended as a diagnostic tool to assess network quality after construction and before substantive analysis.

Belief systems, which can be understood as networks of constraints between political attitudes, are central to our understanding of political attitude formation, polarization, and sorting. In this area of work, there is increasing interest in how individuals perceive of belief systems and their constituent attitude constraints as they exist in the minds of others and more broadly within society. Using large language models (LLMs) as a case study, we explore how our collective perceptions about belief systems are encoded in textual artifacts. As generative models that aggregate the immense amount of text human beings have written about one another, LLMs encode how we view ourselves and others. Existing studies show that LLMs are reasonably performant at guessing an individual’s political identity affiliations based on other known quantities about the individual’s political belief systems (i.e., a form of belief system constraints). As these models are designed for producing the aggregated best-guess of what is a much more multifaceted set of beliefs, such encoding runs the risk of oversimplifying or even caricaturizing human belief systems. In this study, we ask how LLMs’ encoding processes may yield different outputs from how human beings have collectively encoded this type of information, focusing on how well different LLMs’ guesses about human beliefs reflect the level of diversity among our own beliefs about one another. By comparing LLM outputs with responses from human surveys, we show that while LLMs closely resemble human beings in terms of "best guesses" (i.e., distribution means), the distribution of guesses across multiple LLM responses tend to be much less diverse than those from human respondents. Our work has implications for how this increasingly ubiquitous encoding-feedback process entrenches extreme beliefs about the level of political polarization that exists within society.

Narratives of national exceptionalism play a constitutive role in how states interpret their historical purpose and legitimize political action. Existing scholarship tends to conceptualize exceptionalism either as a belief in national uniqueness or as a rhetorical device within political discourse. This paper advances an alternative approach by treating exceptionalism as a network of political ideas – as an underlying semantic infrastructure through which legitimacy is constructed and maintained

Using a corpus of selected speeches by U.S. presidents and Israeli prime ministers from the late twentieth and early twenty-first centuries, the study employs a network-oriented discourse analysis to reconstruct the relational structure of key exceptionalist concepts. Concepts such as mission, destiny, moral responsibility, sacrifice, and historical experience are operationalized as nodes, while their co-occurrence within coded speech segments constitutes edges. This enables the identification of centrality patterns and clusters of co-occurring concepts within the network structure that stabilize and reproduce exceptionalist narratives across time and political contexts.

The analysis demonstrates that exceptionalism functions less as a singular ideological claim than as a relational architecture of meaning. While both the United States and Israel mobilize exceptionalist language to articulate moral authority and historical purpose, the structure of the underlying conceptual networks differs in systematic ways. These differences reflect distinct historical trajectories, collective memory regimes, and political traditions, resulting in divergent configurations of legitimacy.

By conceptualizing and visualizing exceptionalism as an ideational network, the paper contributes to research on political discourse, national identity, and the symbolic foundations of political legitimacy. More broadly, it demonstrates how network approaches can be extended to the analysis of political meaning structures, capturing the structured yet dynamic nature of political meaning-making.

Democratic theory has long linked citizens’ access to evidence and information with the ideal of higher-quality political discussion. Yet information-rich environments may also raise the cognitive demands of participation, potentially discouraging broad engagement. This tension raises a fundamental question for research on political networks: how do information environments shape the structure of political discussion networks?

This study examines whether variation in the informational density of journalism reorganizes the structure of online political discussion networks. We analyze reply networks constructed from nearly one million comments posted on 4,873 stories published by The New York Times (2014–2022). Approximately half of these stories employ data journalism features—including statistical information, visualizations, and external information sources—creating a natural comparison between high- and low-density information environments.

For each story we construct directed interaction networks in which nodes represent commenters and edges represent reply relationships. We then examine how information density relates to key structural properties of discussion networks, including participation scale, interaction intensity, clustering, participation inequality, and community structure.

Results reveal a consistent structural pattern. High-density information environments attract fewer participants, suggesting that complex informational contexts act as a participation filter. Yet the discussions that do emerge exhibit more cohesive and interactive network structures, characterized by stronger clustering and higher interaction intensity among participants. In effect, higher informational density appears to trade participation breadth for structural engagement. 

By examining nearly 5,000 discussion networks across different information environments, this study provides large-scale evidence of how media environments shape discussion networks. These findings suggest that information environments do more than inform citizens—they shape the structure of political discussion networks, influencing both who participates and how they interact.

A central concern for democratic deliberation is whether citizens share a common informational foundation or sort into separate information environments. The echo chamber thesis argues that online platforms intensify this sorting by segregating publics into ideologically homogeneous groups. Yet empirical work has focused primarily on partisan segregation, overlooking other demographic dimensions, and has largely relied on domain-level measures that may underestimate segregation by obscuring content-level variation within shared sources. We analyze URL-sharing behavior among 1.6 million Twitter users matched to U.S. voter files, providing validated demographic attributes including age, gender, and party affiliation. Our dataset comprises 43 million unique URLs shared between January 2017 and December 2022, spanning politically consequential events such as the 2020 election, the COVID-19 pandemic, and the January 2021 Capitol Hill attack. Using the multigroup Theil H index, a decomposable measure of segregation, we quantify segregation along partisan, age, and gender dimensions at both the domain level and the URL level.

URL-level segregation (H ≈ 0.32–0.40) substantially exceeds domain-level segregation (H ≈ 0.10–0.17) across all dimensions and throughout the observation period, indicating that demographic groups draw on similar sources but select systematically different content within them. Furthermore, age-based segregation rivals or exceeds partisan segregation, reaching H ≈ 0.39 by late 2022. Decomposing partisan segregation by age reveals that older partisans are considerably more segregated than younger ones, suggesting age conditions the extent to which partisan identity structures information consumption. We also identify a structural break in January 2021, after which age and gender segregation increase sharply while partisan segregation plateaus, pointing to a reorganization of the information ecosystem along demographic cleavages beyond partisanship.

These results suggest that relying solely on partisan measures may obscure meaningful variation along other demographic dimensions, underscoring the need for demographically disaggregated approaches that operate at the level of content consumption rather than source exposure alone.

Previous empirical studies of marital status change have tended to treat marital transitions as a dependent variable, focusing on political factors that predict entry into or exit from marriage. However, relatively little work has documented marital transitions as an independent variable in shaping political behavior. We build upon previous political science research documenting the effect of marital status change—marriage, divorce, widowhood–on various political attitudes and behaviors. We identify two pathways for marital transition to influence political behavior: shifts in psychological/emotional states and shifts in social network composition and interactions. Using the ANES 2016-2020-2024 panel dataset to test the effect of marital status change on a range of political outcomes. We employ OLS and logistic regression models with lagged dependent variables, finding that moving from unmarried to married is associated with more liberal attitudes and a greater belief that women experience discrimination in the United States. Respondents who moved from married to divorced are associated with reduced ideological extremity. We find no statistically significant effects (p < .05) for respondents moving from married to widowed; however, suggestive evidence emerges at the p < .10 level for political discussion networks, religious importance, and political affect. We propose two mechanisms through which marital status change exerts its effects: psychological processes and alterations in social networks. Future iterations will include additional model specifications and alternative data sources within the US context. To improve generalizability, we draw on both the British Household Panel Study and the UK Household Longitudinal Study to provide greater leverage and to document similarities and differences in a comparative context.

Background:

The 2021-2022 Chilean Constitutional Convention offers a natural experiment for studying network formation and legislative effectiveness. Delegates were largely unfamiliar, included both career politicians and civically recruited citizens, and required co-sponsorships for proposals. This "tabula rasa" legislature—where networks were not predetermined by party structure—enables us to trace how alliances form and translate into political success.

Research Questions:

We examine three interrelated network dynamics: (1) What factors drive co-sponsorship networks in constituent assemblies? (2) Does network exposure causally influence ideological dynamics? (3) How does network structure predict political success, operationalized as lexical retention in the final constitutional text?

Methods:

Using a novel combination of Valued ERGM, panel regression with fixed effects, and Spatial Durbin Models, we analyze 159 convention delegates across 5 commissions over 91 temporal periods.

Findings:

- Network Gatekeeping (1, Valued ERGM): While ideological homophily is the dominant predictor of co-sponsorship (expected baseline), we uncover a novel pattern: delegates with prior institutional experience and lawyers co-sponsor less with peers sharing these attributes (negative homophily). Rather than clustering densely, these "privileged" actors disperse across coalitions—suggesting strategic gatekeeping behavior where they mentor novices and maintain leadership through representational diversity.

- Selection ≠ Influence (2, Panel Regression): Despite strong correlation between network exposure and ideological positions in pooled OLS (β=+0.033, p<0.001), fixed-effects models show null causal effects (β=+0.0004, p=0.91). Falsification tests fail—future networks predict past ideology—confirming endogenous selection: delegates choose co-sponsors aligned with preexisting positions, rather than converging toward their co-authors.

- Collective Legislative Success (3, Spatial Durbin): Lexical retention is not an individual achievement but a network phenomenon. The SDM (AIC=-383.35) dominates OLS/SAR/SEM models. Spatial autocorrelation (ρ=0.997) indicates that delegate success depends overwhelmingly on co-author characteristics and collective success—success "spills over" through co-authorship ties. Ideological consistency and intra-coalitional ties predict retention; cross-coalitional bridges reduce it.

Conclusions:

Networks simultaneously structure both social selection and collective action in constitutional design. Understanding power dynamics in deliberative assemblies and network effects on outcomes has broader relevance for comparative legislative studies. We are expanding the dataset to include all seven commissions, testing whether gatekeeping dynamics and spatial dependence patterns generalize across different thematic domains.

Major national security crises simultaneously expand state power and compress civil liberties, yet how organized civil society responds through judicial channels remains empirically underexplored. This study examines whether and how the 9/11 attacks restructured interest group coalition behavior in Supreme Court amicus curiae networks, asking whether civil liberties advocacy became more ideologically polarized or generated cross-ideological “strange bedfellows” in response to aggrandized executive power. I leverage the ACNet dataset, which records amicus co-signers in Supreme Court cases and includes ideology scores for organizations, combined with case-level data from the Supreme Court Database. Treating 9/11 as a temporal discontinuity, I construct four networks (pre- and post-9/11, for both civil liberties cases and control cases in economic activity) and compare network-level properties including density, modularity, and cross-ideological coalition frequency. I suggest two competing hypothesis; 1) on a polarization hypothesis, heightened national security salience would sort organizations into opposing ideological camps, hardening division within judicial advocacy networks, and 2) otherwise, on a strange bedfellows hypothesis, resource-poor civil liberties groups responded strategically by forming broader, cross-ideological coalitions to signal credibility to justices facing an adverse political environment. I utilize TERGMs to model coalition dynamics, with placebo tests on non-related case groups. This paper contributes to debates on how civil society defends constitutional constraints under pressure.

How does legislative coordination change when executives lose part of their capacity to manage coalition support? This paper addresses that question through a bipartite network analysis of the Brazilian Chamber of Deputies, using Brazil as an informative case of coalition presidentialism under conditions of high party fragmentation. Coordination is defined here as the organized alignment of legislators' votes across recurring issue contexts, measured through the cohesiveness and stability of community structure within issue-specific networks. Rather than reducing legislative behavior to a single ideological dimension, the paper models the chamber as a bipartite structure linking deputies to specific roll-call events, preserving issue-level variation that aggregate party-discipline scores and single-dimensional scaling methods suppress.

The empirical design compares the 54th Legislature (2011–2015) and the 55th Legislature (2015–2019), bracketing Constitutional Amendment 86/2015, which made individual parliamentary budget amendments subject to mandatory execution and reduced executive discretion over their release. Roll-call events are classified into two theoretically motivated domains: economic-fiscal votes, where executive distributive leverage operates most directly, and rights/moral-regulation votes, where ideological and cross-party incentives predominate. Issue-specific weighted networks, with edge weights defined by co-voting frequency within each domain, are derived through bipartite projection to enable community detection in unimodal graphs. The central expectation is that reduced executive control is associated with more differentiated, policy-contingent legislative alignments, particularly in the economic-fiscal domain. The paper contributes to the literature on coalition presidentialism and legislative networks by showing dimensions of legislative coordination that single-dimensional scaling methods structurally suppress.

Data about political acquaintance networks are frequently embedded in publicly available biographical text. However, using these texts to generate a dataset is labor intensive and often requires specialized language skills. At the same time, earlier computational tools such as BERT require fine tuning for each task, which is costly for network datasets that consist of multiple linked variables. In this paper, I demonstrate that open-weight decoder based language models can be deployed in a modular text extraction pipeline to generate network datasets accurately and cheaply. The approach requires no fine tuning, meaning that additional variables can be extracted from the text corpus as needed by the researcher. Similarly, the decoder pipeline can be used to collect network data in diverse contexts with minimal adaptation. I demonstrate the feasibility of this approach by generating two large network datasets from Wikipedia biographical text: a Russian elite network spanning 1990-2020 (13,000 actors), and a Nigerian elite network spanning 1980-2000 (1050 actors). Extraction accuracy is measured on a random sample of the resulting data.

Extracting relational and event data from historical printed sources at scale has remained a central but unsolved measurement challenge in political science. This paper introduces an open-source Python package that lowers this barrier by articulating three standard tools---layout-preserving optical character recognition (HOCR), retrieval-augmented generation (RAG), and schema-constrained large language model outputs---into a single, easy-to-use pipeline that converts unstructured printed sources into network-ready data. The package is designed around transparency, reproducibility, and debugability: it links each extracted record to its source passage, saves each pipeline step's inputs and outputs for debugging and reuse, supports both cloud-based and on-device models, and facilitates the assignment of source passages to human labelers and their systematic retrieval for inter- and intra-coder consistency assessment and LLM benchmarking. Evaluated against a hand-labeled ground truth corpus of 52 volumes, 303 chunks, 23,829 ties, and 5,386 events (across multiple labelers and using a continuous similarity threshold) the pipeline exhibits a clear asymmetry across extraction tasks. For relation extraction, precision ($\approx$1.0), recall ($\approx$0.46), and F1 ($\approx$0.62) all stabilize across thresholds up to 0.75, collapsing only above that point; the limiting factor is recall, reflecting that ChatGPT extracts a clean but incomplete subset of the roughly 24,000 human-labeled relations. For event extraction, the pattern reverses: recall ($\approx$1.0) and F1 ($\approx$0.64) stabilize across thresholds up to 0.4, while precision remains low throughout ($\approx$0.48) up to 0.5, all three collapsing above those points; the limiting factor is precision, reflecting that ChatGPT extracts more than twice as many events as human labelers (5,914 vs.\ 2,793) with substantial noise: a hallucination red flag. The lower scores for events (a quintuple) as compared to relations (a triplet) reflect their greater complexity. Applied to Colombian genealogical and biographical sources spanning five centuries, the pipeline reconstructs a large-scale elite kinship network in which elected candidates appear more betweenness-central and have higher ego-network density than non-elected ones, consistent with network social capital arguments. The package offers a transferable measurement infrastructure for any research agenda requiring relational or event data from printed sources.

This study examines the extent to which far-right parties influence the agenda and increase their political power. We focus on the religiously oriented New Welfare Party (YRP), which maintains ideological affinity with the governing Justice and Development Party, and the nationalist (anti-immigrant) Victory Party (ZP), a splinter from the junior coalition partner, the Nationalist Action Party. In addition to representing two sides of the far-right, these parties are crucial as the nationalist ZP, which remained in opposition and did not form an electoral alliance, and the religiously oriented YRP, which ultimately joined the incumbents' People’s Alliance despite its opposition origins.

The study employs a multi-stage temporal design to investigate the diffusion of ideas from far-right political actors to parliamentary members. In the first phase, we create a novel corpus by scraping text data from the official YouTube channels of political parties, their manifestos, and parliamentary minutes, both before and after the 2023 elections. By constructing a shared latent space through BERT-derived embeddings of the corpora, this framework utilizes monthly temporal aggregates to isolate genuine rhetorical diffusion from baseline ideological congruence while mitigating data irregularity.

At the macro level, we track semantic convergence between outsider parties’ digital output and mainstream parliamentary factions. At the micro level, we employ TERGM to analyze the adoption of ideas by Members of Parliament (MPs). This relational framework treats MPs and YouTube channels as nodes in a bipartite semantic network, where edges are defined by cosine similarity thresholds within the BERT-derived vector space.

The model incorporates individual-level and geospatial covariates to analyze diffusion, including the MP’s electoral margin, the district-level performance of the outsider party, formal alliances, and shared biographical/political career attributes of MPs. Overall, this research reveals the topology of influence by far-right actors in contemporary political communication during the 2023 general elections in Turkey.

Toxic-language Based Echo Chambers on YouTube: Analyzing Illicit Behavior with Inferential Network Analysis

Many social media users express distrust toward mainstream news media, perceiving journalistic output as biased or unrepresentative. This distrust creates opportunities for social media influencers to position themselves as more authentic sources of political information and to offer alternative perspectives on political issues. As a result, social media spaces often function as informal and affectively charged environments in which users’ express emotions, seek validation, and receive emotionally similar responses from like-minded others. Despite growing interest in emotions in politics, the mechanisms through which political actors create and cultivate shared affective identities within networked environments remain insufficiently understood.

Addressing this gap, this study introduces Affective Sorting, a novel concept and methodological framework, for measuring emotional homogeneity and clustering online. Using YouTube as a case study, I identify leading US news outlets and political influencers to analyse comments from the most-commented videos uploaded in 2025. Emotional scores are generated using distributional word embeddings, emotional homogeneity is assessed through a networked co-commenter analysis, and emotional contagion is examined using time-lagged tests to measure emotions over time. Our results confirm the presence of affective sorting as users display higher emotional similarity to their interaction neighbours than expected under random mixing.

Importantly, we also find that right-leaning influencer spaces reduce negative emotional clustering contrary to popular belief. By conceptualizing affective sorting as a framework for measuring emotional alignment, this study advances debates on affective polarization, demonstrating how emotional clustering underpins political communication on social media.

AI is rapidly becoming a key player in almost all aspects of political activity in society. In political campaigns, voters are targeted through social media, and opinions are influenced by messages designed to go viral, lending them a semblance of authenticity. These developments raise critical concerns related to transparency, fairness, and the integrity of democratic processes. In societies, where divisions run deep, the stakes of AI deployment are particularly high. The relationship between AI and democratic governance is complex; as governments increasingly utilize AI for decision-making, the need for independent and transparent oversight mechanisms becomes essential. This paper investigates the relationship between AI enabled social media used by political parties and political polarization through a cross-country panel regression analysis. The study uses AI readiness (AI Readiness is an index developed to measure a country's preparedness to adopt and implement artificial intelligence across sectors and evaluates factors like digital infrastructure, education, governance, and innovation capacity, governance quality) as a proxy for use of AI. The analysis aims to assess whether increased institutional readiness for AI along with social media contributes to democratic fragmentation. The paper also examines the national AI strategies of selected countries to assess how they address the polarising effects of AI deployment, using Large Language Model for systematic document analysis. A manual review of these strategies is subsequently conducted to cross-validate the findings and ensure intercoder reliability. It further explores pathways for effective governance that prioritize a human-centric approach, ultimately aiming to ensure that the integration of AI into democratic systems enhances, rather than undermines, democratic integrity.

In the trade literature it is common to distinguish between trade issues and non-trade issues, trade goals and non-trade goals, trade concerns and non-trade concerns or trade objectives and non-trade objectives. These issues refer nowadays to specific issue areas such as human rights, labor rights or environmental protection. Whereas one stream of the literature considers such non-trade issues as something new threatening to overburden traditional trade negotiations focusing on trade liberalization, a second stream of the literature argues that trade agreements have always been about more than trade, including objectives such as international peace or societal welfare. This paper sets out to historicise the debate on non-trade issues in trade by analysing the diffusion of non-trade issues across trade agreements in the long-term perspective. How do non-trade issues diffuse between trade agreements? Is there a difference between specific issue areas? Which network configurations favor or hinder the diffusion of non-trade issues? Theoretically, the paper draws on the historical institutionalism and specifically the hypothesis of Karl Polanyi and Douglass North that economic issues area always embedded in wider societal institutions. Methodologically, it uses the Collection of Consolidated Treaty Series as a data source to reconstruct the evolution and structures of the trade network and the diffusion of non-trade issues between 1648 and 1919. By means of temporal network analysis techniques, the paper differentiates between channels of contagion to establish to what extent non-trade issues have been prevalent in historical trade agreements from the beginning of the modern state system since the Westphalian peace treaty. The results are of particular interests for scholars in applied network analysis, international relations and international political economy.

This ongoing project investigates the applications of semantic change detection to the analysis of American political concepts. Semantic shift detection, or lexical semantic change, studies the trajectories of word meanings as they occur in diachronic texts. The phenomenon of word meaning shifts is of interest to linguists, conceptual historians, and social scientists who use language as a window to trace the evolution of human and social thought. Computational approaches to lexical semantic change as a theoretical problem have flourished over the past decade thanks to the advancements in natural language processing, but applications to historical and cultural analysis remain underexplored. We analyze the diachronic semantic evolution of core American political concepts, such as equality and liberalism, using a large corpus spanning two centuries. We employ and compare complementary embedding methods: 1) static embeddings trained per decade as a baseline; 2) contextual embeddings from domain adaptation of transformer encoder models. For contextual embeddings, we conduct domain adaptation using BERT and MacBERTh, pretrained on contemporary and historical texts respectively. To examine semantic evolution, we construct semantic association graphs and track how community structure, degree centrality, and network cohesion change over time. Our results reveal historically coherent patterns. Equality's neighborhood undergoes discursive integration in the 1960s and 1970s as racial, gender, and legal lexicons converge. Liberalism displays an ideological drift from classical to modern liberalism. We introduce an evaluation framework grounded in theory-informed historical criteria and implemented through network measures. We apply this framework to evaluate and compare domain adaptation strategies, including models pretrained on contemporary and historical texts. We demonstrate that non-domain-adapted BERT fails to recover these historical signals, making domain adaptation a necessary step.

Public policy typically involves multiple, interrelated policy issues, which means understanding how specific issues emerge, gain dominance, and become salient is important to understanding the policy process. While existing policy diffusion literature has examined the process by which policies diffuse from one organization to another, less attention has been given to asking how and why certain policy issues become salient and are adopted by different organizations. Based on the policy documents in the U.S. on a range of policy domains, we address the following questions: 1) How does the social prominence of organizations that previously adopted a policy issue affect its subsequent diffusion? 2) How does a policy issue’s position within the broader issue network—its proximity and relationships to neighboring issues—shape its subsequent diffusion? To address these questions, we built a dataset of policy documents from the Overton database. We employ Poisson quasi-maximum likelihood regression with issue and year fixed effects using the number of policy documents addressing a given policy issue in year t+1 as our dependent variable. We find that, on the social side, policy issues are more likely to diffuse when they have been used by organizations that consistently engage with an issue across domains, while social prominence—whether within- or cross-domain—is negatively associated with diffusion. On the ideational side, issues that are embedded in semantically coherent clusters of related concepts are more likely to diffuse. Taken together, these results indicate that the future diffusion of policy issues is conditioned not only by who uses an issue but also by how the issue is situated ideationally.

Hegemony is one of the most important concepts in qualitative studies of political discourse, as it allows scholars to show how power is maintained through the production of discourses that are widely accepted because they are treated as commonsensical or natural. However, this concept has rarely been used in quantitative discourse analysis. This paper aims to fill this gap by offering a novel operationalization of hegemony drawing on discourse network analysis.

I argue that hegemonic claims meet two criteria corresponding to different aspects of being widely accepted in a divided society: (1) they are used by people who disagree on other issues, and (2) they are rarely criticized, as they are taken for granted.

To operationalize these criteria, I represent the debate as a bipartite signed network of actors and claims connected by relations of agreement. Criterion (1) is operationalized as being endorsed by members of different coalitions (defined as communities within the projected actor network). Criterion (2) is operationalized as having relatively low negative degree (disagreement) in the actor-claim network, excluding claims with low positive degree, as in their case the absence of criticism may stem from insignificance rather than hegemony.

This approach is applied to the 1994 debate on abortion law in the Polish parliament. Two claims that meet both criteria were identified: "abortion is always a tragedy for a woman" and "abortion is morally bad". These results suggest that in the 1990s, anti-abortion discourse was hegemonic in Poland, and even those who tried to liberalize abortion law adhered to it in order to be treated as legitimate participants of public debate.

This research shows that discourse network analysis can detect structures of discursive power, such as hegemony, by studying discursive relations shaped by these structures, especially where content-focused methods struggle to do so.

Contemporary AI governance has been shaped by a small cluster of influential actors whose normative frameworks have diffused unevenly across the emergent international regulatory landscape. This article analyses the network of corporate and intergovernmental documents produced between 2022 and 2025, including governance frameworks from leading AI developers, the Bletchley Declaration (2023), the revised OECD AI Recommendation (2024) and the EU AI Act (2024), as a structured field of discourse exchange in which particular governance templates achieved variable institutional uptake. Drawing on comparative critical discourse analysis, we identify a coherent governance grammar, characterised by possibilistic futurity, normative hierarchy, evaluative apparatus, exception procedure, and transcendent legitimation, that originated in corporate texts and subsequently travelled into public instruments with differential fidelity. We conceptualise this as a process of asymmetric discourse diffusion: corporate actors occupied structurally central positions within the emerging AI governance network, transmitting a techno-functionalist constitutional logic that marginalised present distributional harms in favour of speculative catastrophic futures, naturalised expert authority through biosafety-style architectures and normalised exceptional intervention as routine administration. The Bletchley Declaration reproduced this grammar most faithfully, whilst the OECD and EU instruments partially reworked or displaced it, reflecting the mediating effects of institutional position and prior normative commitments. These findings suggest that the architecture of AI governance cannot be understood without mapping the relational structure of its producing actors; and that private authority, rather than public deliberation, has occupied the network's centre of gravity.

Political actors agree or disagree with other actors on policy beliefs. When aggregated into a policy subsystem, advocacy coalitions with distinct belief systems emerge from actors’ individual policy belief portfolios. Discourse network analysis measures coalitions by considering actors’ stated beliefs. But policy beliefs differ in how important they are in structuring coalitions. To understand the ideational “glue” that binds coalitions together or keeps them apart in any given subsystem, we must identify the joint subset of beliefs that is structurally most important for the coalition structure. We call this subset the backbone of a policy debate and distinguish it from its complement, the set of redundant beliefs. To identify the backbone and redundant set, we introduce a penalized spectral loss function and a custom simulated annealing algorithm to identify the backbone and redundant belief sets by combinatorial optimization. The approach is illustrated using the discourse network of German pension politics.

A central question in computational social science is whether online traces allow us to infer offline attitudes. We address this by linking a large nationwide survey with a Twitter panel dataset of 6,463 respondents. The survey captures COVID-19 vaccine uptake likelihood, asked before the vaccine became available, on a five-point scale. The Twitter panel includes authored tweets, network structure, and simulated feed exposure.

Our core question is: when does online trace data improve individual-level prediction of offline attitudes, and for whom? Unlike prior work relying on aggregate associations, we examine prediction at the individual level by asking whether a person's online environment meaningfully predicts their own survey-reported attitudes.

Demographic features alone yield ~42% exact-category accuracy on the five-point vaccine intention scale, with 62% of predictions falling within one step of ground truth. Adding authored tweets produces little overall improvement, but heterogeneity across subgroups tells a richer story. Twitter data improves predictions for male and Democrat respondents while worsening them for Republicans — suggesting that predictability depends on how consistently and expressively individuals engage online.

This work will compare three distinct data modalities: authored tweets (self-expression), network structure (social and ideological affiliation), and simulated feed exposure (informational environment). These correspond to different theoretical mechanisms and allow us to determine the types of online data that offer the most signal in predicting offline behavior for different demographic groups.

This project contributes a methodological shift toward individual-level prediction, a focus on heterogeneous predictability, and a theoretical framework linking trace data modalities to offline behavior and attitudes.

Social media user behaviour related to critical global issues, such as anthropogenic climate change, is governed not only by interpersonal exchanges, but also by targeted promotional content representing various strategic interests. Nevertheless, latent interactions among actors who publish content through collaborations enabled by business-oriented interfaces, such as Meta Business Suite, remain largely unexplored in current literature. This study adopts a tripartite affiliations framework for modelling such latent structures formed through social media partnership advertisements related to climate action. Using Exponential Random Graph Models (ERGMs), we evaluate whether such partnerships are driven by concentrated tie formation or topological dispersion, and how these structures allow a subset of actors to gain disproportionate control over content visibility and audience engagement. Using a network constructed from a sample of sponsored Facebook pages (n=8,139) related to climate action, we tested two hypothesised models—a concentrated market and a dispersed market—against a baseline model. A binary-coded nodefactor term was used to assess the extent to which a selected subset of high-degree nodes drives tie formation. Results showed that the top nodes in the concentrated model (60 partners, 60 pages and 20 categories) primarily account for partnership affiliations (AIC: 176,076, BIC: 176,196). This model outperformed the best-fitting dispersed market (20, 20, and 100 high degree partners, pages and categories) hypothesised to account for tie formation (AIC: 183,377, BIC: 183,497). A combined ego-network of the dominant ‘core’ of the concentrated model comprised 69.07% of all nodes and 70.87% of all edges. This core exercises disproportionate control over visibility and audience engagement (Cliff’s Delta > -0.94) for audience engagement compared to the rest of the network, suggesting structural barriers to market entry. We identify such concentration as an attention oligopsony—a market where a core group consisting of a limited number of buyers (partners), pages and categories account for tie formation, dominating audience outreach.

Dyadic representation refers to the degree in which elected legislators represent the interests and opinions of their constituents (Weissberg, 1978). This paper seeks to assess how closely the online Twitter/X profiles of MPs in the UK align with their constituencies beyond party affiliation. Estimating local constituency position along two left/right ideological dimensions (economic and social) and their most important issue (MII) using multilevel regression with poststratification (MRP), this study tests the within-party responsiveness of MP Twitter/X profiles to their constituencies along three digital dimensions: (1) follower networks; (2) positive engagement networks; and (3) tweet content. Results indicate that when controlling for party affiliation, MP follower networks are positively associated with constituency left/right position along a social dimension but negatively associated along an economic one. However, no statistically significant association could be found with MP positive engagement networks or tweet content beyond party affiliation and time-period effects. The findings of this study reaffirm the staunchly partisan nature of Twitter/X usage by UK MPs and the prevailing ideological responsiveness of their follower networks beyond the dominance of the party line.