Last year global deaths from terrorism increased by over 20% (GTI 2024), amplifying the need for careful scientific studies of this pernicious problem. Traditional scholarly approaches to terrorism frame violence as instrumental – the result of an individual-level, rational choice (Arendt 1969). We apply a social network framework to better understand the phenomenon of radical violence, which is helpful at identifying flows in relationships between groups, organizations, and nations (Chua 2024). Terrorism research, however, has been comparatively slow to adopt social network analysis, because relationships between groups are hard to observe (McMillan, Felmlee, & Braines 2020). Bipartite terrorist studies apply social network analysis to actors connected indirectly by the attributes of their actions, such as their 1) attack locations, 2) attack target, and 3) methods of violence. Several advantages include: 1) the ease of visualizing terrorist group two-mode networks, 2) identifying central actors, locations, methods, targets, and 3) detecting terrorist network subgroups. Bipartite networks also can capture largely invisible relationships between terrorist groups to understand better which groups are allies, splinters, and adversaries (Yarlagadda, Felmlee, Verma, & Gartner 2018). Such an approach uncovers the critical underlying, structural dynamics that drive effective counter terrorism policy (Corradi, Felmlee, & Gartner 2024; Verma et al. 2019). We discuss how to apply this approach to address critical terrorism study questions. We provide illustrations using data from multiple countries, demonstrating the insights gained from using bipartite analyses, and showing how this approach complements traditional terrorism studies, rational choice narratives, as well as newer artificial intelligence research.

In its report dated March 27, 2017, Transnational Crime and the Developing World, Global Financial Integrity estimated the revenues of the illicit drug market to range between $426 billion and $652 billion, making it the second most lucrative illicit market in the world. This figure, largely driven by traditional drug trafficking, is also influenced by virtual drug markets on the darknet, which have been in existence for over a decade. Yet, this area remains relatively underexplored in humanities research.

Started in 2023, my investigation relies on more than forty semi-structured interviews conducted with darknet users via specialized forums and cryptomarkets—an empirical dataset that is nearly unparalleled in social sciences research on the darknet at an international level (as confirmed by scholars such as James Martin, Rasmus Munksgaard, and Kyung-Shick Choi). By adopting an ethnographic approach, I contextualized these interviews with in situ observations of platform dynamics and user interactions.

The existence of such networks, which manage to endure over time, raises questions about how relationships between their various actors are formed and maintained. It also implies the establishment of a stable digital infrastructure that is not only technical but also social, and even ideological.

My contribution will illustrate how these criminals succeed in forging lasting connections that go beyond mere illegal transactions for financial gain. It will then demonstrate how they contribute to disrupting the conventional understanding of drug trafficking.

In the outlaw biker world, established members act as mentors to novice affiliates. Mentorship may be limited to subcultural mores, but may also include criminal tutelage. This study applies a new measure of criminal mentorship based on observed patterns of co-offending to examine the prevalence of criminal mentorship in the OMCG member population, and the extent to which that mentorship aligns with the club’s formal organization. To ascertain the importance of formal hierarchy and informal social status we estimate a series of logistic regression quadratic assignment procedure (LR-QAP) models including measures of dyadic similarity in formal rank and degree centrality (Krajewski, Dellaposta and Felmlee, 2022). We next describe the distribution, stability and diversity of OMCG members’ co-offender choice, using the co-offending stability measure (CSM) (McGloin et al, 2008) and weighted co-offending ego network diversity (Adamse, Blokland and Eichelsheim, 2024). Finally, we apply the TF*IDF dyadic difference (TIDD) (Adamse, Blokland and Eichelsheim, 2024) to assess the relative importance of each member of a co-offending dyad in the others’ co-offending ego network. Preliminary findings show ample co-offending among outlaw bikers, yet little overall co-offender stability. Co-offender stability however, shows substantial individual variation, and TIDD findings suggest criminal mentorship among OMCGs.

Human trafficking is considered a significant global crisis, with the 2024 UNODC Global Report on Trafficking in Persons estimating that 27 million individuals are exploited for labour, services, and commercial sex. Due to the hidden and decentralized nature of trafficking operations, the available data on these networks is highly limited, sparse, and fragmented. This lack of comprehensive data complicates efforts to understand trafficking operations and analyze their dynamics. To mitigate these challenges, we utilize a key tool in Social Network Analysis—link prediction methods—to infer missing or future links within a trafficking network with the aim of providing a more complete network structure for further analysis. Our approach introduces a novel learning-based link prediction algorithm that applies a graph-based deep learning model to uncover hidden connections in a real-world human trafficking dataset. We compare the model’s performance against traditional link prediction methods, such as similarity-based measures and probabilistic methods. By identifying missing connections, our approach reconstructs a more complete network representation. This enriched network model would enable law enforcement agencies to prioritize high-risk connections, track evolving trafficking patterns, and disrupt the network more effectively. Additionally, policymakers could leverage these insights to implement more targeted and data-driven intervention strategies. Our findings highlight the advantage of learning-based link prediction approaches over traditional methods in analyzing trafficking networks, providing a powerful support tool for strengthening anti-trafficking efforts.

Criminal desistance—the process by which individuals cease engaging in criminal behavior—remains a critical challenge in criminology and public policy. This study applies the Ising model, network comparison tests, and econometric regressions to analyze different pathways of recidivism and desistance using a dataset of more than 17,000 incarcerated individuals assessed through a validated risk factor instrument. The dataset includes 43 binary risk indicators covering criminal history, personality traits, education, substance abuse, social background, among other dimensions.

By modeling the statistical dependencies between risk factors using the Ising model, we identify key activation patterns that distinguish risk profiles. We further apply network comparison techniques to examine whether different subgroups exhibit structurally distinct risk networks. Finally, we incorporate logistic regression models to assess how these network structures predict recidivism outcomes. This multi-method approach allows us to move beyond traditional summative risk assessments by revealing interaction effects and latent structures within risk factors.

Preliminary findings suggest that certain risk factors act as structural hubs maintaining criminal behavior, while others serve as critical tipping points for desistance. This integrated network-based and econometric approach provides a novel framework for understanding the mechanisms underlying desistance and has direct implications for tailoring rehabilitation policies to different risk profiles.

By leveraging advanced statistical and network models, this study contributes to criminology and public policy by offering a data-driven perspective on recidivism prevention, highlighting the complex interdependencies shaping criminal trajectories.