Transitivity is important for diffusion processes on networks because a propensity for transitive closure leads to network clustering. Human social networks are characterized by high levels of transitive triangles relative to random graphs of the same density. However, measuring transitivity is difficult when collecting field data using egocentric designs. Homophily is a ubiquitous social process in human social relations and is easily measured with egocentric sampling. Homophily and transitivity are typically presented as potential confounds—that some particular network behavior may be caused by either homophily or transitivity. There is surprisingly little work describing the relationship between the two.

Here we ask whether a preference for homophilous ties can directly cause transitivity in social networks in the absence of a psychological preference for structural balance. To evaluate this, we used exponential random graph models to simulate an ensemble of group-structured networks with varying probabilities that an edge will connect members of the same group, and test whether transitivity emerges endogenously from only this preference for homophily. We find that indeed homophily can be sufficient to generate substantial transitivity in our simulated graphs. However, we also find that the degree to which homophily causes transitivity depends on network density: homophily is more likely to generate transitivity in higher-density networks, but has little-to-no effect on transitivity in lower-density networks. Thus, the effect of homophily on endogenous transitivity requires sufficient in-group connectivity so that new in-group ties have a relatively high probability of closing triangles.

This work has strong implications for understanding diffusion processes on networks such as the transmission dynamics of infectious disease. We apply these insights to simulating landscape-scale retrovirus transmission using an egocentric network data from western Uganda.

The minimal dominating set (MDS) is a well-established concept in network controllability and has been successfully applied in various domains, including sensor placement, network resilience, and epidemic containment. In this study, we adapt the local-improvement MDS routine and explore its potential for enhancing seed selection for influence maximisation in multilayer networks. We employ the Linear Threshold Model (LTM), which offers an intuitive representation of influence spread or opinion dynamics by accounting for peer influence accumulation. To ensure interpretability, we utilise rank-refining seed selection methods, with the results further filtered with MDS. Our findings reveal that incorporating MDS into the seed selection process improves spread only within a specific range of situations. Notably, the improvement is observed for larger seed set budgets, lower activation thresholds, and when an ”AND” strategy is used to aggregate influence across network layers. This scenario reflects situations where an individual does not require the majority of their acquaintances to hold a target opinion, but must be influenced across all social circles.

A preprint of the paper can be found here: https://arxiv.org/abs/2502.15236

This article advances the understanding of burnout contagion by proposing that burnout can spread across formal organizational groups, not just within them. Drawing on Job Demand-Resources and network theories, we argue that burnout propagates via intergroup work interactions, through changes in job demands, resources, and emotional contagion. Using data from 1,881 employees and 8 million e-mail exchanges in a South American university, we examine how burnout originating from intergroup collaborators predicts individual burnout. Our findings reveal that intergroup burnout contagion explains unique variance in individual-level exhaustion and disengagement beyond group-level contagion, with intergroup relationships accounting for up to 19% of variance in disengagement compared to 14% for within groups. Mediation analyses highlight the distinct pathways of job demands and emotional contagion in transmitting burnout across groups, but not job resources. These results underscore the importance of examining burnout as a networked phenomenon that extends beyond formal group boundaries, offering insights into how contemporary workplaces can address burnout in interconnected and cross-functional environments.

This study introduces a computational model integrating the homophic principle both as a structural efect (network clustering by sociodemographic similarity) and as a confirmation bias (preferential influence from socially proximate peers) to explain the bimodal success rates observed in innovation adoption. Building on critiques of structural reductionism in classical diffusion models (Goldberg, 2021), we propose that preference heterogeneity—not just network topology—drives phase transitions between adoption failure and success. Using ecological networks constructed from the American Trends Panel (ATP) dataset, we simulate diffusion dynamics where agents adopt innovations based on:

1. Utility thresholds (q_i), representing individual resistance levels, and

2. Minimum social proximity (h), modulating peer influence validity.

3. Agents adopt if an innovation’s intrinsic utility (Gamma) exceeds q_i, or if social reinforcement from peers within distance h surpasses adoption thresholds (tau_i). In this sense, social closeness activates threshold contagion.

Our framework challenges assumptions in prior works (Centola & Macy, 2007; Tur et al., 2024) by replacing synthetic networks with empirically grounded configurations (McPherson & Smith, 2019). Simulations on synthetic networks like Watts-Strogatz, scale-free, and Small-World SDA, show the emergence of phase transitions at critical values of h, demonstrating how homophily amplifies confirmation bias to create non-structural diffusion barriers. For example, at h = 0.2 (moderate social proximity), adoption rates shift abruptly from 18% to 92% as Gamma crosses a critical threshold, mirroring real-world “tipping point” dynamics (Granovetter, 1978).

Methodologically, we advance network science by:

- Estimating homophily parameters (beta_m) from ATP sociodemographics (McPherson & Smith, 2019),

- Creating real-world network topology based on survey data, and imputing preferences based on the individual characteristics of the nodes, and

- Validating through 1,200+ computational experiments on CHPC clusters.

Results of network topology align with empirical observations of cultural differentiation (DellaPosta et al., 2015) and rationalize bimodal market outcomes in technology and public health (Farrell, 1998). Crucially, we show that preference-driven homophily—not preexisting structural segregation—suffices to generate adoption bottlenecks, addressing critiques of earlier models (Goldberg, 2021).

This work bridges micro-level decision-making (rational choice theory) with macro-level diffusion patterns. By incorporating empirical social distances and preference distributions, we provide a paradigm shift from purely structural explanations to ecologically valid models of innovation spread.

- Goldberg, A. (2021). Associative Diffusion and the Pitfalls of Structural Reductionism. American Sociological Review, 86(6):1205–1210.

- Centola, D. and Macy, M. (2007). Complex Contagions and the Weakness of Long Ties. American Journal of Sociology, 113(3):702–734.

- Tur, E. M., Zeppini, P., and Frenken, K. (2024). Diffusion in small worlds with homophily and social reinforcement: A theoretical model. Social Networks, 76:12–21.

- McPherson, M. and Smith, J. A. (2019). Network Effects in Blau Space: Imputing Social Context from Survey Data. Socius: Sociological Research for a Dynamic World, 5:2378023119868591.

- Granovetter, M. (1978). Threshold Models of Collective Behavior. American Journal of Sociology, 83(6):1420–1443.

- DellaPosta, D., Shi, Y., and Macy, M. (2015). Why Do Liberals Drink Lattes? American Journal of Sociology, 120(5):1473–1511.

- Farrell, W. (1998). How hits happen: forecasting predictability in a chaotic marketplace. Harper-Business, New York, 1st ed edition.

Theories of network diffusion often assume that contagions spread symmetrically through undirected ties. However, we show that complex contagions—where adoption requires reinforcement from multiple peers—exhibit emergent directedness, even in undirected networks. Using a novel causal modeling framework, we introduce Causal Tie Importance and Causal Flow Asymmetry to quantify influence propagation. Our analyses reveal that long ties often function as one-way conduits of influence, creating diffusion inequalities. Moreover, as reinforcement thresholds increase, contagion dynamics exhibit a core-periphery inversion, where diffusion is increasingly channeled from the periphery to the core, contradicting conventional assumptions about centrality and influence.

These findings challenge Granovetter’s "strength of weak ties" hypothesis and have direct implications for social capital and innovation diffusion. Applying our framework to empirical networks, we show that the influence of weak ties follows an inverse U-shaped pattern, as seen in a recent large-scale LinkedIn study on job diffusion. Rather than the weakest ties being most effective, moderately strong ties play the largest role in spreading job opportunities, while both very weak and very strong ties are less effective. Our results explain this pattern by showing that medium-strength ties optimally balance bridge formation and reinforcement.

Additionally, we analyze endogenous bridge formation in complex networks, showing that asymmetric bridges form naturally, while strategic interventions (e.g., triadic closure) are needed for integrative, bidirectional influence. These findings offer new insights into diffusion dynamics in complex networks, organizational structures, and policy systems.

A central theme in social network analysis is diffusion—the spread of diseases, information, and behaviors through social ties. Originally introduced by demographers and widely applied by epidemiologists, the basic reproduction number (𝑹0) and its derivations serve as foundational metrics for diffusion processes. Using infectious disease diffusion as an example, this paper describes a mechanism overlooked in most conventional analyses, in which a disease can endogenously “compete” with itself when multiple infectious individuals race to infect the same susceptible individual, thereby reducing the effective reproductive rate. Utilizing an empirically-calibrated network epidemiological model of wild-type COVID-19 diffusion in its early pandemic, we show that the mechanism would be expected to reduce its reproductive rate by an average of 39%. Simulation experiments further identify several different types of endogenous competition mechanisms and their relative effect sizes. We highlight the incorporation of endogenous competition mechanism as a necessary step in realistically modeling diffusion processes.

Understanding optimal strategies for promoting protective behaviors in networked populations is crucial for effective epidemic containment. While diseases spread via simple contagion—where a single contact can transmit infection—protective behaviors like mask-wearing often spread through complex contagion, requiring reinforcement from multiple contacts. This dichotomy presents a strategic trade-off: seeding behavior adoption at nodes with high standard centrality (key in disease spread) may slow the epidemic but hinder the propagation of the protective behavior, whereas seeding at nodes with high complex centrality accelerates behavior adoption but may allow the epidemic to spread unchecked.

In this work, we couple a simple Susceptible-Infected-Susceptible (SIS) epidemic model with a complex contagion model of behavior adoption. We analyze various seeding strategies across single-layer and multiplex network frameworks to assess their effectiveness under different objectives, such as minimizing fatalities or delaying peak infection. Our findings reveal that the optimal seeding strategy is sensitive to the relative time scales of disease and behavior spread and often involves an interpolation between standard and complex centrality measures. These results offer nuanced insights into designing targeted interventions that balance rapid adoption of protective behaviors with effective suppression of epidemic spread.

Social norms are often thought to emerge because they benefit those who follow them. Yet, the persistence of “unpopular norms”—behavioral regularities that endure despite being privately rejected by most—challenges this assumption. Examples include foot binding in historical China, child marriage, and entrenched patterns of bribery and corruption. How can we explain the emergence of such norms?

We argue that the structure of social networks plays a crucial role in the emergence of unpopular norms by magnifying the “majority illusion” paradox: Even when a behavior is globally rare, certain network structures can cause individuals to perceive that most of their neighbors conform to it, creating the false impression that it is more widespread than it truly is. This illusion, when coupled with network externalities, can ripple through the network, pushing the population toward widespread compliance with an unpopular norm.

Our agent-based simulations reveal that network structure not only magnifies this illusion but also facilitates its spread under conditions of incomplete information and simplified decision-making heuristics. While most network configurations allow the majority to resist the spread of an unpopular norm pushed by a few fanatics, specific structures—characterized by fat-tailed degree distributions, disassortative mixing, and strong degree-trait correlation—enable these fanatics to sway the majority into compliance.

To empirically validate this mechanism, we will conduct a large-scale incentivized social network experiment involving an asymmetric coordination game on experimentally manipulated networks. During my talk, I will present the simulation results, describe the experimental design, and—if available—discuss the experimental findings.

Despite a remarkable decline in global incidence over the past two decades, the full eradication of malaria remains a major global health challenge. An estimated 249 million malaria cases still occurred in 85 endemic countries in 2022, according to the World Health Organization, which has set a goal of reducing global incidence by 90% by 2030. Achieving this target requires increased efforts to address persistent local malaria clusters. These are often located in hard-to-reach populations in marginalized areas of the Global South, where conventional large-scale interventions may prove insufficient hence presenting

challenges for public health policies. Effectively fighting malaria in these settings requires context-specific interventions informed by localized knowledge.

Large scale interventions focus on indoor mosquito bites’ preventive strategies such as long-lasting insecticide-treated nets (LLINs) and indoor residual spraying (IRS). Residual malaria transmission take place in outdoor contexts, where exposure occurs during occupational activities, outdoor sleeping, or social gatherings. This challenge is particularly concerning given that effective solutions exist. For example, topical insecticides and repellents, available in various forms such as sprays, lotions, and creams, have been proven to provide cost-effective protection against outdoor biting, and individuals generally acknowledge the effectiveness of these preventive measures. However, an important gap remains between recognition and implementation – the adoption of these protective measures at the community level consistently falls short of the critical mass necessary to achieve meaningful epidemiological impact.

In this work, we show that social network effects can help explain the low adoption of outdoor biting preventive measures in residual malaria epicenters. While previous research in the social sciences and public health has emphasized how networks facilitate the diffusion of positive health practices, our findings reveal that networks can also amplify resistance to adopt new measures that originates from local health beliefs or perceived costs and risks of changing established routines. Through our investigation of two marginalized village populations totalling 352 individuals in Meghalaya, India — a recognized residual malaria epicenter — we identified a dual mechanism combining high-threshold complex contagion and negative influence that explains the persistently low adoption rates of an insecticidal cream designed for outdoor prevention use. By analyzing fieldwork signed network data on the villagers’ health-related peer-to-peer discussions (i.e., villagers with whom one was likely to discuss health issues, as well as villagers with whom one avoided such discussions), we found that overcoming personal resistance to using the cream required a relatively high prevalence of users among their positive ties, which we were able to empirically estimate. However the positive impact provided by such an unlikely situation is offset by observing just one negative tie using the cream: our results provide strong evidence of a negative marginal effect on one’s likelihood of using the cream yielded by the presence of at least one user to whom one reported being negatively tied, so that having a negative tie with a cream user decreased the villagers’ probability of using the cream by 5.5%, net of the other modelled factors.

This study examines the mechanisms of public compliance with stay-at-home orders introduced in response to the Covid-19 pandemic. Various hypothesis have been proposed by scholars to explain compliance patterns as a result of media coverage, policy enforcement and civic capital. However, they fail to explain why some countries such as France had high levels of compliance but low levels of trust in policymakers and the medias. Our hypothesis is that, through their day-to-day interactions, individuals have contributed significantly to the emergence and reinforcement of compliance with the to stay-at-home orders, as a result of social control and peer influence. In the context of France’s 2020 spring lockdown, this article seeks to answer the question of the extent to which social networks and interactions influenced the dynamics of adherence with public health policy.

A data-driven approach was employed to develop agent-based models, with data derived from the Google Community Mobility Reports and social surveys conducted in France. Subsequently, an exploration of compliance mechanisms was conducted through a systematic comparison of model outputs with the actual lockdown compliance curves. Our results reveal significant variation in compliance levels due to social interactions, demonstrating that individual lockdown compliance cannot be accurately understood without considering account social networks and interactions. Our model allows us to better interpret compliance with public health measures as the result of the combined effect of policy, media coverage and social interactions.

The COVID-19 infodemic has underscored the challenge of health misinformation, significantly influencing public health decisions. Despite efforts to counter it, understanding the social mechanisms driving its spread remains a key research gap. This study integrates survey data with agent-based modeling (ABM) to analyze how misinformation propagates within different social network structures and its implications for public health.

A survey conducted in Spain (January–March 2024) gathered responses from 2,200 individuals, assessing health beliefs using the COVID-19 Misinformation Scale (CMS12). Factor analysis identified four misinformation dimensions (Conspiracy, Hoaxes, Vaccines, Fertility), which were used in a k-means clustering to classify respondents into three groups: Informed (49%), Hesitant (30%), and Misinformed (21%). After characterizing these social profiles via logistic regression, an ABM was developed to simulate misinformation spread in four network structures (Regular, Random, Small-World, Scale-Free) using igraph. The model incorporated network size, learning rates, and resistance to opinion change parameters.

Results indicate that misinformation spreads most effectively in scale-free networks due to the presence of highly connected hubs, while small-world networks tend to confine misinformation within local clusters. Hesitant individuals were found to be the most susceptible to misinformation. Overall, our findings suggest that the combination of a highly interconnected network, low resistance to change, and greater learning capacity facilitates the misinformation spreading among hesitant individuals—those with less defined opinions compared to the more stable and ideologically driven positions of the informed and misinformed groups.

These insights highlight the importance of targeting hesitant individuals in health interventions to mitigate misinformation’s impact.