Small farms depend on diverse crop species and varieties to navigate variable socio-ecological conditions. Empirical studies suggest that the structure of seed circulation networks is vital for maintaining a collective pool of crop diversity and ensuring equitable access for farmers. Yet, little is known about which network properties best support diversity in rural communities, and existing models often overlook the complexity of farmers’ seed sourcing and information diffusion behaviors. Combining Agent-Based Modeling (ABM) with network analysis offers a promising approach to address this gap.

In our study, we investigate how individual seed-sourcing and allocation strategies shape and are shaped by network structures driving seed diffusion dynamics, ultimately influencing the collective capacity to sustain crop diversity amid perturbations that may cause local crop extinctions. We introduce an agent-based co-evolutionary network model of farms, where each agent operates within a multi-layer network representing distinct relationships (e.g., familial, commercial). The model integrates crop-variety-specific strategies through probability vectors that determine the likelihood of sourcing seeds from each layer. By incorporating multiple crops and varieties, it captures how the distribution and prevalence of varieties affect seed diversity and availability. Additionally, agents allocate seeds among different uses (planting, sharing, discarding, storing) via imitation and experimentation.

The model iteratively updates agents’ states and reconfigures the network as they seek higher-quality seeds. Our findings offer insights into how micro-level decision-making drives macro-level network evolution, informing targeted interventions to enhance seed circulation networks and bolster agricultural resilience.

Coleman’s theory shows that specific network structures, especially densely connected groups, can build social capital and encourage trust. While he acknowledges that individuals make decisions, most of his work highlights how these structural factors boost cooperation. Our research expands on Coleman' Theory by including an in-depth look at individual differences, including factors like risk tolerance, betrayal sensitivity, and personal experience, which can significantly affect how a person trusts others. We use a Bayesian approach to explain how people update their trust beliefs over time. Every person starts with some initial sense of how trustworthy others might be. Then, as they gather information (e.g., reputation, prior outcomes, or direct observations), they revise those beliefs. However, not everyone updates similarly: one person might trust easily with little evidence, while another demands a lot of proof. Individual traits—like a low tolerance for losses or an intense fear of betrayal—can lead to slower or faster belief changes. In simulations, we show that when these belief-updating styles vary, very different network patterns can emerge—even if everyone begins in the same type of tightly knit group. Dense networks spread trust-related information rapidly, but personal preferences still shape how that information is interpreted. One cluster may become open and trusting, while another remains cautious and fragmented. By adding these individual-level dynamics to Coleman’s focus on structure, our model illustrates why people in similar network settings may follow very different trust trajectories, shedding light on how trust can thrive or falter within seemingly uniform social environments.

In many agent-based models of cultural and opinion dynamics, social influence processes are often based on a single-dimensional homophily. That is, in modeling terms, multiple social characteristics of an actor (e.g., cultural tastes; opinions; gender; race) are considered equally as n-states of a vector, and the overall difference between the vectors of two connected actors in a network generates the process of being more similar (i.e., homophily; similar actors become more similar) or being more dissimilar (i.e., heterophobia; dissimilar actors become more dissimilar). Yet, this assumption overlooks that actors’ perception of social relationships can be hierarchical by their social class, and thereby the feeling of similarity may work differently for actors’ change of cultural/opinion states, by whether the actors are in the same or different classes. In this study, I propose an agent-based model that assumes a two-dimensional horizontal and vertical homophily—instead of the previous one-dimensional homophily. First, based on Blau’s (1977) theoretical idea of inequality and heterogeneity, it is assumed that actors are positioned in a network (or structure) where they feel the hierarchy/non-hierarchy and similarity/dissimilarity of relationships, by their social class and cultural/opinion states, respectively. Second, considering Bourdieu’s concept of distinction (1984), I assume that when actors are in the same social class, their similar cultural/opinion states become more similar (i.e., horizontal homophily), but when their classes are different, their similar cultural/opinion states become dissimilar (i.e., horizontal homophobia). The model shows very different results compared to existing models, implying the unexpected importance of network segregation.

Whilst empirical studies consistently find evidence of micro processes of homophily and social influence, evidence for the corresponding macro-level outcomes such as network fragmentation and polarization remain scarce. We approach this problem by arguing that people organize their relationships around different social foci. Each focus has both a relational fingerprint and a set of attributes which are more relevant within. For example, a sports team promotes the formation of leisure or sporting ties, and relevant attributes may be physical fitness or sporting ambition. We consider a network as multi-relational and the union of several sub-layers. We argue that dynamic processes of homophily and influence are focused and unfold within contextually relevant and focused layers of the overall network.

We expect that these focused micro-processes can produce fragmented and polarized communities within the respective focused layer. However, the overall network could remain sufficiently heterogeneous due to differentiation between the layers. Taking empirical data relating to several layers within a student population, we assess the extent to which each of these is fragmented along the lines of the associated relevant attributes, compared to the overall (unfocused) network. We then conduct a comparative simulation study based on empirically-calibrated ABMs to explore the macro consequences of considering homophily and influence to be either focused or unfocused. Our study sheds light on the role of social foci in masking, but also, potentially, in overcoming network fragmentation and polarization, by providing access to diverse alters in other layers.

This study explores the unmet mental health needs of queer and trans Vietnamese Americans (QTVAs) in California, using an original dataset to develop an agent-based simulation model within a dynamic egocentric network environment. Prior studies found that QTVA individuals have differing levels of social support conditional on belonging to a community-based organization, positioning some to be more susceptible to events that worsen mental health, such as daily heterosexism. Assuming a scenario of “no current support,” an agent may encounter a heterosexist event node at each timestep that may worsen mental health. If the event impacts mental health, an agent’s Kessler Psychological Distress (K6) score is incremented, with a decaying effect applied to additional events to simulate increased resiliency to heterosexism over time. The model was simulated for 30 days across 10,000 runs. K6 scores were assumed to follow a negative exponential distribution with a mean of 9.42 (scores ≥ 5 indicate “moderate mental distress”). Results indicate the initial K6 score increases by approximately 0.71 points over 30 days, with nearly one-fourth of simulations resulting in increases greater than 1 point. A small percentage (2.09%) of K6 scores cross the threshold (≥ 13) of “severe mental distress.” While this represents a worst-case scenario of no current support, the results demonstrate how a vulnerable agent’s mental health may worsen over time without the necessary community and social support resources to address daily heterosexism. Bolstering and complementing community support is needed for the QTVA population to mitigate the mental health harms of heterosexism.