agent-based modeling ( abm )
and
social network analysis ( sna )
simulating social networks — agent-based modeling
what is a social network?
a social network is made up of a finite set ( or sets ) of actors & the relationships these actors embody. they contain patterned relationships that explain how & why people interact, with an emphasis on the relational nature of social life. these networks are dynamic, evolving systems shaped by & responding to social, economic, & political factors.
at their most basic form, social networks are structures made up of actors or entities that exist in the network called nodes. these nodes can represent different levels of organization depending on the level of specificity of interest. for example, they can represent individual people or entire organizations. they can also hold attributes such as group affiliation or location.
the connections or relationships that exist between these nodes are referred to as edges, links, or ties. these can represent different relationships depending on the relationships of interest, such as friendship, collaboration, kinship ties, or general interaction. they can also hold information related to the form of relationship, such as the frequency, intensity, or direction of the relationship. they can be directed, a one-way relationship, or undirected, a mutual relationship, and can hold a weight related to the strength or frequency of the relationship. finally, information ( both physical & non-physical ) can flow across these connections.
understanding social networks is incredibly important as their underlying structure can influence information flow, access to resources, & how behaviors and norms spread. the applications of studying social networks reach into a multitude of fields, highlighting how important empirical & objective study of these networks of interest can be. by mapping & analyzing these networks, we are able to gain critical insight into human interaction and collective outcomes.
an anthropological approach to social networks:
the study of social networks has shifted from describing distinct social groups ( like kinship groups or subcommunities ) to analyzing overlapping, relational patterns of human interaction. this approach allows researchers to examine how social actions are shaped by others & have the ability to shape connections as well. as such, an anthropological approach to social networks should integrate analysis of variables like relationships, group structures, agency, social contexts, & network dynamics.
in particular, john barnes' anthropological work ( 1954 ) highlights this shift in focus. he emphasized mapping individual relationships within broader social networks to understand community dynamics by combining quantitative network mapping with qualitative ethnographic methods to provide a richer context & meaning to the relationships he observed. in his work, he found that individuals were integrated into communities through complex, overlapping network groups rather than through discrete, distinct subgroups, influencing their social status, subgroup affiliations, influence on others, & access to resources.
how can we study social networks?
researching social networks empirically often involves observing & measuring real-world social connections & interactions. this can be done through a variety of methods, often dependent on what field you are approaching your studies from & what types of information you are seeking :
surveys & questionnaires : we can ask people about their connections, who they interact with, & the nature of their relationships directly. doing so can help map out individual networks & provide insight into a person’s perceived ties.
observational studies : instead of directly asking about connections, researchers can instead observe these interactions within groups themselves. this obsevation can provide direct evidence of network formation & individual & group level behaviors.
archival data analysis : existing data sources can be tapped that have implicitly or explicitly recorded social interactions. for example, follower / following relationships on social media platforms, historical data logs, or even co-authorship patterns pulled from academic publication databases. this method is particularly effective for studying historical trends or recreating historical networks. it is also helpful in the study of large-scale networks.
experiments : infrequently, controlled experiments can be designed to study specific aspects of social network dynamics. this can often be challenging in real-world settings.
computational and agent-based modeling : computational modeling involves utilizing computer simulations to study complex systems that might be difficult or even impossible to observe directly in the real world. a powerful form of computational modeling is agent-based modeling.
by utilizing these approaches & even combining them, researchers can gather evidence related to social network structure, function, & their combined influence on individual & group outcomes.
the specifics of agent-based modeling:
as mentioned, computational modeling involves utilizing computer simulations to study complex systems that might be difficult or even impossible to observe directly in the real world. one specific, powerful form of this is agent-based modeling ( abm ). in abm, a system ( like a social network ) is simulated by defining & creating individual agents. when studying social networks, these agents are analogous to nodes & can, therefore, represent anything from an individual person to an entire organization. each agent holds its own rules, behaviors, & ways of interacting with other agents & their environment.
a helpful way to think about abm is to understand that rather than trying to model a system's behavior from the top down, abm builds it from the bottom up. this approach allows us to observe how the collective actions & interactions of many individual agents can lead to emergent patterns & system-wide phenomena. this is particularly helpful when studying social networks, as the decisions of individual nodes affect group-wide patterns while group-wide patterns simultaneously also affect the decisions of these individual nodes. this bi-directional exchange of influence can give rise to complex social, economic, & biological dynamics that might not be obvious from observation or direct questioning alone.
due to the simulatory nature of these models & the control it provides researchers, these models are incredibly useful for exploring "what-if" scenarios & testing hypotheses. however, the amount of control researchers have over these models ( especially those that build these models from scratch ) can also present several inherent challenges & difficulties. one significant hurdle lies in finding the appropriate balance between realism & generalizability in a model. a model that is too simplistic might fail to capture the nuances of the real-world system it is trying to represent which can limit its realism & applications. conversely, a model that tries to incorporate every detail of a system can become overly complex, making it difficult to understand, analyze, & generalize its findings to broader contexts. if a model is overly complex & too specific to a certain population of interest, it may actually be more beneficial to utilize a different method of study such as surveys & questionnaires or observational study to gain insight into the population of interest. a one-to-one map of the world would no longer be helpful as it would no longer function as a map, it would simply just be the world. in this same way, an overly realistic model would not be helpful or effective. achieving a balance that allows for both meaningful insights & broad applicability is a significant challenge in building these models.
another challenge lies in the possibility of introducing bias into simulated results due to the way certain mechanisms within the model are implemented. every decision made during a model's construction should be justified & validated, from how agents behave & interact, to the specific parameters chosen, as each implementation decision can inadvertently embed assumptions or biases that influence outcomes of the model. these implementation choices, even if subtle, could lead to results that reflect a modeler's preconceived notions rather than truly emergent phenomena. ensuring transparency in design & rigorous validation against empirical data is critical to mitigate these risks.
pulling social network data — social network analysis
how can we pull data from social networks objectively?
we’ve briefly talked about different methods of studying social networks, but how can we ensure we are analyzing & understanding them objectively rather than subjectively? once we’ve pulled information related to social networks from methods like surveys & questionnaires, observational studies, archival data, experiments, or computational models, how can we pull standardized measurements from them that we can use to compare to the same network at different points in time or to different networks?
social network analysis ( sna ) offers a systematic and objective way to study these social networks by focusing on the quantifiable patterns of relationships between individuals in them, rather than relying on subjective perceptions. it allows for researchers to move beyond a purely anecdotal understanding of a network to a rigorous measurement of it. once the nodes & ties of interest from a network are mapped using the methods mentioned, researchers can employ mathematical & computational tools to extract standardized metrics. therefore, sna combines empirical, qualitative data that can provide important context to social interactions with quantitative analysis. it employs a relational focus, structural perspective, systemic approach, & quantitative methods & visualizations.
relational focus : sna emphasizes the importance of relationships between nodes rather than just attributes of individual nodes within a network. this reflects the previously discussed holistic, anthropological approach to studying social networks.
structural perspective : sna analyzes overall network structure to identify patterns that influence node & group behaviors & outcomes.
systemic approach : within sna, networks are viewed as integrated systems, not just node aggregations. interactions between nodes create emergent properties that cannot be understood by examining individual nodes alone. this property again reflects the previously discussed holistic, anthropological approach to studying social networks.
quantitative methods & visualizations : sna relies on methods like graph theory, matrix algebra, statistics, & visual representations to provide a rigorous, objective, & quantitative way to analyze networks.
within sna, specific measurements exist that can standardize the quantitative study of networks. here are some specific measures of interest:
I. node degree : the total number of links connected to a node.
II. density* : how closely connected nodes within a network are in relation to the maximum number of possible links for each node.
III. node centrality* : measures of a node’s importance, influence, or level of interaction within a network. degree centrality refers to the number of connections a node has. betweeness centrality refers to a measure of a node’s role as a bridge between other nodes. closeness centrality refers to the average distance from a node to all other nodes in a network.
IV. prestige* : the influence, recognition, or perceived value of a node within a network based on incoming links.
V. network diameter : the longest shortest path between two nodes.
VI. clusters : cohesive units of nodes within a larger network. these groups of nodes are more densely connected to each other than to other nodes in the network.
VII. clustering coefficient : a measure of the tendency for nodes to form clusters.
VIII. community detection algorithms : methods to partition networks into clusters by identifying groups of nodes that are more densely, relationally connected to each other than to other nodes in the network.
* these measures can be applied to individual nodes or can be aggregated across all nodes for a group measure.
reseachers utilizing sna must be sure to clearly define network boundaries & relationships being measured & studied that they are interested in. they must also be sure to include any & all relevant social, cultural, & historical contexts in their analysis. it is the responsibility of a researcher to also ensure the accuracy, validity, & consistency of incorporated data by utlizing rigorous & high-quality collection methods, as well as justifying their choice of analytical methods & measures.
through these types of standardized measures, sna provides objective, comparable data that enables us to identify network patterns & characteristics, track changes in a network over time, & compare the structures of different networks. ultimately, sna provides a powerful lens for data-driven insights into social behavior & collective outcomes.
what are some common objective measures of networks?