Statistical Network Modeling
Social networks are dynamic, interdependent structures. How can we use statistics to understand how they change and matter for the individuals they connect?
To analyze these data, we develop statistical and computational methods. These can help distinguishing the role of network mechanisms (Stadtfeld & Amati, 2020).
Methods are implemented as open-source software, such as the goldfish package in R (Uzaheta et al, working paper).
The Social Networks Lab develops, extends and applies cutting-edge statistical models for network data.
Social networks are multi-layered, dynamic systems that play a crucial role for the lives of individuals they connect. In our empirical research we often asks questions about how social networks emerge and change, and how individuals’ positions in these networks matter for very personal outcomes, such as well-being, identities and behavior. But there are some statistical challenges. Connections between individuals and individual outcomes in social networks are not independent observations, like we often assume in standard statistical models. The presence of one tie can indeed alter the probability of another tie, and individuals are commonly influenced by those they are connected with. This observational dependence is not merely a nuisance but lies at the core of what networks researchers in the social sciences are interested in: how much do shared contexts, shared friends, prior interactions, or popularity increase the chance of two individuals to connect? How much do people change in response to their peers’ identities and behaviors?
Members of the SN Lab work on a couple of projects related to network modeling. Most visible is our work on Dynamic Network Actor Models and the goldfish package in R. Other projects are concerned with improving data collection strategies, developing frameworks for micro-macro analyses, comparing existing methods and models, and modeling of network data collected at different temporal and contextual modalities. Network modeling research often starts with a concrete empirical question. From there, we develop and improve statistical approaches and computational methods to provide answers to it.