Webinar: Johan Henrik Koskinen (SU) February 14

The first webinar of the spring 2025 will be given by Johan Henrik Koskinen (SU).

When: Friday February 14, 11.00 – 11.45

Where: Zoom link: https://stockholmuniversity.zoom.us/j/65344723930

Titel: An outlook to applications of network models, the nature of network dependencies, and inference for a multivariate social influence model

Abstract: Representing social interaction between individuals as graphs has a long history in the social and behavioural sciences. I will provide some illustrations of what different types of data and research questions that network analysis has been applied to. Statistically, the main challenge in modelling network data lies in the complicated dependencies induced by network ties being cross-classified by their constituent nodes. As an example, we will consider how dependencies in the massively multivariate binary data of network graphs can be modelled using interactions in a log-linear framework, the so-called exponential (family) random graph model (ERGM). I will then consider a particular instance of such a model for inferring social influence on different attitudes to modern contraceptives in a rural African village. This multivariate social influence model can be seen as a generalisation of regular graphical models for multiway contingency tables. We will see how a standard statistical model that assume independence between subjects risks inferring associations between outcomes when in fact people that interact with each other socially tend to also have the same opinions.