Latent network modelling for networks based on multiple (imperfect) reports: An introduction to the VIMuRe package in R and Python
Eleanor Power, Jonathan Cardoso Silva, Daniel Redhead
In this workshop, we will introduce the new package "VIMuRe" (Variational Inference for Multiply-Reported network data). This package (with versions in both R and Python) provides a principled way to deal with social network data gathered through self reports, where multiple individuals comment on what should nominally be the same relationship (e.g., through "double sampling" or "cognitive social structure" designs). Often, such reports do not agree (i.e., they show low concordance), so the question is then how to integrate these distinct perspectives. VIMuRe is a Bayesian latent network model that uses the many reports to model the "true" underlying latent network, estimating in the process the tendency for "mutuality" (for people to report balanced, reciprocal relationships across prompts) and the tendency of reporters to over- or under-report ties. We will give a practical hands-on introduction to VIMuRe using open datasets. We will help practitioners through the various decisions that need to be made in fitting the model and outline what can be done with the various outputs. More on the package can be found at:https://latentnetworks.github.io/vimure/