Title: Introducing 'ludwig' - A Package for the Statistical Examination of
Undirected Binary Networks
Author: Georg Hosoya
Affiliation: Freie Universität Berlin
Abstract:
Undirected binary networks are interesting models for examining the stochastic
associations between a set of objects that can take on two states. In
psychology, these objects may be symptoms, items or the joint occurrence of
words in data created by the pile-sort task. The package 'ludwig' is similar in
scope to the package 'IsingFit' by van Borkulo, Epskamp and Robitzsch. However,
the estimation method is inspired by an article by Strauss (1992), in which is
described, how binary network models could be fit using the pseudolikelihood
method, which reduces to a special form of multivariate logistic regression. In
this talk, the basic functionality of 'ludwig' (simulating data from networks
and estimating parameters from data) and the respective algorithms will be
explained. 'Ludwig' makes heavy use of glm(), glmnet() and qgraph(). The
motivation of writing this package was to provide an experimental platform for
researchers interested in network models and to implement a platform for further
developments that would be interesting to discuss with the colleagues.