Speaker

Dr. Maarten Marsman
Department of Psychological Methods
University of Amsterdam

Dr. Marsman is an assistant professor at the Psychological Methods Unit of the University of Amsterdam. He received his PhD from the University of Twente in 2014, after which he became a postdoctoral researcher at the Psychological Methods Unit, where he established the Bayesian Graphical Modeling Lab (https://bayesiangraphicalmodeling.com). Dr. Marsman’s research combines psychometric theory with cutting-edge Bayesian inferential and computational methods, focusing on the Bayesian analysis of psychological data in general, and graphical network models in particular. Together with his team, he has worked extensively on developing the Bayesian approach to graphical network modeling of discrete psychometric data. Maarten’s research has been funded by the Dutch Research Council (NWO) and the European Research Council (ERC). In addition to his research, he has contributed to the development of the open-source statistical software JASP, which incorporates frequentist and Bayesian methods for common statistical analyses, and to several R software packages. He is also an enthusiastic teacher of Bayesian and computational statistical methods.

Title

Bayesian Hypothesis Tests for the Robust Analysis of Networks

Abstract

The psychological network analysis literature is booming. A recent review in my lab identified over 1400 papers published between 2010 and 2023 that applied network analysis to cross-sectional psychological data alone. However, while the literature is thriving, methodological innovation is struggling to keep up. In this talk, I will highlight two methodological challenges that stand in the way of network modeling fulfilling its promise. The first challenge is communicating the robustness of network results. Existing methodological approaches have difficulty expressing the uncertainty that underlies their conclusions. But ignoring the uncertainty that underlies our results is risky because it leads to overconfidence. The second challenge is that current approaches to network analysis are ill-equipped to evaluate fundamental hypotheses about the network, focusing instead on exploratory network analysis. My research program develops Bayesian solutions to these (and other) problems, and in this talk I will outline three Bayesian tests for hypotheses about network structure, such as conditional dependence, network clustering, and network density.