Network Psychometrics concerns itself with the network modelling of psychometric variables. We often borrow models and techniques for network modelling from distant fields, such as physics and machine learning. Even though these methods were a necessary tool to start afresh –distancing ourselves from the psychometric status quo– they were not designed for the challenges laid down by psychological data. In modelling psychological data, some pressing issues emerged in recent years. Population heterogeneity, expressing individual differences, and modelling (skewed) ordinal variables are concrete examples. Where traditional psychometric frameworks such as item response theory offer concrete modelling solutions to these challenges, network modelling is still far behind.
Recent years saw a progressive accumulation of formal connections between Markov Random fields (i.e., network models) and traditional psychometric models –such as item response theory and factor models– and more recently, traditional network models –i.e., random graph models. Since classic psychometric and network modelling offers a wealth of solutions to acute modelling challenges, it is opportune to build on these formal connections to advance psychological network modelling. However, network psychometrics has not taken full advantage of what these modelling connections have to offer.
In this talk, I dive into some concrete challenges in psychological network modelling and draft novel solutions that draw on the formal connections between Markov Random Fields and traditional modelling frameworks.