Below you can find all materials for the 3-day workshop we gave at Ghent University in January 2017.
This 3-day workshop covered a 1-day R intro, and then network estimation, network inference, and network accuracy for between-subjects cross-sectional networks and time-series networks. All materials are available for free online, including slides, R syntax, data, and references. You can find the schedule and the abstract of the workshop below.
Day 1: Introduction into R and RStudio
- Introduction: R and Rstudio
- Working with data: objects, vectors and data frames
- Doing analyses: Tables, Testing two Means, Anova and linear regression models
- Plot and report: plot, ggplot2 and RMarkdown
Day 2: cross-sectional group-level networks: estimation, inference, and stability
- Introduction to psychopathological networks
- Network estimation: how to estimate networks with Gaussian & binary variables
- Network inference: how to interpret networks
- Network stability: how accurate and stable are networks estimated
- Advanced methods: e.g. network comparison test, mixed graphical models, causality
Day 3: dynamical network analysis with intra-individual data
- Introduction to dynamical networks
- N=1: introduction to Vector Auto-Regression (VAR) models
- N>1: multilevel extension of VAR models
- Time-varying VAR models
- Exercises & FAQ
The 3-day network workshop starts with a 1-day introduction to R.
Day 2 starts with a conceptual introduction to psychopathological networks — in which we explain the main differences between the network framework and alternatives like the common cause model – and an overview of the prior literature organized into disorders (e.g., depression, PTSD, psychosis, substance abuse, etc.) and topics (e.g., centrality, comorbidity, early warning signals).
The focus of the second day is on group-level networks: what is the symptom network of a group of patients with, for instance, Major Depression? Using packages such as qgraph, bootnet, and IsingFit, we use the free statistical environment R, and a free dataset on Posttraumatic Stress Disorder, to learn the basics about (1) network estimation, (2) network inference, and (3) network stability and accuracy. Network estimation is concerned with the question which types of models are appropriate for our data, such as the Ising Model for binary data or the Gaussian Graphical Model for metric data. In this section, we also discuss how to apply regularization methods to networks in order to avoid estimating false positive associations. The second topic, network inference, covers graph theoretical measures such as centrality that allow us to interpret networks. What symptoms are most connected and relevant with other symptoms in the causal web? Finally, stability and accuracy estimation allows us to gain insight into the robustness of our networks: how likely are they going to be replicated? We conclude the day with advanced methods such as the statistical comparison of networks, the modeling of networks containing different types of variables (mixed graphical models via the R-package mgm), and some considerations about causality.
The focus of the third day is on dynamic time-series models: how do symptoms impact on each other over time? After an introduction into the general modeling framework with some substantive examples of recent papers, we learn to estimate the network model – specifically, the vector auto-regressive (VAR) model – for one participant via the package graphicalVAR. We then go through the assumptions that the VAR model requires such as stationarity and equidistance of measurement points. After that, we discuss the multilevel extension of the VAR model to the case of a group of participants, using the package mlVAR, followed by a discussion of some common problems and advanced techniques. In the afternoon of the second day, we spend about 3.5 hours with R in a practical session, and workshop participants learn to apply the knowledge of day 1 and day 2 to several datasets.