Plotting the Network Structure

Donald R. Williams

Background

A key aspect of psychological networks is visualizing the estimated structure (the “graph” in Gaussian graphical model). To this end, BGGM provides plotting capabilities for essentially all of the methods. For plotting the networks, BGGM is very elementary compared to the capabilities of the package qgraph. However, one advantage of BGGM is that the plots are ggplots which offers quite a bit of flexibility (especially for those familiar with ggplot2).

This vignette also provides the work flow for estimating networks with BGGM

Estimate the Partial Correlations

Select the Network

In this case, we are setting the threshold for the Bayes factor to 3. Hence, only Bayes factors exceeding that value will be included in network

Plot the Network

This will provide the simplest plots. There is one for non-zero relations and another for null effects.

Conditional Dependence Structure

The dependence structure refers to non-zero effects

Customizing the Plot

There are two ways to proceed for customizing the plot. The first is to take the plots object and do your thing in ggplot. Alternatively, there are some arguments that can be used with plot which should make a publication quality network.

It is also possible to change the layout. In my experience, however, using plotting algorithms should be done with caution. This is because it can lead to overinterpreting the placement of nodes, when really this is mere speculation. To avoid this I prefer the circle layout.

Nonetheless, here is an example of a different layout (many are available)

Conditional Independence Structure

The conditional independence structure refers to null effects.