Network analysis - grouping nodes?
I'm learning to perform a Network Analysis in JASP. The video tutorial is pretty clear, except for the step explaining the grouping of nodes into subscales, indicated by colors. I see the groups defined in the corresponding data file, but don't understand how to define groups in my own data set. Is this doable in SPSS datasets, for instance?
Comments
I'll pass the message on to our network expert!
E.J.
Hi Clair,
This should be doable in SPSS or excel!
To color the nodes you need a grouping variable with a specific format.
For example:
If you have 3 groups, and 9 variables (3 per group) your grouping variable would look like:
"$Variable name 1" = "$group name 1"
"$Variable name 2" = "$group name 1"
"$Variable name 3" = "$group name 1"
"$Variable name 4" = "$group name 2"
"$Variable name 5" = "$group name 2"
"$Variable name 6" = "$group name 2"
etc.
Note that "$Variable name" must be exactly equal to the name of a variable in JASP. The variable is always on the left of the = sign, and the group is always on the right of the = sign. You can call the group whatever you like as long as different groups have distinct names.
Once you've listed all variables you can leave any remaining entries as missing values.
If you have any more questions, let me know!
Don
Thank you for the fast reply! That worked.
Thanks to the JASP network team for creating such a nice module for psychometric network analysis that is of great help.
If I understood correctly, then for group naming and selecting color schemes for them, we need to create a column in our original datafile in which we sequentially define which group contains which variable by assigning (using = sign) the variable names (as string) to a given group (again as string).
However, such cluster or community representation in network plots is based on prior theory/knowledge and not even by visual inspection that which nodes go together.
I would be much obliged if some popular community/modularity detection options (such as spinglass, walktrap, clique percolation etc.) may be added to this network analysis module.
This will really be helpful to identify clusters of nodes based on some empirical/statistical grounds. Clique Percolation even handles situations similar to cross-factor loading in EFA, and thus will be even more useful.
If it is difficult and/or time taking and may not be of that priority then, even a R code example template and how to use them in JASP environment may be of help.
I am completely new to R but understand how difficult and time taking it will be to implement any R-based menu/option in any GUI-based software. However, the addition of this option will make the JASP more useful for network analysis compared to any other software.
With warm wishes
Rakesh
Hi Rakesh,
This is certainly a good idea, but at least right now, it is out of the scope of the module. However, for feature requests please create an issue on GitHub (see https://jasp-stats.org/2018/03/29/request-feature-report-bug-jasp/). It is easier to track and divide the work using GitHub.
Best,
Don
Hi Don,
I would agree that community discovery algorithms would be an incredibly useful addition to the network module, but of course I understand that this is probably a huge piece of work! I would however, point out that this request already exists on GitHub:.
Enhancement: network community discovery algorithms walktrap and spinglass (R-package igraph) · Issue #198 · jasp-stats/jasp-issues (github.com)
if I may be so bold, think there are a couple of (hopefully smaller) new features that will really improve the module, namely some comparisons of group structure (when using multiple groups) and the ability to calculate Bridge statistics:
[Feature Request]: adding functionality to compare network structure across groups · Issue #2067 · jasp-stats/jasp-issues (github.com)
[Feature Request]: Bridge statistics · Issue #2117 · jasp-stats/jasp-issues (github.com)
Somebody else also recently requested this feature, here, on the forum
How can I calculate the bridge index in the network analysis? — Forum (cogsci.nl)
Best,
Tarandeep