This topic contains 2 replies, has 1 voice, and was last updated by Vaughan 1 week, 1 day ago.
April 19, 2017 at 07:59 #842
I just read this blog post on an upcoming paper, which was really handy:
In the post, Denny notes:
…these are sufficient to establish that the authors’ conclusion does not apply to state-of-the-art network modeling techniques (e.g., networks estimated using our R-package IsingFit; Van Borkulo et al., 2014).
Does this imply that converting nodes to binary and estimating with Ising network methods (presumably including IsingSampler?) are better alternatives for checking stability across sample than using original ordinal or continuous data with glasso in qgraph?
We want to run a study doing exactly this (checking stability across two samples) so very relevant.
VaughanApril 21, 2017 at 10:30 #844
So in their paper, they draw conclusions about network models per se, and look at 4 different types of models. Ignoring the other 3 models now, the Ising Model replicates remarkably well. Hence we wrote that their conclusion (that networks do not replicate) “does not apply to state-of-the-art network modeling techniques” such as the Ising Model (which is what they tested in their paper).
This is fairly unrelated to binary vs continuous. If the authors had used continuous data estimated a GGM, we would have written that their conclusions regarding state of the art techniques such as the GGM are invalid.
If you want to check similarity of networks across 2 groups (stability for me is a statistical term and implies the stability and accuracy of coefficients in one dataset, which you can do with bootnet), use the structural network invariance test from Claudia van Borkulo’s NetworkComparisonTest package, and correlate the lower or upper triangle of the adjacency matrices (e.g. via Spearman correlations).
EikoApril 21, 2017 at 12:27 #846
Really useful, thank you.