This.

I have first-authored a dozen cross-sectional network paper and used centrality measures in all of them. Centrality can be useful, and I have some considerations that I would like to include in a more formal comment to this interesting paper from @TRSpiller. 0/ https://twitter.com/paytonjjones/status/1317468138330607616

Lots of colleagues have commented on important aspects of centrality, including the translation from nomothetic and idiographic models, I will focus this thread on the issue of disconnectedness. 1/

While centrality is an interesting measure that allows for studying which nodes are the most interconnected in the network, we are not to the point where it having a central node in a network paper necessarily means it is a good target for clinical intervention 2/

I believe we still are in the early phase of network modelling: most papers use existing psychometric tools as input for a network structure, which I believe is a necessary step to build knowledge around what variables to include in a network and how they relate to each other 3/

Centrality is useful in such papers since it highlights which items are most connected (ie contribute the most to the network self-determination), and helps in determine which variables to include in networks (see this from @EikoFried). https://pubmed.ncbi.nlm.nih.gov/26458184/  4/

However, we do not know if central symptoms are "important" from a clinical point of view, which would mean that less central symptoms are "less important". This is potentially harmful for several reasons 5/

First, central symptoms are only "central" in comparison to other variables; if only a given psychometric tool is used, we may never know how an item's centrality compares to others. Once again, the Fried et al. paper illustrates this very well 6/

For this first reason, it may be useful to enlarge the body of knowledge with network structures comparing as many used tool as possible and in as many samples as possible both clinical and subclinical (for psychological constructs, clinical and nonclinical). 7/

Second, the "disconnectedness" of a node is still "unexplored" in networks. In a recent paper currently submitted, I argue that the disconnectedness may be for instance due to floor and ceiling effects that derive from for instance A) validity issues or B) severe samples. 8/

The disconnectedness should also be formally tested, which is a requirement for formally stating that an item is, indeed, less central or disconnected. See work from @wdonald_1985 9/

In conclusion, it is important to understand why certain items are more "disconnected" than others when dealing with centrality. /10