Reading symptoms as a network
For most of the last century, a psychological scale was read as a sum. Nine items, one score, one estimate of severity. That reading rests on a specific assumption — that the items are interchangeable indicators of something underneath them.
Network psychometrics starts somewhere else. It asks which symptoms connect to which, how strongly, and which ones sit at the crossroads. Two people with the same total score can have entirely different structures, and the structure is often where the clinically interesting part lives.
This blog works through that shift — estimation methods like EBICglasso, centrality and bridge symptoms, stability and what it means when a network is unstable, the ergodicity problem of reading group-level structure as an individual's own. Written for researchers and clinicians who want the reasoning, not just the output.
The position here is deliberately unresolved: the latent variable tradition and the network approach answer different questions about the same data. Neither gets declared the winner. Both are worth having in front of you.