Editors’ introduction to the special issue “Network psychometrics in action”: Methodological innovations inspired by empirical problems.

Abstract

Editors’ introduction to the special issue “Network psychometrics in action”: Methodological innovations inspired by empirical problems.

Possible Futures for Network Psychometrics

“…As this Special Issue may well mark the arrival of a new psychometric tradition that forays into that territory, it is important to adequately document its roots before they fade from memory. In this respect, I would like to draw attention to an important historical element behind the development of network approaches as discussed by Marsman and Rhemtulla (2022), namely the influence of Peter Molenaar. Molenaar was the first to understand and appreciate the resemblance between network models used in statistical physics (particularly the Ising model) and psychometric models (particularly, Item Response Theory).…”

“…In this perspective, regularized network analysis defines an explorative search for a restricted model that fits the data best in a clearly defined sense. However, as Marsman and Rhemtulla (2022) suggest, regularized approaches can also be interpreted differently, for instance, as procedures aimed at estimation or inference, and with respect to these goals they can have considerable downsides. For example, regularized estimates are often biased and their sampling distribution is unwieldy.…”

“…In addition, in certain cases, regularized estimation through the graphical lasso can go awry (Williams & Rast, 2020;Wysocki & Rhemtulla, 2021) although I personally find the contrast between results from different simulation procedures hard to make sense of and doubt the generality of some of the claims now in the literature (e.g., compare the results of Williams & Rast, 2020, to those of Isvoranu & Epskamp, 2021. Finally, if the generating model is not sparse but dense, the regularized approach can struggle to approximate the generating model (Epskamp et al, 2017), which is especially problematic if edges that have been suppressed in the estimation process are used inferentially, namely as evidence for the absence of links in the generating network-the well-known credo that absence of evidence is not evidence of absence applies here as it does elsewhere (Marsman & Rhemtulla, 2022;.…”

Measurement Targets for Network Constructs in Psychopathology

“…These models were implemented using regularized estimation techniques that were originally developed to estimate Ising models in high dimensions (van Borkulo et al, 2014;Foygel & Drton, 2010;Meinshausen & Bühlmann, 2006). Since, a plethora of techniques have been developed to estimate networks using various kinds of data, different estimation methods, and different goals (Borsboom et al, 2021;Marsman & Rhemtulla, 2022).…”