centrality
Definition
Centrality refers to the extent to which a node occupies an important or influential position within a network, usually quantified by indices based on the number and strength of its connections or on how often it lies on paths linking other nodes.
In research published in Advances in Psychology, centrality is treated as a property of symptoms within cross-sectional network psychometric models of psychopathology, where each node is an observable symptom and edges are partial correlations estimated as a Gaussian Graphical Model. Identifying central symptoms is often regarded as a basis for selecting intervention targets. A simulation study comparing the EBICglasso and LoGo-TMFG estimators shows that omitted variables can create spurious edges or suppress true ones, and that unmodeled measurement error inflates edge weights and overstates network density. Because centrality is computed from this edge structure, such distortions can lead researchers to single out the wrong nodes as most central. EBICglasso was the more robust method overall, the two methods converged in larger networks, and increasing sample size did not lessen these problems.
Sources: Henry & Ye (2024)
Related Terms
- network psychometrics (1 shared article)
- measurement error (1 shared article)
- omitted variable bias (1 shared article)
Applications
Centrality and Measurement Error
Network psychometric models rarely account for measurement error explicitly. Unreliable measurement inflates edge weights and exaggerates network density, making a symptom network appear more interconnected than it is. These inflated connections can mislead researchers about which symptoms are most central and steer intervention planning toward the wrong targets. Larger samples did not reduce this effect in simulations.
Sources: Henry & Ye (2024)
Centrality and Omitted Variable Bias
Leaving relevant variables out of a cross-sectional symptom network can produce false edges or mask true connections, altering the structure from which centrality indices are derived. As a result, nodes may be wrongly identified as central, with consequences for intervention strategies. EBICglasso withstood this bias better than LoGo-TMFG, though the two methods converged as networks grew larger.
Sources: Henry & Ye (2024)
