measurement error
Definition
Measurement error refers to the discrepancy between observed variable scores and the true underlying values those scores are intended to capture, a problem that is rarely explicitly modeled in network psychometric frameworks. In cross-sectional Gaussian Graphical Models, failing to account for measurement error produces inflated edge weights and an overestimation of network density, making a psychological construct appear more interconnected than it actually is. This distortion carries direct consequences for clinical interpretation: central nodes, such as the most important symptoms in a psychopathological disorder, may be incorrectly identified, potentially leading to misguided intervention strategies. Simulation evidence comparing EBICglasso and LoGo-TMFG shows that EBICglasso tends to be more robust to the negative impacts of mis-measured variables, and that sample size does not attenuate the magnitude of this impact, with large samples performing similarly to smaller ones.
Sources: Henry & Ye (2024)
Related Terms
- network psychometrics (1 shared article)
- centrality (1 shared article)
- omitted variable bias (1 shared article)
Applications
Measurement Error and Omitted Variable Bias
In network psychometric models, measurement error and omitted variable bias are treated as distinct but related sources of model misspecification, each capable of distorting the estimated structure of a Gaussian Graphical Model. Omitted variables can introduce spurious edges or suppress true connections, while measurement error inflates edge weights and network density. Simulation results indicate that EBICglasso and LoGo-TMFG respond similarly to both problems as network size increases, suggesting that neither source of misspecification is trivially absorbed by larger model configurations.
Sources: Henry & Ye (2024)
Measurement Error and Network Density
When measurement error is present but unmodeled in cross-sectional network psychometric models, the estimated network density is systematically overestimated. This occurs because unreliable items introduce variance that is redistributed across estimated edges rather than isolated in a measurement component, producing a denser partial correlation structure than the true data-generating process warrants. The result is a network representation that overstates the degree of interconnection among psychological symptoms.
Sources: Henry & Ye (2024)
