The distinction tested in the study is technically important but does not mean that the same symptoms cannot be measured with a latent-variable approach. A latent model asks whether a smaller number of underlying dimensions, such as fear, catastrophizing, anxiety and avoidance, explain patterns of responses across questionnaire items. A psychometric network model instead represents each measured symptom as a node and estimates conditional associations among those nodes, allowing the influence of one symptom to depend on the others. Four latent specifications were tested: one factor, two factors, four factors and a second-order model. A confirmatory network model was compared with them using the same observations and standard fit indices, including comparative fit index (CFI), Tucker-Lewis index (TLI), root mean square error of approximation (RMSEA), Akaike information criterion (AIC), confidence intervals and cross-validation results.

Among the tested specifications, the confirmatory network model achieved the best relative fit to the observed covariance structure. It produced higher CFI and TLI values, a lower RMSEA with non-overlapping confidence intervals and a lower AIC than the alternatives. The researchers also repeated key ordinal-data analyses with two software packages, lavaan and psychonetrics, contrasting their results for the latent and network approaches. The network result indicates that a conditional symptom-to-symptom representation described the data more closely than the tested factor structures; it does not establish that one symptom causes another, nor does it demonstrate that any node is a causal treatment target.

The comparison also changed which questionnaire items appeared important. Several symptoms that would have been removed under the four-factor solution emerged as clinically meaningful “bridge” symptoms in the network, meaning they were connected to multiple symptom clusters rather than belonging to only one isolated domain. Within the multidimensional framework, the most parsimonious solution was a two-factor structure with omega equal to 0.90. Its dimensions were labeled Fear and Catastrophizing about Pain, and Anxiety and Avoidance Response to Pain. That structure showed the strongest congruence across samples, even though the network model fit the observed data better overall.

Generalizability was a major limitation. Measurement invariance and network-structure invariance were examined across sex, age and clinical status, and both forms of invariance broke down between the clinical and community samples. In other words, the CPASS did not behave identically in the two populations: relationships among items, and potentially the meaning of measured differences, depended on context. The authors therefore advised interpreting scores in light of the setting in which they were obtained. The findings support the broader methodological conclusion that choosing between latent-variable and network approaches can materially affect conclusions about symptom organization. For pediatric pain anxiety, a conditional network offered the more faithful statistical representation in this sample, while a two-factor solution remained useful for parsimonious measurement. Any suggestion that the network’s central or bridge symptoms should guide intervention remains a hypothesis for future research rather than a demonstrated clinical effect.

Subject of Research: Comparison of latent-variable and network models of pediatric pain anxiety in Spanish clinical and community samples

Article Title: Should child pain anxiety be modeled as a multidimensional latent construct or as a network of interrelated symptoms? Apsychometric comparison in spanish clinical and community populations

Article References: Huidobro-Labarga, B., López-Dolado, E., Aguirre-Sánchez, E., Alvarez de Mon Soto, M., & Castañeda-Vozmediano, R. (2026). Should child pain anxiety be modeled as a multidimensional latent construct or as a network of interrelated symptoms? Apsychometric comparison in spanish clinical and community populations. BMC Psychology. https://doi.org/10.1186/s40359-026-05580-5

Image Credits: AI Generated

DOI: 10.1186/s40359-026-05580-5

Keywords: pediatric pain anxiety, psychometric network model, latent variable model, CPASS, children and adolescents, measurement invariance, clinical psychology, confirmatory factor analysis, Should, child, pain, anxiety