Modelling stress constructs with biomarkers: the importance of the measurement model

Document Type


Publication details

Bradbury, J 2013, 'Modelling stress constructs with biomarkers: the importance of the measurement model', Clinical and Experimental Medical Sciences, vol. 1, no. 5, pp. 197-216.

Article available on Open Access

Peer Reviewed



Background: The transactional model of stress describes a psychological and physiological stress response that is elicited when an environmental demand is perceived to outweigh the ability to cope with the demand. With perception at its core, this theory explains large variances in human stress responses. A frequently cited measure of stress perception is the Perceived Stress Scale (PSS). A two-step approach to structural equation modelling (SEM) necessitates that a valid measurement model for stress be first established through confirmatory factor analysis (CFA), and then the impact of stress on the biomarkers be assessed.

Methods: The aims of this study were to explore and confirm the factorial structure of the 10-item PSS (PSS-10) in a sample of healthy Australians (n = 194) and to crossvalidate it in an independent sample (n = 117) of healthy Australians with chronic work stress, and to explore the impact of stress on the production of the ex vivo stimulated secretion of proinflammatory cytokines, interleukins IL-1β, IL-6 and tumour necrosis factor (TNFα), using SEM. As only Sample 1 had data on both PSS scores and proinflammatory cytokine secretion, SEM would be confined to Sample 1. 198 J. Bradbury.

Results: Perceived Stress was found to have different factorial structures in the different samples. In Sample 1, it was best represented by three correlated factors; Overwhelmed (items 2, 10, 6, 8), Coping (items 4, 7, 5), and Emotional Reactivity (items 1, 9, 3). This three-factor model for was not cross-validated in Sample 2, which fit the two-factor model reported in the literature. The SEM suggested that the three factors predicted significant differential effects on proinflammatory cytokines that were less evident for the two-factor model.

Conclusions: These findings demonstrate that if the measurement model does not adequately represent the relationship structures within the data, differential effects on biomarkers may be diluted by larger measurement errors. This analysis has highlighted the importance of testing the assumptions of the measurement model used to represent stress in human populations. Further research is required to determine whether the construct known as perceived stress comprises two or three inherent factors in population norms.