Psychometric properties and factor structure of the Swedish version of the Perceived Stress Scale

Abstract

Background: The Perceived Stress Scale (PSS) measures general stress and the Swedish version, although used in several studies, has not been extensively evaluated for psychometric properties. Aims: This study aimed to investigate psychometric properties and the factor solution of the Swedish 14-item version when used with two samples, namely a mixed Internet sample of women and men (n = 171) and another of women with stress-related disorders (n = 84). Classical test theory, including confirmatory factor analysis, was employed. Results: The factor structure supported a two-factor model for the PSS and confirmed other language versions of the PSS, although one items showed a low item-total correlation. The PSS showed to be feasible with the investigated samples and the results indicated no ceiling or floor effects and good internal consistency of the PSS. Several aspects of construct validity were shown. An association of − 0.66 between the PSS and a measure of coping indicated good concurrent validity. Criterion validity was demonstrated through a statistically significant difference (P < 0.001) between the women with stress-related disorders and the Internet sample. Predictive validity of the PSS could be demonstrated in a short-term perspective. Based on the sample with stress-related disorders, sensitivity to change was shown through a statistically significant stress reduction (P < 0.001) from entering work rehabilitation to discharge. Conclusions: The Swedish version of the PSS showed satisfactory psychometric properties and may be recommended for use with people with and without known stress-related disorders.

• Coping
• Homogeneity
• Reliability
• Validity

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

Notes

Notes

a) The same models were also estimated using WLSMV and setting all response variables as categorical. These estimations gave the same results; only the RMSEA was a bit higher. The method was not used for invariance testing because some of the response variables were not populated on all rating alternatives for all response variables.

b) Cohen & Williamson suggested a 10-item version (7) that excluded four items (items 4, 5, 12 and 13). Estimating a model with this version of the inventory resulted in acceptable fit, RMSEA = 0.08, CFI = 0.96. Items 4, 5 and 13 did not create problems for the models estimated based on data from the Swedish version.