Introduction: The Perceived Stress Scale (PSS-10) is one of the most widely used self-report instruments for measuring stress. This study aimed to provide a comprehensive psychometric evaluation of the PSS-10 by employing a multi-method approach integrating five advanced techniques, including structural equation modeling, item response theory, exploratory graph analysis, network analysis, and machine learning.
Methods: In this psychometric study, a convenience sample of 1278 students from Islamic Azad University, Garmsar Branch, was recruited. Data were analyzed using confirmatory factor analysis models (CFA, ESEM, ESEM-within-CFA), item response theory models (GRM, GPCM, MPGCM), exploratory graph analysis (EGA), network analysis (including centrality indices), and the random forest algorithm. Key indices from each approach were integrated to rank items according to their performance, thereby identifying core, complementary, and revision-recommended items.
Results: The ESEM-within-CFA model demonstrated excellent fit (RMSEA = .000; CFI = 1.00) and was selected as the final model. Independent analyses, including IRT (MGPCM), EGA, and network analysis, consistently supported a two-factor structure for the PSS-10 (Perceived Helplessness and Perceived Self-Efficacy). Based on the integrated ranking, items 10 and 2 were identified as core items; items 1, 3, 8, and 9 as complementary items; and items 4, 6, 7, and especially item 5 as requiring revision.
Discussion and conclusion: The findings not only support the strong construct validity of the PSS-10 but also demonstrate that a multi-method approach is a powerful strategy for precisely identifying key items and optimizing psychometric instruments.