Psychometric properties of the Post-Traumatic Growth Inventory-Short Form (PTGI-SF) in Chilean population exposed to the February 27th earthquake: analysis through Exploratory Structural Equation Modeling

Introduction The Post-Traumatic Growth Inventory-Short Form (PTGI-SF) has shown substantial cross-cultural variability in its factorial structure, yet contemporary latent-variable validation evidence remains limited for Chilean disaster-exposed populations. Methods This study evaluated the psychometric properties of the PTGI-SF in 578 community-dwelling adults from three affected localities (Cauquenes, Pelluhue, and Tongoy) exposed to the February 27, 2010, Chile earthquake. Internal structure was examined using Exploratory Structural Equation Modeling (ESEM), reliability was estimated with McDonald's ordinal omega, and validity evidence was examined via ROC analysis and correlations with posttraumatic stress symptoms assessed using the SPRINT-E. Factorial similarity in loading patterns across localities was described using Tucker's congruence coefficients, whereas measurement equivalence was tested through multigroup CFA measurement invariance (MI) analyses. Results ESEM supported a culturally specific three-factor solution (Appreciation of Life, Spiritual Change, and Self-Perception), diverging from the original five-factor model. Reliability was high across localities (Omega = 0.91–0.95). Model fit varied by context, with acceptable fit in Tongoy (RMSEA = 0.08; CFI/TLI = 0.99), mixed-to-poor fit in Pelluhue (RMSEA = 0.14; CFI = 0.98; TLI = 0.95), and clearly unacceptable fit in Cauquenes (RMSEA = 0.22; CFI = 0.92; TLI = 0.80). Discriminative validity relative to severe PTSD symptomatology (SPRINT-E ≥7 intense symptoms) was moderate (AUC = 0.703; sensitivity = 90.91%; specificity = 48.69%), and concurrent validity showed a significant positive association with PTSD symptoms ( r = 0.288, p < 0.001). Discussion Measurement invariance analyses supported configural and metric invariance across localities, whereas scalar invariance could not be established due to inadmissible solutions, indicating that latent mean comparisons across communities are not supported.