Mapping psychological heterogeneity in military cadets: a multi-method framework combining network analysis, latent...

Military training imposes sustained psychological demands, yet cadet cohorts are typically treated as homogeneous, which obscures individual differences in distress, resilience, and adaptation. This study examined psychological adaptation and its heterogeneity across an initial training period.

Ninety-two cadets completed the DASS-21, the CD-RISC-25, and personality measures on three occasions spanning five months. A multi-method framework combined four complementary analyses: longitudinal measurement invariance testing through confirmatory factor analysis and Tucker's congruence coefficients, psychological network estimation with graphical LASSO regularisation, latent profile analysis, and logistic regression nomograms developed as clinical screening tools. A pre-specified sensitivity analysis replaced the baseline dynamic indicators with T2 minus T0 change scores while keeping trait level inputs unchanged.

The CD-RISC-25 demonstrated excellent factorial stability (mean congruence = .971), whereas the DASS-21 showed fair stability (mean congruence = .867). Depression and stress formed the strongest partial correlation edge (r = .54, bootstrap inclusion = 100%), with resilience acting as a bridging node between distress and personality domains. Latent profile analysis identified five cadet profiles with high classification entropy (.973) and significant differentiation across 17 outcome variables. The change score solution again yielded five profiles (entropy = .955) but showed only moderate concordance with the primary solution (adjusted Rand index = .245). The academic performance nomogram achieved the highest cross-validated accuracy (AUC = .682).

Baseline and trajectory typologies capture complementary, non-redundant facets of cadet heterogeneity, supporting an integrated account of adaptation to military training and offering practical tools for early identification of at risk cadets, although predictive accuracy remains modest.