Quiet and Passive Quitting Predict Work Engagement in a Machine Learning-Enhanced SEM Analysis

In structural equation modeling (SEM), model fit is commonly evaluated on a single sample, which limits evidence about out-of-sample generalisability in HR research. This study adapts repeated cross-validation to SEM and introduces a normalised generalisation gap (NG) index to quantify fit deterioration on unseen data relative to training fit. Using employee survey data (N = 1,040), we estimated a model in which quiet quitting and passive quitting predict work engagement and evaluated RMSEA across repeated 10×10 folds. In-sample fit indices were stable across training subsamples, yet test-set fit was consistently weaker. The NG index operationalises this decline as a standardised ratio, enabling transparent reporting of overfitting risk and model transportability. The procedure supports more robust evaluation of measurement models and structural relations in people analytics applications.