SEM-Based Out-of-Sample Predictions

Predictive modeling is becoming more popular in psychological science. Machine learning techniques have been used to develop prediction rules based on items of tests. However, this approach does not take into account that these are noisy indicators the constructs they intend measure. Structural equation account. Several authors concluded it impossible make out-of-sample predictions a reflective structural model. We show possible such and we R-code do so. With two empirical examples, SEM-based can outperform linear regression models. three simulation studies, further investigate rule its robustness comparison with using regularized regression. conclude new robust against violation normality assumption but sensitive model misspecification.