Fig. 4: Idiographic models achieve higher group level prediction accuracy than nomothetic models.

a, b CAT-DI prediction accuracy across all individuals in the test set as measured by MAPE (a) and R² (b) across all individuals for different models and latent depression traits. The dotted line in B indicates 70% prediction accuracy and bars indicate 95% confidence intervals of R². c R² versus the number of weeks ahead we are predicting from the last observation in the training set. The dotted line indicates 70% prediction accuracy. Bars indicate 95% confidence intervals of R². d, e log2 fold change in CAT-DI prediction accuracy, as measured by MAPE (d) and R² (e), of feature-based model over the baseline model. Negative log2 fold change in MAPE and positive log2 fold change in R² mean that the feature-based model performs better than the baseline model. A log2 fold change in MAPE of −1 means that the prediction error of the baseline model is twice as large as that of the feature-based model. The dotted line indicates the log2 fold change for the best and worst performing model/latent trait combination. Features were imputed with Autocomplete and CAT-DI was modeled using a logistic elastic net regression. MAPE: mean absolute percent error. LOCF last observation carried forward. CS(xdf) Cubic spline with x degrees of freedom. CS(cv) best-fitting cubic spline according to leave-one-out cross-validation.