A comparative study of variable selection methods in the context of developing psychiatric screening instruments.

Area of Science:

• Psychometrics and statistical modeling for psychiatric diagnostics.
• Development of efficient screening instruments for mental health conditions.

Background:

• Screening instruments for psychiatric disorders require careful item selection from existing questionnaires.
• Key challenges in psychiatric data include correlations, unobserved variables, missing data, and case prevalence.

Purpose of the Study:

• To investigate the impact of psychiatric data characteristics on variable selection methods.
• To compare the performance of various selection methods regarding accuracy and prediction error.

Main Methods:

• Simulation studies were conducted to evaluate methods like LASSO, Elastic Net, Classification and Regression Tree, Random Forest, and t-test.
• The study assessed performance with complete and incomplete datasets, considering factors like missing data patterns.

Main Results:

• LASSO and Elastic Net demonstrated superior performance in variable selection and prediction with complete data.
• Random Forest showed bias in imputation for incomplete data, affecting variable importance ranking.

Conclusions:

• Imputed-LASSO, combining Random Forest imputation and LASSO, effectively addresses bias from missing data.
• This approach provides a simple and efficient method for item selection in psychiatric screening instruments.