Machine learning item selection for short scale construction: A proof-of-concept using the SIMS.

Area of Science:

• Psychological assessment
• Machine learning applications
• Psychometrics

Background:

• Traditional psychometric methods are used for developing short forms of psychological tests.
• The Structured Inventory of Malingered Symptomatology (SIMS) is a validated measure of symptom validity, frequently used in forensic evaluations.

Purpose of the Study:

• To demonstrate machine learning (ML) as a viable alternative to traditional psychometric techniques for creating concise psychological assessments.
• To develop a short form of the SIMS using ML-based item selection.

Main Methods:

• Integrated data from 329 participants across multiple datasets for SIMS research.
• Applied state-of-the-art ML feature selection techniques to identify optimal items for the short form.
• Utilized wrapper methods for evaluating global item value.

Main Results:

• Achieved a 72% reduction in the SIMS scale length.
• The developed short form captured 92% of the original SIMS's variance.
• The new short form comprises 21 items.

Conclusions:

• ML-based item selection is a promising alternative to conventional psychometric methods for developing psychological test short forms.
• ML techniques, particularly wrapper methods, can efficiently identify items that preserve the global characteristics of the original scale.