An item sorting heuristic to derive equivalent parallel test versions from multivariate items.

Area of Science:

• Psychometrics
• Cognitive Science
• Data Science

Background:

• Developing parallel test versions with equivalent difficulty is crucial for accurate measurement.
• Multivariate items, common in language and image data, pose challenges for creating parallel tests.

Purpose of the Study:

• To propose and validate a heuristic method for identifying and selecting similar multivariate items.
• To generate equivalent parallel test versions that adhere to classical test theory principles.

Main Methods:

• A heuristic involving correlation inspection, outlier identification, and principal component analysis (PCA).
• Biplot generation for grouping items based on principal components (PCs).
• Assignment of items to parallel test versions and subsequent checks for equivalence, parallelism, reliability, and internal consistency.

Main Results:

• Successfully derived four parallel test versions, each with 20 items, from a pool of 116 items using the proposed heuristic.
• Demonstrated the heuristic's effectiveness in generating test versions that meet classical test theory requirements.

Conclusions:

• The proposed heuristic provides a systematic approach to creating equivalent parallel test versions with multivariate items.
• This method facilitates robust test construction by considering multiple variables simultaneously.