Columnwise neural imputation for incomplete ordinal psychometric data

Area of Science:

• Psychometrics
• Artificial Intelligence
• Data Science

Background:

• Missing data are common in psychological and educational assessments.
• Traditional imputation methods like listwise deletion and mean imputation can reduce research validity.
• Artificial neural networks show promise in inferring missing values.

Purpose of the Study:

• To introduce the Columnwise Neural Imputation (COLNI) algorithm for imputing missing ordinal data in psychometric assessments.
• To evaluate COLNI's performance against conventional imputation methods.

Main Methods:

• Developed the COLNI algorithm, an artificial neural network-based approach for imputing missing ordinal responses.
• Conducted simulation studies comparing COLNI with traditional methods.
• Validated COLNI using empirical data from the Short Dark Triad test.

Main Results:

• COLNI demonstrated superior accuracy in recovering item means, inter-item correlations, and person/item parameters under the multidimensional graded response model.
• Empirical evaluation confirmed COLNI's effectiveness in a real-world multidimensional setting.

Conclusions:

• COLNI is an effective and accurate method for imputing missing ordinal data in psychometric research.
• The study provides guidelines for implementing COLNI and suggests future research directions for artificial neural network-based imputation.