Missing data in a multi-item instrument were best handled by multiple imputation at the item score level

Area of Science:

• Statistics
• Biostatistics
• Psychometrics

Background:

• Complete-case analysis is common despite available advanced methods like multiple imputation (MI).
• Handling missing data in multi-item instruments is crucial for accurate statistical modeling.
• The performance of various missing data handling techniques needs thorough evaluation.

Purpose of the Study:

• To compare the performance of simple and advanced missing data handling methods.
• To evaluate techniques for missing item scores in multi-item instruments.
• To identify optimal strategies for different missing data scenarios.

Main Methods:

• Simulated real-life missing data situations in a multi-item covariate for linear regression.
• Applied various missing data mechanisms with increasing percentages of missingness.
• Compared fitted regression coefficients using bias and coverage as performance metrics.

Main Results:

• Mean imputation resulted in biased estimates when over 10% of data were missing.
• Multiple imputation (MI) applied to individual item scores outperformed total score methods when >25% of subjects had missing items.
• MI demonstrated superior performance in handling substantial missing data.

Conclusions:

• Recommend using multiple imputation (MI) on item scores for accurate regression model estimates.
• Advise against using any form of mean imputation for handling missing data.
• Multiple imputation is the preferred method for missing data in multi-item instruments.