Comparing Imputation Methods for Trait Estimation Using the Rating Scale Model

Area of Science:

• Psychometrics
• Statistical modeling
• Data analysis

Background:

• Handling missing data is crucial for accurate questionnaire analysis.
• Traditional imputation methods may introduce bias or reduce precision.
• Understanding the impact of missingness on trait estimation is essential.

Purpose of the Study:

• To evaluate four distinct methods for handling missing data in discrete-choice questionnaires.
• To compare the performance of these methods under varying conditions of missingness and questionnaire length.
• To identify the most effective strategies for recovering accurate trait estimates.

Main Methods:

• A simulation study was employed to assess method performance.
• Four missing data handling techniques were examined: ignoring missingness, nearest-neighbor hot deck, multiple hot deck imputation, and semi-parametric multiple imputation.
• Data were simulated with varying questionnaire lengths (10, 20, 41 items) and missingness percentages (10%, 25%, 40%) under the assumption of data missing completely at random.

Main Results:

• Ignoring missing data and semi-parametric imputation demonstrated superior performance in recovering known trait levels across all simulated conditions.
• Semi-parametric multiple imputation yielded the most precise trait estimates among the evaluated methods.
• The study confirmed that ignoring missingness generally results in unbiased trait estimates, supporting Rasch measurement principles.

Conclusions:

• Both ignoring missing data and semi-parametric imputation are effective strategies for handling missing data in polytomous questionnaires scored with the rating scale model.
• Semi-parametric imputation offers enhanced precision for trait estimation.
• The findings underscore the importance of specific objectivity in Rasch measurement, particularly when dealing with incomplete data.