The Impact of Missing Values and Single Imputation upon Rasch Analysis Outcomes: A Simulation Study

Area of Science:

• Psychometrics
• Statistical Analysis
• Educational Measurement

Background:

• Missing data is prevalent in psychometric and educational research.
• Imputation techniques are increasingly used to handle missing data.
• The robustness of imputation strategies under various data conditions in Rasch analysis requires investigation.

Purpose of the Study:

• To evaluate the performance of different imputation strategies in Rasch analysis.
• To assess the impact of missingness extent, type, local item dependencies (LID), differential item functioning (DIF), and misfit on imputation robustness.
• To compare imputation techniques against direct analysis of missing data.

Main Methods:

• Simulated four datasets representing varying data complexities (LID, DIF).
• Introduced missing values at random (MAR) and completely at random (MCAR) with increasing proportions.
• Applied four imputation techniques prior to Rasch analysis.
• Compared deviations in statistical estimates and goodness-of-fit measures.

Main Results:

• Imputation strategies performed well when missingness was below 15%.
• Direct analysis of data with missing values outperformed imputation methods in recovering statistical estimates.
• The presence of LID and DIF influenced the performance of imputation techniques.

Conclusions:

• Direct analysis of missing values is the optimal strategy for Rasch analysis.
• If imputation is necessary, the expectation-maximization algorithm is recommended.
• Careful consideration of missing data handling is crucial for valid Rasch model results.