On Modeling Missing Data of an Incomplete Design in the CFA Framework.

Area of Science:

• Statistics
• Psychometrics
• Data Analysis

Background:

• Missing data is a common challenge in dataset analysis, potentially compromising structural validity.
• Traditional methods may yield inaccurate results when significant data is absent without replacement.

Purpose of the Study:

• To investigate the efficacy of specialized confirmatory factor analysis (CFA) models in analyzing datasets with high rates of missing data.
• To compare the performance of a missing data CFA model and a semi-hierarchical CFA model against a standard one-factor model.

Main Methods:

• Employed two types of CFA models: a missing data CFA model and a semi-hierarchical CFA model, both incorporating a latent variable for missing data.
• Investigated simulated binary data, comparing model fit and accuracy of factor loading estimations.
• Assessed model performance for subgroups with and without data omissions.

Main Results:

• Modeling missing data effectively mitigated negative impacts on model fit compared to the regular one-factor model.
• The semi-hierarchical CFA model demonstrated superior accuracy in estimating factor loadings.
• While factor loadings were estimated with comparable sizes for items with and without omissions, the semi-hierarchical model showed a tendency to underestimate expected values.

Conclusions:

• Specialized CFA models, particularly the semi-hierarchical approach, can yield valid results even with extensive missing data.
• The semi-hierarchical CFA model offers improved accuracy for factor loading estimation in the presence of missing data.
• Further refinement may be needed to address the underestimation of factor loadings observed in the semi-hierarchical model.