Fitting Ordinal Factor Analysis Models With Missing Data: A Comparison Between Pairwise Deletion and Multiple...

Area of Science:

• Statistics
• Psychometrics
• Data Analysis

Background:

• Missing data is common in statistical analyses.
• Ordinal factor analysis is used for categorical data.
• Pairwise deletion (PD) is a default method for handling missing data, but its performance is not always optimal.

Purpose of the Study:

• To compare the performance of pairwise deletion (PD) and multiple imputation (MI) for handling missing data in ordinal factor analysis.
• To determine which method yields parameter estimates and model fit indices closer to complete data results.
• To evaluate performance under various conditions, including missing completely at random (MCAR) and missing at random (MAR) data.

Main Methods:

• Ordinal factor analysis models were used.
• Pairwise deletion (PD) and multiple imputation (MI) were compared.
• Simulations were conducted across different conditions: response categories, sample size, missingness percentage, and model misfit.

Main Results:

• Both PD and MI produced similar parameter estimates to complete data under MCAR conditions.
• PD parameter estimates were severely biased under MAR conditions.
• MI yielded comparable parameter estimates to complete data when missingness was below 50%, but fit indices (χ², RMSEA, WRMR) indicated poorer fit.

Conclusions:

• Multiple imputation (MI) is recommended for applied researchers analyzing ordinal factor models with missing data.
• Researchers should interpret model fit using TLI and CFI indices when using MI, as other indices may suggest worse fit than reality.