Multiple Imputation with Factor Scores: A Practical Approach for Handling Simultaneous Missingness Across Items in...

Area of Science:

• Psychometrics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Nonignorable missingness in intensive longitudinal data can arise from latent factors, causing simultaneous item nonresponse.
• Existing methods like listwise deletion and standard multiple imputation may not adequately address this complex missingness pattern.

Purpose of the Study:

• To propose and evaluate a novel multiple imputation (MI) strategy, MI-FS (Multiple Imputation with Factor Scores), designed for nonignorable missing data.
• To compare the performance of MI-FS against listwise deletion (LD), MI with manifest variables (MI-MV), and partial MI with manifest variables (PMI-MV) using Monte Carlo simulations.

Main Methods:

• Developed MI-FS, incorporating factor scores, lag/lead variables, and missing data indicators into the imputation model.
• Conducted a Monte Carlo simulation study within the framework of Process Factor Analysis (PFA).
• Compared MI-FS, LD, MI-MV, and PMI-MV across various simulated conditions.

Main Results:

• Multiple imputation (MI) based methods generally outperformed listwise deletion (LD).
• MI-FS demonstrated lower Root Mean Square Errors (RMSEs) and superior coverage rates for auto-regression (AR) parameters compared to MI-MV.
• PMI-MV and MI-MV showed higher coverage rates for most parameters than MI-FS, excluding AR parameters.

Conclusions:

• MI-FS offers an effective approach for handling nonignorable missing data, particularly for auto-regression parameters in intensive longitudinal studies.
• The choice between MI-FS, MI-MV, and PMI-MV depends on the specific parameters of interest and desired performance characteristics.
• Recommendations are provided for integrating factor scores into MI processes for improved statistical inference.