Missing data in longitudinal studies: cross-sectional multiple imputation provides similar estimates to...

Area of Science:

• Statistics
• Psychometrics
• Longitudinal Data Analysis

Background:

• Longitudinal studies are crucial for understanding developmental processes.
• Missing data in longitudinal datasets can bias parameter estimates.
• Nonmonotone missingness patterns pose particular challenges for data analysis.

Purpose of the Study:

• To explore the validity of cross-sectional multiple imputation for latent growth curve models.
• To assess parameter estimation accuracy in longitudinal data with nonmonotone missingness.
• To compare multiple imputation with other missing data handling techniques.

Main Methods:

• A simulated longitudinal dataset (N=5000) with nonmonotone missingness (5%-20%) was used.
• Latent growth curve models were analyzed using listwise deletion, full-information maximum likelihood, and multiple imputation.
• Analysis of variance compared parameter estimates from full data versus missing data approaches.

Main Results:

• Multiple imputation produced significantly lower slope variance compared to complete data.
• No significant differences were found between multiple imputation and full-information maximum likelihood estimates.
• Listwise deletion was not explicitly compared in the results section but implied as a less favorable method.

Conclusions:

• Cross-sectional multiple imputation is a potentially valid method for longitudinal data with nonmonotone missingness.
• Estimates from multiple imputation are comparable to those from full-information maximum likelihood.
• Further research is recommended to confirm the robustness of this imputation method.