Handling Missing Data in the Modeling of Intensive Longitudinal Data.

Area of Science:

• Statistics
• Longitudinal Data Analysis
• Psychometrics

Background:

• Handling missing data is crucial in statistical modeling.
• Few studies evaluate methods for intensive longitudinal data.
• Multiple imputation (MI) is a common technique for missing data.

Purpose of the Study:

• To compare two multiple imputation (MI) approaches for multivariate time-series models with intensive longitudinal data.
• To investigate the performance of full MI versus partial MI under various missing data mechanisms.
• To illustrate the application of these MI methods using an empirical dataset.

Main Methods:

• Comparison of a full MI approach (imputing all variables) and a partial MI approach (imputing covariates, using FIML for dependent variables).
• Application to multivariate time-series models.
• Utilized an empirical dataset examining child influences on parental conflict over 15 days.

Main Results:

• Under correctly specified models, the partial MI approach demonstrated superior overall estimation results.
• The study identified strengths and limitations of both full and partial MI techniques.

Conclusions:

• Partial multiple imputation offers a robust strategy for handling missing data in intensive longitudinal time-series analyses.
• The findings provide guidance for researchers selecting appropriate missing data handling techniques for complex longitudinal datasets.