Multiple imputation of missing data in multilevel ecological momentary assessments: an example using smoking...

Area of Science:

• Behavioral Science
• Biostatistics
• Psychology

Background:

• Digital technology facilitates intensive longitudinal data (ILD) collection, such as ecological momentary assessments (EMAs), for behavior change studies.
• ILD often exhibits multilevel structures and missing data, necessitating robust statistical methods for accurate analysis.
• Multiple imputation is a validated technique for handling missing data in ILD, provided the imputation model reflects time dependencies.

Purpose of the Study:

• To demonstrate the importance of accounting for multilevel structures in ILD when performing multiple imputation.
• To compare the performance of multilevel multiple imputation (multilevel MI) against methods that ignore these structures.
• To illustrate the application of multilevel MI using empirical EMA data from a tobacco cessation study.

Main Methods:

• A Monte Carlo simulation study was conducted to compare multilevel MI with other imputation approaches.
• Empirical ecological momentary assessment (EMA) data from a tobacco cessation study were utilized.
• The study evaluated the implications of distinguishing between participant- and study-initiated EMAs.

Main Results:

• Multilevel MI demonstrated superior performance when handling multilevel ILD structures compared to methods that do not account for them.
• The simulation results underscore the necessity of incorporating multilevel properties into imputation models for accurate parameter estimation.
• Analysis of tobacco cessation EMA data revealed distinct patterns in affective dynamics and urge based on EMA initiation type.

Conclusions:

• Properly accounting for multilevel structures in ILD via multilevel MI is crucial for valid analysis of behavior change data.
• The findings emphasize the need for advanced statistical techniques that accommodate the complexity of ILD.
• Differentiating EMA initiation types provides nuanced insights into individual affective dynamics and urges in cessation studies.