Evaluation of supplemental samples in longitudinal research with non-normal missing data.

Area of Science:

• Statistics
• Longitudinal Data Analysis
• Research Methodology

Background:

• Missing data is a frequent challenge in longitudinal research.
• Existing guidance primarily focuses on analysis-stage missing data handling.
• Less attention is given to using supplemental samples to manage attrition.

Purpose of the Study:

• To evaluate the impact of supplemental samples on longitudinal data analysis.
• To assess these effects specifically for non-normally distributed data.
• To compare the efficacy of refreshment versus replacement sampling approaches.

Main Methods:

• Simulations were conducted to analyze longitudinal data with missingness.
• Two supplemental sampling strategies were investigated: refreshment and replacement.
• The refreshment approach involves random selection of new participants.
• The replacement approach uses auxiliary variables to select new participants.

Main Results:

• Simulation results indicate refreshment samples effectively mitigate attrition.
• Refreshment samples demonstrated potential to reduce bias and increase statistical power.
• Replacement samples, conversely, led to biased parameter estimates.
• The effectiveness was observed in the context of non-normally distributed data.

Conclusions:

• The addition of refreshment samples is an effective strategy for handling attrition in longitudinal research.
• Replacement samples are not recommended due to their tendency to introduce bias.
• Researchers should carefully consider the refreshment approach when dealing with missing data and attrition.