Analyzing individual growth with clustered longitudinal data: A comparison between model-based and design-based...

Area of Science:

• Statistics
• Psychometrics
• Longitudinal Data Analysis

Background:

• Clustered longitudinal data, common in social sciences, requires careful statistical modeling.
• Ignoring dependency within clusters (e.g., students within schools) can bias growth estimates.
• Multilevel latent growth curve models (MLGCM) and maximum models (MM) are advanced techniques for such data.

Purpose of the Study:

• To evaluate the performance of model-based (MLGCM, MM) and design-based (D-LGCM) approaches for analyzing clustered longitudinal data.
• To assess the accuracy of parameter estimates for intraindividual growth and interindividual variability.
• To demonstrate the impact of ignoring individual dependency using single-level latent growth curve models (SLGCM).

Main Methods:

• A Monte Carlo simulation study was conducted.
• Examined the effects of varying numbers of clusters (NC) and cluster sizes (CS).
• Compared MLGCM, MM, D-LGCM, and SLGCM in their ability to provide unbiased and efficient estimates.

Main Results:

• MLGCM, MM, and D-LGCM provide unbiased estimates of intraindividual growth.
• MLGCM and MM yield accurate parameter estimates and standard errors for interindividual variability.
• D-LGCM and SLGCM produce confounded estimates of interindividual variability, mixing individual and cluster variance.

Conclusions:

• Researchers can use MLGCM, MM, or D-LGCM when intraindividual growth is the primary focus.
• For interindividual variability, MLGCM and MM are recommended; D-LGCM and SLGCM are inappropriate.
• D-LGCM is not a suitable alternative for analyzing clustered longitudinal data when interindividual variability is of interest.