Effects of Level 1 Variance on Level 2 Variance in Multilevel Model (Mixed Model) for Growth Repeated Data

• Activity Number: 475
• Type: Contributed
• Date/Time: Wednesday, August 1, 2012 : 8:30 AM to 10:20 AM
• Sponsor: Section on Statistics in Epidemiology
• Author(s): Chong Yau Fu*+ and Hsin-Yi Huang
• Companies: National Yang Ming University and National Yang Ming University
• Keywords: multilevel model (mixed model) ; variance components ; variance partition coefficient ; growth repeated data

Abstract

The growth repeated data in a 2-level multilevel model, except the measurement error, other variance components have their important cast in modeling. The data dependence or heterogeneity between individuals refers level 2 random intercept [R-I] or random coefficient [R-C]. Level 1 variance components have serial correlation ([S-C]) (or auto-correlation), heteroscedasticity variance ([H-C]), and constant variance ([C-V]). This study proposed concerns with the confounding effect of level 1 variance on level 2 variance. Based on the illustration of mouse's tumor growth data, this study conduct three simulations. The results reveal that more strong [S-C] together with [C-V] will more likely to be 2-level [R-C] or with a larger VPC (variance partition coefficient), particularly for small number of time points. A strong 1-level [H-V] model commonly dominates 2-level [R-C] model, even [S-C] effect added.