Comparing Repeated-Measures ANOVA and Multilevel Growth Curve Modeling: A Simulation Study

Open Access
Liu, Siwei
Graduate Program:
Human Development and Family Studies
Master of Science
Document Type:
Master Thesis
Date of Defense:
November 12, 2008
Committee Members:
  • Michael J Rovine, Thesis Advisor
  • Peter Cm Molenaar, Thesis Advisor
  • Repeated-Measures ANOVA
  • Growth Curve Modeling
  • Longitudinal Data Analysis
Currently in developmental research, there seems to be a trend for multilevel growth curve modeling to take the place of the classic repeated-measures ANOVA as the first choice for analyzing longitudinal data. The preference for growth curve modeling is so strong that it is not uncommon to see researchers fit growth curve models without considering its appropriateness. This is problematic because the repeated-measures ANOVA model and the growth curve model are sub-models in the general linear mixed model family and are complementary to each other. This paper compares the two methods in analyzing simulated data that is assumed to come from a repeated-measures study with five equally spaced occasions and show a linear increase pattern. The simulation involves varying effect size, intraclass correlation size, and sample size. It is shown that repeated-measures ANOVA models generally have better fit when the error structure is selected properly. The growth curve model fits the data better only when sample size is not small and the variance of slopes is not small, even when the error structure is simulated to show a “growth curve” pattern. The results cast doubt on the popular practice of using growth curve modeling for longitudinal data without comparing the fit of different models. A general procedure of model selection is discussed.