Figure 3.

Models were run on simulated data to determine the best method for testing heteroscedasticity in longitudinal data. a) The simple linear model (Model 1.1) was applied on simulated data to generate absolute residuals that measured interindividual variability. The association between absolute residuals from the linear regression and age was further estimated by b) a simple linear regression (Model 2.1), c) a random intercept and slope model (Model 2.2), and d) a random intercept model (Model 2.3). e) Although the random intercept and slope model (Model 1.2) best fitted the simulated data. f) The absolute residuals from Model 1.2 captured intraindividual variability that did not change with age. g) The random intercept model (Model 1.3) was also tested for the simulated data. h) But absolute residuals from Model 1.3 were not associated with the age.