Skip to main content

Table 2 Absolute values of relative bias of intercept (int) and slope per analysis model: findings over 1000 replications in the simulation for a sample size of 1000 and a high and low degree of separation between classes in Scenario 1

From: Variance constraints strongly influenced model performance in growth mixture modeling: a simulation and empirical study

Model   Intercepts Slopes
   Class 1 Class 2 Class 3 Class 1 Class 2 Class 3
High separation and N = 1000
 0 Nothing constrained .00 .00 .00 .00 .00 .00
 1A Residual variance time constrained .00 .01 .00 .00 .02 .00
 1B Residual variance classes constrained .00 .00 .00 .00 .00 .00
 1C Random effects constrained .02 .02 .01 .00 .01 .02
 2A Residual variance constrained time and classes .00 .00 .00 .00 .00 .00
 2B Random effects and residual variance time constrained .00 .01 .00 .00 .04 .07
 2C Random effects and residual variance classes constrained .23 .17 .01 .10 .17 .29
Low separation and N = 1000
 0 Nothing constrained .00 .00 .00 .00 .01 .01
 1A Residual variance time constrained .00 .01 .00 .00 .08 .08
 1B Residual variance classes constrained .00 .01 .00 .00 .04 .00
 1C Random effects constrained .01 .01 .02 .01 .01 .05
 2A Residual variance constrained time and classes .00 .01 .00 .00 .06 .00
 2B Random effects and residual variance time constrained .01 .01 .01 .01 .03 .21
 2C Random effects and residual variance classes constrained -a – – – – –
  1. Relative bias: bias divided by true population value
  2. Codes for relative bias: Bold ≥ .1
  3. The variance ratio in Scenario 1 is 1:3 across classes and over time points, which means that the variance in the first class/time-point is three times smaller compared to the last class/time-point
  4. aThe bias could not be calculated, since the simulated classes were not recovered
Back to article page

BMC Medical Research Methodology

ISSN: 1471-2288

Contact us