Performance of mixed effects models in the analysis of mediated longitudinal data

Table 3 Impact of distributional assumptions.

Simulated Data		Mediated SEM			Mixed Model without Mediator
Distribution	Non-normal Residual Error	Bias (%)	Coverage Probability (%)	Power (%)	Bias (%)	Coverage Probability (%)	Power (%)
Uniform	Measurement	0.7	96	82	-0.8	96	81
	Measurement & Structural	1.4	97	87	1.6	96	85
Log-normal	Measurement	-1.3	94	82	-1.3	94	80
	Measurement & Structural	-0.7	95	82	-0.7	95	81
Contaminated Normal	Measurement	-0.2	95	18	0.07	95	17
	Measurement & Structural	9.5	96	8	11.1	95	8
Fleishman/Mattson 1	Measurement & Structural	-2.9	94	79	-3.0	94	78
Fleishman/Mattson 2	Measurement & Structural	-2.4	94	80	-2.6	94	78

The impact of distributional assumptions on model performance in evaluating the total effect of the main independent variable on the random intercept.
Based on 1000 simulated datasets with a medium-small effect size, equally distributed and a sample size of 250.
The Fleishman/Mattson 1 distribution is moderately skewed and slightly kurtotic.
The Fleishman/Mattson 2 distribution is highly skewed and kurtotic.

ISSN: 1471-2288