The Use of Mixed Models for the Analysis of Mediated Data with Time-Dependent Predictors

Table 1

Performance of SEM and linear mixed model assuming total effect of main independent variable is constant when true underlying effects are small for early time points and small to moderate for late time points.

Simulated data scenarios				Constant effect SEM			Constant effect LMM
Time point	Effect size	Sample size	Effect distribution	Bias (%)	Coverage probability (%)	Power (%)	Bias (%)	Coverage probability (%)	Power (%)
𝑡1∶𝑡3	0.05	350	Equal	180	67	65	171	71	62
𝑡4∶𝑡6	0.25	350	Equal	−44	56	65	−45	53	62
𝑡1∶𝑡3	0.05	350	Primarily indirect	179	70	60	164	73	55
𝑡4∶𝑡6	0.25	350	Primarily indirect	−44	55	60	−47	51	55
𝑡1∶𝑡3	0.05	350	Primarily direct	173	69	66	169	69	65
𝑡4∶𝑡6	0.25	350	Primarily direct	−45	50	66	−46	49	65
𝑡1∶𝑡3	0.05	100	Equal	168	86	24	159	88	22
𝑡4∶𝑡6	0.25	100	Equal	−46	82	24	−48	81	22
𝑡1∶𝑡3	0.05	200	Equal	178	80	41	169	81	38
𝑡4∶𝑡6	0.25	200	Equal	−44	73	41	−46	70	38
𝑡1∶𝑡3	0.05	400	Equal	175	65	67	165	69	64
𝑡4∶𝑡6	0.25	400	Equal	−45	49	67	−47	46	64
𝑡1∶𝑡3	0.05	350	Equal	395	3	97	369	2	94
𝑡4∶𝑡6	0.5	350	Equal	−50	16	97	−53	21	94
𝑡1∶𝑡3	0.05	400	Equal	−148	73	8	−137	76	6
𝑡4∶𝑡6	−0.11	400	Equal	−78	63	8	−83	60	6

Based on 1000 simulated datasets.