The mixed model for repeated measures for cluster randomized trials: a simulation study investigating bias and type I error with missing continuous data

Table 4 Coverage values for treatment effect (difference at the fourth time point) with 30% missing data in the same direction for each of the three simulation methods^a

k, clusters per arm	m, subjects per cluster	Method 1		Method 2		Method 3
k, clusters per arm	m, subjects per cluster	ICC = 0.01	ICC = 0.1	ICC = 0.01	ICC = 0.1	ICC = 0.01	ICC = 0.1
5	10	95.1	92.8	95.2	93.3	95.6	94.0
	20	92.1	93.3	94.4	92.9	95.3	92.0
	50	94.2	91.5	94.4	92.6	95.6	93.3
10	10	95.1	93.6	95.7	94.8	95.6	93.5
	20	94.1	94.8	95.6	94.6	95.3	90.7
	50	94.2	93.2	94.0	94.5	94.8	91.4
20	10	94.8	95.9	94.0	94.3	95.9	91.8
	20	94.4	94.2	94.6	93.8	94.4	92.9
	50	94.1	94.1	94.8	92.9	93.1	89.7
50	10	94.8	94.4	94.7	95.5	94.1	91.6
	20	94.2	94.5	94.8	95.3	95.1	90.3
	50	94.2	93.2	95.9	93.4	95.0	92.9

^aSimulation method 1 = compound symmetry; method 2 = Toeplitz; method 3 = random intercepts and slopes
ICC intracluster correlation

ISSN: 1745-6215