Table 4.

Performance of tests in analyzing 10000 datasets, case 2 data

Model	Proportion of datasets for which null hypothesis is rejected
Model	t test (individual observations)	Wilcoxon (individual observations)	t test (means)	Wilcoxon (means)	LMM	GEE	Datta and Satten method
1	0.001	0.002	0.052	0.049	0.051	0.079	0.042
2	0.202	0.197	0.626	0.621	0.718	0.738	0.469
3	0.005	0.002	0.039	0.051	0.047	0.062	0.042
4	0.191	0.783	0.438	0.647	0.446	0.526	0.877
5	0.005	0.006	0.048	0.039	0.050	0.116	0.027
6	0.420	0.405	0.709	0.645	0.868	0.895	0.363
7	0.007	0.006	0.028	0.040	0.038	0.091	0.027
8	0.603	0.919	0.730	0.722	0.796	0.843	0.800

Table summarizes the results (proportion of datasets for which the null hypothesis is rejected) when clusters contain observations from both groups; data were obtained for 10000 simulated datasets for each of the eight models described in Table 1 and for each of the analysis methods in Table 2. A significance level of 5% (i.e. p < 0.05) was used to determine whether to reject the null hypothesis in all cases.