Use of the bootstrap in analysing cost data from cluster randomised trials: some simulation results

Table 6 Control ICC = 0.25, 24 clusters of size 25 per arm.

How does the ICC change as a result of the intervention?		Between-cluster distribution combination¹	Rejection Rate (%)²
			Huber-White		Single Bootstrap		Double Bootstrap
			Lower	Upper	Lower	Upper	Lower	Upper
No change	-	N, N	2.76	2.73	3.31	3.27	3.12	3.06
		LN, LN	2.28	2.32	5.77	5.55	5.30	5.16
		N, LN	1.38	4.08	3.42	4.43	3.00	4.30
Double		N, N	2.84	2.66	3.30	3.20	2.91	3.09
		LN, LN	2.03	2.17	5.55	5.81	5.07	5.25
		N, LN	1.41	4.13	3.60	4.45	3.18	4.21
		N, N	2.63	2.74	3.12	3.22	2.88	2.98
		LN, LN	0.84	5.38	3.84	7.90	3.43	7.39
		N, LN	0.71	6.53	3.25	6.20	2.94	5.89
Halve		N, N	2.56	2.65	3.16	3.19	2.99	2.84
		LN, LN	1.98	2.38	4.96	5.32	4.57	4.70
		N, LN	1.47	4.21	3.45	4.41	3.12	4.11
		N, N	2.56	2.79	3.09	3.25	2.91	3.10
		LN, LN	4.26	1.39	6.70	4.54	6.16	4.15
		N, LN	1.79	3.36	3.24	3.73	2.95	3.57

¹ Control arm is first, followed by intervention arm. N, N denotes normal distribution in both arms; LN, LN denotes lognormal distribution in both arms; N, LN denotes normal distribution in control arm and lognormal distribution in intervention arm
² Entries in bold indicate where double bootstrap method achieved rejection rate closer to nominal 2.5% than Huber-White method.

ISSN: 1472-6963