Table 3.

Bias of estimated treatment means and their difference (averaged over nodes of each tree) and coverage probabilities of naïve and bootstrap 95% intervals, based on 1000 simulations trials of nontrivial trees, with 100 bootstrap iterations per trial. r is the number of replicates of a 3⁴ design.

		Bias of naïve estimates of means and difference			Coverage probabilities of 95%
					naïve intervals			bootstrap intervals
r	Expt	μ(t, 0)	μ(t, 1)	d(t)	μ(t, 0)	μ(t, 1)	d(t)	μ(t, 0)	μ(t, 1)	d(t)
2	M1-Gi	6.8E-3	−2.2E-2	−2.9E-2	0.821	0.811	0.818	0.892	0.955	0.934
	M1-Gs	4.4E-3	−1.9E-2	−2.3E-2	0.819	0.800	0.857	0.907	0.952	0.935
	M2-Gi	6.8E-3	−2.0E-2	−2.6E-2	0.835	0.846	0.836	0.937	0.947	0.941
	M2-Gs	3.5E-4	−1.5E-2	−1.5E-2	0.871	0.861	0.907	0.953	0.965	0.942
4	M1-Gi	3.0E-2	−1.6E-2	−1.9E-2	0.880	0.874	0.889	0.903	0.972	0.957
	M1-Gs	3.7E-3	−1.5E-2	−1.9E-2	0.869	0.862	0.888	0.916	0.967	0.955
	M2-Gi	1.1E-3	−7.5E-3	−8.6E-3	0.896	0.915	0.911	0.966	0.967	0.963
	M2-Gs	−3.8E-3	−9.4E-3	−5.7E-3	0.888	0.913	0.916	0.968	0.973	0.950