. 2006 Dec 5;7:529. doi: 10.1186/1471-2105-7-529

Table 2.

Verification of accuracy of confidence intervals

	k = 1.960	k = 2.576
\| $\hat{Δ}$ _triplet- Δ\| > k·σ( $\hat{Δ}$ _triplet)	0.95129 ± 0.00067	0.99062 ± 0.00030
\| $\hat{Δ}$ _pairwise- Δ\| > k·σ_bootstrap( $\hat{Δ}$ _pairwise)	0.9511 ± 0.0020	0.99001 ± 0.00091
\| $\hat{Δ}$ _pairwise- Δ\| > k·σ( $\hat{Δ}$ _triplet)	0.94641 ± 0.00070	0.98896 ± 0.00032
\| $\hat{Δ}$ _pairwise- Δ\| > k· $\tilde{σ}$ ( $\hat{Δ}$ _pairwise)	0.94808 ± 0.00069	0.98953 ± 0.00032
\| $\hat{Δ}$ _pairwise- Δ\| > k·σ_ind( $\hat{Δ}$ _pairwise)	0.98137 ± 0.00042	0.99774 ± 0.00015

Comparison among the different methods to estimate the variance of the two estimators $\hat{Δ}$ _tripletand $\hat{Δ}$ _pairwise, resulting from a simulation using updated Dayhoff matrices over 400,000 proteins triplets, except for the bootstrapping method, based on 40,000 samples. The first column tests the 95% confidence interval, the second the 99% confidence interval.