Notice that because like all percentile intervals, this assumes that when is rescaled to normal, its mean bias and median bias are identical.

In which case:

When is unbiased we would expect (on average) that half (0.5) your bootstrap estimates are more than your sample estimate, then p_{b} = p(^{*} > ) = 0.5

On the other hand, if was positively biased, we might expect (say) p_{b} = p(^{*} > ) = 0.6