I computed the bootstrap estimate of bias as the mean of the bootstrap distribution minus the estimate for the original data. Then, the bootstrap bias-corrected estimator was computed as the original estimator minus the estimate of bias. The bias-corrected and accelerated (BCa) confidence interval was also computed manually, but equivalent results were obtained via the
I noticed that in some cases, the bias-corrected point estimate is situated outside the BCa interval. Is this possible? How are the bias-corrected point estimate and BCa confidence interval related? Is it because the way bias is computed differs?