Statistics Glossary → BCa bootstrap confidence interval
BCa bootstrap confidence intervalconcept
A bias-corrected and accelerated confidence interval that shifts the percentile limits to account for bias and skewness in the bootstrap distribution.
The percentile bootstrap interval takes the α/2 and 1 - α/2 percentiles of the statistics computed from thousands of resamples. It works well when the bootstrap distribution is symmetric and unbiased around the true value. Efron's BCa method adds two corrections: a bias correction z₀ (computed from the share of bootstrap estimates that fall below the original estimate) and an acceleration a (which measures how the standard error of the statistic changes with the parameter value and is usually estimated by the jackknife). Together they shift the percentiles used, for example from 2.5% and 97.5% to something like 1.8% and 96.4%.
For skewed statistics (variances, correlations, proportions) the BCa interval has theoretically better coverage than the percentile interval. On the other hand, in very small samples z₀ and a can be estimated unstably and the interval can shift unexpectedly; several thousand bootstrap replicates are usually needed to estimate the extreme percentiles reliably. Simulation studies of mediation analysis have shown that BC and BCa intervals can inflate the Type I error rate, while the percentile interval is better balanced.
An everyday example
In one neighbourhood, monthly electricity bills of 25 households are very skewed: most are 400-600 TL, but a few homes exceed 3000 TL. The bootstrap distribution of the mean bill is right-skewed too. The percentile interval gives 520-1180 TL, while the BCa interval accounts for the skewness and shifts the limits slightly upward to 560-1290 TL.
In MerQur
In the Bootstrap CI analysis it is chosen as the primary interval with the 'CI method' option (Percentile, BCa); the other intervals are still reported in extra rows. It appears as 'Bootstrap CI type' (Bias-corrected BC / Percentile / BCa) in mediation analysis and as the 'BCa confidence interval' check box in PLS-SEM; in correlation analysis 'Extra outputs: Bootstrap CI' gives percentile and BCa intervals for each pair.
Analyses where this concept appears
Bootstrap CICorrelation AnalysisMediation AnalysisPLS-SEM (Partial Least Squares Path Modeling)