Hi everyone,
I'm running a regression model (multiple predictors) with bootstrap resampling.
I asked for Bias-Corrected and Accelerated confidence interval (as usually preferred to a percentile based one).
However, in one of the coefficient the BCa 99% CI contains the 0, while the p-value is inferior to the significance level (p
Calculate the BCA for a range of alpha levels and plot alpha vs the confidence limit (upper or lower as appropriate) and interpolate this table of numbers at the value of your hypothesized value. Graphically you are finding the alpha where the plot of alpha vs limit crosses a vertical line at the hypothesized value.
Quite analytically consuming but such a brilliant solution!
You can even test the interpolated value and eventually fine-tune it to the desired decimal with some more runs!
Yes, you're absolutely right!
But having a sort of "percentile-based" p-value also when BCA is used, it makes also the not expert reader more able to easily understand the results even if it's not experienced in bootstrap.
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero