BCa bootstrap confidence levels

[Bram Hogendoorn]

Hi all,

I am bootstrapping a program and would like to obtain the confidence intervals for various levels, for example, 95% and 99%. In a normal situation this is easy. You simply ask to report the results anew:

bootstrap, level(95): reg y x
bootstrap, level(99)

However, I would like to do this for the BCa bootstrap. Let's repeat:

bootstrap, level(95) bca: reg y x
estat bootstrap, all

bootstrap, level(99)
estat bootstrap, all

This seems not to work, because the -estat bootstrap- reverts to the initial level. Is it necessary to run the entire bootstrap again, to obtain the various BCa confidence intervals?

[Shiraz Johnson]

would it work to do something like foreach ci in 95 99 {bootstrap, level(`ci') etc...} ? I don't know if this would help with your problem but it would make using multiple confidence intervals simpler imo

[Leonardo Guizzetti]

The two components of the BCa computes are stored in -e(z0)- and -e(accel)-, so if you really wanted you can use these in combination with the bootstrap samples to calculate a different confidence interval. It's probably just easier to re-run the bootstrap estimation with the new desired limit than to manually perform the calculation yourself. I think Stata doesn't allow this because the bootstrap samples are discarded by default, and based on a test, doesn't use that information anyhow.

[Bram Hogendoorn]

Thank you, Leonardo. It seems that using the -z0- and -accel- estimates are the only way to avoid running the bootstrap again.