I ran into this problem myself. It appears that -bootstrap- uses
-_pctile- (a built-in programmer command) to calculate the percentiles
used to define the limits of the bootstrap percentile and
bias-corrected percentile confidence intervals. By default, -_pctile-
uses a different means from -centile- to calculate the percentiles. If you type
. _pctile bs1, p(2.5 97.5)
. return li
instead of
. centile bs1, centile(.025, .975)
then you should be able to reproduce the bootstrap bias-corrected
limits. Conversely, typing
. _pctile bs1, p(2.5 97.5) altdef
should yield the same percentiles as -centile-.
What I cannot tell you is whether the default -pctile- method or the
-centile- method is more compatible with Efron's confidence interval
estimation procedure or whether it makes any real difference which
percentile method is used.
--
Jim
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
