R help - Apr 2004 - bcanon bug???

> in some cases the bootstrap estimates for the
> highest confidence level are obviously wrong:
> 0.05/x    -1.33318711781158
> 0.025    -1.09386961626344
> 0.05    -0.868768454819694
> 0.1    -0.652649120212949
> 0.16    0.0453058935697044
> 0.84    0.104645679223495
> 0.9    0.213385388775905
> 0.95    0.368838243723381
> 0.975    0.615868703042345
> 1-0.05/x    -1.33318711781158
> where the estimate for the highest confidence level is simply 
> the copy of that for the lowest.
Which is not quite so obviously wrong:
it is perfectly right if x = 0.1

You did not tell us the value of x.

This would mean an entry

0.5   -1.333

which does not quite fit into the rest of your series,
but maybe what you are bootstrapping is not quite as 
continuous as it seems?

  Lutz

> You did not tell us the value of x.
:-)

x is an even integer (6,8,10). With x=6, you get

alpha=0.008333333,[default values],0.9916667

This is the reason for why the result I get sometimes is so odd.
>but maybe what you are bootstrapping is not quite as
>continuous as it seems?
Sorry for not being specific. I guess the statistic I'm testing is quite
continuous: I'm actually bootstraping the difference in the loglikelihood of
two nonnested model, where the dependent is always the same one, but the
independent is transformed in different ways (raw "linear" value, log
value,
or power transformed value). As I'm comparing more than two transforms it
comes the need to adjust the confidence levels (I might also include other
transforms with a perceptually relevant meaning).

    Bruno