R help - Apr 2003 - bca ci's and NaN's in boot.out

Dear All,

I am trying to use the bca.ci function on a boot.out object which consists
a few NaN's and I want to ignore those NaN's, and get a ci
only for the "normal" values. 
boot.out$t has R number of values for 3000 different statistics, so when I
use boot.ci(boot.out, index=i) and i happens to be a column in boot.out$t
with some NaN's in there I get an error message. I tried na.omit for the
problematic columns, but bca.ci needs a boot.out object so I cannot feed
in a single column. The number of NaN's varies, and for some statistics 
there isn't any.

Interestingly the standard errors that the boot() calculates and returns
there is an automatic NA exclusion.
(A side question, in S+ the se's are saved in boot.out$statistics$SE, but
in R boot.out I couldn't find them. Are they saved somewhere in boot.out?)

Any idea would be greatly appreaciated!

Kati

___
Katalin Csillery
Division of Biological Sciences
University of Montana, Missoula MT 59801
Phone: 406 243 6106, E-mail: csillery at selway.umt.edu

Katalin  -

I think that calling  bca.ci()  with argument  type="reg"
in addition to the arguments you've used already MIGHT do
the trick.

This argument will be passed to  empinf() and it ought
to guarantee that the method  empinf.reg() is used.  That
method looks as though it will tolerate NAs.  I've never
used any of this stuff, so I don't know.  I'm just looking
at the function definitions for  bca.ci(), empinf() and
empinf.reg(), inf.jack(), usual.jack(), positive.jack().
These last four functions are called by  empinf() under
various conditions.

For more details, I encourage you to look at the function
definitions yourself, and try to trace exactly which ones
will be called for the particular set of arguments you are
using.  To see a function definition, type the name of the
function at the command prompt without any parentheses ()
after the function name.

It's quite possible that you are testing the package with
a combination of arguments that has never been used before,
and that some default setting isn't quite doing its job.
I'm afraid I will leave the more detailed investigation to
you.

HTH  -  tom blackwell  -  u michigan medical school  -  ann arbor  -

On Thu, 24 Apr 2003, Katalin  Csillery wrote:
> Dear All,
>
> I am trying to use the bca.ci function on a boot.out object
> which consists a few NaN's and I want to ignore those NaN's,
> and get a ci only for the "normal" values.
>
> boot.out$t has R number of values for 3000 different statistics,
> so when I use boot.ci(boot.out, index=i) and i happens to be a
> column in boot.out$t with some NaN's in there I get an error
> message. I tried na.omit for the problematic columns, but bca.ci
> needs a boot.out object so I cannot feed in a single column.
> The number of NaN's varies, and for some statistics
> there isn't any.
>
> Interestingly the standard errors that the boot() calculates
> and returns there is an automatic NA exclusion.
>
> (A side question, in S+ the se's are saved in boot.out$statistics$SE,
but
> in R boot.out I couldn't find them. Are they saved somewhere in
boot.out?)
>
> Any idea would be greatly appreaciated!
>
> Kati
> ___
> Katalin Csillery
> Division of Biological Sciences
> University of Montana, Missoula MT 59801
> Phone: 406 243 6106, E-mail: csillery at selway.umt.edu