[R] inv.logit() from "boot" ..[was "proportions confidence.."]

Martin Maechler maechler at stat.math.ethz.ch
Tue Jul 13 14:54:43 CEST 2004

>>>>> "Spencer" == Spencer Graves <spencer.graves at pdf.com>
>>>>>     on Mon, 12 Jul 2004 10:44:25 -0700 writes:

    Spencer>       Please see: Brown, Cai and DasGupta (2001)
    Spencer> Statistical Science, 16: 101-133 and (2002) Annals
    Spencer> of Statistics, 30: 160-2001

    Spencer>       They show that the actual coverage
    Spencer> probability of the standard approximate confidence
    Spencer> intervals for a binomial proportion are quite poor,
    Spencer> while the standard asymptotic theory applied to
    Spencer> logits produces rather better answers.

    Spencer>       I would expect "confint.glm" in library(MASS)
    Spencer> to give decent results, possibly the best available
    Spencer> without a very careful study of this particular
    Spencer> question.  Consider the following:

    Spencer> library(MASS)# needed for confint.glm
    Spencer> library(boot)# needed for inv.logit

    Spencer>  <.............>

well, did you ever look at the definition of inv.logit() ?

doing it for you:

      inv.logit <- function(x) plogis(x)

hence I'd think "boot" is not really needed.

And, BTW, consequently the  logit() function in the 'boot'
package is the same as  qlogis() 
[and could well be defined that way instead of slightly more complicated].

I'll add a \concept{logit} to the plogis / qlogis help page and
a \note{} telling this to those who forget..

Martin Maechler, ETH Zurich

More information about the R-help mailing list