R devel - Aug 2009 - power.t.test (PR#13891)

Full_Name: Michael J. Lew
Version: R version 2.9.1 (2009-06-26)
OS: OS X
Submission from: (NULL) (210.49.195.149)


The default function power.t.test gives inaccurate values, particularly with
large sig.level.
There are two reasons for me to propose that power.t.test is inaccurate: 

1. When sig.level approaches 1, so should the power. However,
power.t.test('n'=5,'d'=1,'sig.level'=1) yields a power
of 0.9430769. I've
compared it to the equivalent function in the 'pwr' package, pwr.t.test,
and
while power.t.test gives a smooth function with respect to sig.level, it
deivates systematically more and more as sig.level is increased.
pwr.t.test('n'=5,'d'=1,'sig.level'=1) gives the correct
power of 1.

2. I've run empirical simulations and found that the power predicted by
pwr.t.test is correct and so it is clear that power.t.test is inaccurate.

At 00:35 14/08/2009, michaell at unimelb.edu.au wrote:
>Full_Name: Michael J. Lew
>Version: R version 2.9.1 (2009-06-26)
>OS: OS X
>Submission from: (NULL) (210.49.195.149)
>
>
>The default function power.t.test gives inaccurate values, particularly with
>large sig.level.
>There are two reasons for me to propose that power.t.test is inaccurate:
>
>1. When sig.level approaches 1, so should the power. However,
>power.t.test('n'=5,'d'=1,'sig.level'=1) yields a
power of 0.9430769. I've
>compared it to the equivalent function in the 'pwr' package,
pwr.t.test, and
>while power.t.test gives a smooth function with respect to sig.level, it
>deivates systematically more and more as sig.level is increased.
>pwr.t.test('n'=5,'d'=1,'sig.level'=1) gives the
correct power of 1.
I think you need to set strict=TRUE when your example gives a power of 1 here.

>2. I've run empirical simulations and found that the power predicted by
>pwr.t.test is correct and so it is clear that power.t.test is inaccurate.
>
>______________________________________________
>R-devel at r-project.org mailing list
>https://stat.ethz.ch/mailman/listinfo/r-devel
Michael Dewey
http://www.aghmed.fsnet.co.uk