Power calculation using pwr.t.test()

[Jokel Meyer]

Dear R experts,

I have conducted a power calculation in order to estimate the number of
subjects needed to detect an effect size of d=0.28 (cohen's d) for a
difference between two independent groups (alpha level should be 0.05 and
the effect should be detected with 80% probability).
The results from the code below indicates that I would need n=400 subjects
(200 in each group). This is seems so incredibly high that I mistrust my
results & wanted to ask whether I miscalculated n?

library(pwr)
pwr.t.test(d=0.28,sig.level=0.05,power=0.8,type="two.sample",alternative="two.sided")

Many thanks for you help!
Jokel

        [[alternative HTML version deleted]]

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

[Uwe Ligges-3]

On 24.06.2012 17:41, Jokel Meyer wrote:

> Dear R experts,
>
> I have conducted a power calculation in order to estimate the number of
> subjects needed to detect an effect size of d=0.28 (cohen's d) for a
> difference between two independent groups (alpha level should be 0.05 and
> the effect should be detected with 80% probability).
> The results from the code below indicates that I would need n=400 subjects
> (200 in each group). This is seems so incredibly high that I mistrust my
> results & wanted to ask whether I miscalculated n?
>
> library(pwr)
> pwr.t.test(d=0.28,sig.level=0.05,power=0.8,type="two.sample",alternative="two.sided")

Good luck this is an open source software where you can easily review
the code and check if it is right. If it is not, I'd propose to send a
patch to the package maintainer.

Best,
Uwe Ligges

> Many thanks for you help!
> Jokel
>
> [[alternative HTML version deleted]]
>
> ______________________________________________
> [hidden email] mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.

[David Winsemius]

On Jun 24, 2012, at 11:41 AM, Jokel Meyer wrote:

> Dear R experts,
>
> I have conducted a power calculation in order to estimate the number  
> of
> subjects needed to detect an effect size of d=0.28 (cohen's d) for a
> difference between two independent groups (alpha level should be  
> 0.05 and
> the effect should be detected with 80% probability).
> The results from the code below indicates that I would need n=400  
> subjects
> (200 in each group). This is seems so incredibly high that I  
> mistrust my
> results & wanted to ask whether I miscalculated n?
>
> library(pwr)
> pwr
> .t
> .test
> (d
> =
> 0.28
> ,sig.level=0.05,power=0.8,type="two.sample",alternative="two.sided")

Your reaction is typical for inexperienced researchers who conduct  
power analyses. An effect size of .28 is considered a "small effect  
size" so it should not seem unreasonable that you would need 200 per  
group.  As a quick check I Googled for an online sample size  
calculator and the first one I used reported the need for 202 per  
group for a two sided test at those levels.

--

David Winsemius, MD
West Hartford, CT

______________________________________________
[hidden email] mailing list
https://stat.ethz.ch/mailman/listinfo/r-help
PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
and provide commented, minimal, self-contained, reproducible code.