Re: Effect Size and Sample Size

On Thu, 17 Jan 2008 12:33:29 -0800 (PST), beginner1.mat@xxxxxxxxxxx
wrote:

Thank you all for the very informative responses! I guess I do still
have the same question. It was posed that I should first determine
the effect size I wish to detect and then determine the sample size.
This is what brought up the original post in the first place. I was
determinig my effect size using Cohen's d - (difference between groups/
pool sd), where the data was based off of a N=4 (2 per group).

Okay. That is not really "reasonable".

If you don't have, historically, 3 dozen measurements
of what-ever-it-is, are you ready to do an experiment
with it?

Use what you know about the measurement, and what
constitutes "usual variations" in it. Or *ought* to constitute
"usual variations." If you don't have any idea, then -- for
experiments that I think of -- you are not ready to do any
formal design. Pilot.


Because I did not think it would be appropriate to look at weather the
data was normally distributed (i.e., kurtosis, skewness) as it was
only 2 per group, I was wondering if this was indeed the case?
Perhaps the answer lies somewhere in this post, but it wasn't very
clear to me. Given that I have the data of N=4, would anyone argue
that I should not use Cohen's d to calculate effect size? I think the
argument would be in that because there are only 2 per group, the true
mean and true variability in the data is not established, thus may
affect the effect size calculation. I would like to collect more
data, and assuming the data fits the assumptions of a t-test, I would
run a t-test on a larger sample size. But, I would like to calculate
a sample size to do that, and since I do not have a larger sample to
pull means and sds from, I am trying to use the N=4 that I currently
have. Any recommendations?

--
Rich Ulrich
http://www.pitt.edu/~wpilib/index.html
.