Hello all,
First of all thanks to all of you who got back to me on my original query
about statistical power.
I was wondering if any of you would mind letting me know what you think
about the following reasoning. I found a book "Design Sensitivity" by mark
W Lipsey, that has easy to use tables for calculating statistical power.
It taught me that what I need to know is (I think);
a) What I want my alpha to be (this is traditionally 0.05)
b) What I'd like my power to be (seems to be a tradition of accepting 0.8
and over)
c) What I'd like my effect size to be (according to Cohen, 0.2 is small,
0.5 is medium and 0.8 is large) - This area seems like a minefield, but
because my research is exploratory rather than experimental I was happy to
go with Cohen's averages. I don't know if this is ill advised or not.
When I used the tables in Lipsey I found that if I was doing a t-test I'd
need about 26 people in each group to achieve a power of 0.8 and to be
able to detect "large" effects, I'd need about 65 in each group to be able
to detect "medium" effects (power 0.8) and I'd need about 400 in each
group to be able to detect "small" effects.
It was quite a sobering exercise on my behalf! I would really appreciate
it if anyone could let me know if they think I have came to a reasonable
conclusion regarding my sample size.
Thank you in advance for your help,
Brian
---------------------------------------------------
Dr. Brian McMillan
School of Psychology
University of Leeds
Leeds, LS2 9JT UK
phone: 0113 3435714
fax: 0113 2431751
www.psyc.leeds.ac.uk
|