This may help. Jacob Cohen wrote a book a while back called "Statistical
Power for the Behavioral Sciences" (now published by Lawrence Erlbaum
Associates, 1988). He defined the term "effect size", which for the two
group t-test is the difference between the means divided by the pooled
standard deviation: (m1-m0)/SDp. (In other words, the size of the
difference between the means relative to the SD.) He gives the ranges and
his logic behind what he considers small, medium, and large effect sizes.
Small EF = about .1 to .2; medium EF= about .35 to .5; and large EF =
anything greater than .75. So, your pathologist is certainly only looking
for "colossal" effect sizes.
Cohen states (and I think we'd all agree) that the study sample size should
be large enough to detect the SMALLEST effect size that is reasonably or
logistically possible.
If your pathologist has no clue of the size of the effect of COX2 that he is
looking for, then the most prudent approach is to use the largest sample
size that he has funds to cover (e.g., 20). If his primary interest is in
designing a study that is inexpensive, tell him that by looking for 2
standard deviations, he only needs to run 4 animals per group.
Paul
-----Original Message-----
From: Ricardo Ovaldia [mailto:[email protected]]
Sent: Tuesday, February 18, 2003 12:11 PM
To: [email protected]
Subject: st: Sample size for ttest - (am I wrong?)
Dear all,
I feel like such an idiot and I need some help on this
non-Stata problem. I was in a meeting this morning
with a pathologist that is submitting a proposal to
conduct an experiment to compare the concentration of
a biochemical (Cox-2) in two different tissue cell
lines. He wants to compare mean concentrations so I
suggested a t-test, after a monotone transformation if
needed. He agreed and proceeded to tell me that
because of monetary constrains, he could only have 20
observations per group and wants to know, what is the
minimum difference that he can detect with 80% power?
I told him that I need some idea of the means and SD
of the measurement and he said that he did not know
because this was a new method, and furthermore I did
not need this information because his friend (another
"statistician") told him that with 8 observations per
group he could detect a difference of 1.5 standard
deviations with 80% power. I felt humiliated and
embarrassed. Does anyone know how to do this, or is
this guy wrong?
Thank you,
Ricardo
=====
Ricardo Ovaldia, MS
Statistician
Oklahoma City, OK
__________________________________________________
Do you Yahoo!?
Yahoo! Shopping - Send Flowers for Valentine's Day
http://shopping.yahoo.com
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
