Leman leman at
Fri Sep 17 12:09:50 EST 1999

On Fri, 17 Sep 1999 14:17:19 +0100, Marcel Hoffmann
<marcel_hoffmann at> wrote:

>have three sets of data each containing 3 values and I want to compare
>these sets two by two to determine if they are significantly different
>from each other. Should I use an ANOVA or Student test. What is for
>instance the minimal number of observations for these tests ?
>Any help would be appreciated.

Strictly speaking, one should not use Student t-test on more than one
pair of data sets. Otherwise, there is a correction co-efficient you
must use in estimating your P values, and I don't remember what it is
(actually, there are 2 or 3 of them -- depending of how safe you want
to feel). ANOVA seems to be "what the doctor prescribed" for your
task. As for the minimal number of observations per group, AFAIK it's
2. You should remember, though that your SEM is inversly proportional
to the SQRT of n, so the more samples the better. I seem to remember
also that it doesn't matter once your n is more than 30. I post this
reply just in case nobody else does. I have to admit, my understanding
of statistics is far from perfect, so please don't flame me if I
interpreted anything wrong.
BTW, there must be some groups in the "sci" hierarchy dedicated to



More information about the Methods mailing list