Posted by
David Hitchin on
Aug 28, 2006; 7:31am
URL: http://spssx-discussion.165.s1.nabble.com/significantly-not-significant-tp1070571p1070585.html
Rohan Lulham <
[hidden email]> asked:
>
> This maybe be a naive question, but was wondering how to test if the
> difference between to means are significantly similar (e.g.
> "significantly" not significant differences).
There are specific methods for this very problem, which you will find in
the literature under the heading "bio-equivalence". These tests are
used, for example, when other companies are making a pharmaceutical
which is licensed and has specific published qualities. Your company
finds a way to make the same drug via another method, and the question
is whether yours is sufficiently similar to the standard to be marketed
as (medically) the same thing.
The fundamentals are that FIRST you decide how similar these should be
(within 1%, 2% or whatever value you choose of the standard on a
particular biochemical or other test) and you decide how sure you must
be that your version doesn't fall outside of that range. That enables
you to determine the sample size that you need, and when you have
completed the test the confidence interval for the mean of your data
should lie within the limits which you have decided for acceptability.
Note that this experiment was designed BEFORE the data values were
collected.
If you collected data for another purpose and happened to notice that
the means of two samples were very similar, then a post-hoc test would
need very different methods.
David Hitchin