W have a fairly straightforward experiment comprising six treatments: an
untreated control plus five doses of a chemical. Each treatment is
replicated 20 times, but unfortunately we have some missing values.
Some treatments have no missing values, whilst others have up to four.
The William's test in SAS, which apparently is only approximate unequal
replication, gives us a p value of exactly 0.0500 for one of the doses,
and we would really like to know what the outcome would be if we carried
out an exact test. Having studied the SAS manual and one of William's
papers, we have come to the conclusion that writing a program to handle
the case of unequal replication will me a major undertaking. So, can
anyone help me by putting me in touch with a person, or software
product, that has already managed to do this. A google search has
unearthed nothing, so we are not sure that it's possible to do this. I
realise that there are other tests that we could use, but circumstances
beyond my control are forcing me towards William's test.
Sincerely,
Peter Chapman
|