[R] Repeated measures 2-way anova -- robustness question

[Alpert, William]

I'm a journalist, wondering what questions to ask about a study that
contrasted the impact on serum cholesterol of two drugs.  This was a
40 dog study: 5 treatment blocks of 4 dogs each, randomized to: a
control block, two blocks at different doses of drug1 and two at
corresponding doses of drug2.  Analysis was 2-factor repeated measures
ANOVA on treatment group and sampling time.  Linear contrast tests
looked at differences between the drugs, and between doses, and found
p<.001 for cholesterol differences between one dosage block of drug1,
the control block and the similarly-dosed block of drug2.

Data on one of the drug blocks was actually just 3 dogs, because one
dropped out.  One critic has also claimed that the population variance
of cholesterol levels in these dogs is so wide as to cast suspicion on
estimates based on the sample variance in these small blocks.

Are there any repeated-measure ANOVA mavins who could suggest whether
the approach is sufficiently robust to violations of assumptions to
preserve the validity of those impressivly low p-values ?

Bill Alpert
william.alpert at barrons.com