Any thoughts on the following would be greatly appreciated.
I have been asked to comment on a situation where an experiment has been
designed on the basis of power calculations using one-tailed test to
detect a beneficial effect of a treatment. Whilst the experimenters
would anticipate either a beneficial effect or no effect, the hypothesis
of an adverse effect cannot be totally ruled out. However, an adverse
effect is of no inferential interest; the treatment will be adopted if
there is a significant beneficial effect, whereas if there is no
significant effect or an adverse effect it will not be adopted.
Normally in scientific work I would certainly recommend use of a
two-tailed test where there was any doubt about the direction of the
effect, but in this case there are strong animal welfare reasons for
wishing to maximise power with the minimum number of replicates. Given
these welfare issues and the lack of inferential interest in an adverse
effect, is there a case for allowing the use of a one-tailed test?
Thanks,
Steve Langton
*******************************************************************
Steve Langton E-mail: [log in to unmask]
Central Science Lab Phone: 01 904 462092
Sand Hutton Fax: 01 904 462111
York, YO41 1LZ, UK.
*******************************************************************
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|