I must agree with the previous posters that there are far too many
examples of inappropriate, misused, and misleading statistical analyses in
published research. As an ecologist I am more aware of examples in this
field than in medicine, but the impact that this unfortunate situation has
on the direction of future research and the expenditure of funds for
corrective measures in either field can be frightening.
I believe that one way to address this problem is in the peer review
process. If every journal editor were to include for each article, one
reviewer who's primary purpose is to determine if the statistical
analyses were appropriate for the experimental design and data, and were
done correctly, the editor can then give the authors opportunity to
rectify inappropriate or incorrect analyses.
Further corrective measures need to be taken at a more basic level.
While it probably isn't necessary for every researcher to become an
accomplished statistician, a certain level of statistical sophistication
is desriable. For today's students that are tomorrow's research
scientists, a certain amount of course work in statistics should be a
requirement, especially at the Ph.D. level. In addition to basic
classical statistics, this should include some form of instruction in
experimental design, as well as training on when and how to use
non-parametric statistics. For current researchers, they really owe it to
not only themselves, but to the readers and decision makers who will be
reacting to their published research, to attain this same level of
statistical competence, through some form of personal development.
Jesse M. Purvis, Ph.D.
elcoyotero at aol.com