| Following message was posted by Phil Pichotta on September 29, 1998 at 11:24:36: in reply to alternative statistical approaches? posted by Roger PPM on September 23, 1998 at 11:52:55: |
I agree with RogerPPM's comments. I would like to add that we as a statistician need to assess the likelihood that the assumptions will be met and how they would affect the results. For example, a crossover study may be a very efficient design (rather than an alternative analysis) but most designs require an assumption of equal carryover. If this assumption cannot be justified by external sources or internally, then the study will be flawed. A crossover study can also be subject to questions of validity if there are many dropouts. A client may be led to a crossover because it holds the promise of fewer patients and thus be cheaper, but it may be a false hope. How many times do I want to struggle trying to salvage such studies when the key assumptions don't appear to be valid? In some cases you may need to present alternatives but at the same time evaluate their chances of success.Dr. Gardenier commented that the client deserves to be informed if there are ways to get an almost-as-good result for substantially less investment of time, money, or effort. While I agree with that, there are other statements that it would be unethical if the most valid wasn't used in human experiments.
A committee's desire us to be good statisticians is admirable. However, I am still concerned about making these desires into ethical statements.
| Following replies were posted: |