I think the guideline: B.1. "Offer an analysis of alternative statistical
approaches so the best approach can be taken."

I think this is often not appropriate for at least two reasons.
1) In many cases, when dealing with an internal or external client, the best
analysis is the simplest. This is the one that they can understand and hence
will take action on. A complicated analysis only confuses them or they
perceive it as a "black box" that only you understand, so they don't (fully)
trust the answers and won't (fully) back them. I have seen this happen over
and over again, so that, even though I have a Ph.D in Statistics, even if I do
a complicated analysis to satisfy myself that the answers I am getting are on
the right track, I present only the simplest analysis (primarily graphics) to
the client.

2) In addition, when presenting two "competing" analysis, the client is
tempted to choose the one that suits their interests best. They can't really
judge the analyses on their technical merits, and since you presented both of
them to them, they will assume they are equally valid technically, so they
will take the one that gives them the answer they want.

I believe a better approach is to ask enough questions about the data
collection methods and the process/situation to determine which set of
assumptions for the potential analyses is most appropriate. That, and
consideration of the technical level of the client and how they intend to use
the results should lead you to the most appropriate analysis. I would not let
the client choose between two analyses.

Dr. Gardenier's comment(9/23/98). At least three comments have objected to this guideline for various reasons. The Committee has been pondering the issue. Basically, we feel that the client or employer deserves to be informed if there are ways to get a much better result with only a little more investment, an almost-as-good result using substantially less investment of time, money, or effort. Similarly, the client or employer should understand the projected error bounds of a study and the possibility that results may be indeterminate or otherwise may turn out differently than expected. Accepting those principles, we have to agree that there are circumstances where presentation of alternatives may be unnecessary. We are seeking a way to encourage presentation of alternatives without implying that it is always and everywhere unethical to fail to present them.

• Re: alternative statistical approaches? Phil Pichotta 11:24:36 9/29/98 (1)
  • Re: alternative statistical approaches? John Gardenier 09:34:56 9/30/98 (0)