>
>A doctor writes:
>Hello, are you a statistician? Look, I've just had this paper referred
>back by a journal. They want some p-values. Can I do a t-test?
As I am sure you appreciate, what statistics does is provide a process
which relates a set of numbers which may or may not bear the
relationship you hope to an assumed property of reality and a decision
(I use this in the widest sense to include the decision to adopt a
belief). At either end of this process reside a theory of measurement
and a belief that a decision can be transformed into an appropriate
action. These are not aspects of statistics but they are very much the
province of science.
I would never suggest that statisticians should not take a lively
interest in these areas. However, the suggestion appeared to be made
that statistics might somehow 'solve' this problem in the social
sciences. Also it seemed to be implied that the gap between a theory of
what constitutes reality and the outcome of a measurement process might
be qualitatively different in social sciences from other sciences. This
is not so - the gap is always there.
At least the first two examples you quote are concerned with the
consistency of the truly statistical part of the process. The Bayesian
comment I am not sure about. I haven't really thought deeply about
Bayesian attitudes to this problem, but I had noted that Bayesians
rather neatly avoid the problem of converting decisions to action by
resolving never to let the result of a statistical calculation outside
their own head. There it resides as a personal belief which guides their
actions via whatever processes evolution has provided to link belief and
action. But this is probably not what you are discussing.
******************************************************
Please think before you press the 'Reply' button! Note that if you press
the 'Reply' button your message will go the individual who posted this message
not to the list. With many mailers you will have a 'Reply-to-All' button that
will send automatically to the list address of <[log in to unmask]>. The
Radstats list is set up for public discussion so please be generous with your
thoughts and share them us all.
*******************************************************
|