Re: study criticism
- From: Richard Ulrich <Rich.Ulrich@xxxxxxxxxxx>
- Date: Tue, 17 Jan 2006 20:47:47 -0500
On Sun, 15 Jan 2006 23:42:57 +0100, Georg Hintermaier
<georg_hintermaier@xxxxxxxxx> wrote:
> On Sat, 14 Jan 2006 19:13:06 -0500, Richard Ulrich
> <Rich.Ulrich@xxxxxxxxxxx> wrote:
>
> >On Sat, 14 Jan 2006 13:24:53 +0100, Georg Hintermaier
> ><georg_hintermaier@xxxxxxxxx> wrote:
> >
[snip]
RU > >
> >I believe that you are using the word "correlation" in
> >a particular way, and one that does not say anything
> >about the "15 parameters" and their variation between
> >subjects, or variation within subjects. If your 15 parameters
> >are not intercorrelated, I wonder whether you are measuring
> >anything useful at all.
GH >
> He actually started with 40 parameters. Checking them
> for correclation with one he took a priori as a "gold standard".
> Then he eliminated most of them because they did not correlate
> or were redundant or for unspecified reasons.
> I just wanted to emphasize that he does not clame to see a
> trend within all of the subjects to respond in the same way
> to the treatments. But the focus of the study is this one
> person (out of 30) that responded the way he liked it.
It *sounds* like he should be computing a
composite score.
It *sounds* like his winnowing produces fewer
'positive cases' than a 5% level test would produce
by chance.
[snip]
RU >
> >This is a poor experimental procedure, in most cases,
> >to assume that the Placebo effect has to be zero.
> > - The ordinary *reason* to have a placebo is to control
> >for the natural changes in the course of the experiment.
GH >
> Interesting.
> He has two controls: no treatment and placebo.
> I would say the placebo is really to identify effects from treatment.
It sounds like one reasonable analysis might compare the
Treatment to Placebo, while using No-Tr as a covariate.
>
> >
> >> So in one individual of the 30 he finds 4 parameters that
> >> are positive under this criteria.
> >>
> >> Questions:
> >> a) shouldn't there be some sort of multiple testing correction?
> >> The more individuals and parameter he is looking at the
> >> higher is the probability of finding something?
Correct, if you can figure out how. I think you have
a very bad start, and should start over. I think 'correction'
will leave you nothing, from what you have described.
[snip]
--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.
- References:
- study criticism
- From: Georg Hintermaier
- Re: study criticism
- From: Richard Ulrich
- Re: study criticism
- From: Georg Hintermaier
- study criticism
- Prev by Date: Re: Non-Linear Regression
- Next by Date: How to show sum(xi^2) >= (sum(xi))^2
- Previous by thread: Re: study criticism
- Next by thread: Re: probability problem
- Index(es):
Relevant Pages
|
