On 16 Jan 2003 10:23:59 -0800, [EMAIL PROTECTED] (Conn, Judith) wrote:

> I don't know the specifics but I thought you had to take into account that
> having multiple tests of hypotheses on the same data and would get false
> positives cause multiple tests.  Or am I thinking of something else?  Judy
> Conn
[snip:  my comment, which quoted another comment. Neither
comment reveals to me the immediate salience of this
comment about multiple tests.]

Judy, and everyone,

I occasionally top-post when I write e-mail, but it does
not work well in writing to net-groups.  What works most
often, it seems to me, is the Quote followed by the Comment.

It also does not work very well when the original poster
has substance written into the Subject line which is not 
repeated in the text -- the Subject line was easy to 
overlook or to ignore, in my experience,  even when my
Newsreader wrote it out with the message.  These days,
I have it an inch or two above the text that is supposed to
be relevant.

Anyway, it seems to me that Judy's comment is 
directed to the Subject line.

> Subject: Re: validity of secondary outcomes with very large sample size

Yes, one ought to take multiple testing into account.

But if I piggyback my Question onto your expensive data 
collection,  and my 'study'  questions are largely unrelated 
to yours, can't I start with a fresh 5%?  - folks do this, and
I can't object.

And, taking it further, each of us 'piggybackers' (including
the original investigator)  may want to use a fresh 5%  level 
for two or three other questions that are "unrelated"  to some
extent ....  I think it is okay for a first one, "totally unrelated",
but it is not okay for #100  if I don't say *something*  about
multiplicity.   

One attempt to control for  *excessive*  multiplicity is to
insist that the *real*  questions have to be documented in
the original grant application.  (Especially, it is hazardous
to are safe-inference making if the hypotheses are generated
by the peeks at 1000s of data items.)  The expectations should
be laid out in advance, along with (if possible) the hierarchy
of decision making:  For instance, if Overall-test #1  does 
not 'reject',  then the Subtests 1a, 1b, 1c   are going to be 
regarded as  exploratory and doubtful.

(This strategy can be undermined if interim research changes
the scientific priorities, in the years between funding and 
publication.)

(I suspect that PhDs  are more successful than MDs or 
other clinicians, when it comes to expressing their *doubts* 
about  'marginal'  clinical findings.)

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to