On Thu, 20 Mar 2003 21:10:37 GMT, Syrahz Derzai <[EMAIL PROTECTED]>
wrote:

> Rich Ulrich <[EMAIL PROTECTED]> wrote:
> >  On 19 Mar 2003 10:58:15 -0800, [EMAIL PROTECTED] (dennis roberts) wrote:
> > > for the above statement you are suggesting that the p value gives you 
> > > information on the "strength" of evidence for the possible fact that the 
> > > effect is NOT absolutely 0 ... 
> >  
> >  No, no, no.   You don't have to invent your naive straw-man for
> >  what I was "suggesting", since  I was explicit in my NEXT line.
> >  
> >     "- the strength of the evidence?  Whether   *chance*  
> >     might be sufficient to account for what has been observed?"
> >  
> >  Not 'the fact that the effect is NOT absolutely 0'.  
> >  But
> >  'whether the occurrence is unlikely by chance.'
> 
> But then it requires a leap of faith to attribute observations to
> something else rather than chance merely because it is "unlikely by
> chance".

Yes and No.

Controlled studies are set up precisely so that
we can attribute something-in-particular.  Study
manipulations are not always perfectly "blind"  to
participants and to raters, etc.,  but the randomization
removes the most notorious  of the outside influences.
 - That is, there are certain biases that scientists do
know about; so it is poor practice to ignore them, if
there is way we know to do better.  (And we do have
to take into account the new problems that we might
learn about.)


Certainly, in non-controlled studies, there's a 'leap
of faith' -- What is it that connects the hypothesize
item to the outcome, besides some number?  That's
why an education leading to research needs to 
emphasize the philosophical roots.  

'Correlation is not causation.'    
The better scientists are the ones who
know this in their bones, so they are more careful 
to describe and account for whatever else might exist. 

Then, as Herman might say, there is a decision to
make -- What are the consequences?
Sometimes, there is mainly an up-side in plunking
for slight evidence, or very little evidence at all.
(But this branches into wider, metaphysical issues,
so I will stop here.)

-- 
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