On 15 Mar 2004 19:58:03 -0800, [EMAIL PROTECTED] (Alexei) wrote:

> 
> I have a very simple question related to the combination of two
> results that can, in general, be correlated.
> 
> It's natural to assume that combination C of two measurements A and B
> must be within the interval between the values A and B. Is this always
> true?

What is the definition of "combination"?

It is certainly not true about the simplest 
combination, A+B .  

I don't think a version of a generalized mean of two
numbers would be acceptable, if it popped up a value 
outside the interval.


> On the other hand, two measurements A and B can be the values that are
> higher than the true value T. Then if the process of combination
> reduces the error or uncertainty of the measurement, the combined
> value C will get more close to the true value T and even may get
> outside the interval of the values between A and B.

If you take the average of independent biased estimates,
sharing the same bias, the precision of the estimate
will improve.  But *precision*  is not *accuracy*.  
It will get no closer to T.    (Increasing a sample size may
reduce the bias of a biased estimator, such as, the
variance estimated by dividing by N.  But that was not
the question.)

You are mis-interpreting something; or 
you are reading something written badly; or 
your text is grossly wrong.


This is a distinction between 'reliability' (where measures
are reproducible) and 'validity'  (where measures are right).


> 
> So which statement is the correct one?
> I would really appreciate any help on this problem.
> Thank you.

-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
 - I need a new job, after March 31.  Openings? -
.
.
=================================================================
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