posted and e-mailed.

On 19 Nov 2003 13:56:23 -0800, [EMAIL PROTECTED] (Dennis Roberts) wrote:

[ snip, the original Q]
> 
> some on various lists have claimed that the coefficient of variation ... 
> which is usually expressed as a % of the following ratio: SD / mean ... is 
> a useful tool
> 
> i have NEVER been convinced that this is so
[ snip, rest]

I was surprised when I first met the CoV  as a measure
of reliability.  I thought the folks must be kidding or mistaken.
If I remember right, I said something embarrassingly ignorant.

Then they pointed out that it was the very efficient and effective 
descriptor of measurement accuracy for, say, chemical assays
and bioassays:  Trace amounts are measured with 10% accuracy, 
or 2% accuracy, or whatever.  That is  *the*  conventional  usage.

So, the CoV  is a very popular measure for  a lot of people --
even though it was hardly mentioned in the biostatistical training
that I received, and I wouldn't be surprised if it is missing entirely
from books on psych-statistics. 

As I  implied in my post of yesterday, it is especially relevant 
when taking the logarithm is relevant -- perhaps  I should have 
stated it that way.  I said something about "log-normal"  but that
is not the important facet. 


-- 
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
"Taxes are the price we pay for civilization." 
.
.
=================================================================
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