On 19 Nov 2003 08:53:23 -0800, [EMAIL PROTECTED] (Paul Smith) wrote:

> Dear All
> 
> Some books say that the coefficient of variation can only be used if
> the random variable is positive. I would like to know why cannot one
> extend the use of the same coefficient with not necessarily positive
> random variables but with positive mean. Could somebody here please
> help me?
> 

To my way of thinking, the CoV  is mainly of value when
the underlying data are log-normal.  So, yes, the values
will be positive.  - I do not recall reading about it that way, but
what you describe sounds  like admirable  advice.

That is, the SD  needs to *stay*  proportional to the mean, 
or there is hardly any value in talking about its value, right?

You could compute a ratio and call it a CoV.   But will you have
anything worthwhile, if you have values that are negative, and
not natural amounts of something?

So, I guess the reason that you can't, is if  your journal 
editors and grant reviewers are apt to see it that way, too.

If someone ever has a good use for a CoV  on number-line
data from some source,  I figure they could  argue for the 
exception for that source.  This is a practical matter.


-- 
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