On Fri, 27 Feb 2004 08:13:09 -0500, Bruce Weaver <[EMAIL PROTECTED]>
wrote:
[ ... ]
> 
> We must be thinking about different formulae.  I was 
> referring to this computational formula for the sum of 
> squared deviations about the mean (SS):
> 
> 
>       SS = sum[X^2] - (sum[X])^2/n
> 

That  *is*  the formula with the bad reputation.
It should not be used as a computational formula 
for computers -- unless there is some degree of
protection against 'bad data'.

It works okay for a lot of simple datasets; it only
fails on occasional sets of 'real'  data.

The person who wrote the program generally knows
to subtract off your huge constants; that can be 
protection, too.  But I've seen someone use a 
stat-pack 'date'  variable  in regression,
to control for 'week' -- when the dates were internally
held as the time since some 19th century zero-point,
as measured in seconds or maybe milliseconds.  

I don't want my stat-pack to blow up for that, 


-- 
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