Re: prediction den.list-d Digest V2005 #28

[hoeve899@wxs.nl]

On 18 Jul 2005 at 9:27, John wrote:

> I found Petter Ogland's comments regarding regression to have some
> weaknesses, and Doug Hensler has raised part of the point in his post
> (below).   Regression is fraught with difficulty when it comes to
> prediction, and, as we know,  Dr. Deming was fond of saying that 'Management
> is prediction."  

True to the point of steering.


> 
> >From Lewis and Shewhart we see the elements of prediction as first a theory,
> then some some experience, and a degree of belief.  Shewhart showed that the
> evidence (to be most useful) must have temporal spread.  

You can calculate the correlation with a formula (familiar with you 
all?)  to see in what measure it correlated towards the lineair 
regression line. The result of this correlation between 1 and 0 is the 
percentage that you can rely on the resulting line as a signal for 
further prediction when continued there where there are no 
measurements already.

Maybe it would be good to take this a percentage from the length of 
the line allowing only the percentage als allowable lengthening as a 
trustworthy predication and also having the precentage as the 
unsureness measure.

that means:

if the line has 100 points of measurements and the correlation 
formula shows a value of 0,60 then you are only allowed to lenghten 
the predictive length for 60 pointlengths. Also when the line would be 
for example y=10 + 2*X it should show from point 100 untill point 160 
the spreading range y = 10 + 3,2  *X     3,2 =  2 + 2*0,60 = 2+1,2 = 
3,2 )at point 160 and from point 100 and to point 160 the spreading 
range y =10 + 0,8 * X 0,8 = 2- 2*0,60 = 0,8.

when the correlation shows another value then the 0,60 and the 
spreading has to be adjusted. 

put it in an ongoing refresh of the added values and you have a 
scheme which you could also relate to time, having even more 
trustworthyness on prediction.

Could be worth a try out, while management of the variation is as 
important.

greetings,

Henri