[Met_help] Re: PointStat results over multiple times combined (UNCLASSIFIED)

[Tressa Fowler]

Barb,

Correlation (and some other stats like mean squared error and  
variance) are accumulated from the sums of squares in the point-stat  
output, so there is no need to return to the original data.

This confusion is fairly common among those familiar with the  
equations for correlation that involve sample means and sample  
standard deviations. Those values cannot be used directly to get an  
overall correlation. However, there are equivalent formulas for  
correlation using sums of squares and sample sizes, which can be added  
up over a bunch of forecasts to yield the appropriate values for  
correlation and mean squared error.

Tressa

On Jan 7, 2009, at 4:29 PM, Sauter, Barbara (Civ, ARL/CISD) wrote:

> Classification:  UNCLASSIFIED
> Caveats: NONE
>
> Wonderful.  Thanks for the prompt and clear response.  I had thought  
> the
> original forecast and observation data needed to be used to calculate
> overall correlation coefficients, but I'm no statistician.
>
> Barb
>
> Classification:  UNCLASSIFIED
> Caveats: NONE
>