Percent error or percent difference?  
*

*Using either percent error or percent difference, I want to compare one of my measured values from a set to the mean of the set. By reading the Wikipedia article on percent difference, it's still not quite clear which I should choose.

*What if I wanted to compare an instantaneous slope to the slope of a linear regression, do I use percent error or percent difference?


If it isn't clear yet, my misunderstanding lies in the comparison of one value from a set to some computed aggregated value from that set.
Thanks in advance.
-JP
 A: Well, neither of them. And you are trying to bite it from a wrong side.
Namely, you need some hypothesis -- to what end you are comparing individual measurements to the mean? This is important, since this determines what method you should use; for instance:


*

*you may have a million normally distributed numbers and 3 outliers and you want to find them -- this way a Z-score may be a good idea;

*you may have some numbers form unknown distribution and want to find some outliers -- then you should think about some IQR-based Z alternative or some other nonparametric methods;

*you may want to check if your sample is from a certain distribution -- making a qqplot is a some way to go.


Without it, you will just get 15.3%, 93.4%, 7.532% or some any other number that will be equally useless regardless of being percent error or percent difference.
A: I do not necessary agree with the Wikipedia article, however, of the two,  Percent Error would be a better statistic since you would are essentially treating your average as the  "theoretical" value.  The article seems to state (though not clearly) that "Percent Difference" involves two experimental values x1 and x2, which you do not have.
-Ralph Winters
