For a TGraph, all weights are equal to 1 and the chisquare is simply the sum of (y-f(fx))^2
ok…maybe that is not a question directly relevant with root procedure… but, what is quality of such test as compared to computing with known data points error?
In this case the fit is OK, except that the computed chisquare has no statistical meaning. It is only the sum of the squares
ok. i understand… but is there a possibility to make statistical
sense of such a test? is there a way to define data point error having only data points and fitted points?