I fitted a graph in ROOT and I display test statistics. I discover a chi-square/ndf value close to 0, but the fit looks pretty good, why is this? It also says Prob 1, what does that mean? Any help is appreciated.
alpha_0_beta_and_C01.cxx (2.1 KB)
alpha_0_beta_and_C01.txt (556 Bytes)
alpha_0_beta_and_C01.pdf (18.8 KB)
_ROOT Version: 6.23/01
This means that the fit is “too good”. Looking at the
.pdf, I see that the errors of all points are a factor 5 to 10 too large. These errors are not statistical errors (= due to fluctuations), because 1/3 would need to miss the fitted curve (with 1-sigma errors, 68% confidence level, ~ 32% need to be outside. This is not completely statistically sound, e.g. because the curve also has errors, but if you want to check visually that all is ok, you can use this rule of thumb.).
We can therefore conclude that the errors are either wrong, or they include systematic errors, which the fitter mistakes for statistical errors.
Yes, the machine where I got the data fluctuated, so the errors are systematic. Is there any way to fix the problem with the fitter, or is it really a problem at all?
The fitter is correct. What you should do (if you are interested in the statistical uncertainties of the parameters) is to put only the statistical uncertainties in the error bars.