Under estimation of error in roofit


I am using UML fit and splot to fit data. I run this fitter 100 times to obtain pull distribution. After 100 runs I fit a Gaussian to resultant parameter to get the fit result and resultant parameter fit looks OK.
I have problem with error and pull distribution. Somehow error is under estimated, due to which the width of pull distribution is greater than 1. Is there any way to improve the error? (have already tried MINOS(), minuit2).
I have also attached the code.
fit.C (12.6 KB)


Do you have a weighted data set ? This could explain the under-estimation


Yes, I have weighted data set. I attached one dataset. Last column in this file are the weights.

wght-_1.txt (117.1 KB)

You should then fit with Sumw2(true)


I am already using SumW2.

I understan, but can happen that with weights the errors are not properly estimated, since it is just an approximation.
Also you can’t use Minos with aeivhts

But, if I am not using minos my fits go worse.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.