Roofit: Pull distribution of simultaneous fits

Dear roofit experts,

I want to check the robustness of a simultaneous fit of 6 independent data samples by using the pull method.

I use the RooMCStudy class to generate N new samples from the six original ones:

[color=#4040FF]RooMCStudy mcStudy(myPDF,myPDF,x,"","");
RooDataSet newData = (RooDataSet) mcStudy.genData(i);[/color] x6

(Since I use a simultaneous fit, I cannot use the generateAndFit function from RooMCStudy).

Then, I fit these new samples with the same PDF and store the final parameters values.

The resulting pull distributions are good (mu = 0 & sig = 1) for most of the variables except all the variables related to Signal amplitude (I don’t fit the direct signal amplitude, but combination variables as asymmetries and yield ratio) for which I found a width which is around 0.7.

I try to fit the direct signal amplitudes instead of combination one, or only fit 4 indepedendent samples. I obtain the same results.

Anyone have any idea of the origin of this 1/sqrt(2) factor ?




Could be that the number of signal events is not fluctuated correctly when you are generating the toys. Are you sure you are taking into account the fluctuations in the total number of events ?