ROOFIT binned likelyhood and Goodness of Fit

Roofit and RooStats
1

Dear all,

I have some questions concerning binned likelyhood and goodness of fit tests.

I know that when one performs a Binned Likelyhood Fit using ROOSTAT one can use -2 * (the minimum value of negative log-likelihood) to get information on goodness of fit (as described in S. Baker and R. D. Cousins, Nucl. Instrum. Meth. 221 (1984) 437).

I have a few questions :

-My data are imported in ROOFIT from a ROOT TH1D histogram so I guess ROOFIT is performing a binned likelyhood fit. The only thind I could find on ROOFIT is that if the data is binned a binned likelyhood is automaticaly performed. Am I correct?

  • Assuming the fit is indeed a binned maximum likelyhood fit, I couldn’t find anywhere whether ROOFIT is using this Baker & Cousins prescription or not for Binned Likelyhood just like ROOSTAT does. In other words, can I use -2 * (the minimum value of negative log-likelihood) to get information on goodness of fit in ROOFIT as well?

Thanks a lot for your help.

Cheers,

MM

2

Hi,

  • Yes if the data set is binned RooFIt performs automatically a binned likelihood fit

  • No, you cannot use the NLL value directly computed from RooFit to get the Baker-Cousins log-likelihood ratio.
    this is possible if you are using TH1::Fit but not in RooFit, since an offset term (not dependent on the parameter) is missing. What you can do is to transform your pdf in a TF1 and use the function
    TH1::Chisquare with the option “L” , see
    root.cern.ch/doc/master/classTH … 4fd63e2cb4
    This is possible only in the latest ROOT release (6.08 or master)

Best Regards

Lorenzo