Explanation of createNLL ... 2

Roofit and RooStats
1

I would kindly ask you if you guys can please answer to my post:

Explanation of createNLL …
by mauroroot » Sat Jan 19, 2013 13:41 in Stat and Math Tool Support 022
by mauroroot
Sat Jan 19, 2013 13:41

Many thanks,

  • Mauro.
2

Hi Mauro,

“createNLL” does what you would expect it to do. It creates a RooNLLVar object that can compute the log likelihood “-log(L)” of your pdf. Of course, you have to take into consideration constant terms, like log(n!) in a Poisson extended term (for example), which are not included. In principle, this might cause the log likelihood to be negative for more complicated pdfs, but that doesn’t matter since the maximum likelihood estimate will be the same with or without the constant terms.

Cheers,
Gabriel

3

Many thanks Gabriel !
So then I have one more specific question:
Values of NLL which are more and more negative (far on the left-hand side of an x-axes)
correspond to a higher or lower likelihood ?

In other words:
point (a) with respect to point (b) on the NLL axes is more probable or less probable ?
-(a)------------------------------0|--------------------(b)-----> NLL

Thanks,

  • Mauro.
4

Since we are talking about -log(L), lower values of this expression correspond to a higher value for L, the original likelihood function. As such, lower values of -log(L) make the estimate more probable.

Cheers,
Gabriel