Does fAmin give -2*log(Likelihood)?

Roofit and RooStats
1

Hi,
I fit a 1D histogram with binning likelihood method (option “L”). Everything is OK before I want to extract the loglikelihood value. The method itself uses the function “FitUtil::EvaluatePoissonLogL” (see http://root.cern.ch/root/html/TH1.html) and the FCN is -loglike. In order to get the mimimun value of FCN, I use the code like “gMinuit -> fAmin”. A message (see http://root.cern.ch/root/roottalk/roottalk00/1963.html) said it gave “-2*loglike”). So the problems are

  1. Is FCN -log(L) or -2*log(L)
  2. Question 1) is connected with the error estimation, as for -log(L), the SetErrorDef should be 0.5, while 1 for -2*log(L).
  3. fAmin gives “-2*log(L)” or not?

Could anyone help me?

Thanks!
Xia Ligang

2

Hi,

If you are using option “L” when fitting an histogram in the current version of ROOT, 5.34, gamin is = - logLikelihood.
Moreover, the log-likelihood is a likelihood ration with respect a saturated model, as described in this paper,
physics.ucla.edu/~cousins/st … urated.pdf

So by doing a 2 * gamin you’ll have a chi2

Best Regards

Lorenzo