SetGaussBkgGaussEff in TRolke

Roofit and RooStats
1

I am starting to use TRolke and SetGaussBkgGaussEff does not perform as I expect. In particular, the upper limit returned does not seem to depend on the uncertainty on the background estimate. SetGaussBkgKnownEff does seem to work as I expect.

I am using Root 5.26/00 on Mac OS 10.6.2

Here is my test program:

[quote]{
TRolke tr;

Double_t alpha = 0.8;
tr.SetCL(alpha);

Double_t events = 182;
Double_t background = 182;
Double_t backErr = 0;
Double_t efficiency = 1;
Double_t effErr = 0;
Double_t UL;

tr.SetGaussBkgGaussEff(events, background, efficiency, effErr, backErr);
UL = tr.GetUpperLimit();
cout << "GaussBkgGaussEff: Observed = " << events << " Background = "
<< background << " +/- " << backErr << " upper limit = " << UL << endl;

tr.SetGaussBkgKnownEff(events, background, backErr, efficiency);
UL = tr.GetUpperLimit();
cout << "GaussBkgKnownEff: Observed = " << events << " Background = "
<< background << " +/- " << backErr << " upper limit = " << UL << endl;

backErr = 14;

tr.SetGaussBkgGaussEff(events, background, efficiency, effErr, backErr);
UL = tr.GetUpperLimit();
cout << "GaussBkgGaussEff: Observed = " << events << " Background = "
<< background << " +/- " << backErr << " upper limit = " << UL << endl;

tr.SetGaussBkgKnownEff(events, background, backErr, efficiency);
UL = tr.GetUpperLimit();
cout << "GaussBkgKnownEff: Observed = " << events << " Background = "
<< background << " +/- " << backErr << " upper limit = " << UL << endl;
}
[/quote]

When I run it, the upper limit using GaussBkgGaussEff does not increase when the uncertainty on the background increases.

Anyone have any suggestions? Thanks,

    - Chris
2

Hi,
the TRolke class is having few bugs (see for example savannah.cern.ch/bugs/?55718 ) and the result is not always stable. I am expecting the author of the class to improve this for the next release.
Also, in your case if you put a small error in the efficiency (like 0.0001) you get the right result.

I would try to use the ProfileLikelihoodCalculator class of RooStats.
See as example the roostats tutorials rs101 or rs501

root.cern.ch/root/html/tutorials … ple.C.html

root.cern.ch/root/html/tutorials … mit.C.html

Best Regards

Lorenzo