Discrepancy between FC and CLs

Roofit and RooStats
1

Dear experts,

we are setting an upper limit on BR for a given decay mode. We used the ProfileLikelihoodCalculator as well as a CLs (hybrid) approach, obtaining compatible results with and without systematic uncertainties.

In the last days we also tried to get the Feldman Cousing interval on BR. If systematic uncertainties are not taken into account, the FC upper limits are close to the ones calculated with the other strategies. When we include the systematic uncertainties, the observed and expected upper limits are respectively ~30% and ~50% greater for FC, which is quite surprising.

We are using the FeldmanCousins class (following the many examples available), passing to it the same ModelConfig used for PLC and CLs. We tried to vary the number of bins and toys without any results.

The discrepancy just comes out when systematic uncertainties are considered and we cannot understand the source. (maybe an issue in the way the FeldmanCousins class deals with the nuisance parameters?)

Should we use another class to handle with systematic uncertainties in the FC scheme?
Othe suggesionts/ideas?

Thanks.

2

Hi,

There could be some discrepancy between CLs and FC, depending also if you have a deficit or an excess. To understand better this problem, I would need to have the workspace you are using

Cheers

Lorenzo

3

Hi Lorenzo,

thanks for the answer, you have a PM.

Cheers,
Davide.