Jumping expected limit using parametric model

[EtienneDreyer]

Dear RooFit/Stat-ers,

Following the popular RooStats tutorial I’ve been trying to set upper limits on signal strength in a spectrum fitted with a parametric bkg+sig model.

Problem: The observed limit seems correct and the expected limit also except that it always jumps up when there is a deficit in data (signal strength wants to be negative).

To elucidate my problem, I’ve attached a standalone example macro where you can check the following:

• Signal strength is capped positive when setting limits
• bkg and sig normalizations are controlled by small parameters
• order of null and alternate models in AsymptoticCalculator

Since I’m new to RooStats, I don’t have much confidence that I’ve implemented this correctly except that the signal strength scan and the observed limits behave as expected. It appears to me that somehow the background model is becoming ill-defined in the case of a deficit.

Please let me know what you think.
Thanks!
Etienne

minimalRooStats.C (5.7 KB)

[Axel]

@moneta maybe you can help here?

[EtienneDreyer]

Hey y’all, @nicolas.berger found the problem, and yes it was a one-liner:

//inverter.SetFixedScan(0.,1.,0.001);      //old
inverter.SetFixedScan(100,0.01, 10, true); //new

I am having a hard time recalling where I picked up the notion, but I thought the argument takes (min, max, step) which is clearly not the definition. Corrected macro attached.

minimalRooStats.C (5.8 KB)