I have a technical question about the implementation of systematic uncertainties in various RooStats classes that might turn out to be a conceptual question as well. I am following an example for creating simple models with one channel, e.g.
I have noticed that the RooStats::HistFactory::Sample class has many options and methods for adding systematic uncertainties, while the RooStats::HistFactory::Measurement class has fewer methods and the RooStats::HistFactory::Channel class has none. Why is this? For example, the method AddHistoSys() is only implemented for the Sample class, but not for Channel or Measurement.
Suppose I have a total systematic uncertainty on my measured data that is dependent on the variable I am binning in. How would I go about adding this to the Channel class?
The Sample class is the lower one and represents for example the signal histogram or the background histograms. Therefore there you are going to specify the systematic uncertainty for the histogram shape (HistoSys).
The Channel contains just a collection of Sample and there is no need to add additional uncertainties.
Some common uncertainties will appear when combining different Channels in a Measurement.
There one can specify global uncertainties across the channels such as the luminosity one.
I hope this is clearer. See maybe existing example in the HistFactory users guide,