ShapeSys question

[pmalecki0]

Dear Experts!

I am fitting a model in which one of the samples has an bin-by-bin uncorrelated uncertainty (which is not statistical uncertainty, but a systematic one derived from external package). For this I am using ShapeSys - I hope it is the right thing to do.
My question is - how should I define the contents of the corresponding ShapeSys histogram?
Since the output shows "gamma_* " nuisance parameters corresponding to this ShapeSys, I believe these are multiplicative.
So, should the content of the given ShapeSys bin be relative error? I.e. if the bin content is 0 then there is no uncertainty and if it is 1 then it is 100%?

Best regards,
Paweł

[alexander-held]

Hi Paweł, shapesys indeed implements the functionality you are looking for. It works like staterror when it comes to specifying the "data" field: you provide the absolute uncertainties. See also the relevant documentation here: https://pyhf.readthedocs.io/en/v0.7.4/likelihood.html#uncorrelated-shape-shapesys.

What is multiplicative about this is that the value of the nuisance parameter multiplies the event yield in the respective bin, so for a bin with nominal prediction of 10 events and a shapesys nuisance parameter fitted as 1.1, you would have a post-fit yield of 11 events. The naming of the parameter is up to you: it will be called "gamma_*" only if you give it that name. That prefix is a common one though, given that shapsys parameters use a Poisson constraint term.