CADI:
https://cms.cern.ch/iCMS/analysisadmin/cadilines?line=HIG-20-008&tp=an&id=2332&ancode=HIG-20-008
Color Code
Color |
Meaning |
BLACK |
Question or comment from ARC/conveneers |
RED |
Authors know but not yet added/implemented |
GREEN |
Answer from Authors |
ORANGE |
Authors working on this item |
BLUE |
authors answered but might need iteration with ARC |
Comments for data-card
NP naming: in order for the NP to completely follow the HComb convention, it would be nice if you
could add _hzz to the analysis-specific nuisances names
done
Lumi NP: you mention that for the lumi you are using a weighted average of the three years as NP.
However, this is not taking into account the partial correlations among different years.
Also, the value quoted by the Lumi POG for the full Run II uncertainty on the luminosity should be 1.8%,
somewhat different than the 2.4% you are quoting.
Please consider using the correct value of 1.8%, this should be fine as you are considering a single signal
process inclusive on the three years.
OTOH note that when using this 1.8% you would not be considering the partial correlations among
years, but this should not have any substantial
impact on the analysis results.
I changed to the recommended luminosity uncertainty of 1.8%.
Blinded limits: since you are using HybridNew to extract blinded limits, you should generate an asimov
toy first (with GenerateOnly --saveToys)
and use -D <file_with_toys.root>:toys/toy_asimov to have a blind limit, otherwise you are getting post-
fit expected, which already uses data.
done
CMS_H_mean_e NP: In the ZJpsi->2e2mu and ZJpsi->4mu cards (so the separate ones) there are param
lines in the datacards that are not associated with any params that actually exist in the input workspace.
For example there is a parameter CMS_H_mean_e in the workspace, and a param line CMS_H_mean
param 124.690 0.047 Please fix this and use consistent naming between datacard and workspace.
done
Background: Both in the AN and in the presentations given at HZZ/HIG PAG meetings you mention that
this analysis uses "data-driven background". OTOH you also quote three MC samples used to model the
background. Our understanding is that the background is not really data driven, but rather parametrized
with an analytical expression using these MC samples. Can you please clarify on this?
The background is dominated by associated production which is sampled from the side-bands.
In addition, possible peaking background is estimated from dedicated MC samples and is found
to be negligible (far less than 1 event). These backgrounds are not included in the fit. Hence, we
characterize the procedure as purely data driven.
Also, you have a NP associated to the background (CMS_bkg_frac param) only in the JJ4mu card. I think
you should have NPs associated to the background in all your cards.
The parameters of the background shape functions are free to float. Therefore, they are not
accounted for as NP – the inclusion of the background fraction which is floated in the JJ card was
a mistake (double counting).
text2workspace: if you are using any particular model, please upload it to the repository.
It would be nice, just for completeness, if you could also share the text2worskpace commands you use
to generate the workspaces starting from the datacards, especially if they contain any particular option
for the POIs ranges and/or re-definitions.
done
-- HimalAcharya - 2021-02-17