In the attached plot, it seems the scanning is reasonable. But the Black curve has several intersection points with the line p=0.05, this makes the returned observed limit quite strange.
So I use UpperLimit() to get the observed limit, I get 4.82286, which would be the last intersection point between the Black curve and p=0.05. But from the scanning plot clearly the observed limit would be ~1.8.
Can we have a way to really return the (first) intersection between the Red curve and p=0.05.
HypoTestInverter_Scan_scan_170.k1.root.pdf (41.9 KB)
What is happening at large values it is p probably due to some fitting problems. Anyway this points at large values are not really interesting, so I would recommend you to reduce in such case the scan to a smaller range.
Howver, I don’t understand much the shape of the CLb curve. In principle CLb is computed on the model which should not have a dependency on the parameter of interest, so it should be approximately constant as function of SigXsecOverSM.
Seeing that shape it is an indication of a problem somewhere in your model description or in the Asimov data set generation
Many thanks for your reply! I checked a bit the log file and wouldn’t see issues about “somewhere in your model description or in the Asimov data set generation”. I attach the workspace, could you help to look a bit ?
170.root (800 KB)
I have tried your workspace using the Asymptotic calculator, this is what I get running the
standard RooStats tutorial macro
The output looks reasonable to me.
Asymptotic_Scan.pdf (15.2 KB)
Many thanks for your reply!
If I use $ROOTSYS/tutorials/roostats/StandardHypoTestInvDemo.C, I still get similar plot as I got before, please see the attachement. May I ask which ROOT version you are using ?
HypoTestInverter_Scan_Asym_CLs_grid_ts3_170.rootAsym_CLs_grid_ts3_170.root.pdf (15 KB)
I am using the master version. However I think nothing changed recently, so the same result should happen with any other recent version (e.g. 6.08 or 6.06).
Which version are you using ?
I am using 5.34/14. So then in this version, the issues could happen ?
This is quite an old version (3 years ago). Few things have been fixed since that time, so I am not surprised. Could you move at least to the latest 5.34 patches if you cannot upgrade to ROOT 6 ?
I tried to install ROOT6, but seems my working server is a bit old …
Configuring for linuxx8664gcc
Checking for GNU Make version >= 3.80 … ok
Checking for C compiler … gcc
Checking for C++ compiler … g++
Checking for linker (LD) … g++
Checking for F77 compiler … gfortran
Checking whether compiler can generate dependencies … yes
Checking whether c++11 mode is supported … no
configure: c++11 mode is requested but the current compiler does not support it.
Would you remind me how to install patches for 5.34/14 ? I searched a bit and still not sure …
Just download 5.34.35 from
root.cern.ch/download/root_v5.3 … rce.tar.gz
and then you can use ./configure; make