Hybrid Calculator on MC only?

Roofit and RooStats
1

Hi,

I want to do a study on MC only where I use the Hybrid Calculator to generate the B and S+B test statistic distributions but with no data (Since we dont have any yet).

I came across this tutorial:
roostats/rs506_HybridCalculator_averageSignificance.C

Which doesn’t use any data since it uses the mean S+B test stat as the data test statistic:

HybridCalculator * hc=new HybridCalculator(*model,*modelBkg,observable);
...
RooRandom::randomGenerator()->SetSeed(0);
HybridResult* hcresult=hc->Calculate(ntoys,usepriors);
HybridPlot* hcplot = hcresult->GetPlot("HybridPlot","Toy MC Q ",100);
double mean_sb_toys_test_stat = hcplot->GetSBmean();
hcresult->SetDataTestStatistics(mean_sb_toys_test_stat);
...

This tutorial does not have any data as an input paramter in the HybridCalculator constructor. I assume this has since then been updated and data is forced to be entered?

Any suggestions? Use a dummy data set and manually set the data test statistics?

Thanks for the help!
Rob

2

Hi,

This tutorial I don;t think exists anymore in RooStats and I think it refers to an old implementation of the HybridCalculator class in RooStats.

If you want to estimate the expected significance, you can use an Asimov data set, that can be generated from your model.
You can use the function AsymptoticCalculator::GenerateAsimovData to generate this special data set.
root.cern.ch/doc/master/classRo … 2977e55abe

Best Regards

Lorenzo

3

Hi,

Thanks for the answer. I am involved in a study which is very low on statistics and we are not in the asymptotic region.

Is it fine to create an asimov set in order to calculate the expected significance and use this asimov dataset in the hybrid calculator? I was under the impression that you can only use the asimov data for the asymptotic region?

Thanks!
Rob

4

Hi,

Yes it should work. When using the hybrid calculator it re-generates internally data sets to compute the test statistic distribution and then observed and expected limits. IN case of an Asimov data set the observed limit should be the same as the expected limit apart from small differences due to different way they are computed.

In case it does not work, please let me know, I will look into this

Lorenzo