Home | News | Documentation | Download

Predict number of events in signal region from sideband

Hi,

Is there any method in RooStat to interpolate the number of events found in the sideband to signal region? In sideband, I find roughly less than 10 events after a set of selection cuts (becomes difficult to fit now). Now I want to predict or expect the number of events to be observed in the signal region based on the events found in the sideband. Can anyone direct me to certain solutions for this?

Thanks

@StephanH can you please help here?

Thank you,
Oksana.

Hi,

I am not sure RooStats can help with this. If you want to extrapolate from side bands, you need some kind of model first. I could e.g. imagine a Monte Carlo simulation with 1000 events in the side band and 100 in the signal region. So you know that the ratio between the two regions is 1/10.
In a second step, RooFit can help you measure a yield in the side band region (let’s say the result is 100). Then, using the Monte Carlo knowledge you can estimate the number in the signal region is 10.

Hi Stephan,

Thank You for the reply and suggestions. As I understand, there will be a need for model first to simulate events. But my doubt is how to choose the model? Also, how to find the ratio of events between two regions initially. You gave an example of the ratio to be 1/10 but, it can be anything.

Thanks

Hi @Dev,

exactly, it can be anything. You didn’t give details about the analysis in question, so it’s hard to say what’s necessary. Maybe you have a colleague who knows the analysis to discuss with?

Hi Stephan,

Thank You! Since some MC samples are missing, I am following data driven method for the analysis to find the branching fraction of process like tau—>mu mu mu . For this, I can follow the ABCD method to predict the expected number of background events in signal region by looking sideband region at first. But for ABCD method the variables should be uncorrelated. But the variables( reconstructed mass and change of energy) that I have chosen are somehow correlated upto some extent. Another possibility is to interpolate the number of events from sideband to signal region. This kind of interpolations are done in previous analyses. So I posted here to look for the possibility of interpolation in root.

Thanks

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.