Feldman Cousins - Setting/Non-Setting CL

[usccaa]

I’ve been adapting the RooStats Feldman-Cousins method documented here by using this tutorial

I’ve been using a simplified method to just understand how it works in Root. I generate a simple 1 event dataset off of a model, and then test it with the model.

Example.C (4.3 KB)
Data_0.root (5.4 KB)

My question is this: if I follow blindly in the tutorial, the tutorial claims to be reproducing the FC neutrino oscillation result by measuring the 90% C.L. I had assumed this was from the line

fc.SetTestSize(.1);

which the documentation specifies “set the size of the test (rate of Type I error) ( Eg. 0.05 for a 95% Confidence Interval)”. But. I notice that there’s also function to set the confidence level explicitly with alot of the same wording as setting the test size.

The example I have has two function FC_Test1() and FC_Test2(). The only difference between them is that FC_Test2() explicitly sets the confidence level via

fc.SetConfidenceLevel(.95);

Test 1 yields a bound of [0.375, 10.875] and Test 2 yields a bound of [0.075, 10.575]. If I comment out the SetTestSize on Test 2 I recover the bounds of [0.375, 10.875]. Is it inappropriate to set it both ways? If so, why?

[Danilo]

Hi,

Thanks for the interesting post. I add in the loop @moneta .

Best,
D

[usccaa]

Thank you. My primary confusion is that I believe that

fc.SetConfidenceLevel(.95);

and

fc.SetTestSize(.05);

should be doing the same thing, so my assumption would be that setting both doesn’t change the result. Yet, what I’m finding is that setting both is changing the result, and only doing one or the other is recovering the same limit.