ROOSTATS: Problem with correlated nuisance parameters

Dear expert,

I have the following problem. In a counting experiment I have many nuisance parameters which I can assume 100% correlated (positively). Under this hypothesis, I can treat the uncertainties on those parameters by rewrite them as a function of a single variable X. With this approach X is the only nuisance parameter I consider (with guass(0,1) as pdf).

However, I don’t see any changes in the significance (null hypothesis) and limits when I include the pdf on X in the likelihood…

I am using the ProfileLikelihoodCalculator.

In case you can take a look I have attached an (very) exemplified version with some fake values to help the debug.

Any ideas?

Many thanks,
test.C (5.17 KB)


By doing a simple fit to your model, your systematic parameter “X” is uncorrelated with your parameter of interest, thus it has no influence, so what you get it is expected.
I suppose you have an error in building your model,

Best Regards