Dear experts,
Maybe it is related with Fit bias in HistFactory.
I found that pull distribution with MINOS error produce a fitter bias.
The pull distribution shows a discontinuous shape at 0. Here is the code to reproduce the problem:
Histfactory_bias.cc (10.3 KB)
This code
- make histogram templates and workspace
- fluctuate the nuisance parameter by Normal distribution. Generate toy MC and fit
- iterate (2) 100000 times and make pull distribution
The pull is calculated by the equation (27) in https://lucdemortier.github.io/assets/papers/cdf5776_pulls.pdf. The pull distribution becomes
You can see some bias near 0.
So, is it expected in Histfactory? Or do I make some mistakes?
Thank you