Different fitting results from two similar approaches

[HIMANSHU_SHARMA]

Hi Experts,
There are two approaches model1.C and model2.C to parameterize a dataset.

• model1: with 3 parameters: Signal1 counts, signal2 counts and bkg counts
• model2: with 2 parameters: fraction fB = sig1/(sig1+sig2) and a fraction fSig=total_sig/(total_sig+bkg)

I prefer to use model1, because by using this, I can also get the results from model2 i.e fractions.
And both results should be the similar (Or maybe I am wrong )
But when I run both of them on different datasets:

For dataset # 1 → templates.root

• model1 gives 3x smaller statistical error on fB than model2 and mean fB from both models also differs. In addition, errors on fSig are different…

For dataset # 2 → templatesDummy.root

• results are similar by both models.

I don’t understand such behavior.
Could someone please look into this?

You can find the dummy data and macros here (ready to run) in the cernbox Link

Thank you very much!

[bellenot]

Maybe @jonas or @moneta can take a look?

[HIMANSHU_SHARMA]

Just to let the experts remind about it, which is still open
@jonas @moneta