Hello,
I have setup a toyMC that works in the following way:
* on generic MC I make the mes histograms in bins of MX for B0-SS, B0-OS
and Bch
* each bin of these histograms is randomized in a poisson way, i.e. it is
replaced by the output of a poisson randomizer with mean the original
value in the bin.
* the fit is performed again.
I did 317 such tests (not that any failed, the loop command was a bit
idiot and can only give 317...). The distribution of the outcome (true
mean value BrBr=0.0118) is in
http://www.slac.stanford.edu/~rfaccini/phys/vub/toyDist.eps
You can see that the statistical error is the same as in data
The pull distribution (BrBr-0.0118)/sigma(BrBr) is shown in
http://www.slac.stanford.edu/~rfaccini/phys/vub/toyPull.eps
where you can see that the error is under control at the 4% level, which
is ok.
There is a shift which is 20% of the error (i.e. 5% on the measurement,
smaller than out systematics), but I do not believe it is significance: it
includes possible fluctuations in the efficiency on cuts that are applied
before the toy, which is therefore correlated among all point...
Ric
|