I am using the TFractionFitter to find the fractions of contribution from different type of jet flavors in data. The data are fitted by 3 templates corresponding to three type of jet flavors. The templates are derived from the Monte Carlo simulation. As I understand, TFractionFitter will take into account the Poisson uncertainties of bins of the templates. Therefore, the errors reported (I use MINOS error) will include the statistical uncertainty from data and the statistical uncertainty from the templates. Is there an implementation in TFractionFitter to separate those uncertainty sources (from data and templates)? Can scaling the templates by high scale factors (say 10 or 20) effectively separate them (In this case, the scaling will make new templates with very high bin content, so eliminate the Poisson uncertainties of bins of the templates)? Or may be there is flag that tells TFractionFitter not to use the template uncertainty?
It is not possible to this in TFractionFitter, but on the other hand if you do not consider the error in the template the problem is simpler and you can probably write the likelihood yourself.
Otherwise, yes, by re-scaling the histogram to very large statistics you get the same result.