Re: confidence region and leasqr()

[Dmitry Shkirmanov]

> It would actually not be correct to use this variance to compute
> confidence intervals with the normal distribution, since it is not the
> true variance, but itself a guess.
If i understood correctly the Delta method is a bad approach.

May be then somebody can say what approach does the CERN's ROOT package 
use use to calculate the confidence interval?
Here is an example ( 
http://root.cern.ch/root/html/tutorials/fit/ConfidenceIntervals.C.html  ),
and here is the sources of GetConfidenceIntervals() function ( 
http://www.slac.stanford.edu/grp/eg/minos/ROOTSYS/pro/math/minuit/src/TFitter.cxx 
)