HELP with errors using correlations

Hi,
I wonder if my code is bugged, please help.
If my fit (based on the Marquardt algorithm)
result for the uncorrelated case
CHI2=SUM_i[(data-theory)/stat error]^2
gives let's say A+/-50
(where I have used only statistical errors)

But for the correlated case, were general chisquared:
CHI2=SUM_ij(data-theory)_i(Vij)^-1(data-theory)_j
where V is the covariance matrix and I add the statistical errors
in quadrature to the diagonal elements
the fit gives A'+/-30

is it physically possible to get smaller errorrs and what does that
mean???
HELP PLEASE!!

.

Relevant Pages