> I would also like to get an estimation of the standard-error on each
of the fittet parameters. I suspect the values in the hessian matrix
can be used for this, however I'm not quite sure how to go about
this. Any help would be apreciated..
For nonlinear regression I recommend using the functions NLFit, then RegTest
(or for weighted regression, WNLFit and WRegTest) from the REGRESS unit.
(or WLSFit from the MODELS unit if you are using one of the predefined
regression models).
RegTest or WLSFit will return the updated variance-covariance matrix V (V =
Vr * H_inv, where Vr is the residual variance and H_inv the inverse
hessian). The estimated error of the I-th parameter is therefore
Sqrt(V[I,I])
If you use Marquardt only, you get H_inv but you need an independent
subroutine to compute Vr:
Vr = Sum(Y[I] - Ycalc[I])^2 / (N - M)
where N is the number of data points and M the number of fitted parameters.
= Owner
= Moderator
= Online