Parameter Errors from lsqcurvefit

Hi Everybody,

I've been using the lsqcurvefit function to fit a Weibull function to a small set of data (4 to 10 data points) I've been using lsqcurvefit as it allows me to constrain the parameters as otherwise the fits become unrealistic with near right-angle turns at a single point. However I need to calculate the errors of the Weibull parameters, especially the error in ymax as this is needed for another graph. I've tried the code below which has been suggested in other threads:

[Q,R] = qr(J,0);

mse = sum(abs(r).^2)/(size(J,1)-size(J,2));

Rinv = inv®;

Sigma = Rinv*Rinv'*mse;

se = sqrt(diag(Sigma));

but this has not produced any numbers for my data.

any suggestions?

Answer 1

[Torsten]

I guess that the formulae above work for unconstrained fitting, but are no longer valid if you constrain your parameters.

By the way: "inv" is a "registered trademark" ?

Best wishes

Torsten.

[Unkle77]

I don't understand why a constrained fitting should result in it not working, surely it would just result in a larger error?

Yeah, no idea why (R) became a registered trade mark character, just for clarity it should be ( R ).