nlinfit vs. lsqcurvefit

[Jason Park]

Well, it's been years since this was posted, so I don't know whether anyone would respond to my question here.

Of course, when you do have active constraints on some
parameters in a model, the simple tools for confidence
estimation are no longer truly appropriate anyway

could you explain more about WHY that is so? and if there is any relevant reference, you could kindly let me know?

I fitted my model to the data with some constraints, and having calculated the covariance matrix estimate using the finite-difference jacobian that the LSQNONLIN yielded, none of the parameter estimates have found significant and I freaked out! Maybe the reason is because the resulting covariance matrix estimate was not calculated at the global minimum due to the constraints, but I'm not very certain about my intuition.

Regards,
Jason

[Donnacha]

And even more years pass...

The statistics toolbox has nlinfit.m and fitnlm.m
The optimization toolbox has lsqcurvefit.m and lsqnonlin.m

Could someone tell me if the previous (12 year old) answer is up to date and if not what are the pros and cons of these fitters?

DD

Post by Jason Park
Well, it's been years since this was posted, so I don't know whether anyone would respond to my question here.

Of course, when you do have active constraints on some
parameters in a model, the simple tools for confidence
estimation are no longer truly appropriate anyway

could you explain more about WHY that is so? and if there is any relevant reference, you could kindly let me know?
I fitted my model to the data with some constraints, and having calculated the covariance matrix estimate using the finite-difference jacobian that the LSQNONLIN yielded, none of the parameter estimates have found significant and I freaked out! Maybe the reason is because the resulting covariance matrix estimate was not calculated at the global minimum due to the constraints, but I'm not very certain about my intuition.
Regards,
Jason