Unreasonably large confidence interval values from nlparci and lsqcurvefit

I am using lsqcurvefit for non-linear fitting and nlparci afterwards to determine the confidence intervals (based on the jacobian output from lsqcurvefit). My confidence intervals come out on the order of 10^6, whereas all of my parameters are on the scale of 1 to 100. I notice that the algorithm cannot seem to move very far from the initial parameters, no matter what they are. Does anyone have any suggestions about what could be going wrong?

Answer 1

[Star Strider]

Without seeing your code (and some representative data) it’s impossible to say with any certainty. One possibility is that your initial parameter estimates are sufficiently far from the desired global minimum that your function is trapped in a local minimum. You could also be constraining your parameters to a region in which your function will not find an optimal solution. The resulting large residuals are creating the large CIs.

Experiment with different initial parameter estimates, and extend the limits of your constraints if you’re using them.