Q: What should I do when the covariance matrix is unreliable?

A: This could happen for various reasons. The most common is because the data are not informative enough to allow reliable estimation of the model parameters. The best way to solve this problem is to simplify the model of the system or redesign the kinetic experiment and collect additional data.

Sometimes this could also happen because the error is misspecified in magnitude (e.g. Absolute Weighting is used and a Fractional Standard Deviation of 20% instead of 3% is given). The use of Relative Weighting should give a better understanding of the true magnitude of the error.

Sometimes the following error message appears - in this example, concerning k(0,2):

WARNING: The following parameter limit(s) constrain further optimization:

Usually, the parameter actually hits the limit. This problem can be easily solved by resetting the Upper and Lower Limits for that parameter.

However, this could occur even if the parameter is not actually at limits (by inspection of the Parameter Window). When this happens, the most likely reason is that the parameter(s) that hit the limits are not estimable from the data. The Statistics Window will contain the standard deviations of the parameters calculated with the parameters at limits held fixed. Possible suggestions to avoid this problem are to actually fix them in the fitting or to make them Bayesian parameters. 

Please see the links:

1. What should I do when the fit goes on forever?
2. When and how should I use Bayesian fitting?
3. How can I calculate Statistics if SAAM II has not converged?

for more information on how to deal with this situation.

<< Back to Support Page