Warning "Problems inverting the matrix. Results could be misleading."

[krinaj]

Hello,

I am trying to evaluate and optimize a design using quasi equilibrium TMDD model in PopED. While evaluating the design, I get the warning message "## Problems inverting the matrix. Results could be misleading.". I tried a few different things, such as playing with tolerance, removing and fixing some fixed and random effects via notfixed_bpop and notfixed_d options, but the warning does not go away.

Can someone please share any tips on what may work to remove this warning? What are the implications if we end up moving forward with ignoring the warning message?

Many thanks in advance,
Krina

[Andrew Hooker]

Hi Krina,

This often happens when the design is not informative enough to give information on all parameters. The message comes from trying to take the inverse of a matrix in the FIM calculation. If the standard inversion methods do not work then PopED used a pseudo inverse (Moore-Penrose generalized inverse), and the warning message is displayed. The idea is that by using the pseudo inverse one should still get information about the parameter that is poorly identified in the model given the design. One can then either fix that/those parameter(s) and redo the calculations, or change the design to make it more informative.

Best regards,
Andy

[krinaj]

Thank you very much, Andy. This is helpful!