fitting with weights given by covariance matrix

Share icon

Hello,

I wonder if there is a possibility to use a full covariance matrix for weighted regression instead of using a 1D wave with values of standard deviation (or inverse of standard deviation square). I couldn't find this information in the manual. I am using Igor Pro 7.

I have data points characterized by significant off-diagonal elements of a covariance matrix (for some points correlation is even ~0.9). Therefore, I believe I should include information about covariance in my analysis. 

I will be grateful for comments.

Regards,

Rafal

 

Thank you for your response. I hope that this feature will appear soon. I don't think it should be hard to implement this, as anyway (at least I believe so) fit optimization is done using matrix formalism. 

December 20, 2021 at 06:42 pm - Permalink

The difficulties have to do as much with the public interface as with the math. We could require you to provide the full covariance matrix, but that's N^2 elements, and that gets big fast. In most cases, I think the off-diagonal elements are likely to be the same for a given diagonal, so perhaps it would be possible to accept a wave containing one element for each non-zero diagonal. But that decreases generality.

In any case, I'm afraid this would have to wait for at least Igor 10.

December 21, 2021 at 09:20 am - Permalink