sign of coefficients in eig

Dear experts,

I am aware that the coefficients in PCA componentes are arbitrary (and can be flipped which still has the same meaning). However, I am still a little puzzled. Is there a rule or convention behind the signs in eigenvectors in the output of the eig function? (and determining which part of the elements gets which sign?) Does the strength (absolute) values of coefficients have any influence?

Thanks very much for your help,

Best

Hans van der Horn

Answer 1

[Hans van der Horn]

I'd like to comment on my own question.

I'm analyzing functional MRI data, and extract the first eigenvector from 4D connectivity matrices. I noticed sign flips now and then, that look strange. Now I'm using eigenshuffle.m by John D'Errico, and this tackles the problem. Great code, thanks so much!