quote:
Hi,
I am reading a pre-print for a colleague who has used a seed pls analysis to study some brain imaging data and I am slightly confused. The main problem is that when they discuss the saliences for the different brain regions, they usually have numbers larger than 1. It was my understanding that the saliences where contained in the orthogonal matrix U, hence between [-1,1], and that the legitimacy of the results could be verified by the bootstrap analysis (using the compare_u matrix). Having inspected the paper further, it turns out that what they call saliences are actually the entries of compare_u, from the bootstrap analysis. Is this valid? Or should they be using the matrix U returned from the SVD to determine the most important brain regions? Any help would be most appreciated.
Thanks,
Jonathan
Hi Jonathan,
If your colleagues are reporting the values from "compare_u" then those are the bootstrap ratios for the singular vector:
bootstrap ratio = (singular vector weight*singular value)/bootstrap estimated standard error
These are roughly equal to a z-score, though we prefer to interpret them akin to how you would a confidence interval
In matrix U, each vector is unit normal so the values should be between [-1 and 1].
Sorry about the terminology confusion. I hope the user guide and help text that is part of the m-files clarifies this further.
Randy