Back to PLS HelpPosted on 11/13/13 14:02:26
Number of posts: 11
I have a reviewer who is arguing that as the three significant LVs in my mean-centred task PLS analysis together explain 99.97% of the crossblock covariance, that the analysis is invalid due to overfitting. They are admittedly unfamiliar with PLS, but I wondered if anyone has insight into whether this is a legitimate criticism?
Replies:
Posted on 11/13/13 15:06:50
Number of posts: 394
I have a reviewer who is arguing that as the three significant LVs in my mean-centred task PLS analysis together explain 99.97% of the crossblock covariance, that the analysis is invalid due to overfitting. They are admittedly unfamiliar with PLS, but I wondered if anyone has insight into whether this is a legitimate criticism?
Its not a valid criticism b/c PLS is not working on total variance but rather the covariance between task/behavior and the brain. Of that covariance strucutre, the LVs will have to account ultimately for all the covariance (100%), with each one accounting for a proportion thereof. The total percentage isn't really useful, but the percentages for each LV can give you a picture of the relative importance of each LV.
Posted on 11/13/13 15:13:41
Number of posts: 11
Hi Randy - thank you very much for your prompt reply! I had thought this was the case, as when specifying a single contrast in a non-rotated seed PLS, the single resulting LV always explains 100% of the crossblock covariance (in this case we are looking at the covariance between the specified contrast and the brain?).
Thank you again,
Victoria
Login to reply to this topic.
