Back to PLS HelpPosted on 07/03/18 14:53:33
Number of posts: 9
Hi,
I recently ran a non-rotated behavioral PLS with a single contrast. It is my first time running a non-rotated PLS, so I'm not sure if the values I am getting are reasonable. The percent crossblock covariance explained by the LV is 100%, which seems unlikely. When I ran a regular behavioral PLS on the two conditions of interest, the main LV accounted for 85.65% of the covariance. I would appreciate any help in interpreting and reporting the covariance explained by the non-rotated PLS.
Thanks,
Kayleigh
Replies:
Posted on 07/03/18 15:46:03
Number of posts: 394
this is expected because with one contrast there is only one source of covariance to consider, so it must account for 100%. That metric is really only relevant when you have more than one contrast and are comparing it relative to one another or to a comparable analysis for another group. Be mindful that this is NOT variance accounted for.
Posted on 07/03/18 15:48:16
Number of posts: 9
Thank you for this quick response. Is there another metric that I should use when reporting the amount of variance the LV accounts for?
Posted on 07/03/18 15:54:57
Number of posts: 394
Thank you for this quick response. Is there another metric that I should use when reporting the amount of variance the LV accounts for?
There really isn't a metric for variance accounted for per se. You could calculate something like Cohen's d but that's not part of the code.
Posted on 07/03/18 16:03:17
Number of posts: 9
OK. I do report the percent crossblock covariance for a regular mean-centered Task PLS -- is this more similar to a percent variance explained or should I avoid reporting this metric in general?
Posted on 07/03/18 21:36:40
Number of posts: 394
OK. I do report the percent crossblock covariance for a regular mean-centered Task PLS -- is this more similar to a percent variance explained or should I avoid reporting this metric in general?
You can report the percent cross-block, but also report the associated singular value as that can be compared between analyses of the same data matrix
Login to reply to this topic.
