Back to PLS HelpPosted on 07/16/15 07:04:35
Number of posts: 8
Dear PLS users,
I got results from structural behavioral PLS (5 behavioral variables and just GM volumes) and I decided to try new split-half reliability option (I chose 100 permutations and 100 splits as recommended). One LV was not significant but the trend could be seen (p=0.108 for permutation test) but p values for brain and behavioral saliences were about p=0.5. The results are at : https://dl.dropboxusercontent.com/u/70539709/split_half_rel_interp.pdf.
I wanted to ask you if I can still work with the first LV or if these results suggests that I shouldnt continue to interpret them at all.
Thank you for any suggestion,
Matyas
Replies:
Posted on 07/17/15 09:56:03
Number of posts: 394
Hi Matyas,
First off, I would ask that you check whether you have downloaded the most recent version of PLS. We posted a bug-fix for the split-half.
Let me explain the split half assessment. It is meant to complement the overall permutation test that we do, which focuses on the singular value for each latent variable (this was where the bug-fix happened). The split-half test assesses whether both sides of the analysis are robust in terms of the their contribution to the effect. So, you should check the overall permutation test and then the split-half results. In the best case scenario, both will line up, but usually you find that this is not perfect. In the case of mismatch, the discordance can be informative. For example, if in the split-half you have significance on the brain side but not the behavior, this would usually indicate instability in behavior side.
First thing, however, is whether you have the latest version of PLS?
Posted on 07/22/15 03:58:50
Number of posts: 8
Hi Randy,
Thanks for the explanation and yes I have the latest PLS version.
Login to reply to this topic.
