Back to PLS HelpPosted on 04/11/14 00:10:47
Number of posts: 10
Hi all,
I wanted to get opinions about what would be the best way to display results from a behavior PLS analysis, whereby I get many voxels with either large positive or large negative bootstrap ratios (BSR). In particular, I was hoping to get some opinions about how to portray the correlation plot (e.g., the plot with correlations for each behavior variable). As I understand it, the directionality of the plot that plsgui produces is relevant for the voxels with positive BSR values, but that for voxels with negative BSR values, the directionality of the correlation reverses from what is shown in the correlation plot (e.g., positive correlations are now negative for those voxels with negative BSR values). So my question really is how does one best convey that the directionality of the correlations reverses for negative versus positive BSR voxels?
Thanks in advance for any thoughts on this,
Mike
Replies:
Posted on 04/11/14 03:08:45
Number of posts: 394
Hi Mike,
This is one of the toughest parts of multivariate analysis to convey. The choice of how to do this will depend on how many behaviors you have. If there are several, you may want to stick with the correlation plot for the behavior side and just show the bootstrap ratio image for the brain side, indicating the positive weights indicate the voxel is postiively correlated with the *entire pattern* on the behavior side, while negative weight indicates a negative correlation with the entire pattern on the behavior side. Its important to emphasize that its the entire pattern b/c the optimization in PLS is done on the entire matrix rather than individual behaviors and voxels. {this holds for interpreting the behavior side too}.
For voxels that are particularly strong, you may want to extrac the actual correlation for that voxel and plot that for illustration/descriptive purposes, but be clear to indicate that is for that purpose only and not for inferences.
hope that helps
Randy
Posted on 04/11/14 03:14:06
Number of posts: 10
Hi Randy,
Thanks very much for the quick reply and your insight on this and this is very helpful for me to have a start thinking about how my results should be displayed and what comments need to go with it for proper interpretation of it.
Thanks again,
Mike
Login to reply to this topic.
