Having trawled through the forum list (I wish it was searchable :)) I found this answer about the colour coding of voxels in the LV images. Q1. Understanding the coloured weights (saliences) across the brain (and how they relate to the scatter plots in the brain score overview window): (a) do BOTH blue and yellow areas represent voxels whose intensities correlate with behaviour, with only the polarity changing (yellow = same polarity as the shown correlations; blue = reversed polarity from the shown correlations) OR (b) do the colours represent a breakdown of the correlations: in this case, I have a negative correlation, so: blue weights representing areas that are associated with large behavioural scores and yellow areas representing areas of low behavioural scores. 2. I guess, more generally, I'm confused about how to interpret the scatter plots in the brain scores overview. What does it mean that *brain scores* are negatively correlated with behaviour? Does that actually tell me anything about how voxel intensities are correlated with behaviour?? A1: answer to my first question is (a):  The saliences tell me where voxels show the same (or opposite) correlation depending on the colour of the salience. Q2: But I would still appreciate some insight as to how to interpret the correlations themselves.  Is it correct to assume that for salient voxels, there is a correlation between BOLD activation and the behavioural measure?  Or do these correlations represent something more obscure: the overall correlations are between brain scores and behaviour, but I'm hazy about exactly what brain scores are and whether they allow me to draw any conclusions back to the actual brain activity A2: The salience/weight for a voxel is proportional to its correlation, or to the difference in the correlation if your behavior profile (weight or LV) indicates a task or group difference. The brain scores are a summary across the entire image and depend, for a given subject, on the balance of activity between positively and negatively weight voxels: more positive brain score, more activity in positively weighted voxels. The correlation between these scores and the behavior you will note is the same pattern as in the behavior LV. In fact, the behavior LV you see in the PLSGUI is essentially a normalized version of the correlations (unit normal vector). The pattern is the multivariate expression of the brain-behavior correlation so it is vital to your interpretation: i.e., is the correlation pattern significant (permutation test) and reliable (bootstrap confidence intervals on the behavior correlation). The voxel weights, and bootstrap ratio, add to this by idenifying the voxels with the more reliable contribution to the pattern. Clarification: So, in my case, I assume that because the face condition doesn't seem to contribute to the LV at all (no correlation, and a weight of near-zero in the behaviour LV plot), and both other conditions are negatively weighted, then yellow would indicate negative correlations and blue positive correlations between brain activity and behaviour. Correct? A: yes - though you may have some voxels that show a weak positive correlation in the face condition HTH

Hi Anna..
It wasn't quite clear which of Randy's responses to the previous questions you needed to have answered in more detail...

so, I've created a 'help' document that outlines the major steps for evaluating behaviour/seed PLS results

the example is for a 2condition block fMRI experiment, but you simply need to expand the logic to apply it to your dataset..

hopefully this will help..
you can download 'BehavPLS_Interpretation.pdf' from the Rotman ftp server:

ftp://ftp.rotman-baycrest.on.ca/pub/Nancy/PLS_HELP/

cheers,
nancy

Grace...
those barplots come from "Windows>Datamat Correlations Response"
nancy