Back to PLS Helptask/group saliences
Posted on 08/20/08 09:00:18
Number of posts: 29
j crofts
Hi, could somebody let me know if I am interpreting my pls correctly. I have the following decomposition A = USV', where U contain my brain saliences, S the singular values (SV's) and V my group saliences. Now, if the permutation tests have returned a small p-value for the first singular value, I can use the first columns of U and V to discern information. For exmple, if V(:,1) = [0.01;0.7;-0.9;-0.03], I can conclude that groups two and three are related to the first SV, similarly I can discern which brain regions are related to the first SV by looking for large values in U(:,1). We may then conclude that brain regions which have large values in U(:,1) are the most active in groups 1 and 2. This process being repeated for each significant SV. Of course not forgetting to use the bootstrap analysis to check the reliability of the results.
One other quick question. In the above the group saliences in V(:,1) for groups 2 and 3 are both large and of opposite signs. What can I conclude, as far as differences between the two groups, from this difference in signs?
Thanks,
Jonathan
One other quick question. In the above the group saliences in V(:,1) for groups 2 and 3 are both large and of opposite signs. What can I conclude, as far as differences between the two groups, from this difference in signs?
Thanks,
Jonathan
Replies:
Untitled Post
Posted on 08/20/08 11:16:13
Number of posts: 394
quote:
Hi, could somebody let me know if I am interpreting my pls correctly. I have the following decomposition A = USV', where U contain my brain saliences, S the singular values (SV's) and V my group saliences. Now, if the permutation tests have returned a small p-value for the first singular value, I can use the first columns of U and V to discern information. For exmple, if V(:,1) = [0.01;0.7;-0.9;-0.03], I can conclude that groups two and three are related to the first SV, similarly I can discern which brain regions are related to the first SV by looking for large values in U(:,1). We may then conclude that brain regions which have large values in U(:,1) are the most active in groups 1 and 2. This process being repeated for each significant SV. Of course not forgetting to use the bootstrap analysis to check the reliability of the results.
One other quick question. In the above the group saliences in V(:,1) for groups 2 and 3 are both large and of opposite signs. What can I conclude, as far as differences between the two groups, from this difference in signs?
Thanks,
Jonathan
Hi Jonathan,Hi, could somebody let me know if I am interpreting my pls correctly. I have the following decomposition A = USV', where U contain my brain saliences, S the singular values (SV's) and V my group saliences. Now, if the permutation tests have returned a small p-value for the first singular value, I can use the first columns of U and V to discern information. For exmple, if V(:,1) = [0.01;0.7;-0.9;-0.03], I can conclude that groups two and three are related to the first SV, similarly I can discern which brain regions are related to the first SV by looking for large values in U(:,1). We may then conclude that brain regions which have large values in U(:,1) are the most active in groups 1 and 2. This process being repeated for each significant SV. Of course not forgetting to use the bootstrap analysis to check the reliability of the results.
One other quick question. In the above the group saliences in V(:,1) for groups 2 and 3 are both large and of opposite signs. What can I conclude, as far as differences between the two groups, from this difference in signs?
Thanks,
Jonathan
The interpretation of the group saliences would be most similar to interpreting a contrasts, thus the vector: [0.01, 0.7,-0.9,-0.03] would represent a contrast between group 2 & 3 and the brain salience would identify the pattern the characterizes this difference.
Randy
Untitled Post
I'm Online
Posted on 08/20/08 11:22:54
Number of posts: 229
also keep in mind that it is important to go back to the data for the best interpretation:
a contrast of -1/+1 could indicate a condition where one group or condition shows 'deactivation' and the other activation, or it could indicate that both show a similar pattern, just different magnitude (e.g, smaller 'activation' in TaskA than TaskB). Either of these two scenarios will produce a -1/+1 contrast
a contrast of -1/+1 could indicate a condition where one group or condition shows 'deactivation' and the other activation, or it could indicate that both show a similar pattern, just different magnitude (e.g, smaller 'activation' in TaskA than TaskB). Either of these two scenarios will produce a -1/+1 contrast
Login to reply to this topic.
