Back to PLS Helpstatistics within a LV
Posted on 09/11/08 13:11:46
Number of posts:
Hi,
In our analysis (mean centered task PLS) three conditions were separated with values of:
condition 1: - .76
condition 2: .56
condition 3: .05.
Thus, conditions 2 and 3 were grouped together and showed a different brain pattern than condition 1. Is there a way to test that conditions 2 and 3 are statistically different to make the claim that condition 3, which is near zero, is not contributing significantly to the brain pattern revealed by the task analysis?
Also, is there a way in a behavioural PLS to test that - if two conditions show both positive correlations but one is smaller than the other - that the two correlations are statistically different?
I ask only because a reviewer wants these stats.
I hope that make sense.
Thanks!
Replies:
Untitled Post
Posted on 09/13/08 09:04:35
Number of posts: 394
Hi Erin,
In both the mean-centred PLS and behavior PLS, you can use the confidence interval estimation provided by bootstrap to make the inferences.
in task PLS under Windows:Task PLS Brain scores with CI
in behavior: Windows:Brain Scores vs Behavior:Correlation Overview or Show Correlation Plot
The error bars on these plots are the 95% confidence interval.
Randy
In both the mean-centred PLS and behavior PLS, you can use the confidence interval estimation provided by bootstrap to make the inferences.
in task PLS under Windows:Task PLS Brain scores with CI
in behavior: Windows:Brain Scores vs Behavior:Correlation Overview or Show Correlation Plot
The error bars on these plots are the 95% confidence interval.
Randy
Untitled Post
Posted on 09/15/08 15:09:28
Number of posts:
Hi Randy,
Thanks for the information; however, I can't seem to find that option. Is it only available in Version 5 (I have version 4)?
Also, I was told that what I could do is extract the brain scores for each participant and do an ANOVA. Can you tell me how to go about doing this?
Thanks,
Erin
Thanks for the information; however, I can't seem to find that option. Is it only available in Version 5 (I have version 4)?
Also, I was told that what I could do is extract the brain scores for each participant and do an ANOVA. Can you tell me how to go about doing this?
Thanks,
Erin
Untitled Post
I'm Online
Posted on 09/15/08 15:13:58
Number of posts: 291
This option is only available when you run bootstrap test for Task PLS.
Untitled Post
Posted on 09/15/08 15:24:44
Number of posts:
Hi Jimmy,
We did run a boostrap test (12 participants). Is there some other reason I'm not finding it?
Thanks,
Erin
We did run a boostrap test (12 participants). Is there some other reason I'm not finding it?
Thanks,
Erin
Untitled Post
I'm Online
Posted on 09/15/08 15:50:07
Number of posts: 291
If you load the result file, 'boot_result' structure should contain fields like 'orig_usc', 'ulusc', 'llusc', etc. If you cannot find those fields, you will have to download the latest version and try again. That feature was add in this June 3, 2008.
t-test using brain score values?
Posted on 09/20/10 18:55:02
Number of posts: 34
quote:
This option is only available when you run bootstrap test for Task PLS.
Hi,
I have a similar question (as the original poster's) regarding the possibility of doing further statistics on the LV's and the answer was that there's a way to do it (comment about "bootstrap test for Task PLS" above). I understand that it's not a simple matter of taking "the average of brain score values for each condition" (as Jimmy replied below)......so what are the values that I need to do a paired-sample t-test (for example) that could confirm what's shown on the LV profile as being "significant" or not (by the CI's) and how can I extract those? Thanks for your help in advance!
Miyoung
----------------------------------------------------------
Task PLS with CI - part 2
mkim
Posted on 11/03/09 16:27:49
Number of posts: 33 mkim posts:
Hi. I ran a Task PLS analysis (with CI) and am trying to re-create the LV profiles in Excel. I have 2 groups with 18 subjects in Group1 and 13 subjects in Group2, and there are 6 conditions. The analysis gave me 12 LV's. I extracted the Brain Score values (b_score) and got 12 columns (for the 12 LV's, I think) and the 186 rows of brain scores represent the number of subjects per condition per group (I think): LV1 (for example) 18 condition1 18 condition2 18 condition3 18 condition4 18 condition5 18 condition6 13 condition1 13 condition2 13 condition3 13 condition4 13 condition5 13 condition6 Is that correct? I averaged the brain score values for each condition but only the Group1 averages matched the Brain Score LV profile......did I miss something? Thanks once again for you help! Miyoung
Replies:
Untitled Post
I'm Online jimmy
Posted on 11/03/09 16:48:34
Number of posts: 137 jimmy replies:
The rows that you mentioned are correct. It's always subject in condition in group.
Brain score in plot "Task PLS with CI" is not just the average of brain score values for each condition, because the 'datamat' here in the calculation should also be mean-centered.. i.e.:
Rather than: task_mean( datamat * brainlv )
It is: task_mean( mean_center( datamat) * brainlv )
That's why I told you to re-run the analysis.
If you still cannot match the numbers, please upload your datamats and result file onto your FTP server, and post the URL here. So I will look into it.
----------------------------------------------------------
This option is only available when you run bootstrap test for Task PLS.
Untitled Post
Posted on 09/20/10 20:30:02
Number of posts: 394
As in Jimmy's reply, the b-scores can be extracted from the results file for further analysis. You can also get the upper and lower bounds for the confidence intervals around the score means from the results files. These are structure arrays so you will need to extract them from the command line rather than the GUI.
Randy
Randy
Untitled Post
I'm Online
Posted on 09/21/10 00:23:12
Number of posts: 291
Hi Miyoung:
First, I suggest that in the future you copy and paste only the link instead of the whole text, which looks much clearer. e.g.:
http://www.rotman-baycrest.on.ca/index.php?action=view_thread&id=927&module=bbmodule&src=%40random45c35fcb17881
Second, let me sort your question. You are trying to ask: "What are the values (guess: result values?) that I need to do a paired-sample t-test (for example) that could confirm what's shown on the LV profile as being "significant" or not (by the CI's), and how can I extract those (guess: result values?)?"
In that particular link, Erin also asked: "I was told that what I could do is extract the brain scores for each participant and do an ANOVA. Can you tell me how to go about doing this?"
One thing that you may want to keep in mind that PLS is a nonparametric multivariate analysis method (aka MANOVA). If you are trying to confirm its result with standard ANOVA or any parametric method, you are probably comparing an apple with an orange.
If you are interested in the significance of the LV, you should check the permutation result. The bootstrap result only tells you the reliability of each voxel.
If you are trying to extract the brain scores of the particular topic (refer to link above), you can load the result file in command window, and the brain score is under "boot_result.orig_usc".
If you are trying to compare the PLS results with standard ANOVA test, or trying to run ANOVA with the PLS results, I would not recommend you do so.
Jimmy
First, I suggest that in the future you copy and paste only the link instead of the whole text, which looks much clearer. e.g.:
http://www.rotman-baycrest.on.ca/index.php?action=view_thread&id=927&module=bbmodule&src=%40random45c35fcb17881
Second, let me sort your question. You are trying to ask: "What are the values (guess: result values?) that I need to do a paired-sample t-test (for example) that could confirm what's shown on the LV profile as being "significant" or not (by the CI's), and how can I extract those (guess: result values?)?"
In that particular link, Erin also asked: "I was told that what I could do is extract the brain scores for each participant and do an ANOVA. Can you tell me how to go about doing this?"
One thing that you may want to keep in mind that PLS is a nonparametric multivariate analysis method (aka MANOVA). If you are trying to confirm its result with standard ANOVA or any parametric method, you are probably comparing an apple with an orange.
If you are interested in the significance of the LV, you should check the permutation result. The bootstrap result only tells you the reliability of each voxel.
If you are trying to extract the brain scores of the particular topic (refer to link above), you can load the result file in command window, and the brain score is under "boot_result.orig_usc".
If you are trying to compare the PLS results with standard ANOVA test, or trying to run ANOVA with the PLS results, I would not recommend you do so.
Jimmy
Login to reply to this topic.
