Back to PLS Help

Update on CI calculations
rmcintosh
Posted on 12/12/14 17:15:40
Number of posts: 385
rmcintosh posts:

Hi all,

We've detected a bug in the calculation of the confidence intervals for non-rotated task PLS and have posted the updated code.  Please download this when you can.  The problem was an incorrect mean centreing in the bootstrap loop so the bounds of the confidence intervals for the scores were incorrect.  

 

thanks!

Randy

Replies:

Trouble to download the latest version
Marine
Posted on 08/22/16 09:27:15
Number of posts: 7
Marine replies:

Dear Professor McIntosh,

Many thanks for this helpful software :-)

However, when I try to download the latest version of PLS on the download page of the website (https://www.rotman-baycrest.on.ca/index.php?section=345), the zip file I have extracted seems to include the Version 6.1311050 instead of the latest one (Version 6.15).

Perhaps I have missed a point about the correct procedure to download ?

 

Thanks in advance for your help,

 

Kind regards,

 

Marine Manard



Untitled Post
rmcintosh
Posted on 08/22/16 16:14:00
Number of posts: 385
rmcintosh replies:

quote:

Dear Professor McIntosh,

Many thanks for this helpful software :-)

However, when I try to download the latest version of PLS on the download page of the website (https://www.rotman-baycrest.on.ca/index.php?section=345), the zip file I have extracted seems to include the Version 6.1311050 instead of the latest one (Version 6.15).

Perhaps I have missed a point about the correct procedure to download ?

 

Thanks in advance for your help,

 

Kind regards,

 

Marine Manard

hi Marine - thanks for pointing this out.  The version you have is indeed the latest.  We neglected to change the version number in the script but what you have is 6.15



Untitled Post
Marine
Posted on 08/22/16 17:24:36
Number of posts: 7
Marine replies:

Many thanks for this fast response !

I was thinking about this option, but when I run the Non-rotated analysis in the new version, the CI bars are still outside the graph. Is my design can cause this problem ? 

Many thanks in advance and sorry for this very basic question...

Have a nice day,

Marine

 



Untitled Post
rmcintosh
Posted on 08/22/16 17:37:41
Number of posts: 385
rmcintosh replies:

quote:

Many thanks for this fast response !

I was thinking about this option, but when I run the Non-rotated analysis in the new version, the CI bars are still outside the graph. Is my design can cause this problem ? 

Many thanks in advance and sorry for this very basic question...

Have a nice day,

Marine

 

The CI's are the location of the upper and lower bounds of the interval and not the distance from the point estimate.  For example, if the mean is 2 and upper limit 5 and lower limit 1, the bars should be set to go up to 5 and down to 1.

 

does that help?  If not, post the number and the design LV and we can figure it out



Untitled Post
Marine
Posted on 08/23/16 03:45:43
Number of posts: 7
Marine replies:

quote:

The CI's are the location of the upper and lower bounds of the interval and not the distance from the point estimate.  For example, if the mean is 2 and upper limit 5 and lower limit 1, the bars should be set to go up to 5 and down to 1.

 

does that help?  If not, post the number and the design LV and we can figure it out

Many thanks again for this response and for your help !

It is ok for me that the CI bars are the upper and lower limits, if it is only a display problem, I can reproduce the figure in excel without difficulty, but I prefer to be sure to use correct values :-)

The values I can see on the PLS graph attached (PLS brain score plot) seem similar to the values found in the boot_result structure, with orig_usc = 22.12 & -24.02; ulusc = 21.26 & -12.70; llusc = 13.09 & -22.62.

The design LV in the result file ("v") is [0.7071; -0.7071] and the singular value ("s") is 32.6285

My design is not optimal, using only four subjects to contrast conditions (it is a pilot study). Therefore, I have used the non-rotated task PLS, using only two conditions with "1 -1" as contrast values, and set permutations to 300 and bootstrap to 30 (more permutations or bootstrap causing an error message, due to the small number of subjects I suppose). I have used the mean centering option of 0 to "boost conditions differences", given I have only one group of participants.

Therefore, if I rightly understood, the last version of PLS should take into account the updates made for the Version 6.14 December 9, 2014

	1.	New:  Fixed bug in using the wrong mean-centering
		approach for confidence interval estimate for
		non-rotated PLS scores

However, the graph obtained (see attached) and the result file for this non-rotated analysis (in the last downloaded version) have the same values and display problems than the same analysis ran in the older version of PLS (Version 6.1311050), so I can reproduce easily the graph on excel with values from the result file but I prefer to be sure that I did not miss an important point (on mean centering for example) and that the display problem is only a display problem ;-)

Many thanks for your help !

Kind regards,

Marine

PS: the graph image was send by dropbox to your mailbox, I don't know why but I can't save the image here directly

 



Untitled Post
rmcintosh
Posted on 08/23/16 06:53:21
Number of posts: 385
rmcintosh replies:

quote:

Many thanks again for this response and for your help !

It is ok for me that the CI bars are the upper and lower limits, if it is only a display problem, I can reproduce the figure in excel without difficulty, but I prefer to be sure to use correct values :-)

The values I can see on the PLS graph attached (PLS brain score plot) seem similar to the values found in the boot_result structure, with orig_usc = 22.12 & -24.02; ulusc = 21.26 & -12.70; llusc = 13.09 & -22.62.

The design LV in the result file ("v") is [0.7071; -0.7071] and the singular value ("s") is 32.6285

My design is not optimal, using only four subjects to contrast conditions (it is a pilot study). Therefore, I have used the non-rotated task PLS, using only two conditions with "1 -1" as contrast values, and set permutations to 300 and bootstrap to 30 (more permutations or bootstrap causing an error message, due to the small number of subjects I suppose). I have used the mean centering option of 0 to "boost conditions differences", given I have only one group of participants.

Therefore, if I rightly understood, the last version of PLS should take into account the updates made for the Version 6.14 December 9, 2014

	1.	New:  Fixed bug in using the wrong mean-centering
		approach for confidence interval estimate for
		non-rotated PLS scores

However, the graph obtained (see attached) and the result file for this non-rotated analysis (in the last downloaded version) have the same values and display problems than the same analysis ran in the older version of PLS (Version 6.1311050), so I can reproduce easily the graph on excel with values from the result file but I prefer to be sure that I did not miss an important point (on mean centering for example) and that the display problem is only a display problem ;-)

Many thanks for your help !

Kind regards,

Marine

PS: the graph image was send by dropbox to your mailbox, I don't know why but I can't save the image here directly

 

can you email the figure to me directly: rmcintosh@research.baycrest.org

 

thanks



Untitled Post
rmcintosh
Posted on 08/23/16 07:38:21
Number of posts: 385
rmcintosh replies:

actually it seems to be fine (I see the figure now).  all good - please continue!

 

randy




Login to reply to this topic.

  • Keep in touch

Enter your email above to receive electronic messages from Baycrest, including invitations to programs and events, newsletters, updates and other communications.
You can unsubscribe at any time.
Please refer to our Privacy Policy or contact us for more details.

  • Follow us on social
  • Facebook
  • Instagram
  • Linkedin
  • Pinterest
  • Twitter
  • YouTube

Contact Us:

3560 Bathurst Street
Toronto, Ontario
Canada M6A 2E1
Phone: (416) 785-2500

Baycrest is an academic health sciences centre fully affiliated with the University of Toronto