Baycrest Research Division - Enriching Care, Enhancing Knowledge, Enlightening Minds

Rotman > Resources > Support + Services

Task PLS confidence intervals

kwiebels
Posted on 07/05/18 05:57:12
Number of posts: 9

kwiebels posts:

Dear PLS experts,

I have run a non-rotated task PLS with four conditions (A,B,C,D), contrasting A and B vs C and D (i.e. the design is 1 1 -1 -1).

The latent variable is (highly) significant, yet the task PLS brain scores plot showing all four conditions going positively (Mean Values = 110 109 105 101). In addition, the four CIs all overlap quite extensively. However, when one looks at the brain scores and does some statistical tests on them, the brain scores for A and B are significantly greater than those of C and D. In addition, bootstrapped CIs of the difference scores do not cross zero (e.g. taking the difference scores of brain scores for i) A vs C and ii)A vs D, and then bootstrapping a CI around these difference scores produces CIs that do not cross zero).

So, my question is whether these CIs of difference scores are more informative than the CIs produced in the plsgui software, as these are measures of across-subject variance, whereas the comparisons of interest are within-subject.

Thanks for your time,

Kristina and Reece

Replies:

Untitled Post

rmcintosh
Posted on 07/05/18 08:48:37
Number of posts: 394

rmcintosh replies:

quote:

Dear PLS experts,

I have run a non-rotated task PLS with four conditions (A,B,C,D), contrasting A and B vs C and D (i.e. the design is 1 1 -1 -1).

Thanks for your time,

Kristina and Reece

Thanks for the insightful post. Please keep in mind that all the analyses you have done are looking at within-subjects effects, which includes the confidence intervals around the brain scores without doing the differneces. The current version of PLS does not do the difference assessment that you did mainly because the permutation test is assessing that as well - though based on null hypothesis testing rather the reliability (which is what the CI calculation is for). If you want use bootstrap to assess the overall differences, then what you did is legitimate.

I am hoping we will be able to get the whole PLS framework updated by the year's end, so if you think this additional feature would be helpful, we can include it.

thanks

Randy

Untitled Post

kwiebels
Posted on 07/09/18 18:26:53
Number of posts: 9

kwiebels replies:

Thanks for your quick response, Randy.

We were mostly interested in clarifying our own understanding of the CIs on task PLS plots when the design is within subjects. It would be great if this could be added to the next version of PLS, but it's easy enough to extract on your own so no worries if it's not included.

All the best,

Kristina and Reece

Baycrest Research Division - Enriching Care, Enhancing Knowledge, Enlightening Minds

Replies:

Contact Us: