Back to PLS HelpPosted on 08/07/13 18:07:57
Number of posts: 2
Hello,
I am interested in calculating effect size for one of my nonsignificant LVs. The crossblock covariance is 13%, which I understand to mean 13% of the crossblock covariance is explained by this LV (please correct me if I'm wrong!). Is this equivalent to effect size, i.e., is it fair to say that this LV had a "medium" effect size?
I hope that makes sense. Thanks!
Meera
Replies:
Posted on 08/08/13 08:34:53
Number of posts: 394
HI Meera - the percent of crossblock covariance is not comparable to effect size since its only relevant to the covariance matrix rather than the entire variance/covariance structure. For example, if you add another contrast or behavior measure to the analysis, the covariance matrix changes and so too does the %'s.
I would suggest you look at the confidence intervals for your scores/correlations to get a better feel for the effect size in the non-significant LV.
Posted on 08/10/13 20:16:46
Number of posts: 2
Great thanks Randy. How do I obtain the actual numeric values for the CIs?
Posted on 08/11/13 14:53:03
Number of posts: 394
Great thanks Randy. How do I obtain the actual numeric values for the CIs?
I guess that would help, eh :)
is the results file from pls there is structure array called boot_result (result.boot_result). In there are the upper and lower limits of the confidence intervals. There are prefaced by 'ul' and 'll' and the rest of the name will depend on the analysis (e.g., for behavior/seed ulcorr and llcorr)
if you type 'help pls_analysis' when you are in matlab it will give you the exact names
Posted on 12/04/15 10:52:32
Number of posts: 3
I guess that would help, eh :)
is the results file from pls there is structure array called boot_result (result.boot_result). In there are the upper and lower limits of the confidence intervals. There are prefaced by 'ul' and 'll' and the rest of the name will depend on the analysis (e.g., for behavior/seed ulcorr and llcorr)
if you type 'help pls_analysis' when you are in matlab it will give you the exact names
Hi Randy,
Just to follow up on this - is there any way to find out the effect size from the datamats/results files? Otherwise, what would you suggest using as an alternative to state the effect size of the results? I've listed the crossblock covariance, but not sure if this is enough.
Thank you in advance for your help!
Posted on 12/04/15 15:07:26
Number of posts: 394
Hi Randy,
Just to follow up on this - is there any way to find out the effect size from the datamats/results files? Otherwise, what would you suggest using as an alternative to state the effect size of the results? I've listed the crossblock covariance, but not sure if this is enough.
Thank you in advance for your help!
The confidence intervals are the best way to convey effect size as it captures the differentiability of the parameter estimates based on the sampling distribution.
Login to reply to this topic.
