Hello,
Can you ever be justified in discussing a non-significant LV?
Regards
Andrew
Hello,
Can you ever be justified in discussing a non-significant LV?
Regards
Andrew
yes, but with all these situations it depends - can you give more information on the analysis and outcome?
It is 3 (condition) by 2 (group) analysis. I have one sig LV explaining ~50% of variance and a second LV explaining ~25% of the variance but only p=0.15. The first LV provides novel evidence for shared neural activity between the groups and two of the conditions. The second LV nicely separates two of the conditions for one of the groups but not the other and reveals activation in regions consistent with previous research.
Would you like anymore information?
Would you like anymore information?
Thanks - the info you gave is fine. What I would suggest is presenting both the p-values and the confidence intervals around the brain scores for both LVs to give an idea of the effect size. You should also run the two groups separately as a sort of test for simple effects that one would do in an ANOVA type analysis. You can map these separate analyses to the full analysis by correlating the brainLVs, which you will need to do at the command-line level. It will be helpful however to show, that for example, in group 1, their dominant effects is split between two LVs for the full group analysis.
You can do this by: (assume Fbrainlv is from the full analysis and g1brainlv and g2brainlv are from the separate group)
Fbrainlv'*g1brainlv
Fbrainlv'*g2brainlv
the products of these will be the cosine (correlation) between analyses, where the rows will be the Full analysis LVs and the columns will be for the separate groups.
Thanks, I'll give this a go.
Hi,
Just returned from overseas and returning to this analysis. I thought I would show you the brain scores for LV2 and get your opinion. Is this reportable, even though the LV is non-sig? The effect seems to be quite large. I have tried running the groups separately but this doesn't "improve" things.
Thanks for your help.
It doesn't seem to allow me to post an image
I wish I could give you a clear answer. The situation you are facing is where you are playing off "significance" vs "reliability". The permutaiton test gives you p=0.15, which by covential thresholds is non-significant, but does mean that only 15% of the time in a random sample do you get this particular effect, so...
this might be helpful: http://amstat.tandfonline.com/doi/pdf/10.1080/00031305.2016.1154108
and then the bootstrap results indicate a reliable difference in one group but not the other. I tend to put more weight on this, but how you convey this in the write up will take some work. You may wish to supplement the PLS with some univariate analyses to characterize these differences more (not as another null hypothesis test)
Do you know of any examples in the literature where they have discussed a similar LV?
Do you know of any examples in the literature where they have discussed a similar LV?
There are but I don't have an example at the tip of my fingers. It might be easiest to do a search of PLS + fMRI to see whats been done.
Baycrest is an academic health sciences centre fully affiliated with the University of Toronto
Privacy Statement - Disclaimer - © 1989-2024 BAYCREST HEALTH SCIENCE. ALL RIGHTS RESERVED