Back to PLS HelpHow many LV's should I keep?
Posted on 08/16/08 11:29:49
Number of posts: 29
j crofts
Hi,
In the literature you suggest that we should remove any LV's such that, the chance of permuted values greater than observed are very large. My question is what counts as very large? In my PLS I have the following values for result.perm_result.sprob = [0, 0.03, 0.17,0.98]. Is 0.17 large?
Thanks,
Jonathan
In the literature you suggest that we should remove any LV's such that, the chance of permuted values greater than observed are very large. My question is what counts as very large? In my PLS I have the following values for result.perm_result.sprob = [0, 0.03, 0.17,0.98]. Is 0.17 large?
Thanks,
Jonathan
Replies:
Untitled Post
Posted on 08/16/08 11:38:02
Number of posts: 394
That is a good question. Keeping in mind what you are assessing with the permutation test and bootstrap, it is possible that an effect with a p-value of 0.17 is reliable by bootstrap. The p-value is derived from assessing how often a random resampling of the data produces a singular value as large as the original sample order. If you are comfortable with an answer of 17% (0.17), then its not too large. If you stick to conventional alpha-levels (0.05, 0.10), then 0.17 is too large. I would strongly suggest you use the combination of permutation, bootstrap and physiological validity to guide your decision.
Randy
Randy
Login to reply to this topic.
