Back to PLS Helpstatistics of the LV
Posted on 01/31/09 12:35:31
Number of posts: 60
Hello,
In order to be able to judge about the significancy of my LVs, 500 permutation test was conducted which means p=0.002 could be the smallest obtainable P value for the LVs.right? So why I got Lv1 with p<0.000 which refering to your previous posting means p<0.0001.This sounds like an impossible issue according to the statistical rules.Does anybody know the reason for that?
Secondly,I could not really understand how slice-timming can affect the PLS result in such a big dimention.I address this problem since by omiting the slice timing,the priority of my patterns has changed.this means that while I eliminated slice-timming from the preprocessing step,the order of the first two significant LVs was switched.Can anybody provide a possible reason for that?or perhaps give a hint to how one can probabely deceit the PLS by using slice-timming?
Thanks!
/Alireza
In order to be able to judge about the significancy of my LVs, 500 permutation test was conducted which means p=0.002 could be the smallest obtainable P value for the LVs.right? So why I got Lv1 with p<0.000 which refering to your previous posting means p<0.0001.This sounds like an impossible issue according to the statistical rules.Does anybody know the reason for that?
Secondly,I could not really understand how slice-timming can affect the PLS result in such a big dimention.I address this problem since by omiting the slice timing,the priority of my patterns has changed.this means that while I eliminated slice-timming from the preprocessing step,the order of the first two significant LVs was switched.Can anybody provide a possible reason for that?or perhaps give a hint to how one can probabely deceit the PLS by using slice-timming?
Thanks!
/Alireza
Replies:
Untitled Post
Posted on 02/02/09 10:42:16
Number of posts: 394
In
order to be able to judge about the significancy of my LVs, 500
permutation test was conducted which means p=0.002 could be the
smallest obtainable P value for the LVs.right? So why I got Lv1 with
p<0.000 which refering to your previous posting means
p<0.0001.This sounds like an impossible issue according to the
statistical rules.Does anybody know the reason for that?
The p-value reported by PLS is the ratio of (the number of times a singular value for a permuted data set is equal to or higher than the original singular value) / number of permutations. If you never get a permuted singular value that meets the criteria, then the p-value is zero. The estimated p-value you quote I think is the estimated precision given 500 permutations.
Secondly,I could not really understand how slice-timming can affect the PLS result in such a big dimention.I address this problem since by omiting the slice timing,the priority of my patterns has changed.this means that while I eliminated slice-timming from the preprocessing step,the order of the first two significant LVs was switched.Can anybody provide a possible reason for that?or perhaps give a hint to how one can probabely deceit the PLS by using slice-timming?
Slice timing per se is not the problem. Its probably the interpolation that takes place which always affects your signal. I suspect the singular values for the first two LVs are probably very close to each other, so slight changes in the image processing stream can change the variance and also the relative ordering of your LVs. It is not a serious problem.
The p-value reported by PLS is the ratio of (the number of times a singular value for a permuted data set is equal to or higher than the original singular value) / number of permutations. If you never get a permuted singular value that meets the criteria, then the p-value is zero. The estimated p-value you quote I think is the estimated precision given 500 permutations.
Secondly,I could not really understand how slice-timming can affect the PLS result in such a big dimention.I address this problem since by omiting the slice timing,the priority of my patterns has changed.this means that while I eliminated slice-timming from the preprocessing step,the order of the first two significant LVs was switched.Can anybody provide a possible reason for that?or perhaps give a hint to how one can probabely deceit the PLS by using slice-timming?
Slice timing per se is not the problem. Its probably the interpolation that takes place which always affects your signal. I suspect the singular values for the first two LVs are probably very close to each other, so slight changes in the image processing stream can change the variance and also the relative ordering of your LVs. It is not a serious problem.
Login to reply to this topic.
