-------- Original Message --------
Subject: PLS and significance test
Date: Tue, 9 Dec 2008 05:37:40 -0800 (PST)
From: Demetrios Halazonetis <[EMAIL PROTECTED]>
To: <[email protected]>

Hi,

We are investigating integration in the craniofacial structures and running
PLS analysis on configurations of landmarks. We have used tpsPLS, MorphoJ
and IMP (PLSMaker). The problem is that although all three software report
the same correlation coefficients between the two blocks of landmarks, the
permutation tests give very different P values. Here is the output for a
test sample of 30 specimens:

tpsPLS:
Number and percent of correlations >= observed:
  Dim.  Observed Count  Percent
     1  0.726880    19     1.90% <----
     2  0.611517   108    10.80%
     3  0.479918   544    54.40%
     4  0.416872   707    70.70%
     5  0.534085    99     9.90%
     6  0.335551   814    81.40%

MorphoJ:
Singular values and pairwise correlations of PLS scores between blocks:
Singular value  P-value (perm.) % total covar.  Correlation     P-value
(perm.)
PLS1    0,00071285      0,0225  64,620  0,72709 0,4386 <-----
PLS2    0,00040780      0,2129  21,148  0,61194 0,4380
PLS3    0,00023018      0,7320  6,738   0,47995 0,7328
PLS4    0,00017020      0,6327  3,684   0,41794 0,5551
PLS5    0,00016025      0,0441  3,266   0,53389 0,5594
PLS6    0,00006549      0,6783  0,545   0,33590 0,9852

IMP:
For the correlations of the SVD scores
      correlation  Exceedances p-value
s1    0.7269       0      0.000 <----
s2    0.6115       0      0.000
s3    0.4799       0      0.000
s4    0.4169       0      0.000
s5    0.5341       0      0.000
s6    0.3356       0      0.000

As you can see, the correlation for dimension 1 is 0.727 in all three
reports, but the P values (next to the arrows) are completely different (and
do not vary much when different number of permutations are used). Any clues?

Thanks,
Demetris Halazonetis and Nikos Gkantidis
Orthodontic Dept.,
Dental School, Athens University,
Greece.



--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org

Reply via email to