Hi again;
I've got more questions to ask about some aspects of the TpsRelw and the
Tpsregr:
The first question is about the alpha: when is desirable to give
greater emphasis to shape variations at larger or smaller spatial scales
(e.g. with alphas of 1 and -1,respectively) ? Does the alpha change only
the values of the relative warps (as I think) or it affects also the
partial warps?
I've read the Tpsregr help file but I still have some doubts about how
to use it to test with this program if it exists change of shape during
growth and/or between sexes, and if there is interaction between them (I
should perform a MANCOVA, shouldn't I?).
I've created a file that contains two columns; the first one with the
values of the centrois size and the other one with the values of sex (1
for females and -1 for males). At this point I don't know if I must
compute the regression choosing both variables at a time only, or I
should run the regression with each of two variables...
Moreover, I'm not sure what the program performs when it's choosen a
nominal variable ( as sex). I think that it should be a MANOVA, but the
report listing is the same than when I choose the centroid size - I mean
the general aspect ( what is performed), not the values.
Despite the sample report listing there are still some aspects that I
need to clarify about it;
The first of all is the result of the test; I suppose that the null
hypothesis must be (in my case) that it doesnt exists shape change
during growth or between sexes. If the p-value of the statistics (the
Wilk's lambda) is < than the significance level I must reject the null
hypothesis.
I don't know if it's possible that change exists -e.g. during growth-
at some spatial scales but not at all of them.;can this be given by the
p-value of the results of the regressions for each shape variable?. I
think that if the p-value is high, it means that shape changes can not
be predicted by growth changes although the R2 is high.
One more question is about the percent unexplained as "an overall
measure of fit". Is this related with the existence, or not, with (e.g.)
shape changes during growth?. I mean, if the percent unexplained is high
( a few percentage of the shape change is explained by the change in
lenght) it is possible that the p-value of the Wilk's lambda is
significative, and therefore that shape change during growth.
The last question is about the Generalized Goodall F-test: and the
permutation test. I think that both of them are a measure of Goodness of
fit. If the p-value is < than the significance level, then there a good
goodness of fit (I think that it's the expected or desirable result).
However I don't know in which situations there can be a lack of fit. In
relation to the permutation test I don't know if it is as well a measure
of goodness of fit or it represents another thing. I don't know neither
what is considered a small percentage.
Well, that's all. I am conscious about the great amount of questions I
am asking, but I do really appreciate some help.
Thanks indeed.
Javier
Mariana ([EMAIL PROTECTED])
Barcelona
(Spain)
Although maybe these questions should be addressed to James Rohlf
because are related to their programs I prefer address them to morhpmet;
In this way some other people may solve their doubts or learn from the
questions (and obviously and mainly from the answers).
==
Replies will be sent to list.
For more information see http://life.bio.sunysb.edu/morph/morphmet.html.
