----- Forwarded message from Paolo Piras <[email protected]> -----
Date: Wed, 26 Jun 2013 12:01:34 -0400
From: Paolo Piras <[email protected]>
Reply-To: Paolo Piras <[email protected]>
Subject: TPSsmall
To: "[email protected]" <[email protected]>
Hi all,
I'm writing a R function that does the same things of TPSsmall
the comparison of geodesic and tangent procrustes distances between specimens
(i.e. nxn matrix) is already implemented in regdist()
of Morpho package;
I want to do the comparison using the **distances from the consensus**. The
comparison is done via linear model on these two vectors
As we all already know in MOST biological applications this exploration is
useful just to detect digitization errors, because, *USUALLY* variation is
rather small in biological data.
This is not true for simulations on very different shapes.
Thus one wants to test if the slope from the model: lm(euclidean~riemannian-1)
(the regression is trough the origin)
is different from 1, i.e. from the complete identity between the two types of
distances.
However I wonder if I can use the function linearHypothesis() from car package
because this statement in the TPSsmall help pages, mainly the last sentence:
"This window displays information about the data file, the consensus
configuration, and a listing of the numerical results produced by the program.
The program presents the minimum, maximum, and means for the Procrustes
distances and the distances in the tangent space. For reference, the maximum
Procrustes distance (p/2) is also displayed. These are followed by Y-intercept
and regression coefficient for the regression through the origin of the tangent
distance on to the Procrustes distance. The uncentered product-moment
correlation coefficient is also given (note: statistical significance test
cannot be applied since the distances are not independent and because the two
distances matrices are simply different mathematical functions of the same set
of data).
Copyright © 1999 by F. James Rohlf"
I think it is referred to test significance of the coefficient PER SE thus if
the slope is different from 0. In this case...one cannot neither test if the
slope is different from 1, is'nt?
Do you think that can be tested?
best
paolo
----- End forwarded message -----
