A quick answer to a question from this afternoon (Vienna time):
I've used both semilandmark metrics (in the same paper, in fact).
There is a different between the apparent (Procrustes) metric
of semilandmark positions on the page and the actual (reduced)
metric the covariance structure must (approximately) obey,
and different displays quite reasonably draw on different choices
in this realm. For instance, permutation tests for group mean difference,
if such a test makes any sense in your particular
scientific context, ought to use distances
normal to the mean outline, as would linear or quadratic discrimination
(studies of local variation _around_ the mean outline).
But shape regressions, relative warp analysis and, of course, actual
thin-plate splines would use both coordinates.
Analogous statements with more coordinates remain true for curve
or surface data in 3D. Regards, Fred B.
--
Replies will be sent to the list.
For more information visit http://www.morphometrics.org
