Arnaud Trollé wrote:
Thank you for your answer,
With regard to what you told, I would just need a few precisions, in:
> to compare INDSCAL solutions only permutations and scalings of
solutions are authorized.
does ''scalings'' involve that the dilatations along the dimensions
are only allowed according to a _same _factor ?
each dimension can be scaled independent from the other dimensions. That
is, if for example, an Indscal solution has vectors a1, a2, a3 for
stimuli coordinates and c1, c2, c3 containing subjects weights, then one
can multiply a1 with any value s, provided that c1 is divided by s^2;
also a2 can be multiplied by any value t, provdied that c2 is divided by
t^2, etc.
Concerning the Tucker coefficient of congruence:
>... where a and b are two vectors of coordinates form the two
different Indscal solutions.
By ``vectors'', did you also include ''arrays ''( of size
nbpoints*nbdimensions in the present case) ?
No
Or must the calculcation be done separately for the couples of
dimensions (dimension 1 config1 ---> dimension 1 config2, (dimension 2
config1 -----> dimension 2 config2)
Yes; specifically, I meant in the example above that, if a second
INDSCAL solution has vectors u1, u2, u3 for stimuli coordinates and v1,
v2, v3 containing subjects weights, then to compare the solutions one
should compute
phi(a1,u1), phi(a2,u2), phi(a3,u3) and phi(c1,v1), phi(c2,v2), phi(c3,v3) .
Does the comparison via this coefficient imply that the 90°rotations,
reflections, dilatations must be all proceeded beforehand ?
Permutations and reflections should be done in advance (if reflections
are forgootten, then phi simply gets a negative sign); dilations are not
necessary, because phi is not sensitve to dilations.
with best regards
Henk
--
Henk A.L. Kiers
Heymans Instituut
University of Groningen
Grote Kruisstraat 2/1
9712 TS Groningen
The Netherlands
tel. +31503636339
fax. +31503636304
email: [EMAIL PROTECTED]
www.gmw.rug.nl/~kiers
----------------------------------------------
CLASS-L list.
Instructions: http://www.classification-society.org/csna/lists.html#class-l
