Date: Tue, 22 Oct 2002 11:43:38 0700
ReplyTo: Michael Healy <healym@earthlink.net>
Sender: "SPSSX(r) Discussion" <SPSSXL@LISTSERV.UGA.EDU>
From: Michael Healy <healym@earthlink.net>
Subject: Statistical Questions about MDS/INDSCAL
Contenttype: text/plain; charset="USASCII"
Hello SPSS list readers,
I have a statistical question about MDS/INDSCAL modeling I hope that
this list might be able to help me with.
I obtained a matrix unconditional INDSCAL model that recovered the
predicted 2dimensional structure. However, the angle of rotation of the
group configuration space was such that the dimensions were not clearly
defined in the CONFIG weights. That is, the group configuration space was
rotated about 45 degrees from Dimension 1 being parallel to the xaxis. My
interest was examining the subject weights in this space in relation to
several external measures. Since Dimension 1 and 2 were not in line with
the x and yaxes, it seems that the Dimension 1 and Dimension 2 weights
would be sharing some of the same information. To correct for this, I
rotated the group configuration space so that Dimension 1 was parallel to
the xaxis and refit the INDSCAL model by telling ALSCAL to use the rotated
configuration and that this configuration was FIXED.
My question is whether doing this rotation/refitting is acceptable or
whether I have introduced some sort of error or bias Into my solution.
Another question I have is that the data I am modeling are essentially
zscores, although I believe them to be only fairly stable estimates of
distances given the data collection method. I am fitting these models
setting at both the RATIO and INTERVAL levels of measurement, and I am
finding that the solutions differ depending upon the measurement
levelespecially in the subject weights. My concern is that the RATIO
level is more appropriate for the data, but because the ratio fits is more
restrictive the difference in the solution/subject weights is reflecting
lack of fit to a greater extent that the interval level solutions. Are
there any guidelines for deciding which level of measurement Is appropriate?
Thanks again for you help and any feedback would be appreciated.
Sincerely,
Mike

Michael R. Healy, M.A., A.B.D
Claremont Graduate University/Pitzer College
Department of Psychology
170 E. Tenth St.
Claremont, CA 917116163
Claremont Memory and Aging Project: 9096074499
http://home.earthlink.net/~healym/healym.htm


