A) I was wondering if anybody know how one can get the following or similar to following statistics that one can get in SAS software when doing Cluster Analysis: 1. RMSSTD (root-mean-square total-sample standard deviation) for measuing homogeneitiy of new clusters 2. SPR (Semipartial R-sqared) for measuring homogeneity of merged clusters 3. RS (R-Squared) for measuing hetrogeneitiy of clusters 4. CD (distance between two clusters)for measuing homogeneitiy of merged clusters

B) Is it appropriate to use categorical independent variables in Discriminant Analysis?

Thanks in advance.

Best regards, Sanjay

This is a multi-part message in MIME format. ------=_NextPart_000_0000_01C5E3F0.C8B2E2B0 Content-Type: text/plain; charset="Windows-1252" Content-Transfer-Encoding: quoted-printable From: Steven R Brown Date: Mon Nov 07 15:52:59 CST 2005 T Q-METHOD@LISTSERV.KENT.EDU Subject: Re.: Use of SPSS I must disagree with some of the presumed advantages of SPSS advanced by David Goldstein. At 05:32 PM 11/6/2005 -0500, David M. Goldsteinwrote: Some of the pluses of using SPSS: It provides some help with the issue of how many factors to extract.=20 Steve,=20 Do you decide how many factors are present only after rotation=20 Particularly in connection with judgmental rotations, the nature of the factors and which to retain is a judgment that emerges as the = factoranalyst interacts with the data. A priori criteria, such as Cattell's scree test or eigenvalues greater than 1.00, are onlyassociated with the statistical properties of the data.It provides somedifferent ways of rotating. I have found that the Equimax method keeps onyielding results which are consistent with judgemental rotation.This has to be mainly accidental.

I understand your point.. However, I keep finding that the Equimax method of rotation keeps on producing the same results as when I use judegmental rotation. The Lipset data is usually used to illustrate the advantages of judgmental rotation. The Equimax rotation produces the same result. Is it possible that when a person is doing judgmental = rotation that the person is doing something similar to what Equimax is doing? In a sense, Equimax rotation could me a model of what is happening with judgmental rotation.

That is, there can be nosystematic connection between equimax or any other automatic rotationprocedure (which responds only to the statistical topography of the data) and judgmental rotation, which mainly responds to content, or totheoretical considerations. The statistical configuration can ofcourse influence the analyst's decision making, but there are too manyother considerations (of which equimax, varimax, and other rotations are oblivious) that will also play a role.

One can obtainfactor scores just like with the other programs. True enough, but PCQ and PQMethod provide analyses of factor scores(based on standard error formulas) that are missing in SPSS, and probablySAS (with which I am not familiar). Steve, I don undertand this statement. Can you explain it a little more. One can analyze the items for patterns over time. I use the k-means cluster analysis program. This is probably similar to the repertory grid people. I'm sure that SPSS can do this, but looking at individual items outside of the context of the overall response that gives them meaning (if thisis in fact what is being done) is to violate the gestalt principle. Steve, The k-means cluster analysis results in clusters of items. In the way that I am doing it, each cluster consists of items which have a similar pattern over the therapy sessions. So there is a getalt involved. One can createnew variables from the factors which result. For example, I created aQ-sort which reflected the degree of conflict between the differentself-images which resulted. The Conflict Q-Sort can then be entered intothe analysis with the other Q-sorts.

I'm not sure what this implies. How exactly was the new Q sortcreated? Steve, This analysis can be seen in the case study which I recently published in Clinical Case Studies. For each item, I looked at the largest difference in factor scores (expressed as a z-score) among the three factors. The factors were interpreted to reflect self-images of the person. For each item, if the largest difference was zero, this shows that the different self-images are the same. If the largest difference was maximal, this would show conflict between the self-images. The z-score different scorses were rank ordered from smallest to largest and this became a Q-sort that reflected conflict amont the self-images.Incidentally, I'm not saying that any of the above statistical strategies(e.g., analyzing items over time) isn't useful for some purpose oranother, but it's equally important to be mindful of the principles fromwhich these practices depart.

[此贴子已经被作者于2005-12-22 11:07:40编辑过]

Can anybody offer some help in the following problem? I want to analize clusters. I know how to do it when data is in the form of variables and subjects, but I have data that is directly in the form of similarity matrices. If I have one similarity matrix for each subject, can I input these matrices as data and perform the analisys from there? How do I do it? How about if I have a similarity matrix that is a summary of all subjects?

Thanks,

Sergio Chaigneau.

Best regards

hi spss listers

i would like to get some help on an output when i do a simple hierarchical cluster analysis:given below I have tried it both on version 10 and 8 for windows

Proximities

Warnings An error was encountered when attempting to open the input file named in the MATRIX subcommand. Check the existence and contents of the matrix input file. This command is not executed.

Cluster

Error # 5260 in column 16. Text: C:\TEMP\spssclus.tmp The file named above is empty or is not an SPSS data file. This command not executed Didier Gerard Rodney Soopr

Amanien