s-news
[Top] [All Lists]

multivariate analysis

To: s-news@wubios.wustl.edu
Subject: multivariate analysis
From: Florin Vaida <vaida@sdac.harvard.edu>
Date: Fri, 23 Mar 2001 15:22:15 -0500 (EST)
Dear all,

I have a statistical question.  Suppose that I collect data on
N individuals, on k items.  E.g., how well individual i answers question
#j.  My response is continuous.  I am interested in
(i) assessing the strength of correlation between the k questions
(ii) `deconvolving' the `common signal' for the k questions, from the
`question-specific' signal.

1. Do you have any advice on approaching the problem, and possible
references?

2. For (i), the determinant det(R) of the sample correlation matrix for
the k items gives a 1-number summary of the strength of correlation.
Bartlett 1954 has proposed a test for noncorrelation (R=I) based on
det(R).  I would be interested in any insight of how good a measure 
of correlation det(R) is, and whether people have looked at CI's for R,
distributions under a correlated model, etc.

3. Related to (ii), suppose now that I have two such N x k tables, i.e.
the same N people x k questions, measured at two points in time.  I'd like
to compare the strength of correlation between the two tables, i.e. some
kind of test based on det(R1)/det(R2).

Any insight is warmly welcome.

Thanks,

        Florin











============================
Florin Vaida
Assist Prof, Dept of Biostatistics
Harvard School of Public Health 
655 Huntington Ave Boston MA 02115
Tel (617) 432-2914


<Prev in Thread] Current Thread [Next in Thread>