"Arthur J. Kendall" schrieb:
...
> If I recall correctly, Q and R factoring the same data matrix have equivalences.
> I just cannot remember what they are right now.

Possibly a much more simple solution than mine. I am not fully aware of the
special properties of a q-factorization, and remember the results of P and Q
factoring should be roughly equivalent according to readings some years ago.
Since in P-factor analysis we weight the measures for each variable with a
different coefficient: accordig to the variables stddev, in Q-factor-analysis
we weight the measures for each person with a different weight: the persons
stddev in answering his questions. Maybe comparing the *centroids* (being
the mean-vectors) of two methods may lead to identical results, I doubt,
whether this is the same with principal components. But - if I remember right,
some authors said, they are similar.

Much more simple is the simple proceeding in the following, which I remembered
after reading the link provided by Hiu Chung Law in this thread; a quick&dirty 
examination on data produced the same result as a classic PCA, even without the 
need of explicitely computing a correlation and a loadings-matrix:

 center each variable over the 150 measures along the row, so that the mean is 0
 normalize data so that the sqsum along a row is 1

 just use this data as a factor-loadings-matrix

 rotate for pca (iterative jacobi rotation between all 150 columns)

 The result is the PCA-Loadingsmatrix

 The pca-rotation is a simple task, which can even be programmed in a
 Excel-macro.

Gottfried

--
Gottfried Helms
Univ. Kassel
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to