Please, replace the previous letter with it.
Dear list members,
Sure, the majority of researchers worldwide acknowledge that Likert
type variables (items) are ordinal scale variables. However, I have
practically and bibliographically detected that:
1. A great deal of consultants advice the performance of a standard
(linear) PCA, based on Pearson correlations, notwithstanding the Likert
scale is ordinal.
2. Many others, by accepting the ordinal nature of ordered categories of
items, suggest use FA technique by conducting on the matrix of polychoric
inter-item correlations rather than on the matrix of Pearson correlations.
i.e U. Lorenzo-Seva and A. Rodriguez-Fornells (2006) as well as U. Lorenzo-
Seva and P.J. Ferrando (2007), suggest the âunrestricted factor analysisâ
technique. As they propose, âpolychoric correlation is advised when the
univariate distributions of ordinal items are asymmetric or with excess of
kurtosis, otherwise if both indices are lower than one in absolute value,
then Pearson correlation is advisedâ.
3. Recently, a great majority of researchers (mainly those at Data Theory
Department, Leiden University are ardent supporters of the Categorical
Principal Component analysis (CatPCA). They excuse the transformation
treatment of any kind of scaling and, more importantly they consider that
non-linearity can effectively be treated.
4. Finally a few recommend use of a combination of the above methods, so
as to confirm results. In fact, I am not so sure how could I excuse such a
recommendation, as the adoption of any of the techniques it presumably
excludes the possibilities of any other.
I would much appreciate to receive your reactions and views.
Thanks a lot
George S.
25-11-2010
Greece
=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD
Locked
