Date:         Thu, 16 Feb 2006 01:38:08 +0100
Reply-To:     "Kooij, A.J. van der" <KOOIJ@fsw.leidenuniv.nl>
Sender:       "SPSSX(r) Discussion" <SPSSX-L@LISTSERV.UGA.EDU>
From:         "Kooij, A.J. van der" <KOOIJ@fsw.leidenuniv.nl>
Subject:      Re: CATPCA - Model Summary - why missing % of Variance?
Content-Type: text/plain; charset="iso-8859-1"

Nico,

 

A note on CATPCA passive and imputation (mode or extra category) options for missing data:

When passive treatment of missing values is applied, the solution is computed using the

non-missing cells of the data matrix. Note that this is different from listwise deletion: 

cases with missing values are not deleted but are included in the analysis, resulting 

in different numbers of variables in effect for the cases (this is where the weight matrix

I mentioned in previous mail comes in). So, in contrast to listwise deletion, all data are used, and in contrast to imputation no data are added/estimated.

If you run analysis with passive missing option and with imputation option, and solutions 

are not very different, this indicates that imputation is "safe".

 

Regards,

Anita van der Kooij

Data Theory Group

Leiden University

 


________________________________

From: SPSSX(r) Discussion on behalf of Nico Peruzzi
Sent: Thu 16/02/2006 00:43
To: SPSSX-L@LISTSERV.UGA.EDU
Subject: Re: CATPCA - Model Summary - why missing % of Variance?



Anita,

Thank you for the reply.  It was situation #1 below.  When I imputed the
missing values, the % of variance appeared.

Using this method is it now safe to interpret the loadings as correlations?

Thank you, Nico

On 2/15/06, Kooij, A.J. van der <KOOIJ@fsw.leidenuniv.nl> wrote:
>
> Two possibe causes:
>
> 1.
>
> % of Variance is not given if there are missing values treated as passive
>
> (passive is default if /MISSING not specified).
>
> With passive missing values the eigenvalues are not eigenvalues of a
> correlation matrix.
>
> The correlation matrix in output is computed from the quantified variables
> with missing values
>
> substituted; note that the eigenvalues of this correlation matrix are not
> the eigenvalues
>
> reported in the Summary table (as is the case if no missing values or if
> missings values imputed).
>
> The eigenvalues in the Summary table are of a matrix not using
> substitution for missing values,
>
> but involves a weights matrix to take into account missing values; this is
> not a correlation matrix.
>
> So, % of Variance is not valid with passive missing values. Also,
> interpretation of loadings as correlation
>
> with components is not valid then (note that loadings can be higher than
> 1).
>
> 2.
>
> When using case weights, sometimes % of Variance is not displayed. This is
> a bug
>
> that is fixed in 14.0.1 (? or next release).
>
>
>
> Regards,
>
> Anita van der Kooij
>
> Data Theory Group
>
> Leiden University
>
>
> ________________________________
>
> From: SPSSX(r) Discussion on behalf of Nico Peruzzi
> Sent: Wed 15/02/2006 21:56
> To: SPSSX-L@LISTSERV.UGA.EDU
> Subject: CATPCA - Model Summary - why missing % of Variance?
>
>
>
> Hi Listers,
>
> After running CATPCA in v13 on mainly ordinal variables (a couple nominal
> and a couple numeric), my model summary shows the dimensions, cronbach's
> alpha, Eigenvalue, but no % of Variance.  Al the references I'm checking
> show this value, so either I screwed something up or I'm missing
> something.
>
> Any thoughts on whay I'm not seeing this?
>
> Thanks, Nico
>
> --
> Nico Peruzzi, Ph.D.
>
>
>
> **********************************************************************
> This email and any files transmitted with it are confidential and
> intended solely for the use of the individual or entity to whom they
> are addressed. If you have received this email in error please notify
> the system manager.
> **********************************************************************
>



--
Nico Peruzzi, Ph.D.