factor analysis

factor analysis

Post by Roland Peter » Mon, 07 Sep 1998 04:00:00



Hi!
Who can help me? I have a problem with the result of a factor analysis.
The output of the correlation matrix is "The matrix is not positive
definite". What can I do?
thx
 
 
 

factor analysis

Post by Joonwook Pa » Tue, 08 Sep 1998 04:00:00


What is the type of your input data?
If it is the text file which converted from word program such as MS WORD.
It might be help to look for characters which you can't see but exists.
If you have SPSS version 6.xx, just View the Output of correlation or
covariance matrix of your data.
Otherwise, It must be better for you to regether the whole data.
I wish it can help you.

Bye.


Quote:

>Hi!
>Who can help me? I have a problem with the result of a factor analysis.
>The output of the correlation matrix is "The matrix is not positive
>definite". What can I do?
>thx


 
 
 

factor analysis

Post by mlu.. » Tue, 08 Sep 1998 04:00:00




>Subject: Re: factor analysis
>Date: 7 Sep 1998 04:19:21 GMT
>Bye.

>>Hi!
>>Who can help me? I have a problem with the result of a factor analysis.
>>The output of the correlation matrix is "The matrix is not positive
>>definite". What can I do?
>>thx

The problem could also be that yoy sample is too small.  Which procedure are
you using? The PCA should wok with even a very small sample, other estimation
procedures have problems with it.

and errrrrrrr... if this should be the problem, and a PCA the solution, don't
put too much weight to the results, since you're sample was that small that
you are looking at a lot of error!

hope it helps

Cheers,

marc

 
 
 

factor analysis

Post by Jan Berger » Tue, 08 Sep 1998 04:00:00



Quote:>Hi!
>Who can help me? I have a problem with the result of a factor analysis.
>The output of the correlation matrix is "The matrix is not positive
>definite". What can I do?
>thx

There are two reasons a matrix can be not positive definite:
1) you used pairwise deletion of missing values
2) there exists collinearity among the variables, reasons:
 a) composite variable made up of two or more component variables
 b) outliers
 c) sample size less than number of observed variables

conclusion: first find the problem and fix  it. then make a new matrix.
Greetings, Jan Bergers

 
 
 

factor analysis

Post by Dicon Montfor » Wed, 09 Sep 1998 04:00:00




>>Hi!
>>Who can help me? I have a problem with the result of a factor analysis.
>>The output of the correlation matrix is "The matrix is not positive
>>definite". What can I do?
>>thx

>There are two reasons a matrix can be not positive definite:
>1) you used pairwise deletion of missing values
>2) there exists collinearity among the variables, reasons:
> a) composite variable made up of two or more component variables
> b) outliers
> c) sample size less than number of observed variables

>conclusion: first find the problem and fix  it. then make a new matrix.
>Greetings, Jan Bergers

It could also be that you're using ipsative data such as preference
scores where the scores add up to the same sum for each person.
If so, you might be able to get away with PCA or unweighted least
squares but interpretation requires an understanding of what you're
doing. Better if you can get the original normative data if there is
any.

Dicon Montford