SPSSX Discussion
Login  Register

Re: SPSS factor scores

Posted by SabatoPsy on Dec 04, 2013; 3:31pm
URL: http://spssx-discussion.165.s1.nabble.com/SPSS-factor-scores-tp5723392p5723418.html

Attached is an AMOS graphics picture of the model with the standardized parameter estimates. The data is from an observational psychology study of participants completing online questionnaires repeatedly over the course of one year. The data is in wide format and thus each case is a participant. I am performing a cross-lagged autoregressive model between gratitude and well-being. Gratitude and well-being are measured the same across 5 time points, each 3 months apart. I was given a dataset with list-wise deletion of the data such that only participants who had all 5 time points are included. I am not sure how many original participants there were, but this dataset has 851. Gratitude is a manifest variable, and well-being a latent variable created from 4 validated scale sum totals:  life satisfaction, two happiness scales, and depression. My concern is that I have standardized estimates connecting the well-being factors across time (i.e., well-being stability paths) that range from .97 - 1.03. These are the largest regression coefficient magnitudes I have ever had, and I am therefore spectacle. I also attached the model fit output.


On Wed, Dec 4, 2013 at 9:37 AM, Maguin, Eugene [via SPSSX Discussion] <[hidden email]> wrote:

Would you describe your model please in enough detail that anybody could recreate your analysis. Or, post the amos syntax or post an image of the model picture if you used the graphic interface to define the model. Gene Maguin

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SabatoPsy
Sent: Tuesday, December 03, 2013 9:04 PM


To: [hidden email]
Subject: Re: SPSS factor scores

 

Good question. No there is not. This is a data set I created from only the complete cases of my study. Because someone mentioned model fit, I am beginning to wonder if that is what is going on. I know I have poor model it. I read online that if you have poor model fit, your model is most likely misspecified, which can lead to biased parameter estimates (i.e. correlations). After all, SEM is solving regression equations simultaneously.  Does this jive with what you know about SEM and AMOS? Again I am sorry that this is leading away from the content of this listserv.

 

On Tue, Dec 3, 2013 at 8:03 PM, Art Kendall [via SPSSX Discussion] <[hidden email]> wrote:

Is there system missing data? (sysmis)



Art Kendall
Social Research Consultants

On 12/3/2013 5:44 PM, SabatoPsy [via SPSSX Discussion] wrote:

I just checked in my SPSS dataset. There is no user-missing data. The indicators correlation between .40 and .75 of one another. Good thinking. I am going to look into trying to print out the correlation matrix from AMOS. That should be helpful in determining what might be going on.

 

On Tue, Dec 3, 2013 at 4:52 PM, Maguin, Eugene [via SPSSX Discussion] <[hidden email]> wrote:

Ok, let’s get the ‘slap your forehead’ question out of the way first. Is missing data correctly accounted for, i.e., is there any possibility that user missing values are being treated as nonmissing? Is there any possibility that one indicator is a near duplicate of another or a composite of several indicators?

 

Let’s assume not. Next. I’m not familiar with Amos since I use mplus but if amos can print out the covariance or correlation matrix for the indicators, does that matrix match the matrix  that spss computes? Amos is FIML and spss is not and the numbers should be very similar unless you have large amounts of missing data that are not missing completely at random and is ‘strongly’ correlated with model covariates. Lastly, if you have a correlation of .99 between factors, the correlations between the two factors’ indicators have to be extremely high, like in the upper 90’s, I’d guess. If this is true, then I’d guess you have variable construction or conceptualization problems.

 

Gene Maguin

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SabatoPsy
Sent: Tuesday, December 03, 2013 4:30 PM


To: [hidden email]
Subject: Re: SPSS factor scores

 

Good questions. The correlations are strange because they are so high. The literature would suggest they be highly correlated (around .70), however, I am getting correlations of .99. Everything you said is true except for good model fit. These are correlations from raw data. I have an SPSS file with no missing data. One of the factor loadings is set to 1.0 for each factor. These are standardized correlation coefficients. The model fit is poor though. I have a CFI of .77 and RMSEA of .16. Could having poor model fit change/bias my correlation estimates?

 

On Tue, Dec 3, 2013 at 4:18 PM, Maguin, Eugene [via SPSSX Discussion] <[hidden email]> wrote:

Let’s back up to where the problem begins, which seems to be at the SEM stage because you are reporting “strange correlations” between factors. So, what’s strange about the correlations? And, just so that we all are on the same page, please clarify your use of ‘correlations’ in the context of an SEM. Affirmatively verify that the following sequence is true. Analyzing raw data not a covariance or correlation matrix. One loading fixed at 1.00 for each factor rather than factor variances fixed at 1.00. Fit of the model is acceptable under current standards. The strange values are standardized values and not unstandardized values.

Gene Maguin

 

 

 

 

 

 

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of SabatoPsy
Sent: Tuesday, December 03, 2013 3:56 PM
To: [hidden email]


Subject: Re: SPSS factor scores

 

Thank you for all of your posts. I looked up online how the factor scores are computed in SPSS in the IBM labyrinth. I got a bunch of matrix algebra and calculus notation that I am not smart enough to follow. I did however do an experiment on my own as was also suggested. I found that there are small differences between factor scores saved via the PCA method of extraction compared to the PAF and ML extraction methods; however, there were negligible differences between the PAF and ML methods. I tested this by correlating different saved factor score variables from the exact same factor structure but with different extraction methods.

 

To address the question of why I am creating the factor(s) may go outside the scope of this listserve, but I will gladly take any advice/knowledge. I am running some SEM models in AMOS and am getting strange correlations between my factors. I wanted a way to test whether indeed these correlations between factors are accurate. I thought that by correlating the saved factor scores in SPSS using the ML extraction method, I should get the same correlations I got in AMOS. I am creating multiple single factors from a multiple set of 4 manifest indicators to create 5 total factors. To my knowledge, an EFA using ML and a CFA create essentially (minus negligible estimation differences) the same factor scores (when you are creating a single factor using all indicators). Therefore correlations between the SPSS factor scores I created and correlations between the latent factors I created in AMOS should be essentially equivalent. However, I am getting substantially different correlations. Does anybody know what is going on... or what I am doing wrong... or why my belief that the correlations should be the same is wrong??? I would greatly appreciate any enlightenment.

 

David Disabato

George Mason University

 

On Tue, Dec 3, 2013 at 12:26 PM, Art Kendall [via SPSSX Discussion] <[hidden email]> wrote:

What is the purpose of your using FACTOR? Data reduction? Creating scales? etc.

What are you going to use the factor scores for?
 

Art Kendall
Social Research Consultants

On 12/3/2013 11:42 AM, SabatoPsy [via SPSSX Discussion] wrote:

Hi Listserve,

I am wondering how the SPSS factor scores in the "Factor" command are calculated. More specifically, does the method of extraction change the calculated factor scores? I want to compare the factor scores created by a PCA compared to an EFA (with ML estimation). Is this possible in SPSS? I know there are three ways of calculating the factor scores (Regression, Bartlett, Anderson-Rubin). Let us assume I tell SPSS to use the same way: Regression. If I tell SPSS to create "Regression" factor scores with a principle component method of extraction, will they be different than the "Regression" factor scores with a maximum likelihood method of extraction?

Thank you,

David Disabato
George Mason University

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

 

Art Kendall
Social Research Consultants

 

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392p5723399.html

To unsubscribe from SPSS factor scores, click here.
NAML



 

--
David J. Disabato

Clinical Psychology Doctoral Student

George Mason University

 

View this message in context: Re: SPSS factor scores


Sent from the SPSSX Discussion mailing list archive at Nabble.com.

 

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392p5723403.html

To unsubscribe from SPSS factor scores, click here.
NAML



 

--
David J. Disabato

Clinical Psychology Doctoral Student

George Mason University

 

View this message in context: Re: SPSS factor scores
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

 

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392p5723406.html

To unsubscribe from SPSS factor scores, click here.
NAML



 

--
David J. Disabato

Clinical Psychology Doctoral Student

George Mason University

 

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392p5723407.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

 

Art Kendall
Social Research Consultants

 

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392p5723409.html

To unsubscribe from SPSS factor scores, click here.
NAML



 

--
David J. Disabato

Clinical Psychology Doctoral Student

George Mason University

 

View this message in context: Re: SPSS factor scores
Sent from the SPSSX Discussion mailing list archive at Nabble.com.



If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/SPSS-factor-scores-tp5723392p5723415.html
To unsubscribe from SPSS factor scores, click here.
NAML



--
David J. Disabato
Clinical Psychology Doctoral Student
George Mason University
[hidden email]

Screenshot 2013-12-04 10.14.01.png (339K) Download Attachment
Screenshot 2013-12-04 10.27.00.png (99K) Download Attachment
Free forum by Nabble Edit this page