Comparing Correlation Matrices

[Salbod]

Good Morning SPSSers,

I want to determine whether the relations among six subscales for the students and parents differ.  A colleague administered a 73 item parenting scale, consisting of six subscales (e.g., subscales A thru F), to 95 8th graders and their parents (just one parent). The school requested that neither students' responses nor their parents' responses be identifiable; that is, the scales were administered without any identification codes. Therefore, the problem is how do I compare two independent matrices?

Green (1992) suggests a SEM solution. He provides a LISREL VI program. I'm not sure how to use convert into AMOS code to solve my particular problem.
I would appreciate if someone can point me in the right direction: reference or example.

TIA

Stephen Salbod, Pace University, NYC



Green, J. A. (1992). Testing whether correlation matrices are different from each other.  Developmental Psychology, 28(2), 215-224.

=====================
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

[Mike]

Hi,

First, I would suggest that you test the equality of covariance matrices
instead of correlation matrices.

Second, Box's M test is test of equality of covariance matrices and is
a available in SPSS in MANOVA and DISCRIMINANT (if that procedure
is still available).  Karl Wuensch briefly covers this on his statistics website
where he reproduces responses by David Nichols and Joop Hox on this
question back in 1994 and 1997; see:
http://core.ecu.edu/psyc/wuenschk/StatHelp/ComparingCorrelationMatrices.htm

Dave notes that if you convert the variables into z-score form, then Box's
M test become a test of equality of correlation matrices.  As Joop points out,
SEM/Covariance Structure Analysis (CSA) is more robust.

Third,  Barbara Byrne who appears to be in the business of teaching people
of how to use the various SEM program to perform different types of SEM
analyses, has an article on how to do testing of equality of covariance in the
journal Structural Equation Modeling.  Here is a link to the abstract of the
article:
http://www.informaworld.com/smpp/content~db=all~content=a785833114
The article was published in 2004 and might have to be modified to take
into account recent developments in SEM.

Fourth, one example of a study that uses equality of covariance structures
to test whether groups receiving different modes of administration of a
scale is this one published in the journal Clinical Trials:

James W Varni, Christine A Limbers, and Daniel A Newman
Using factor analysis to confirm the validity of children's self-reported
health-related quality of life across different modes of administration
Clin Trials April 2009 6: 185-195, doi:10.1177/1740774509102309

The authors work through several different types of analyses to examine
the nature of the underlying relationships.

Others, I'm sure, can provide additional advice.

-Mike Palij
New York University
[hidden email]


----- Original Message -----
From: "Salbod, Mr. Stephen" <[hidden email]>
To: <[hidden email]>
Sent: Friday, June 17, 2011 8:36 AM
Subject: Comparing Correlation Matrices

> Good Morning SPSSers,
>
> I want to determine whether the relations among six subscales for the students and parents differ.  A colleague administered a 73 item parenting scale, consisting of six subscales (e.g., subscales A thru F), to 95 8th graders and their parents (just one parent). The school requested that neither students' responses nor their parents' responses be identifiable; that is, the scales were administered without any identification codes. Therefore, the problem is how do I compare two independent matrices?
>
> Green (1992) suggests a SEM solution. He provides a LISREL VI program. I'm not sure how to use convert into AMOS code to solve my particular problem.
> I would appreciate if someone can point me in the right direction: reference or example.
>
> TIA
>
> Stephen Salbod, Pace University, NYC
>
> Green, J. A. (1992). Testing whether correlation matrices are different from each other.  Developmental Psychology, 28(2), 215-224.

=====================
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

[Salbod]

Thanks, Mike,  for getting back to me so quickly.  I came across the Box's M test, determinant ratios, but I wasn't too sure it would answer the question about the relations among the subscales. Now that you've pointed out, I'll take another look at it through Nicholes' and Hox's eyes.

Green (1992) mentioned the covariance correlation distinction, but dismissed it in his article because claimed that the fit indices were the same regardless of the matix (i.e., scale invariant, p. 217).

The Bryne's article was off my radar.  Now that you pointed it out, that is an article that I do want to check out.

Thank you for the mini-course :)

All the Best, Steve




-----Original Message-----
From: Mike Palij [mailto:[hidden email]]
Sent: Friday, June 17, 2011 9:38 AM
To: Salbod, Mr. Stephen; [hidden email]
Cc: Mike Palij
Subject: Re: Comparing Correlation Matrices

Hi,

First, I would suggest that you test the equality of covariance matrices instead of correlation matrices.

Second, Box's M test is test of equality of covariance matrices and is a available in SPSS in MANOVA and DISCRIMINANT (if that procedure is still available).  Karl Wuensch briefly covers this on his statistics website where he reproduces responses by David Nichols and Joop Hox on this question back in 1994 and 1997; see:
http://core.ecu.edu/psyc/wuenschk/StatHelp/ComparingCorrelationMatrices.htm

Dave notes that if you convert the variables into z-score form, then Box's M test become a test of equality of correlation matrices.  As Joop points out, SEM/Covariance Structure Analysis (CSA) is more robust.

Third,  Barbara Byrne who appears to be in the business of teaching people of how to use the various SEM program to perform different types of SEM analyses, has an article on how to do testing of equality of covariance in the journal Structural Equation Modeling.  Here is a link to the abstract of the
article:
http://www.informaworld.com/smpp/content~db=all~content=a785833114
The article was published in 2004 and might have to be modified to take into account recent developments in SEM.

Fourth, one example of a study that uses equality of covariance structures to test whether groups receiving different modes of administration of a scale is this one published in the journal Clinical Trials:

James W Varni, Christine A Limbers, and Daniel A Newman Using factor analysis to confirm the validity of children's self-reported health-related quality of life across different modes of administration Clin Trials April 2009 6: 185-195, doi:10.1177/1740774509102309

The authors work through several different types of analyses to examine the nature of the underlying relationships.

Others, I'm sure, can provide additional advice.

-Mike Palij
New York University
[hidden email]


----- Original Message -----
From: "Salbod, Mr. Stephen" <[hidden email]>
To: <[hidden email]>
Sent: Friday, June 17, 2011 8:36 AM
Subject: Comparing Correlation Matrices

> Good Morning SPSSers,
>
> I want to determine whether the relations among six subscales for the students and parents differ.  A colleague administered a 73 item parenting scale, consisting of six subscales (e.g., subscales A thru F), to 95 8th graders and their parents (just one parent). The school requested that neither students' responses nor their parents' responses be identifiable; that is, the scales were administered without any identification codes. Therefore, the problem is how do I compare two independent matrices?
>
> Green (1992) suggests a SEM solution. He provides a LISREL VI program. I'm not sure how to use convert into AMOS code to solve my particular problem.
> I would appreciate if someone can point me in the right direction: reference or example.
>
> TIA
>
> Stephen Salbod, Pace University, NYC
>
> Green, J. A. (1992). Testing whether correlation matrices are different from each other.  Developmental Psychology, 28(2), 215-224.

=====================
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

[David Marso]

"Green (1992) mentioned the covariance correlation distinction, but dismissed it in his article because claimed that the fit indices were the same regardless of the matix (i.e., scale invariant, p. 217)."

Having identical Covariance matrices implies identical Correlation matrices (But NOT vice versa).
In SEM one could fit a multigroup model and examine a sequence of parameter equality hypotheses.
First of all are the coefficients relating the latent variables to the measured variables equal?
Second, are the covariances of the latent variables equal?
OTOH:  With N=95 in each group you don't have sufficient sample size to do this.
Have you examined the reliabilities of the 6 subscales?
If they are low it would hardly make sense to compare their covariances.
---

 

Salbod, Mr. Stephen wrote

Thanks, Mike,  for getting back to me so quickly.  I came across the Box's M test, determinant ratios, but I wasn't too sure it would answer the question about the relations among the subscales. Now that you've pointed out, I'll take another look at it through Nicholes' and Hox's eyes.

Green (1992) mentioned the covariance correlation distinction, but dismissed it in his article because claimed that the fit indices were the same regardless of the matix (i.e., scale invariant, p. 217).

The Bryne's article was off my radar.  Now that you pointed it out, that is an article that I do want to check out.

Thank you for the mini-course :)

All the Best, Steve




-----Original Message-----
From: Mike Palij [mailto:[hidden email]]
Sent: Friday, June 17, 2011 9:38 AM
To: Salbod, Mr. Stephen; [hidden email]
Cc: Mike Palij
Subject: Re: Comparing Correlation Matrices

Hi,

First, I would suggest that you test the equality of covariance matrices instead of correlation matrices.

Second, Box's M test is test of equality of covariance matrices and is a available in SPSS in MANOVA and DISCRIMINANT (if that procedure is still available).  Karl Wuensch briefly covers this on his statistics website where he reproduces responses by David Nichols and Joop Hox on this question back in 1994 and 1997; see:
http://core.ecu.edu/psyc/wuenschk/StatHelp/ComparingCorrelationMatrices.htm

Dave notes that if you convert the variables into z-score form, then Box's M test become a test of equality of correlation matrices.  As Joop points out, SEM/Covariance Structure Analysis (CSA) is more robust.

Third,  Barbara Byrne who appears to be in the business of teaching people of how to use the various SEM program to perform different types of SEM analyses, has an article on how to do testing of equality of covariance in the journal Structural Equation Modeling.  Here is a link to the abstract of the
article:
http://www.informaworld.com/smpp/content~db=all~content=a785833114
The article was published in 2004 and might have to be modified to take into account recent developments in SEM.

Fourth, one example of a study that uses equality of covariance structures to test whether groups receiving different modes of administration of a scale is this one published in the journal Clinical Trials:

James W Varni, Christine A Limbers, and Daniel A Newman Using factor analysis to confirm the validity of children's self-reported health-related quality of life across different modes of administration Clin Trials April 2009 6: 185-195, doi:10.1177/1740774509102309

The authors work through several different types of analyses to examine the nature of the underlying relationships.

Others, I'm sure, can provide additional advice.

-Mike Palij
New York University
[hidden email]


----- Original Message -----
From: "Salbod, Mr. Stephen" <[hidden email]>
To: <[hidden email]>
Sent: Friday, June 17, 2011 8:36 AM
Subject: Comparing Correlation Matrices


> Good Morning SPSSers,
>
> I want to determine whether the relations among six subscales for the students and parents differ.  A colleague administered a 73 item parenting scale, consisting of six subscales (e.g., subscales A thru F), to 95 8th graders and their parents (just one parent). The school requested that neither students' responses nor their parents' responses be identifiable; that is, the scales were administered without any identification codes. Therefore, the problem is how do I compare two independent matrices?
>
> Green (1992) suggests a SEM solution. He provides a LISREL VI program. I'm not sure how to use convert into AMOS code to solve my particular problem.
> I would appreciate if someone can point me in the right direction: reference or example.
>
> TIA
>
> Stephen Salbod, Pace University, NYC
>
> Green, J. A. (1992). Testing whether correlation matrices are different from each other.  Developmental Psychology, 28(2), 215-224.

=====================
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

[Dawn Wiest]

Hello –

After importing data from a different program into SPSS, the leading zeros on a string variable that I use as a key variable across multiple data sets are lost. I need to recapture those leading zeros.

This is an ID variable that should have nine digits. When the data are exported, only the non-zero digits remain so that, for example, ID 000022122 becomes 22122; ID 000000011 becomes 11; ID 077777777 becomes 77777777, and so on.

Is there a way to replenish the leading zeros on this variable so that the total length of each of the resulting values is 9 digits?

Thank you in advance for your help

Dawn

 

Dawn Wiest, Ph.D.
Senior Research Analyst
Research Center - 7th Floor
American College of Physicians
190 North Independence Mall West Philadelphia, PA 19106-1572

[hidden email]
[phone]

[Maguin, Eugene]

Dawn,

 

Are you sure that ID remains a string variable once imported? The behavior you describe seems consistent with the import of a numeric variable or a string variable being converted to numeric in the import process.

 

Gene Maguin

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dawn Wiest
Sent: Friday, June 17, 2011 3:38 PM
To: [hidden email]
Subject: String variable - leading zeros

Hello –

After importing data from a different program into SPSS, the leading zeros on a string variable that I use as a key variable across multiple data sets are lost. I need to recapture those leading zeros.

This is an ID variable that should have nine digits. When the data are exported, only the non-zero digits remain so that, for example, ID 000022122 becomes 22122; ID 000000011 becomes 11; ID 077777777 becomes 77777777, and so on.

Is there a way to replenish the leading zeros on this variable so that the total length of each of the resulting values is 9 digits?

Thank you in advance for your help

Dawn

 

Dawn Wiest, Ph.D.
Senior Research Analyst
Research Center - 7th Floor
American College of Physicians
190 North Independence Mall West Philadelphia, PA 19106-1572

[hidden email]
[phone]

[ViAnn Beadle]

If you need the leading zeroes than you must be doing a string compare since 12 and 012 are the same number. So you need to convert the number back to a string using the string function and specifying the N format for the number.

string stringvar(a9).

compute stringvar=string(numbervar, n9).

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dawn Wiest
Sent: Friday, June 17, 2011 1:38 PM
To: [hidden email]
Subject: String variable - leading zeros

 

Hello –

After importing data from a different program into SPSS, the leading zeros on a string variable that I use as a key variable across multiple data sets are lost. I need to recapture those leading zeros.

This is an ID variable that should have nine digits. When the data are exported, only the non-zero digits remain so that, for example, ID 000022122 becomes 22122; ID 000000011 becomes 11; ID 077777777 becomes 77777777, and so on.

Is there a way to replenish the leading zeros on this variable so that the total length of each of the resulting values is 9 digits?

Thank you in advance for your help

Dawn

 

Dawn Wiest, Ph.D.
Senior Research Analyst
Research Center - 7th Floor
American College of Physicians
190 North Independence Mall West Philadelphia, PA 19106-1572

[hidden email]
[phone]

[Dawn Wiest]

Gene, you are correct: the variable was imported as a numeric variable. I applied the syntax ViAnn recommended and it worked beautifully. Thanks!

>>> Gene Maguin <[hidden email]> 6/17/2011 3:58 PM >>>

Dawn,

 

Are you sure that ID remains a string variable once imported? The behavior you describe seems consistent with the import of a numeric variable or a string variable being converted to numeric in the import process.

 

Gene Maguin

 

If you need the leading zeroes than you must be doing a string compare since 12 and 012 are the same number. So you need to convert the number back to a string using the string function and specifying the N format for the number.

string stringvar(a9).

compute stringvar=string(numbervar, n9).

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Dawn Wiest
Sent: Friday, June 17, 2011 3:38 PM
To: [hidden email]
Subject: String variable - leading zeros

Hello –

After importing data from a different program into SPSS, the leading zeros on a string variable that I use as a key variable across multiple data sets are lost. I need to recapture those leading zeros.

This is an ID variable that should have nine digits. When the data are exported, only the non-zero digits remain so that, for example, ID 000022122 becomes 22122; ID 000000011 becomes 11; ID 077777777 becomes 77777777, and so on.

Is there a way to replenish the leading zeros on this variable so that the total length of each of the resulting values is 9 digits?

Thank you in advance for your help

Dawn

 

Dawn Wiest, Ph.D.
Senior Research Analyst
Research Center - 7th Floor
American College of Physicians
190 North Independence Mall West Philadelphia, PA 19106-1572

[hidden email]
[phone]