Comparing 2 groups of variables

[Mark W. Andrews]

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 

Mark W. Andrews
Senior Study Director
Synovate 
7600 Leesburg Pike, East Building, Suite 110
Falls Church, VA 22043
Phone   [phone]
FAX      [phone]
Email    [hidden email]
Web     www.synovate.com

 

[Poes, Matthew Joseph]

Can you give more detail?  Are you saying that you have a group of, say, 5 predictor variables, and you want to know what the association is of those 5 variables on an outcome variable Y, and then what total variance is explained by 3 of the 5 variables, while still accounting for the common variance in the other two variables?  Or are you asking what amount of variance in a group of 5 variables is explained by a separate group of 3 variables?  Or, are you asking this in more scalar terms, and wanting to know which variables account for the greatest amount of variance within the collective latent variable that they collectively make up.  In the latter, we have the same 5 variables, combined and they explain mood stability, and you want to know what amount of the total explainability 3 of the 5 variables have?  Thanks.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Mark W. Andrews
Sent: Wednesday, December 14, 2011 12:49 PM
To: [hidden email]
Subject: Comparing 2 groups of variables

 

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 

Mark W. Andrews
Senior Study Director
Synovate 
7600 Leesburg Pike, East Building, Suite 110
Falls Church, VA 22043
Phone   [phone]
FAX      [phone]
Email    [hidden email]
Web     www.synovate.com

 

[Mark W. Andrews]

I have 17 attributes (11 point scale). We want to reduce it to 10 attributes. Someone asked how much we are losing by reducing the number of attributes. I was thinking that the math would be similar to the accumulated variance explained figures you get from a factor analysis. Instead of factors, however, I would be looking at attribute variables.

 

From: Poes, Matthew Joseph [mailto:[hidden email]]
Sent: Wednesday, December 14, 2011 2:05 PM
To: Mark Andrews ([hidden email]); [hidden email]
Subject: RE: Comparing 2 groups of variables

 

Can you give more detail?  Are you saying that you have a group of, say, 5 predictor variables, and you want to know what the association is of those 5 variables on an outcome variable Y, and then what total variance is explained by 3 of the 5 variables, while still accounting for the common variance in the other two variables?  Or are you asking what amount of variance in a group of 5 variables is explained by a separate group of 3 variables?  Or, are you asking this in more scalar terms, and wanting to know which variables account for the greatest amount of variance within the collective latent variable that they collectively make up.  In the latter, we have the same 5 variables, combined and they explain mood stability, and you want to know what amount of the total explainability 3 of the 5 variables have?  Thanks.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

 

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Mark W. Andrews
Sent: Wednesday, December 14, 2011 12:49 PM
To: [hidden email]
Subject: Comparing 2 groups of variables

 

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 

Mark W. Andrews
Senior Study Director
Synovate 
7600 Leesburg Pike, East Building, Suite 110
Falls Church, VA 22043
Phone   [phone]
FAX      [phone]
Email    [hidden email]
Web     www.synovate.com

 

[Rich Ulrich]

This is a pretty opaque statement of a problem.

The simplest way that I read it reminds me of the origins
of factor analysis - looking at a correlation matrix, and looking
at the matrix after partialing out one or more variables.

Thus - if you have (say) 10 variables, v1 to v10, and you want
to know whether v1 to v5 "account for the useful variance", you
could look at the correlation matrix of v6 to v10 which is produced by
"partialing out" v1 to v5.  Do the off-diagonals differ from zero,
by more than random deviations, or has all the "useful variance"
been removed?

James Stieger's routines probably include a test of residual r's.

If this isn't what you have in mind, try again.

--
Rich Ulrich

---

Date: Wed, 14 Dec 2011 12:49:00 -0600
From: [hidden email]
Subject: Comparing 2 groups of variables
To: [hidden email]

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 [...]

[Poes, Matthew Joseph]

Sum the squared factor loadings.  The Factor loadings are the correlations to the latent variable.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

 

From: [hidden email] [mailto:[hidden email]]
Sent: Wednesday, December 14, 2011 1:14 PM
To: Poes, Matthew Joseph; [hidden email]
Subject: RE: Comparing 2 groups of variables

 

I have 17 attributes (11 point scale). We want to reduce it to 10 attributes. Someone asked how much we are losing by reducing the number of attributes. I was thinking that the math would be similar to the accumulated variance explained figures you get from a factor analysis. Instead of factors, however, I would be looking at attribute variables.

 

From: Poes, Matthew Joseph [hidden email]
Sent: Wednesday, December 14, 2011 2:05 PM
To: Mark Andrews ([hidden email]); [hidden email]
Subject: RE: Comparing 2 groups of variables

 

Can you give more detail?  Are you saying that you have a group of, say, 5 predictor variables, and you want to know what the association is of those 5 variables on an outcome variable Y, and then what total variance is explained by 3 of the 5 variables, while still accounting for the common variance in the other two variables?  Or are you asking what amount of variance in a group of 5 variables is explained by a separate group of 3 variables?  Or, are you asking this in more scalar terms, and wanting to know which variables account for the greatest amount of variance within the collective latent variable that they collectively make up.  In the latter, we have the same 5 variables, combined and they explain mood stability, and you want to know what amount of the total explainability 3 of the 5 variables have?  Thanks.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

 

From: SPSSX(r) Discussion [hidden email] On Behalf Of Mark W. Andrews
Sent: Wednesday, December 14, 2011 12:49 PM
To: [hidden email]
Subject: Comparing 2 groups of variables

 

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 

Mark W. Andrews
Senior Study Director
Synovate 
7600 Leesburg Pike, East Building, Suite 110
Falls Church, VA 22043
Phone   [phone]
FAX      [phone]
Email    [hidden email]
Web     www.synovate.com

 

[Art Kendall]

You say you have attribute variables but say you have an 11 point response scale.
Are your values in the response scale "not severely discrepant from ordinal level"?

Is this a summative scale where the 11 variables are repeated measure of some construct? E.g., an attitude?
Is the idea to have just 10 for reasons of respondent burden?

If these are correct assumptions see RELIABILITY.  just to be safe drop any item whose removal would increase reliability.
keep doing that until the removal of no item increase reliability.

Then drop the item for which its removal would make the least decrement in reliability.
keep doing that until you are down to 10 or you may have to keep more than 10 if removal of an item makes a serious decrement in reliability.

HTH


On 12/14/2011 2:14 PM, Mark W. Andrews wrote:

I have 17 attributes (11 point scale). We want to reduce it to 10 attributes. Someone asked how much we are losing by reducing the number of attributes. I was thinking that the math would be similar to the accumulated variance explained figures you get from a factor analysis. Instead of factors, however, I would be looking at attribute variables.

 

From: Poes, Matthew Joseph [[hidden email]]
Sent: Wednesday, December 14, 2011 2:05 PM
To: Mark Andrews ([hidden email]); [hidden email]
Subject: RE: Comparing 2 groups of variables

 

Can you give more detail?  Are you saying that you have a group of, say, 5 predictor variables, and you want to know what the association is of those 5 variables on an outcome variable Y, and then what total variance is explained by 3 of the 5 variables, while still accounting for the common variance in the other two variables?  Or are you asking what amount of variance in a group of 5 variables is explained by a separate group of 3 variables?  Or, are you asking this in more scalar terms, and wanting to know which variables account for the greatest amount of variance within the collective latent variable that they collectively make up.  In the latter, we have the same 5 variables, combined and they explain mood stability, and you want to know what amount of the total explainability 3 of the 5 variables have?  Thanks.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

 

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Mark W. Andrews
Sent: Wednesday, December 14, 2011 12:49 PM
To: [hidden email]
Subject: Comparing 2 groups of variables

 

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 

Mark W. Andrews
Senior Study Director
Synovate 
7600 Leesburg Pike, East Building, Suite 110
Falls Church, VA 22043
Phone   [phone]
FAX      [phone]
Email    [hidden email]
Web     www.synovate.com

 

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

[Mark W. Andrews]

Thanks for the response.

 

I am not sure about "not severely discrepant from ordinal level”.

 

Regarding intent, the attributes are not intended to measure a single concept. The model for reduction was mixed. We did a content review and focused on reducing redundancy. No further reduction is required. We just want to quantify the impact.

 

Mark

 

 

 

From: Art Kendall [mailto:[hidden email]]
Sent: Wednesday, December 14, 2011 2:35 PM
To: Mark Andrews ([hidden email])
Cc: [hidden email]
Subject: Re: [SPSSX-L] Comparing 2 groups of variables

 

You say you have attribute variables but say you have an 11 point response scale.
Are your values in the response scale "not severely discrepant from ordinal level"?

Is this a summative scale where the 11 variables are repeated measure of some construct? E.g., an attitude?
Is the idea to have just 10 for reasons of respondent burden?

If these are correct assumptions see RELIABILITY.  just to be safe drop any item whose removal would increase reliability.
keep doing that until the removal of no item increase reliability.

Then drop the item for which its removal would make the least decrement in reliability.
keep doing that until you are down to 10 or you may have to keep more than 10 if removal of an item makes a serious decrement in reliability.

HTH

Art Kendall

Social Research Consultants

On 12/14/2011 2:14 PM, Mark W. Andrews wrote:

I have 17 attributes (11 point scale). We want to reduce it to 10 attributes. Someone asked how much we are losing by reducing the number of attributes. I was thinking that the math would be similar to the accumulated variance explained figures you get from a factor analysis. Instead of factors, however, I would be looking at attribute variables.

 

From: Poes, Matthew Joseph [[hidden email]]
Sent: Wednesday, December 14, 2011 2:05 PM
To: Mark Andrews ([hidden email]); [hidden email]
Subject: RE: Comparing 2 groups of variables

 

Can you give more detail?  Are you saying that you have a group of, say, 5 predictor variables, and you want to know what the association is of those 5 variables on an outcome variable Y, and then what total variance is explained by 3 of the 5 variables, while still accounting for the common variance in the other two variables?  Or are you asking what amount of variance in a group of 5 variables is explained by a separate group of 3 variables?  Or, are you asking this in more scalar terms, and wanting to know which variables account for the greatest amount of variance within the collective latent variable that they collectively make up.  In the latter, we have the same 5 variables, combined and they explain mood stability, and you want to know what amount of the total explainability 3 of the 5 variables have?  Thanks.

 

Matthew J Poes

Research Data Specialist

Center for Prevention Research and Development

University of Illinois

 

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Mark W. Andrews
Sent: Wednesday, December 14, 2011 12:49 PM
To: [hidden email]
Subject: Comparing 2 groups of variables

 

I am trying figure out how much of the variance within a group of variables is explained by the a subgroup of variables. Can anyone point me in the right direction?

 

Mark W. Andrews
Senior Study Director
Synovate 
7600 Leesburg Pike, East Building, Suite 110
Falls Church, VA 22043
Phone   [phone]
FAX      [phone]
Email    [hidden email]
Web     www.synovate.com