Doing EFA after multiple imputation

[ziweiguan]

After multiple imputation, one get a number of data sets, and EFA is not such a procedure to use multiple data sets and finally give pooled estimates.
I found a very interesting discussion in 2013,
http://spssx-discussion.1045642.n5.nabble.com/Multiple-Imputation-tp4994372p5722178.html
where Bruce cited the Graham 2009 paper: "The EM covariance matrix is also an excellent basis for exploratory factor analysis with missing data."

I have two questions:
1. SPSS missing data analysis - EM has an option: save completed data.
Graham pointed out that the single imputed data set based on EM algorithm is not recommended
unless random error residuals are added.

Is there a good way to add the random error residuals?

2. How to use the EM covariance matrix for EFA?
I cannot find an option that I can input the matrix to do EFA, but I guess there must be a way.
Can anyone give me some hints?

Thanks!

[David Marso]

1. See RV.NORMAL function under COMPUTE command.
2. See MATRIX subcommand under FACTOR command.
All in the FM near your F1 key!!!
---

ziweiguan wrote

After multiple imputation, one get a number of data sets, and EFA is not such a procedure to use multiple data sets and finally give pooled estimates.
I found a very interesting discussion in 2013,
http://spssx-discussion.1045642.n5.nabble.com/Multiple-Imputation-tp4994372p5722178.html
where Bruce cited the Graham 2009 paper: "The EM covariance matrix is also an excellent basis for exploratory factor analysis with missing data."

I have two questions:
1. SPSS missing data analysis - EM has an option: save completed data.
Graham pointed out that the single imputed data set based on EM algorithm is not recommended
unless random error residuals are added.

Is there a good way to add the random error residuals?

2. How to use the EM covariance matrix for EFA?
I cannot find an option that I can input the matrix to do EFA, but I guess there must be a way.
Can anyone give me some hints?

Thanks!

[Bruce Weaver]

Re your first point, even if you did add random error residuals, would you do it multiple times?  If you only did it once, you'd still have single imputation, not multiple imputation.  Also, the FACTOR procedure is not in the list of "procedures that support pooling" (http://www-01.ibm.com/support/knowledgecenter/SSLVMB_21.0.0/com.ibm.spss.statistics.help/mi_analysis.htm), so you'd still not be much further ahead.  

Re your second point, as the next post in the thread you linked points out, one of the obstacles to using the EM covariances as input to FACTOR is that the MVA command does not have a MATRIX subcommand, so it does not allow you to generate directly the matrix data set that FACTOR wants as input.  I wrote a couple macros that can be used (with the MVA command sandwiched between them) to generate matrix datasets containing the EM covariances and correlations that can be used as input for FACTOR (and RELIABILITY).  See this article for details & examples:

  http://tqmp.org/RegularArticles/vol10-2/p143/p143.pdf

The macro definitions and other supplementary materials can be downloaded here:

  https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/spss/my-spss-page/emcorr

HTH.

ziweiguan wrote

After multiple imputation, one get a number of data sets, and EFA is not such a procedure to use multiple data sets and finally give pooled estimates.
I found a very interesting discussion in 2013,
http://spssx-discussion.1045642.n5.nabble.com/Multiple-Imputation-tp4994372p5722178.html
where Bruce cited the Graham 2009 paper: "The EM covariance matrix is also an excellent basis for exploratory factor analysis with missing data."

I have two questions:
1. SPSS missing data analysis - EM has an option: save completed data.
Graham pointed out that the single imputed data set based on EM algorithm is not recommended
unless random error residuals are added.

Is there a good way to add the random error residuals?

2. How to use the EM covariance matrix for EFA?
I cannot find an option that I can input the matrix to do EFA, but I guess there must be a way.
Can anyone give me some hints?

Thanks!

[ziweiguan]

Hi Bruce

Thank you for your reply.
Because I only want to do EFA after the multiple imputation, and I do not do hypothesis testing, a single imputed data set is okay. Graham pointed out in his paper that "this data set is very useful for analyses that do not require hypothesis testing, such as coefficient alpha analysis and exploratory factor analysis".

I will look at the links in your reply regarding my second point. It is a better solution. Thank you for providing your macro.

ziweiguan

[ziweiguan]

Hi David

Thank you for your input. Would you please explain a little bit more how to use the compute / rv.normal to add random errors to the EM output data set? Thanks!

David Marso wrote

1. See RV.NORMAL function under COMPUTE command.
2. See MATRIX subcommand under FACTOR command.
All in the FM near your F1 key!!!
---

[David Marso]

What aspects of this suggestion do you find mysterious?
I don't believe there is much to explain!
---

ziweiguan wrote

Hi David

Thank you for your input. Would you please explain a little bit more how to use the compute / rv.normal to add random errors to the EM output data set? Thanks!

David Marso wrote

1. See RV.NORMAL function under COMPUTE command.
2. See MATRIX subcommand under FACTOR command.
All in the FM near your F1 key!!!
---

[Maguin, Eugene]

Yes, but the most important questions are a) should something be added and b) if yes, how much should be added.

How do we know that this current MI version (23 for me) needs to have something added?

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of David Marso
Sent: Thursday, July 16, 2015 4:05 PM
To: [hidden email]
Subject: Re: Doing EFA after multiple imputation

What aspects of this suggestion do you find mysterious?
I don't believe there is much to explain!
---

ziweiguan wrote

> Hi David
>
> Thank you for your input. Would you please explain a little bit more
> how to use the compute / rv.normal to add random errors to the EM
> output data set? Thanks!
> David Marso wrote
>> 1. See RV.NORMAL function under COMPUTE command.
>> 2. See MATRIX subcommand under FACTOR command.
>> All in the FM near your F1 key!!!
>> ---

-----
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
---
"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Doing-EFA-after-multiple-imputation-tp5730158p5730163.html
Sent from the SPSSX Discussion mailing list archive at Nabble.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

=====================
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]

Indeed, and not really our responsibility/concern to address such things.
I think OP has some serious weekend research to do.
--

Maguin, Eugene wrote

Yes, but the most important questions are a) should something be added and b) if yes, how much should be added.

How do we know that this current MI version (23 for me) needs to have something added?

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of David Marso
Sent: Thursday, July 16, 2015 4:05 PM
To: [hidden email]
Subject: Re: Doing EFA after multiple imputation

What aspects of this suggestion do you find mysterious?
I don't believe there is much to explain!
---

ziweiguan wrote
> Hi David
>
> Thank you for your input. Would you please explain a little bit more
> how to use the compute / rv.normal to add random errors to the EM
> output data set? Thanks!
> David Marso wrote
>> 1. See RV.NORMAL function under COMPUTE command.
>> 2. See MATRIX subcommand under FACTOR command.
>> All in the FM near your F1 key!!!
>> ---





-----
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
---
"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Doing-EFA-after-multiple-imputation-tp5730158p5730163.html
Sent from the SPSSX Discussion mailing list archive at Nabble.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

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

[Bruce Weaver]

IMO, you should not use the values imputed by MVA, either with or without random error added to the imputed values.  For what you want to do, just use the matrix of EM covariances (or correlations) as input to the FACTOR command.  

ziweiguan wrote

Hi David

Thank you for your input. Would you please explain a little bit more how to use the compute / rv.normal to add random errors to the EM output data set? Thanks!

David Marso wrote

1. See RV.NORMAL function under COMPUTE command.
2. See MATRIX subcommand under FACTOR command.
All in the FM near your F1 key!!!
---

[ziweiguan]

Hi Bruce

Thank you. This afternoon I downloaded your macro and syntax file and gave a try. Very nice. The result I got is basically the same when I used only complete cases. I only have <0.2% missing.

Bruce Weaver wrote

IMO, you should not use the values imputed by MVA, either with or without random error added to the imputed values.  For what you want to do, just use the matrix of EM covariances (or correlations) as input to the FACTOR command.