Quantcast

EM in MVA question

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

EM in MVA question

Salbod
I want to use EM algorithm from MVA to handle missing data (e.g., 9 cases are missing 3 items) in this distribution of 14-item scale scores. I plan to save the imputed values for subsequent reliability analysis.

14-Item Scale
        Missing  Frequency Percent Valid Percent Cumulative Percent
         .00 747 86.5 86.5 86.5
        1.00 34 3.9 3.9 90.4
        2.00 9 1.0 1.0 91.4
        3.00 9 1.0 1.0 92.5
        4.00 1 .1 .1 92.6
        5.00 5 .6 .6 93.2
        7.00 1 .1 .1 93.3
        8.00 1 .1 .1 93.4
        9.00 3 .3 .3 93.8
        10.00 1 .1 .1 93.9
        11.00 2 .2 .2 94.1
        13.00 1 .1 .1 94.2
        14.00 50 5.8 5.8 100.0

        Total 864 100.0 100.0

Do I blindly subject the data to EM or do I edit the data, excluding those cases that I feel have too many missing values, before applying EM?

Thank you, Stephen Salbod, Pace University, NYC
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: EM in MVA question

Bruce Weaver
Administrator
Hello Stephen.  I would NOT recommend analyzing imputed values from MVA in the manner you describe.  What you ought to do instead, IMO, is use the matrix of EM correlations as input to the RELIABILITY procedure.  Unfortunately, MVA does not have a /MATRIX sub-command, and so does not write out that matrix in a form that can be read by RELIABILITY (or FACTOR).  To get around that problem, I wrote a couple macros that you may find useful.

http://dx.doi.org/10.20982/tqmp.10.2.p143 -- the TQMP article
https://sites.google.com/a/lakeheadu.ca/bweaver/Home/statistics/spss/my-spss-page/emcorr

Regarding your other question, I think I would include all of the cases as input to MVA.  But as a check, you could repeat the analysis excluding cases with "too much missing data", and eyeball it to see if the final story differs in any substantial way.  I suspect it won't.  

HTH.

Salbod wrote
I want to use EM algorithm from MVA to handle missing data (e.g., 9 cases are missing 3 items) in this distribution of 14-item scale scores. I plan to save the imputed values for subsequent reliability analysis.

14-Item Scale
        Missing  Frequency Percent Valid Percent Cumulative Percent
         .00 747 86.5 86.5 86.5
        1.00 34 3.9 3.9 90.4
        2.00 9 1.0 1.0 91.4
        3.00 9 1.0 1.0 92.5
        4.00 1 .1 .1 92.6
        5.00 5 .6 .6 93.2
        7.00 1 .1 .1 93.3
        8.00 1 .1 .1 93.4
        9.00 3 .3 .3 93.8
        10.00 1 .1 .1 93.9
        11.00 2 .2 .2 94.1
        13.00 1 .1 .1 94.2
        14.00 50 5.8 5.8 100.0

        Total 864 100.0 100.0

Do I blindly subject the data to EM or do I edit the data, excluding those cases that I feel have too many missing values, before applying EM?

Thank you, Stephen Salbod, Pace University, NYC
--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

NOTE: My Hotmail account is not monitored regularly.
To send me an e-mail, please use the address shown above.
Reply | Threaded
Open this post in threaded view
|  
Report Content as Inappropriate

Re: EM in MVA question

Salbod
Dear Bruce,

Thanks for the quick response and the link. I will follow your suggestions. --Steve



Loading...