Missing Values Analysis EM

classic Classic list List threaded Threaded
1 message Options
Reply | Threaded
Open this post in threaded view
|

Missing Values Analysis EM

Alice Sullivan
Dear all,

 

I am trying to use EM (Expectation-Maximization) to impute the missing
values in a set of 9 test scores. Is there anyone out there who can help
me with this? I am using SPSS 12.01.

 

The descriptives for the test scores are as follows:

 

Descriptive Statistics

 

 

N

Minimum

Maximum

Mean

Std. Deviation

1T Problem Arithmetic Test score

10816

0

10

5.19

2.440

1T Southgate Group Reading Test score

10842

0

30

23.76

6.764

1T Draw-a-man test score

10641

0

53

24.02

6.955

1S Total score on Copying Designs Test

10808

0

12

7.08

1.949

2T Verbal score on general ability test

10604

0

40

22.53

9.174

2T Non verbal score on gen ability test

10604

0

40

21.26

7.385

2T Total score on general ability test

10604

0

80

43.79

15.710

2T Reading comprehension test score

10602

0

35

16.27

6.151

2T Mathematics test score

10598

0

40

17.13

10.254

Valid N (listwise)

9365

 

 

 

 

 

 

Under 'Distribution' there are 3 options: 1. normal, 2. mixed normal,
and 3. Student's t. For a mixed normal assumption, you have to specify
the proportion and the standard deviation ratio. For Student's t
distribution, you must specify the degrees of freedom. How do I decide
which one to choose?

 

I started out by trying option 1 - normal. Here is the syntax:

 

MVA

  n90 n92 n1840 n914 n917 n920 n923 n926 n457  

  /EM ( TOLERANCE=0.001 CONVERGENCE=0.0001 OUTFILE='F:\Work
Alice\coeducation vs single sex\academic attainment\test scores EM
imputation.sav' ) .

EXECUTE.

 

When I looked at the descriptives for the resulting dataset, I found
that some of the variables now included minus values, whereas all the
original variables had ranges starting with zero. Why would this be?

 

Also, I can't work out how to get the serial number included in the new
outfile - without which I can't reattach the new variables to the rest
of my dataset.

 

Many thanks in advance for any responses!

Alice

 

Dr Alice Sullivan

Institute of Education

London

 

020 76126661

[hidden email]

 

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