Google Groups no longer supports new Usenet posts or subscriptions. Historical content remains viewable.
Dismiss

Unrecoverable error trying to get residuals in mixed model

3 views
Skip to first unread message

mishery

unread,
May 9, 2008, 8:05:13 AM5/9/08
to
Hi. When I try and get residuals out of a mixed model in SPSS 15
(patched) I get an error message about an unrecoverable error and the
processor being unavailable. The actual model runs and produces output
but no residuals are written to the data file.

Any ideas folks?

Syntax is below and first and last few lines of the data. (There are
9030 rows of data, by the way).

- Mike

Mixed LnRT with respprop
/FIXED = respprop
/PRINT = SOLUTION TESTCOV
/RANDOM = INTERCEPT | SUBJECT(subj) COVTYPE(UN)
/RANDOM = INTERCEPT | SUBJECT(item) COVTYPE(UN)
/SAVE = resid(LnRTx) .

subj item LnRT respprop
1 1 7.10661 .03333
1 2 6.74993 .26667
1 3 7.13569 .46667
...
...
30 299 6.55108 .46667
30 300 7.17166 .46667
30 301 7.25418 .50000

Andrew

unread,
May 14, 2008, 6:32:58 PM5/14/08
to
Have you tried saving other variables?

If you get could get it to save the predicted values, you could
calculate the residuals yourself - not a fix but a useable work
around.

Andrew

mishery

unread,
May 15, 2008, 8:20:44 AM5/15/08
to
On May 14, 11:32 pm, Andrew <Andrew.Mackin...@anu.edu.au> wrote:
> Have you tried saving other variables?

Thank you for the suggestion. Hadn't thought of this, but
unfortunately just tried this but same thing happens.

I think I will try and contact SPSS themselves as I have posted this
problem a couple of times here and elsewhere and your response was the
first I have received.

Andrew

unread,
May 15, 2008, 6:24:16 PM5/15/08
to
A couple more thoughts.

First, COVTYPE(UN) is probably generating a warning because there is
only one variance component. I'd change this to COVTYPE(ID) to get
rid of it.
I'd try deconstructing the model to see whether removing components of
it allow SPSS to save the residuals. As far as I can see, there is
nothing wrong with the model, but understanding what SPSS chokes on
may help.

I'd also look for empty cells in the design or cells with zero
variance - once again, shouldn't be a problem but the program may not
like it.

Final suggestion: run the model in SAS! It's always interesting to
see if the benchmark program has the same problem.

Best wishes,

Andrew

Kylie

unread,
May 15, 2008, 8:06:28 PM5/15/08
to
Hi Andrew,

It looks like this has been documented on the SPSS support site (login
with username/password guest/guest to http://support.spss.com):

Problem Description: I'm running the Linear Mixed Models (MIXED)
procedure in SPSS with a multilevel model. I have two RANDOM
subcommands in my model. It seems to work fine as long as I don't try
to save anything like predicted values or residuals, but when I do, I
get an error and the SPSS processor terminates. If I also ask to print
certain things, it seems to hang and wipes out my working data file,
and I have to use the Task Manager to close the program. Is there a
way to work around this problem?

Resolution Description:
The problem has been fixed for Release 16. You can work around this by
computing a constant variable equal to 1 and using this as a random
effect instead of a random intercept. The model fitted is the same and
the procedure should function appropriately. We apologize for any
resulting inconvenience.

Hope this helps.

Kylie.

0 new messages