Warnings from mixed linear models

[Suh-Ing Amy Hsieh]

Hi all:
I am running mixed linear models for the project of a longitudinal dataset
using SPSS. While I used compound symetry for error covariances, it
appeared "The final Hessian matrix is not positive definite although all
convergence criteria are satisfied. The MIXED procedure continues despite
this warning. Validity of subseuent results cannot be ascertained." Do any
factors cause this problem? Any solutions can be used to solve this problem.
Thanks.
Amy

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

[Maguin, Eugene]

All,

I'm using the following casetovars statement to restructure a dataset. In
terms of the variables used, Record is A9, Code is A2 and the variables I
restrucure are A80.

*  LENGTH RANGES FROM 614 TO 3794.
CASESTOVARS /ID=RECORD/FIXED=CODE/SEPARATOR="".

Anyway, I get one listing of the following warning message.

Warnings
Cases are out of order. The first sequence error was found at case 81. The
result data may not be correct.


I'm not sure whether the case 81 referred to is based on the input file. If
it is, case 81 is the last record in a set of records. However, the set of
records to which it belongs is not the first set of records, rather it is
the 2nd or third set of records.

Has anyone seen this warning before? Under what circumstances?

Thanks, Gene Maguin

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

[peter link]

Hi Amy -

This generally happens when one or more of the covariance components are on
the edge of the parameter space.  So, for you, I mean that one of your
variances is probably close to zero.

A few things to try...

(1) Increase the maximum number of iterations
(2) Increase the maximum number of step-halvings
(3) Simplify your model

(1) and (2) are found by clicking on the 'estimation' button.

For (3), if the slope/intercept variance is close to zero, make it a fixed
parameter rather than random.

Hope this helps.  Ask if you have other questions.

Peter Link
VA San Diego Healthcare System

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]]On Behalf Of
Suh-Ing Amy Hsieh
Sent: Monday, April 28, 2008 7:18 PM
To: [hidden email]
Subject: Warnings from mixed linear models


Hi all:
I am running mixed linear models for the project of a longitudinal dataset
using SPSS. While I used compound symetry for error covariances, it
appeared "The final Hessian matrix is not positive definite although all
convergence criteria are satisfied. The MIXED procedure continues despite
this warning. Validity of subseuent results cannot be ascertained." Do any
factors cause this problem? Any solutions can be used to solve this problem.
Thanks.
Amy

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

[Richard Ristow]

At 09:30 AM 4/29/2008, Gene Maguin wrote:

>I'm using the following casetovars statement to restructure a dataset. In
>terms of the variables used, Record is A9, Code is A2 and the variables I
>restrucure are A80.
>
>CASESTOVARS /ID=RECORD/FIXED=CODE/SEPARATOR="".
>
>Anyway, I get one listing of the following warning message.
>
>Warnings
>>Cases are out of order. The first sequence error was found at case 81. The
>>result data may not be correct.
>
>I'm not sure whether the case 81 referred to is based on the input file.

It seems to be based on the input file, and to result from a
violation of the requirement, "CASESTOVARS expects the data to be
sorted by SPLIT FILE variables and then by ID variables" (p.192, in
the v.14 Command Syntax Reference.)

What's actually enforced appears not to be this requirement, but the
lesser one that the file be *grouped* by the ID variables, where the
groups need not be in ID-variable sort order. See test, below:
|-----------------------------|---------------------------|
|Output Created               |30-APR-2008 15:00:11       |
|-----------------------------|---------------------------|
Record   Letter CODE

Roman    A      RR
Roman    B      RR
Roman    C      RR
Greek    Alpha  GG
Roman    D      RR
Greek    Beta   GG
Greek    Gamma  GG

Number of cases read:  7    Number of cases listed:  7


CASESTOVARS /ID=RECORD/FIXED=CODE/SEPARATOR="".

Cases to Variables
|-----------------------------|---------------------------|
|Output Created               |30-APR-2008 15:00:11       |
|-----------------------------|---------------------------|
Warnings
|---------------------------------------------------------|
|Cases are out of order. The first sequence error was     |
|found at case 5. The result data may not be correct.     |
|---------------------------------------------------------|

Generated Variables
|---------|-------|
|Original |Result |
|Variabl  |-------|
|e        |Name   |
|-------|-|-------|
|Letter |1|Letter1|
|       |2|Letter2|
|       |3|Letter3|
|-------|-|-------|

Processing Statistics
|---------------|---|
|Cases In       |7  |
|Cases Out      |4  |
|---------------|---|
|Cases In/Cases |1.8|
|Out            |   |
|---------------|---|
|Variables In   |3  |
|Variables Out  |5  |
|---------------|---|
|Index Values   |3  |
|---------------|---|

LIST.

List
|-----------------------------|---------------------------|
|Output Created               |30-APR-2008 15:00:11       |
|-----------------------------|---------------------------|
Record   CODE Letter1 Letter2 Letter3

Roman    RR   A       B       C
Greek    GG   Alpha
Roman    RR   D
Greek    GG   Beta    Gamma

Number of cases read:  4    Number of cases listed:  4

============================
APPENDIX: Test data and code
============================
*  ................................................................. .
*  .................   Test data               ..................... .
*  ....  Letter names in selected alphabets. Taken from        ..... .
*  ....  http://www.merriam-webster.com/mw/table/alphabet.htm  ..... .
*  ................................................................. .
*  ................................................................. .
DATA LIST LIST /
     Record   #Order Letter
    (A8,      F3,    A6).
BEGIN DATA
     Roman       1   A
     Roman       2   B
     Roman       3   C
     Greek       1   Alpha
     Roman       4   D
     Greek       2   Beta
     Greek       3   Gamma
END DATA.

*  ................................................................. .
*  ....  Recode, to Gene's variable list                       ..... .
*  "Record is A9, Code is A2 and the variables I restrucure are A80.".
*  ................................................................. .

STRING CODE (A2).
RECODE Record ('Hebrew' = 'HH')
               ('Roman'  = 'RR')
               ('Greek'  = 'GG')
               ('Arabic' = 'AA') INTO CODE.

LIST.
CASESTOVARS /ID=RECORD/FIXED=CODE/SEPARATOR="".
LIST.

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