mixed model question

mixed model question

Post by Floris de Lan » Sat, 28 Jun 2003 19:12:32



Dear SPSS-ers,

I have a question regarding the use of the mixed-model approach and
nested designs.
In my dataset there are 2 groups each with 16 subjects, and each
subject performed 2 different tasks (fixed effect) in which a variable
was manipulated (7 levels: random effect). Then, there are multiple
observations per subject in all these levels of all tasks.
Previously, with just one group, I used univariate GLM with subject as
a random factor. But now, I cannot do that since subject is nested
within group, so I cannot enter subject and group independently.
Therefore, I though the Linear Mixed Model in SPSS 11.0 would be a
solution. The problem is that I don't understand how to carry out this
procedure; I would like to be able to test for group differences, and
whether there are any task*group differences.

I hope my story is not to complicatedly stated, and any help is
greatly appreciated!

Kind regards,
Floris de Lange

 
 
 

mixed model question

Post by Bruce Weave » Sat, 28 Jun 2003 21:13:00



> Dear SPSS-ers,

> I have a question regarding the use of the mixed-model approach and
> nested designs.
> In my dataset there are 2 groups each with 16 subjects, and each
> subject performed 2 different tasks (fixed effect) in which a variable
> was manipulated (7 levels: random effect). Then, there are multiple
> observations per subject in all these levels of all tasks.
> Previously, with just one group, I used univariate GLM with subject as
> a random factor. But now, I cannot do that since subject is nested
> within group, so I cannot enter subject and group independently.
> Therefore, I though the Linear Mixed Model in SPSS 11.0 would be a
> solution. The problem is that I don't understand how to carry out this
> procedure; I would like to be able to test for group differences, and
> whether there are any task*group differences.

> I hope my story is not to complicatedly stated, and any help is
> greatly appreciated!

> Kind regards,
> Floris de Lange

Hi Floris,
        You can indicate nesting of subjects within groups as
follows:  SUBJ(GRP).  Does that solve your problem?

Cheers,
Bruce
--
Bruce Weaver

www.angelfire.com/wv/bwhomedir/

 
 
 

mixed model question

Post by Floris de Lan » Fri, 04 Jul 2003 18:40:11


Dear Bruce,

Thanks a lot for your answer. I'm afraid it hasn't solved my problem
yet though.
I have been able to build a nested term (in this case SUBJ(GROUP)) as
a Random factor, as you suggested; furthermore, I have specified
Rotation as a random factor and Task as a fixed factor (so it's a
2x2x7 design; 2 groups, 2 tasks, each task having 7 levels). What I
would like to test for, is the significance of Task, Group, and
Rotation effects, and possible interactions between these effects.

But when I inspect the output, I only get a Type III test for Task
(the fixed effect). For the random factors, it only supplies the
covariance parameters. So how I can I test for the significance of the
random effects? And is it possible to test for an effect of Group on
itself, since it is a nested term?

Thanks again for any insights on this issue.

Yours,
Floris



> > Dear SPSS-ers,

> > I have a question regarding the use of the mixed-model approach and
> > nested designs.
> > In my dataset there are 2 groups each with 16 subjects, and each
> > subject performed 2 different tasks (fixed effect) in which a variable
> > was manipulated (7 levels: random effect). Then, there are multiple
> > observations per subject in all these levels of all tasks.
> > Previously, with just one group, I used univariate GLM with subject as
> > a random factor. But now, I cannot do that since subject is nested
> > within group, so I cannot enter subject and group independently.
> > Therefore, I though the Linear Mixed Model in SPSS 11.0 would be a
> > solution. The problem is that I don't understand how to carry out this
> > procedure; I would like to be able to test for group differences, and
> > whether there are any task*group differences.

> > I hope my story is not to complicatedly stated, and any help is
> > greatly appreciated!

> > Kind regards,
> > Floris de Lange

> Hi Floris,
>    You can indicate nesting of subjects within groups as
> follows:  SUBJ(GRP).  Does that solve your problem?

> Cheers,
> Bruce

 
 
 

mixed model question

Post by Bruce Weave » Fri, 04 Jul 2003 21:45:11



> Dear Bruce,

> Thanks a lot for your answer. I'm afraid it hasn't solved my problem
> yet though.
> I have been able to build a nested term (in this case SUBJ(GROUP)) as
> a Random factor, as you suggested; furthermore, I have specified
> Rotation as a random factor and Task as a fixed factor (so it's a
> 2x2x7 design; 2 groups, 2 tasks, each task having 7 levels). What I
> would like to test for, is the significance of Task, Group, and
> Rotation effects, and possible interactions between these effects.

> But when I inspect the output, I only get a Type III test for Task
> (the fixed effect). For the random factors, it only supplies the
> covariance parameters. So how I can I test for the significance of the
> random effects? And is it possible to test for an effect of Group on
> itself, since it is a nested term?

> Thanks again for any insights on this issue.

> Yours,
> Floris

--- snip ----

Hi Floris.  If you post your syntax (for the GLM), maybe I,
or someone else, can spot the problem.  And remind us of the
format of your data file--i.e., is it one row per
measurement (LONG file) with variables to code for Subject,
and all the other variables?

Cheers,
Bruce

 
 
 

mixed model question

Post by Floris de Lan » Fri, 25 Jul 2003 17:36:17


Hi Bruce,

I'll briefly describe the setup of the file and syntax.
The way the file is organized is one row per measurement.
Subno   Group   Task   Angle   RT
1       1       1      30      1260
1       1       1      30      1410
1       1       1      90      1700

etc.

There are two groups, both having 16 subjects. So I numbered the
subjectno from 1 to 32, subjects 1..16 adhering to group 1, and
subjects 17..32 adhering to group 2. All subjects have repeated
observations (10 or so) for each factor of each task. There are two
tasks (labeled 1 and 2), and for each of these tasks, there is an
Angle specified, which has 7 levels (0..180 in 30 degrees steps). The
dependent variable is RT (reaction time, measured in msec).

Task is a fixed factor, since it can only have two levels, task1 or
task2. Same applies to Group. Angle on the other hand, is a random
factor, since the levels we chose for angle are not the only levels
that one can choose, it's just some levels of this ordinal variable.
And subject is also a random factor, since subjects are drawn at
random from both groups. (so is there a problem here already, when
subject is a random factor, and group a fixed factor, in building a
nested term of the two?)

I would like to test for possible group differences, group*task
interactions, and group*task*angle interactions.

The syntax for the GLM is the following:

MIXED
 rt  BY angle group task subno
 /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)
 SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0,
ABSOLUTE)
 PCONVERGE(0.000001, ABSOLUTE)
 /FIXED = task  | SSTYPE(3)
 /METHOD = REML
/RANDOM = subno(group) angle  | COVTYPE(ID) .

Thanks again for your help!
Cheers,
Floris

Quote:

> Hi Floris.  If you post your syntax (for the GLM), maybe I,
> or someone else, can spot the problem.  And remind us of the
> format of your data file--i.e., is it one row per
> measurement (LONG file) with variables to code for Subject,
> and all the other variables?

> Cheers,
> Bruce

 
 
 

1. mixed model question

Dear Listserv Users:

I am fitting a mixed model to some data that contains repeated measurements on the same set of experimental units over time. The experimental units were divided into several treatment groups.  Each unit was in the same treatment group through the course of the experiment.  I believe there to be random variation related to each of the experimental units.  In PROC MIXED, do I need to include both a REPEATED line to specify the subject of the repeated measures, as well as a RANDOM line to specify random variation associated with the experimental units?

e.g.

data treatment;

input unit treatment y;

datalines

1 5 12

1 5 23

1 5 42

2 10 17

2 10 29

2 10 52 (etc.)

;

proc mixed;

     class treatment unit;

     /*unit is the class variable identifying each experimental unit with in a treatment*/

     model y=treatment;

     repeated /type=cs subject=unit;

    random unit;

run;

Please forgive me if this seems an elementary question; I am just getting aquainted with the mixed procedure (the actual data set is quite large, with multiple explanatory variables, both categorical and continuous). Thank you in advance for your kind help.

Best regards,

Kimberly Austin

---------------------------------
Do You Yahoo!?
Yahoo! Finance - Get real-time stock quotes

2. Smalltalk Developers

3. explained variance in mixed models (was re: proc mixed)

4. Issues With the Sprint Backbone

5. A question concerning mixed linear models

6. Appointment Calendar: How to change week numbering?

7. Question regarding mixture models or cure models or split population models using SAS

8. How to make a standard field compulsory ?

9. SAS PROC MIXED memory overload in cross-classified model,

10. PROC MIXED memory overload in cross-classified model