Mixed Model Planned Comparisons
Locked
 
|
Hello,
I have a dataset with two time points (pre-intervention and post-intervention) and three groups (two different intervention groups and one control group) and am using a Mixed Model to compare the effectiveness of the two interventions relative to the control group and relative to each other. Thus, the first 'step' is to compare one intervention with the control group; the next step is to compare the other intervention to the control group, and the last step is to compare the two interventions to each other. At the moment the syntax I am using is this: MIXED ECBIINTENSITY BY Condition2 Time /CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=Condition2 Time Condition2*Time | SSTYPE(3) /METHOD=REML /PRINT=SOLUTION /RANDOM=INTERCEPT | SUBJECT(School) COVTYPE(VC) /RANDOM=INTERCEPT | SUBJECT(RCodeNum*School) COVTYPE(VC) /EMMEANS=TABLES(Condition2) /EMMEANS=TABLES(Time) /EMMEANS=TABLES(Condition2*Time) COMPARE(Condition2) /EMMEANS=TABLES(Condition2*Time) COMPARE(Time). where ECBIINTENSITY is the dependent variable and Condition2 represents the three groups. However this syntax is not giving me what I want as I am less interested in the main/interaction effects than a comparison between the three groups. Could someone please help me in putting together syntax that will enable me to do this? Thank you very much. |
|
|
IMO, you'd be much better off using the TEST subcommand in MIXED to specify your contrasts of interest. I cannot comment on your MIXED code because too much information is lacking around your study design. For example, you have told us nothing about the randomization process, exactly how the intervention was administered and by whom, sample size at various levels (schools, teachers, classes, students), and the list goes on and on. If you would like help on the model parameterization, you'll need to elaborate on how the study was set up and carried out. And what is "RCodeNum"?
Ryan
On Sun, Jan 22, 2012 at 6:50 AM, Duncombe <[hidden email]> wrote: Hello, |
|
|
Sorry for the lack of information. The participants were
randomised by school into a behaviour-focused parenting intervention, an
emotion-focused parenting intervention and a 12-month waitlist control group.
The interventions are designed to prevent the development of conduct disorder in
kids and was administered by a child and adolescent mental health service based
across 43 schools in total. There were an average of 9 children from each
school. The children were aged 5-9 years and came from prep (the first year of
schooling in Australia), grade 1 and grade 2. We only used school as a random
effect (not class or teacher as, at most, three children would have come from
the one classroom). The sample size by condition was 91 (emotion-focused
parenting group), 116 (behaviour-focused parenting group) and 112 (waitlist
control).
RCodeNum is where the child code number (that identifies each
child in the study) has been ranked by school. It was recommended to do this (by
Heck et al) as it produces faster results. When this is done it changes the SPSS
variable name to RCodeNum.
This was the syntax I used (after the date has been
restructured), which does not control for school. However, in most cases there
was a school effect for our dependent variables.
MIXED ECBIINTENSITY BY Condition2
Time
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=Condition2 Time Condition2*Time | SSTYPE(3) /METHOD=REML /PRINT=SOLUTION /RANDOM=INTERCEPT | SUBJECT(CodeNumb) COVTYPE(VC) /EMMEANS=TABLES(Condition2) /EMMEANS=TABLES(Time) /EMMEANS=TABLES(Condition2*Time) COMPARE(Condition2) /EMMEANS=TABLES(Condition2*Time) COMPARE(Time). And
below is the syntax I used when controlling for school with the ranked ID code
(RCodeNum) which is faster
MIXED ECBIINTENSITY BY Condition2
Time
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=Condition2 Time Condition2*Time | SSTYPE(3) /METHOD=REML /PRINT=SOLUTION /RANDOM=INTERCEPT | SUBJECT(School) COVTYPE(VC) /RANDOM=INTERCEPT | SUBJECT(RCodeNum*School) COVTYPE(VC) /EMMEANS=TABLES(Condition2) /EMMEANS=TABLES(Time) /EMMEANS=TABLES(Condition2*Time) COMPARE(Condition2) /EMMEANS=TABLES(Condition2*Time) COMPARE(Time). Then,
I used the below syntax to control for differences in grade and time difference
(where ChildAgeDiff is the difference in age for the child at time one, which is
the pre-intervention time-point) and time 3, which is the 6 month
post-intervention time point).
MIXED ECBIINTENSITY BY Condition2
Time WITH Grade ChildAgeDiff3
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE) /FIXED=Condition2 Time Grade ChildAgeDiff3 Condition2*Time | SSTYPE(3) /METHOD=REML /PRINT=SOLUTION /RANDOM=INTERCEPT | SUBJECT(School) COVTYPE(VC) /RANDOM=INTERCEPT | SUBJECT(RCodeNum*School) COVTYPE(VC) /EMMEANS=TABLES(Condition2) /EMMEANS=TABLES(Time) /EMMEANS=TABLES(Condition2*Time) COMPARE(Condition2) /EMMEANS=TABLES(Condition2*Time) COMPARE(Time). I hope
this makes sense. It was helpful to know that I should be using the TEST
subcommand in MIXED to specify my contrasts of interest.
Thank
you.
From: R B [via SPSSX Discussion] [mailto:[hidden email]] Sent: Monday, 23 January 2012 1:20 AM To: Duncombe Subject: Re: Mixed Model Planned Comparisons IMO, you'd be much better off using the TEST subcommand in MIXED to specify
your contrasts of interest.��
I cannot comment on your MIXED code because too much information is lacking
around your study design. For example, you have told us nothing about the
randomization process, exactly how the intervention was administered and by
whom, sample size at various levels (schools, teachers, classes, students), and
the list goes on and on. If you would like help on the model parameterization,
you'll need to elaborate on how the study was set up and carried out. And what
is "RCodeNum"?
Ryan On Sun, Jan 22, 2012 at 6:50 AM, Duncombe <[hidden email]>
wrote: Hello, If you reply to this email, your message will be
added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/Mixed-Model-Planned-Comparisons-tp5163929p5164097.html
No virus found in this
message. |
|
|
Okay. You have now provided enough information for me to provide you with some untested MIXED code. Your explanation suggests the following dataset structure: ID School Condition Time Y
1 1 1 1 24 1 1 1 2 13 2 1 1 1 22 2 1 1 2 15 3 1 1 1 33
3 1 1 2 20 . . 9 1 2 1 36 9 1 2 2 35 .. 1 2 2 1 39
1 2 2 2 23 2 2 2 1 56 2 2 2 2 45 3 2 2 1 36 3 2 2 2 33
. . 9 2 2 1 53 9 2 2 2 50 .. 1 2 3 1 34 1 2 3 2 20
2 2 3 1 53 2 2 3 2 41 3 2 3 1 34 3 2 3 2 31 . . 9 2 3 1 54
9 2 3 2 50 .. Note that ID=RCodeNum, Condition=Condition2, and Y=ECBIINTENSITY. Here's the untested code for you to ponder:
mixed Y by Condition Time /fixed=Condition Time Condition*Time /method=reml /print=solution /random=time | subject(School) COVTYPE(un) /repeated=time | subject(ID*School) COVTYPE(un)
/test 'Mean(C1,T1)' intercept 1 Condition 1 0 0 Time 1 0 Condition*Time 1 0 0 0 0 0 /test 'Mean(C1,T2)'
intercept 1 Condition 1 0 0 Time 0 1 Condition*Time 0 0 0 1 0 0 /test 'Mean(C1,T1) MINUS Mean(C1,T2)' intercept 0
Condition 0 0 0 Time 1 -1 Condition*Time 1 0 0 -1 0 0 /test 'Mean(C2,T1)' intercept 1 Condition 0 1 0
Time 1 0 Condition*Time 0 1 0 0 0 0 /test 'Mean(C2,T2)' intercept 1 Condition 0 1 0 Time 0 1 Condition*Time 0 0 0 0 1 0
/test 'Mean(C2,T1) MINUS Mean(C2,T2)' intercept 0 Condition 0 0 0 Time 1 -1 Condition*Time 0 1 0 0 -1 0 /test 'Mean(C3,T1)'
intercept 1 Condition 0 0 1 Time 1 0 Condition*Time 0 0 1 0 0 0 /test 'Mean(C3,T2)' intercept 1 Condition 0 0 1
Time 0 1 Condition*Time 0 0 0 0 0 1 /test 'Mean(C3,T1) MINUS Mean(C3,T2)' intercept 0 Condition 0 0 0 Time 1 -1
Condition*Time 0 0 1 0 0 -1 /test 'C1 Diff MINUS C2 Diff' intercept 0 Condition 0 0 0 Time 0 0 Condition*Time 1 -1 0 -1 1 0
/test 'C1 Diff MINUS C3 Diff' intercept 0 Condition 0 0 0 Time 0 0 Condition*Time 1 0 -1 -1 0 1 /test 'C2 Diff MINUS C3 Diff'
intercept 0 Condition 0 0 0 Time 0 0 Condition*Time 0 1 -1 0 -1 1. On Sun, Jan 22, 2012 at 7:39 PM, Duncombe <[hidden email]> wrote:
|
|
|
One small correction. In my dataset illustration, ID 9 for School 1 should be associated with condition 1. In other words, randomization of condition was performed at the school level. Anyway, everything else seems in order. Write back if you have specific questions. Ryan
|
|
|
Thank you so much for your help. I will test it
out.
Melissa From: R B [via SPSSX Discussion] [mailto:[hidden email]] Sent: Monday, 23 January 2012 4:42 PM To: Duncombe Subject: Re: Mixed Model Planned Comparisons One small correction. In my dataset illustration, ID 9 for School 1 should
be associated with condition 1. In other words, randomization of condition was
performed at the school level. Anyway, everything else seems in order. Write
back if you have specific questions.
Ryan
If you reply to this email, your message will be
added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/Mixed-Model-Planned-Comparisons-tp5163929p5165333.html
No virus found in this
message. |
|
|
In reply to this post by Ryan
Hi,
Thank you for your assistance. I have tried running the syntax
however I get a message that (most of) the custom hypothesis tests will not be
performed because the L matrix is not estimable. Could you please help me with
this as well?
Thank you.
Melissa From: R B [via SPSSX Discussion] [mailto:[hidden email]] Sent: Monday, 23 January 2012 4:42 PM To: Duncombe Subject: Re: Mixed Model Planned Comparisons One small correction. In my dataset illustration, ID 9 for School 1 should
be associated with condition 1. In other words, randomization of condition was
performed at the school level. Anyway, everything else seems in order. Write
back if you have specific questions.
Ryan
If you reply to this email, your message will be
added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/Mixed-Model-Planned-Comparisons-tp5163929p5165333.html
No virus found in this
message. |
|
|
Please provide the mixed code (including the TEST statements) you adapted from mine. There's likely a simple explanation. Ryan
|
|
|
I decided to review the TEST code myself. I see the error I made. Rather than explain the error, I just provide the corrected code for now (BELOW). As before, feel free to write back if you have any questions.
Ryan --
/test 'Mean(C1,T1)' intercept 1 Condition 1 0 0 Time 1 0 Condition*Time 1 0 0 0 0 0
/test 'Mean(C1,T2)' intercept 1 Condition 1 0 0 Time 0 1 Condition*Time 0 1 0 0 0 0 /test 'Mean(C1,T1) MINUS Mean(C1,T2)'
intercept 0 Condition 0 0 0 Time 1 -1 Condition*Time 1 -1 0 0 0 0 /test 'Mean(C2,T1)' intercept 1 Condition 0 1 0
Time 1 0 Condition*Time 0 0 1 0 0 0 /test 'Mean(C2,T2)' intercept 1 Condition 0 1 0 Time 0 1 Condition*Time 0 0 0 1 0 0
/test 'Mean(C2,T1) MINUS Mean(C2,T2)' intercept 0 Condition 0 0 0 Time 1 -1 Condition*Time 0 0 1 -1 0 0 /test 'Mean(C3,T1)'
intercept 1 Condition 0 0 1 Time 1 0 Condition*Time 0 0 0 0 1 0 /test 'Mean(C3,T2)' intercept 1 Condition 0 0 1
Time 0 1 Condition*Time 0 0 0 0 0 1 /test 'Mean(C3,T1) MINUS Mean(C3,T2)' intercept 0 Condition 0 0 0 Time 1 -1
Condition*Time 0 0 0 0 1 -1 /test 'C1 Diff MINUS C2 Diff' intercept 0 Condition 0 0 0 Time 0 0 Condition*Time 1 -1 -1 1 0 0
/test 'C1 Diff MINUS C3 Diff' intercept 0 Condition 0 0 0 Time 0 0 Condition*Time 1 -1 0 0 -1 1 /test 'C2 Diff MINUS C3 Diff'
intercept 0 Condition 0 0 0 Time 0 0 Condition*Time 0 0 1 -1 -1 1. On Mon, Jan 23, 2012 at 9:02 AM, <[hidden email]> wrote:
|
|
|
Thank you! That worked perfectly! From: R B [via SPSSX Discussion] [mailto:[hidden email]] Sent: Tuesday, 24 January 2012 10:39 AM To: Duncombe Subject: Re: Mixed Model Planned Comparisons Ryan
-- /test 'Mean(C1,T1)'
intercept 1
Condition 1 0 0
Time 1 0
Condition*Time 1 0 0 0 0 0
/test 'Mean(C1,T2)'
intercept 1
Condition 1 0 0
Time 0 1
Condition*Time 0 1 0 0 0
0
/test 'Mean(C1,T1) MINUS Mean(C1,T2)'
intercept 0
Condition 0 0 0
Time 1 -1
Condition*Time 1 -1 0 0 0 0
/test 'Mean(C2,T1)'
intercept 1
Condition 0 1 0
Time 1 0
Condition*Time 0 0 1 0 0 0
/test 'Mean(C2,T2)'
intercept 1
Condition 0 1 0
Time 0 1
Condition*Time 0 0 0 1 0 0
/test 'Mean(C2,T1) MINUS Mean(C2,T2)'
intercept 0
Condition 0 0 0
Time 1 -1
Condition*Time 0 0 1 -1 0 0
/test 'Mean(C3,T1)'
intercept 1
Condition 0 0 1
Time 1 0
Condition*Time 0 0 0 0 1 0
/test 'Mean(C3,T2)'
intercept 1
Condition 0 0 1
Time 0 1
Condition*Time 0 0 0 0 0 1
/test 'Mean(C3,T1) MINUS Mean(C3,T2)'
intercept 0
Condition 0 0 0
Time 1 -1
Condition*Time 0 0 0 0 1 -1
/test 'C1 Diff MINUS C2 Diff'
intercept 0
Condition 0 0 0
Time 0 0
Condition*Time 1 -1 -1 1 0 0
/test 'C1 Diff MINUS C3 Diff'
intercept 0
Condition 0 0 0
Time 0 0
Condition*Time 1 -1 0 0 -1 1
/test 'C2 Diff MINUS C3 Diff'
intercept 0
Condition 0 0 0
Time 0 0
Condition*Time 0 0 1 -1 -1 1.
On Mon, Jan 23, 2012 at 9:02 AM, <[hidden email]> wrote:
If you reply to this email, your message will be
added to the discussion below: http://spssx-discussion.1045642.n5.nabble.com/Mixed-Model-Planned-Comparisons-tp5163929p5198973.html
No virus found in this
message. |
«
Return to SPSSX Discussion
|
1 view|%1 views
| Free forum by Nabble | Edit this page |