Fwd: Confusion with Linear Mixed Models - Repeated Measures

[Mark Webb-5]

Data
Repeated - before/after with same/matching respondents.
Data in Long Format as required by LMM SPSS version 18.
Aim
Has there been a significant improvement in the Education Index From T1 to T2 ?
Time = T1 T2
EduIndex = Education Attitude Scale - range 0-1
Id = Matching students
MIXED EduIndex BY Time
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,
ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=Time | SSTYPE(3)
/METHOD=REML
/PRINT=R SOLUTION TESTCOV
/REPEATED=Time | SUBJECT(Id) COVTYPE(UN).
Estimates of Fixed Effectsb
Parameter	Estimate	Std. Error	df	t	Sig.
Intercept	.953282	.004487	209.095	212.432	.000
[Time=1]	.006556	.006002	199.520	1.092	.276
[Time=2]	0	0	.	.	.
Conclusion
0.276 is greater than 0.05 [95% Confidence Interval] so no significant change in eduindex over time period.
Is this correct ?
################################################################################
Now I want to see if there is any significant difference between Schools [1 numeric varaible 8 schools]
How do I do this ?
Do I change
/FIXED=Time | SSTYPE(3)
to
/FIXED=School | SSTYPE(3)
i.e.
MIXED EduIndex BY School
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,
ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=School | SSTYPE(3)
/METHOD=REML
/PRINT=SOLUTION
/REPEATED=Time | SUBJECT(Id) COVTYPE(UN).
Here is the output which I struggle to interpret.
Why is school 8 output redundant ?
How do I determine which schools have improved/declined & which represent
significant changes?
Estimates of Fixed Effectsb
Parameter	Estimate	Std. Error	df	t	Sig.
Intercept	.967253	.009116	194.946	106.106	.000
[School=1]	.006861	.011456	197.440	.599	.550
[School=2]	-.023379	.012097	193.892	-1.933	.055
[School=3]	-.025070	.013936	192.233	-1.799	.074
[School=4]	-.022792	.011507	193.454	-1.981	.049
[School=5]	-.022165	.017332	195.677	-1.279	.202
[School=6]	-.000200	.012465	191.551	-.016	.987
[School=7]	-.008814	.012479	195.325	-.706	.481
[School=8]	0	0	.	.	.

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

[Ryan]

Mark,

 

I visualize your data set in veritcal format as follows:

 

ID   School    Time   EduIndex

1       1            1     .45 

1       1            2     .24   

2       1            1     .89  

2       1            2     .54

.

.

.

10      2            1    .24

10      2            2    .31

.

.
.

 

If you were to treat time as a categorical variable, then one possible parameterization would be:

 

MIXED EduIndex BY School Time
  /FIXED=School Time School*Time | SSTYPE(3)
  /METHOD=REML
  /PRINT=SOLUTION
  /REPEATED=Time | SUBJECT(ID) COVTYPE(CS).

 

You can invoke the TEST subcommand to examine if the mean change from time 1 to time 2 is significantly different between each pair of schools. Here are a couple of examples, which assume that you have two time pionts and 8 schools:

 

/TEST = 'Difference in Mean Change Between School 1 and 2'

             School*Time -1  1 

                                 1 -1

                                 0  0

                                 0  0

                                 0  0

                                 0  0

                                 0  0

                                 0  0

 

/TEST = 'Difference in Mean Change Between School 1 and 3'

             School*Time -1  1 

                                 0  0

                                 1 -1

                                 0  0

                                 0  0

                                 0  0

                                 0  0

                                 0  0

 

 

I have many questions about your design and data, but instead getting into the weeds, I decided to just narrow in on one of your questions. Having said that, one question that I just can't let go of is the term "matching" that you used a couple times in your post. Exactly what do you mean by "matching"? Are you referring to the fact that post scores of individuals are matched to their prescores? Or were subjects actually matched on key characteristics?

 

Ryan

On Tue, Nov 9, 2010 at 5:19 AM, Mark Webb <[hidden email]> wrote:

Data
Repeated - before/after with same/matching respondents.
Data in Long Format as required by LMM SPSS version 18.
Aim
Has there been a significant improvement in the Education Index From T1 to T2 ?
Time = T1 T2
EduIndex = Education Attitude Scale - range 0-1
Id = Matching students
MIXED EduIndex BY Time
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,
ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=Time | SSTYPE(3)
/METHOD=REML
/PRINT=R SOLUTION TESTCOV
/REPEATED=Time | SUBJECT(Id) COVTYPE(UN).
Estimates of Fixed Effectsb
Parameter	Estimate	Std. Error	df	t	Sig.
Intercept	.953282	.004487	209.095	212.432	.000
[Time=1]	.006556	.006002	199.520	1.092	.276
[Time=2]	0	0	.	.	.
Conclusion
0.276 is greater than 0.05 [95% Confidence Interval] so no significant change in eduindex over time period.
Is this correct ?
################################################################################
	Now I want to see if there is any significant difference between Schools [1 numeric varaible 8 schools]
How do I do this ?
Do I change
/FIXED=Time | SSTYPE(3)
to
/FIXED=School | SSTYPE(3)
i.e.
MIXED EduIndex BY School
/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,
ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)
/FIXED=School | SSTYPE(3)
/METHOD=REML
/PRINT=SOLUTION
/REPEATED=Time | SUBJECT(Id) COVTYPE(UN).
Here is the output which I struggle to interpret.
Why is school 8 output redundant ?
How do I determine which schools have improved/declined & which represent
significant changes?
Estimates of Fixed Effectsb
Parameter	Estimate	Std. Error	df	t	Sig.
Intercept	.967253	.009116	194.946	106.106	.000
[School=1]	.006861	.011456	197.440	.599	.550
[School=2]	-.023379	.012097	193.892	-1.933	.055
[School=3]	-.025070	.013936	192.233	-1.799	.074
[School=4]	-.022792	.011507	193.454	-1.981	.049
[School=5]	-.022165	.017332	195.677	-1.279	.202
[School=6]	-.000200	.012465	191.551	-.016	.987
[School=7]	-.008814	.012479	195.325	-.706	.481
[School=8]	0	0	.	.	.

--
Mark Webb

Line [phone]
Cell [phone]
Fax to email [phone]
Skype  webbmark
Email  [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