SPSSX Discussion

Help with interaction with contrasts

Classic

List

Threaded

7 messages Options

charla3@aol.com

Help with interaction with contrasts

Hi- I am having trouble with the syntax for a model of condition by time,
with condition having special contrasts. Here's what I have:

2 time points
4 conditions

If I want to see the difference between the conditions for a single time
point,I use the following contrasts:

0 0 1 -1
1 -1 0 0
1/3 1/3 1/3 -1

in a model like this:

GLM outcome BY COND
/CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).

If I simply wanted the COND by TIME effects, I would use the following model:

MIXED outcome BY COND TIME
/FIXED COND TIME COND*TIME
/REPEATED= TIME | SUBJECT (ID).

But, I need to model the interaction of the contrasts by time. I can't seem
to get the model right.

Any suggestions? Any references? Any links?

THANKS!!!

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

Re: Help with interaction with contrasts

Charla,

It's certainly possible to perform a statistical test on the mean
difference between any 2 cells (e.g., group 1, time 1 versus group 2,
time 1). Your MIXED code needs a little tweaking, but you're certainly
headed in the right direction. The TEST subcommand will be key to
performing the sort of contrast you desire. I can write back later
tonight with the solution, if you're still interested.

Ryan

On Wed, Mar 30, 2011 at 11:01 AM, Charla Lopez <[hidden email]> wrote:

> Hi- I am having trouble with the syntax for a model of condition by time,
> with condition having special contrasts. Here's what I have:
>
> 2 time points
> 4 conditions
>
> If I want to see the difference between the conditions for a single time
> point,I use the following contrasts:
>
> 0 0 1 -1
> 1 -1 0 0
> 1/3 1/3 1/3 -1
>
> in a model like this:
>
> GLM outcome BY COND
> /CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).
>
> If I simply wanted the COND by TIME effects, I would use the following model:
>
> MIXED outcome BY COND TIME
> /FIXED COND TIME COND*TIME
> /REPEATED= TIME | SUBJECT (ID).
>
> But, I need to model the interaction of the contrasts by time. I can't seem
> to get the model right.
>
> Any suggestions? Any references? Any links?
>
> THANKS!!!
>
> =====================
> 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

Ryan

Re: Help with interaction with contrasts

Charla,

To obtain the contrast of interest (third TEST statement), you need to
take the difference of the terms from the first two TEST statements.
Also, take note that I specified the residual covariance structure on
the REPEATED statement. No time to explain further.

MIXED Y BY COND TIME
/FIXED=COND TIME COND*TIME | SSTYPE(3)
/METHOD=REML
/REPEATED=TIME | SUBJECT(ID) COVTYPE(UN)
/EMMEANS=TABLES(COND*TIME)
/TEST = 'G1,T1' Intercept 1 COND 1 0 TIME 1 0 COND*TIME 1 0 0 0
/TEST = 'G2,T1' Intercept 1 COND 0 1 TIME 1 0 COND*TIME 0 0 1 0
/TEST = 'G1,T1 minus G2,T1' Intercept 0 COND 1 -1 TIME 0 0
COND*TIME 1 0 -1 0.

HTH,

Ryan

On Wed, Mar 30, 2011 at 6:28 PM, R B <[hidden email]> wrote:

> Charla,
>
> It's certainly possible to perform a statistical test on the mean
> difference between any 2 cells (e.g., group 1, time 1 versus group 2,
> time 1). Your MIXED code needs a little tweaking, but you're certainly
> headed in the right direction. The TEST subcommand will be key to
> performing the sort of contrast you desire. I can write back later
> tonight with the solution, if you're still interested.
>
> Ryan
>
> On Wed, Mar 30, 2011 at 11:01 AM, Charla Lopez <[hidden email]> wrote:
>> Hi- I am having trouble with the syntax for a model of condition by time,
>> with condition having special contrasts. Here's what I have:
>>
>> 2 time points
>> 4 conditions
>>
>> If I want to see the difference between the conditions for a single time
>> point,I use the following contrasts:
>>
>> 0 0 1 -1
>> 1 -1 0 0
>> 1/3 1/3 1/3 -1
>>
>> in a model like this:
>>
>> GLM outcome BY COND
>> /CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).
>>
>> If I simply wanted the COND by TIME effects, I would use the following model:
>>
>> MIXED outcome BY COND TIME
>> /FIXED COND TIME COND*TIME
>> /REPEATED= TIME | SUBJECT (ID).
>>
>> But, I need to model the interaction of the contrasts by time. I can't seem
>> to get the model right.
>>
>> Any suggestions? Any references? Any links?
>>
>> THANKS!!!
>>
>> =====================
>> 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
>>
>

charla3@aol.com

Re: Help with interaction with contrasts

In reply to this post by charla3@aol.com

Thanks for your help Ryan. I am still unable to run the model. I get the
message "Custom hypothesis test X will not be preformed because the L matrix
is not estimable."

Here's a statement of the problem: I have 4 treatment groups and 2
measurement points of a single dependent variable. I am trying to run a
model with special contrasts so I can assess "contrast by time." All 4
groups represent treatment conditions with a combination of any set of 3
components. Here is the setup:

Cond 1) B
Cond 2) AB
Cond 3) ABC
Cond 4) A

These are the 3 contrasts of interest:
1) Condition 3 v. 4
2) Condition 2 v. 1
3) Conditions 1, 2, and 4 v. 3

A single time point model looks like this:

GLM DV BY COND
/CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).

I need to use the same contrasts in a repeated measures model to assess
"condition by time."

Any suggestions for how to model this?

THANKS!

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

Swank, Paul R

Re: Help with interaction with contrasts

What transformation are you using in the repeated statement?

Proc glm;
Class group time;;
Model y0 y1 y2 y3 = group|time;
Contrast "group1,2 vs 3,4" group 1 1 -1 -1;
Repeated time 4 profile / summary;

Profile gives you the nth successive difference in time so that the first is time 2 - time 1, etc. The contrast for that differences is printed as part of the summary.

Dr. Paul R. Swank,
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center-Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Charla Lopez
Sent: Tuesday, April 05, 2011 5:12 PM
To: [hidden email]
Subject: Re: Help with interaction with contrasts

Thanks for your help Ryan. I am still unable to run the model. I get the
message "Custom hypothesis test X will not be preformed because the L matrix
is not estimable."

Here's a statement of the problem: I have 4 treatment groups and 2
measurement points of a single dependent variable. I am trying to run a
model with special contrasts so I can assess "contrast by time." All 4
groups represent treatment conditions with a combination of any set of 3
components. Here is the setup:

Cond 1) B
Cond 2) AB
Cond 3) ABC
Cond 4) A

These are the 3 contrasts of interest:
1) Condition 3 v. 4
2) Condition 2 v. 1
3) Conditions 1, 2, and 4 v. 3

A single time point model looks like this:

GLM DV BY COND
/CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).

I need to use the same contrasts in a repeated measures model to assess
"condition by time."

Any suggestions for how to model this?

THANKS!

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

Swank, Paul R

Re: Help with interaction with contrasts

In reply to this post by charla3@aol.com

Sorry, that was sas code, my mistake.

Dr. Paul R. Swank,
Professor and Director of Research
Children's Learning Institute
University of Texas Health Science Center-Houston

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Charla Lopez
Sent: Tuesday, April 05, 2011 5:12 PM
To: [hidden email]
Subject: Re: Help with interaction with contrasts

Thanks for your help Ryan. I am still unable to run the model. I get the
message "Custom hypothesis test X will not be preformed because the L matrix
is not estimable."

Here's a statement of the problem: I have 4 treatment groups and 2
measurement points of a single dependent variable. I am trying to run a
model with special contrasts so I can assess "contrast by time." All 4
groups represent treatment conditions with a combination of any set of 3
components. Here is the setup:

Cond 1) B
Cond 2) AB
Cond 3) ABC
Cond 4) A

These are the 3 contrasts of interest:
1) Condition 3 v. 4
2) Condition 2 v. 1
3) Conditions 1, 2, and 4 v. 3

A single time point model looks like this:

GLM DV BY COND
/CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).

I need to use the same contrasts in a repeated measures model to assess
"condition by time."

Any suggestions for how to model this?

THANKS!

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

Ryan

Re: Help with interaction with contrasts

In reply to this post by charla3@aol.com

Charla,

Sorry it has taken a while to respond. It's been on of those weeks.
Anyway, I didn't realize you had 4 conditions. Ok. Assuming you're
still interested in using the MIXED procedure, the following untested
sample TEST statements should work:

MIXED Y BY COND TIME
/FIXED=COND TIME COND*TIME | SSTYPE(3)
/METHOD=REML
/REPEATED=TIME | SUBJECT(ID) COVTYPE(UN)
/TEST = 'G1,T1' Intercept 1 COND 1 0 0 0 TIME 1 0 COND*TIME 1 0 0 0 0 0 0 0
/TEST = 'G2,T1' Intercept 1 COND 0 1 0 0 TIME 1 0 COND*TIME 0 0 1 0 0 0 0 0
/TEST = 'G3,T1' Intercept 1 COND 0 0 1 0 TIME 1 0 COND*TIME 0 0 0 0 1 0 0 0
/TEST = 'G4,T1' Intercept 1 COND 0 0 0 1 TIME 1 0 COND*TIME 0 0 0 0 0 0 1 0
/TEST = 'G1,T2' Intercept 1 COND 1 0 0 0 TIME 0 1 COND*TIME 0 1 0 0 0 0 0 0
/TEST = 'G2,T2' Intercept 1 COND 0 1 0 0 TIME 0 1 COND*TIME 0 0 0 1 0 0 0 0
/TEST = 'G3,T2' Intercept 1 COND 0 0 1 0 TIME 0 1 COND*TIME 0 0 0 0 0 1 0 0
/TEST = 'G4,T2' Intercept 1 COND 0 0 0 1 TIME 0 1 COND*TIME 0 0 0 0 0 0 0 1
/TEST = 'G1,T1 minus G2,T1' Intercept 0 COND 1 -1 0 0 TIME 0 0
COND*TIME 1 0 -1 0 0 0 0 0
/TEST = 'G1,T1 minus G3,T1' Intercept 0 COND 1 0 -1 0 TIME 0 0
COND*TIME 1 0 0 0 -1 0 0 0
/TEST = 'G1,T1 minus G4,T1' Intercept 0 COND 1 0 0 -1 TIME 0 0
COND*TIME 1 0 0 0 0 0 -1 0
/TEST = 'G1,T2 minus G2,T2' Intercept 0 COND 1 -1 0 0 TIME 0 0
COND*TIME 0 1 0 -1 0 0 0 0
/TEST = 'G1,T2 minus G3,T2' Intercept 0 COND 1 0 -1 0 TIME 0 0
COND*TIME 0 1 0 0 0 -1 0 0
/TEST = 'G1,T2 minus G4,T2' Intercept 0 COND 1 0 0 -1 TIME 0 0
COND*TIME 0 1 0 0 0 0 0 -1
/TEST = 'G1 and G2 AT time 1' Intercept 2 COND 1 1 0 0 TIME 2 0
COND*TIME 1 0 1 0 0 0 0 0 DIVISOR=2
/TEST = 'G3 and G4 AT time 1' Intercept 2 COND 0 0 1 1 TIME 2 0
COND*TIME 0 0 0 0 1 0 1 0 DIVISOR=2
/TEST = 'G1,G2 minus G3,G4 AT time 1 ' Intercept 0 COND 1 1 -1 -1
TIME 0 0 COND*TIME 1 0 1 0 -1 0 -1 0 DIVISOR=2
/TEST = 'Diff in Change between G1 and G2 from t1 to t2' COND*TIME 1
-1 -1 1 0 0 0 0
/TEST = 'G1 across all time points' Intercept 2 COND 2 0 0 0 TIME 1 1
COND*TIME 1 1 0 0 0 0 0 0 DIVISOR=2
/TEST = 'G2 across all time points' Intercept 2 COND 0 2 0 0 TIME 1 1
COND*TIME 0 0 1 1 0 0 0 0 DIVISOR=2
/TEST = 'G3 across all time points' Intercept 2 COND 0 0 2 0 TIME 1 1
COND*TIME 0 0 0 0 1 1 0 0 DIVISOR=2
/TEST = 'G4 across all time points' Intercept 2 COND 0 0 0 2 TIME 1 1
COND*TIME 0 0 0 0 0 0 1 1 DIVISOR=2
/TEST = 'G1,G2 across all time points' Intercept 4 COND 2 2 0 0 TIME
2 2 COND*TIME 1 1 1 1 0 0 0 0 DIVISOR=4
/TEST = 'G3,G4 across all time points' Intercept 4 COND 0 0 2 2 TIME
2 2 COND*TIME 0 0 0 0 1 1 1 1 DIVISOR=4.

HTH,

Ryan
--

On Tue, Apr 5, 2011 at 6:11 PM, Charla Lopez <[hidden email]> wrote:

> Thanks for your help Ryan. I am still unable to run the model. I get the
> message "Custom hypothesis test X will not be preformed because the L matrix
> is not estimable."
>
> Here's a statement of the problem: I have 4 treatment groups and 2
> measurement points of a single dependent variable. I am trying to run a
> model with special contrasts so I can assess "contrast by time." All 4
> groups represent treatment conditions with a combination of any set of 3
> components. Here is the setup:
>
> Cond 1) B
> Cond 2) AB
> Cond 3) ABC
> Cond 4) A
>
> These are the 3 contrasts of interest:
> 1) Condition 3 v. 4
> 2) Condition 2 v. 1
> 3) Conditions 1, 2, and 4 v. 3
>
> A single time point model looks like this:
>
> GLM DV BY COND
> /CONTRAST (COND) SPECIAL (1 1 1 1 0 0 1 -1 1 -1 0 0 1/3 1/3 1/3 -1).
>
> I need to use the same contrasts in a repeated measures model to assess
> "condition by time."
>
> Any suggestions for how to model this?
>
> THANKS!
>
> =====================
> 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
>