MIXED interaction with contrasts

Hi there-
I would like to find out whether there is an interaction between treatment
group and time, using a special contrast of treatment group.  In my example,
treatment group has 3 levels, and time has 5 levels.  I am able to get the
output for the custom hypotheses, but the interaction effects seem to be
based on the default coding for treatment group.

Here is my syntax:

MIXED WEEKLYCIGS by GROUP
 with  WEEK
 /FIXED =
 GROUP
 WEEK
 GROUP*WEEK
 | SSTYPE (3)
/test 'G1 v G3' group -1/2 0 1/2
/test 'G2 v G1 and G3 ' GROUP -1/2 1 -1/2
 /PRINT SOLUTION LMATRIX/METHOD = REML
 /RANDOM INTERCEPT |SUBJECT (ID).

Any advice?

Charla

=====================
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]

Charla,

I realize this response is late, but I just came across this post. If
you treat "Week" as a categorical variable, then you should be able to
test if there is a difference in linear trends between Group 1 and
Group 3 by adding the following TEST statement:

  /TEST = 'Linear Trend Contrast 1 v 3' Group*Week  -3 -1  1  3

      0  0  0  0

      3  1 -1 -3

If you wanted to test if there is a difference in linear trends
between Group 2 and Groups 1 and 3, then you would add the following
TEST statement:

  /TEST = 'Linear Trend Contrast 2 v 1 and 3' Group*Week  1.5  0.5 -0.5 -1.5

             -3.0 -1.0  1.0  3.0

              1.5  0.5 -0.5 -1.5

These recommended TEST statements are untested, so you will need to
confirm that these are correct.

On another note, I suggest you consider accounting for the possible
residual covariance due to repeated measures by adding a REPEATED
statement (e.g., unstructured, autoregressive).

HTH,

Ryan

On Tue, Aug 10, 2010 at 11:40 AM, Charla Lopez <[hidden email]> wrote:

> Hi there-
> I would like to find out whether there is an interaction between treatment
> group and time, using a special contrast of treatment group.  In my example,
> treatment group has 3 levels, and time has 5 levels.  I am able to get the
> output for the custom hypotheses, but the interaction effects seem to be
> based on the default coding for treatment group.
>
> Here is my syntax:
>
> MIXED WEEKLYCIGS by GROUP
>  with  WEEK
>  /FIXED =
>  GROUP
>  WEEK
>  GROUP*WEEK
>  | SSTYPE (3)
> /test 'G1 v G3' group -1/2 0 1/2
> /test 'G2 v G1 and G3 ' GROUP -1/2 1 -1/2
>  /PRINT SOLUTION LMATRIX/METHOD = REML
>  /RANDOM INTERCEPT |SUBJECT (ID).
>
> Any advice?
>
> Charla
>
> =====================
> 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]

All,

 

If the OP were to treat the time variable as a continuous variable (as appears to be the case), then one way to test for group differences in slopes using the TEST subcommand in MIXED would be:

MIXED y BY Group WITH Time
  /FIXED=Group Group*Time | NOINT SSTYPE(3)
  /METHOD=REML
  /PRINT=DESCRIPTIVES  SOLUTION
  /REPEATED=Time | SUBJECT(ID) COVTYPE(UN)
  /TEST = "Diff in Slopes Between G1 and G2" Group*Time 1 -1 0

  /TEST = "Diff in Slopes Between G2 and G3" Group*Time 0 1 -1 

  /TEST = "Diff in Slopes Between G1 and G2,G3" Group*Time 1 -.5 -.5

  /TEST = "Diff in Slopes Between G2 and G1,G3" Group*Time -.5 1 -.5.

 

Take note that the FIXED statement includes the effects Group and Group*Time along with the NOINT specification. The grand intercept and the time main effect have been excluded. These specifications result in an intercept and slope term for each group, allowing one to use the TEST subcommand to test for differences in slopes.

 

I include several TEST statements to demonstrate how one would test a couple of simple and complex contrasts.

 

HTH,

 

Ryan

 

p.s. It's certainly possible that a RANDOM statement could be added and/or the REPEATED statement could be modified depending on the data.

 

On Sat, Aug 14, 2010 at 12:24 AM, R B <[hidden email]> wrote:

Charla,

I realize this response is late, but I just came across this post. If
you treat "Week" as a categorical variable, then you should be able to
test if there is a difference in linear trends between Group 1 and
Group 3 by adding the following TEST statement:

 /TEST = 'Linear Trend Contrast 1 v 3' Group*Week  -3 -1  1  3

     0  0  0  0

     3  1 -1 -3

If you wanted to test if there is a difference in linear trends
between Group 2 and Groups 1 and 3, then you would add the following
TEST statement:

 /TEST = 'Linear Trend Contrast 2 v 1 and 3' Group*Week  1.5  0.5 -0.5 -1.5

            -3.0 -1.0  1.0  3.0

             1.5  0.5 -0.5 -1.5

These recommended TEST statements are untested, so you will need to
confirm that these are correct.

On another note, I suggest you consider accounting for the possible
residual covariance due to repeated measures by adding a REPEATED
statement (e.g., unstructured, autoregressive).

HTH,

Ryan

On Tue, Aug 10, 2010 at 11:40 AM, Charla Lopez <[hidden email]> wrote:
> Hi there-
> I would like to find out whether there is an interaction between treatment
> group and time, using a special contrast of treatment group.  In my example,
> treatment group has 3 levels, and time has 5 levels.  I am able to get the
> output for the custom hypotheses, but the interaction effects seem to be
> based on the default coding for treatment group.
>
> Here is my syntax:
>
> MIXED WEEKLYCIGS by GROUP
>  with  WEEK
>  /FIXED =
>  GROUP
>  WEEK
>  GROUP*WEEK
>  | SSTYPE (3)
> /test 'G1 v G3' group -1/2 0 1/2
> /test 'G2 v G1 and G3 ' GROUP -1/2 1 -1/2
>  /PRINT SOLUTION LMATRIX/METHOD = REML
>  /RANDOM INTERCEPT |SUBJECT (ID).
>
> Any advice?
>
> Charla
>
> =====================
> 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
>