MANCOVA with mixed-model repeated measure

Dear All,

I have a question about the use of ANCOVA. In a repeated measure
design with three levels, (neuropsychlogical domains: memory,
executive, attention) on the within subject factor and three levels on
the between subject factor (Controls, Group2, Group3), i want to
correct for psychopathological symptoms which were present only in
Group2 and Group3, but not in the control group.

Thus, I added for all controls zero on the symptom varibale and tried
to use the following syntax.

GLM
  np_memo np_exec np_atten BY group WITH symptoms
  /WSFACTOR = domain 3 Polynomial
  /METHOD = SSTYPE(3)
  /EMMEANS = TABLES(group) WITH(symptoms=MEAN) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(group*domain) WITH(symptoms=MEAN)
COMPARE(eproject) ADJ(BONFERRONI)
   /PRINT = DESCRIPTIVE ETASQ PARAMETER HOMOGENEITY GEF
  /CRITERIA = ALPHA(.05)
  /WSDESIGN = domain
  /DESIGN = eproject*cor_pans_pos eproject.

I have the following questions:
1. Did produce this syntax covariates which were estimated separate
for each group and each neuropsychological domain?
2. Were main effects and interactions in the output based on adjusted
means corrected for theinfluence of the covariate?
3. Is it possible with the EMMEANS subcommand to compare the adjusted
means of the groups. Only group2 were compared against group3 in my
outputs? Is it possible to use a more hirarchical procedure, where
first the effects of the covariates were eleminated and then the
groups were compared against each other? How can I do this with
SPSS?
4. Is it possible to save the adjusted values for each case in the
data sheet?

Thank you very much for your reply,
best regards

Ingo

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

[Maguin, Eugene]

Ingo,

Are you saying that the psychopathology score every member of the control
group was 0.0 or that a psychopathology measure was not administered to the
control group and, since you had to put something in for the controls, you
put in a score of 0. If the latter, do you really believe that every control
would have gotten a score of 0 had the measure been administered?

Gene Maguin

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

On Wed, 30 Apr 2008 04:04:13 -0400, [hidden email]
<[hidden email]> wrote:

>Dear All,
>
>I have a question about the use of ANCOVA. In a repeated measure
>design with three levels, (neuropsychlogical domains: memory,
>executive, attention) on the within subject factor and three levels on
>the between subject factor (Controls, Group2, Group3), i want to
>correct for psychopathological symptoms which were present only in
>Group2 and Group3, but not in the control group.
>
>Thus, I added for all controls zero on the symptom varibale and tried
>to use the following syntax.
>
>GLM
>  np_memo np_exec np_atten BY group WITH symptoms
>  /WSFACTOR = domain 3 Polynomial
>  /METHOD = SSTYPE(3)
>  /EMMEANS = TABLES(group) WITH(symptoms=MEAN) COMPARE ADJ(BONFERRONI)
>  /EMMEANS = TABLES(group*domain) WITH(symptoms=MEAN)
>COMPARE(eproject) ADJ(BONFERRONI)
>   /PRINT = DESCRIPTIVE ETASQ PARAMETER HOMOGENEITY GEF
>  /CRITERIA = ALPHA(.05)
>  /WSDESIGN = domain
>  /DESIGN = eproject*cor_pans_pos eproject.
>
>I have the following questions:
>1. Did produce this syntax covariates which were estimated separate
>for each group and each neuropsychological domain?
>2. Were main effects and interactions in the output based on adjusted
>means corrected for theinfluence of the covariate?
>3. Is it possible with the EMMEANS subcommand to compare the adjusted
>means of the groups. Only group2 were compared against group3 in my
>outputs? Is it possible to use a more hirarchical procedure, where
>first the effects of the covariates were eleminated and then the
>groups were compared against each other? How can I do this with
>SPSS?
>4. Is it possible to save the adjusted values for each case in the
>data sheet?
>
>Thank you very much for your reply,
>best regards
>
>Ingo
>
>=====================
>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

Dear all,

Sorry, the submitted command syntax needs some correction:

GLM
  np_memo np_exec np_atten BY group WITH symptoms
  /WSFACTOR = domain 3 Polynomial
  /METHOD = SSTYPE(3)
  /EMMEANS = TABLES(group) WITH(symptoms=MEAN) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(group*domain)WITH(symptoms=MEAN)COMPARE(group)
   ADJ(BONFERRONI)
  /PRINT = DESCRIPTIVE ETASQ PARAMETER HOMOGENEITY GEF
  /CRITERIA = ALPHA(.05)
  /WSDESIGN = domain
  /DESIGN = group*symptoms group.

I did not find a simple solution for my problem in SPSS, so
I calculated simple linear regression analysis for all dependent variables
with variable "symptoms" as predictor within each clinical group (group1 and
group2). I get a linear equitation for each dependent variable and each
group: Y = B*X + c + e, where Y is the dependent variable, x the symptoms, B
the estimated coefficient,c the constant, and e the residual. Then I
adjusted all dependent variables by subtracting the estimated effect from
the empirical data by calculating: Ya = Y - B*X , where Ya is the adjusted
domain score. For the control group I copy the original data. Finally with
the adjusted variables I conduct a simple MANOVA:

np_memo_a np_exec_a np_atten_a BY group
  /WSFACTOR = domain 3 Polynomial
  /METHOD = SSTYPE(3)
  /EMMEANS = TABLES(group) COMPARE ADJ(BONFERRONI)
  /EMMEANS = TABLES(group*domain)COMPARE(group)
   ADJ(BONFERRONI)
  /PRINT = DESCRIPTIVE ETASQ PARAMETER HOMOGENEITY GEF
  /CRITERIA = ALPHA(.05)
  /WSDESIGN = domain
  /DESIGN = group.

I do not know, whether this procedure was correct?
Any advice appreciated!

Thanks, Ingo

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

On Wed, 30 Apr 2008 14:17:47 -0400, Gene Maguin <[hidden email]> wrote:

>Ingo,
>
>Are you saying that the psychopathology score every member of the control
>group was 0.0 or that a psychopathology measure was not administered to the
>control group and, since you had to put something in for the controls, you
>put in a score of 0. If the latter, do you really believe that every control
>would have gotten a score of 0 had the measure been administered?
>
>Gene Maguin
>
>=====================
>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

Gene,

I forgot to give some information. The psychopathology score was not
obtained in the control group, but every member of the control group
received another instrument that measures the same construct but with much
more sensitivity. All control subjects, who scored on this scale, were
excluded. Therefore it might be possible to assume that every member of the
control group was 0 on this psychopathology score.

Thank you for the opportunity to clarify this.

Ingo

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

[Richard Ristow]

On Wed, 30 Apr 2008 14:17:47 -0400, Gene Maguin <[hidden email]> asked:

>>Are you saying that the psychopathology score every member of the
>>control group was 0.0?

At 01:40 PM 5/2/2008, Ingo Frommann responded:

>Every member of the control group received another instrument that
>measures the same construct but with much more sensitivity. All
>control subjects, who scored on this scale, were excluded. Therefore
>it might be possible to assume that every member of the control
>group was 0 on this psychopathology score.

That means the psychopathology measure is heavily confounded with
anything else that differs between the control and subject groups.
Selecting only controls without psychopathology makes this worse.

You also have,

>   /WSFACTOR = domain 3 Polynomial

Your within-subjects factor ("neuropsychlogical domains: memory,
executive, attention") is really three measures, a multivariate
analysis, rather than a repeated-measures analysis of the same
quantity. There are others here who can tell better than I what you
should be doing, but I'm pretty sure you shouldn't be doing
'/WSFACTOR'. Certainly not with polynomial contrast, which makes no
sense with a categorical factor.

Has you had a statistical review of the study design?

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