2 ways MANOVA? or what? and how...

[antonio ml]

Please help.
    I've got the following scenario
I've got a sample of 7 patients (I'll get more patients  in the future,
maybe about 12 to 15 pats), and I process the sample in 5 deferents
ways, like 4 treatments .
I measure 3 different things (interval level, 0 to 100) for each
sample*treatment (3 independent variables: v1, v2,...)

something like this:
trat;pat;v1;v2;v3;v4
1;1;2;4;0;2
1;2;4;10;4;5
1;3;5,4;12,4;1,4;11
1;4;4;3;2;0
1;5;1;1;0;0
1;6;4;3;0;4
1;7;0;3;0;0
2;1;2,9;2,2;5,1;14,5
2;2;1;1;23,3;16,5
2;3;23,7;17,5;11,34;23,6
2;4;3,6;5,4;0,9;21,4
2;5;0;0,7;3,4;0,7
2;6;;;;
2;7;0;1,27;1,27;2,5
3;1;2,7;2,7;10,8;13,5
3;2;46,8;1,8;30,3;3,7
3;3;7,4;9,26;20,4;20,4
3;4;15;7;28;7
3;5;21,3;8,73;6,67;7,77
3;6;19,3;11,7;3,88;3,88
3;7;1,9;5,7;0,9;3,8

¿What analysis should I apply if I want to study relation among
different treatments and also among the variables (v1,...v4) for each
treatment?
I'm not rely interested in the factor 'patient' but I guess I have to
count it some how.
At th beginning I thought of a 2 ways MANOVA, one of the factor as
random effcet (patient) and the other (treatment) as a fixed effect.

Any help will be very much appreciated, I'm lost.

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

Antonio,

There are several ways to analyze your data. I'd assume that trat is not a
random factor.

One way is manova:
GLM v1 to v4 by trat.

Another way is mixed (which requires a data restructure vis varstocases). V
is the restructured v1-v4 and measure is the measure id 1-4.
Mixed v by measure trat/fixed=measure trat measure*trat/
   repeated=measure | subject(pat) covtype(un).

The key question, I think, is the structure of the residual covariance
matrix. If the residual matrix has compount symmetry (covtype=cs) the
analysis is like a repeated measures. The other extreme is an unstructured
matrix (covtype=un). The flexibility of mixed is that different covariance
matrix structures can be tested, which, if a more structured type (fewer
degrees of freedom) can be justified, a more powerful test results.

Gene Maguin


>>I've got a sample of 7 patients (I'll get more patients  in the future,
maybe about 12 to 15 pats), and I process the sample in 5 deferents
ways, like 4 treatments .
I measure 3 different things (interval level, 0 to 100) for each
sample*treatment (3 independent variables: v1, v2,...)

something like this:
trat;pat;v1;v2;v3;v4
1;1;2;4;0;2
1;2;4;10;4;5
1;3;5,4;12,4;1,4;11
1;4;4;3;2;0
1;5;1;1;0;0
1;6;4;3;0;4
1;7;0;3;0;0
2;1;2,9;2,2;5,1;14,5
2;2;1;1;23,3;16,5
2;3;23,7;17,5;11,34;23,6
2;4;3,6;5,4;0,9;21,4
2;5;0;0,7;3,4;0,7
2;6;;;;
2;7;0;1,27;1,27;2,5
3;1;2,7;2,7;10,8;13,5
3;2;46,8;1,8;30,3;3,7
3;3;7,4;9,26;20,4;20,4
3;4;15;7;28;7
3;5;21,3;8,73;6,67;7,77
3;6;19,3;11,7;3,88;3,88
3;7;1,9;5,7;0,9;3,8

¿What analysis should I apply if I want to study relation among
different treatments and also among the variables (v1,...v4) for each
treatment?
I'm not rely interested in the factor 'patient' but I guess I have to
count it some how.
At th beginning I thought of a 2 ways MANOVA, one of the factor as
random effcet (patient) and the other (treatment) as a fixed effect.

Any help will be very much appreciated, I'm lost.

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

[antonio ml]

Thank gene

'Trat' is not  a randon factor, this are all the treatments we want to
test, there's no more.

when you say to restructure data ¿is like this?:
trat;pat;measure ;V
1;1;2;v1
1;2;4;v1
1;3;5,4;v1
1;4;4;v1
1;5;1;v1
1;6;4;v1
1;7;0;v1
2;1;2,9;v1
2;2;1;v1
2;3;23,7;v1
2;4;3,6;v1
2;5;0;v1
2;6;;v1
2;7;0;v1
3;1;2,7;v1
3;2;46,8;v1
3;3;7,4;v1
3;4;15;v1
3;5;21,3;v1
3;6;19,3;v1
3;7;1,9;v1
1;1;4;v2
1;2;10;v2
1;3;12,4;v2
1;4;3;v2
1;5;1;v2
1;6;3;v2
1;7;3;v2
2;1;2,2;v2
2;2;1;v2
2;3;17,5;v2
2;4;5,4;v2
2;5;0,7;v2
2;6;;v2
2;7;1,27;v2
3;1;2,7;v2
3;2;1,8;v2
3;3;9,26;v2
3;4;7;v2
3;5;8,73;v2
3;6;11,7;v2
3;7;5,7;v2
1;1;0;v3
1;2;4;v3
1;3;1,4;v3
1;4;2;v3
1;5;0;v3
1;6;0;v3
1;7;0;v3
2;1;5,1;v3
2;2;23,3;v3
2;3;11,34;v3
2;4;0,9;v3
2;5;3,4;v3
2;6;;v3
2;7;1,27;v3
3;1;10,8;v3
3;2;30,3;v3
3;3;20,4;v3
3;4;28;v3
3;5;6,67;v3
3;6;3,88;v3
3;7;0,9;v3
1;1;2;v4
1;2;5;v4
1;3;11;v4
1;4;0;v4
1;5;0;v4
1;6;4;v4
1;7;0;v4
2;1;14,5;v4
2;2;16,5;v4
2;3;23,6;v4
2;4;21,4;v4
2;5;0,7;v4
2;6;;v4
2;7;2,5;v4
3;1;13,5;v4
3;2;3,7;v4
3;3;20,4;v4
3;4;7;v4
3;5;7,77;v4
3;6;3,88;v4
3;7;3,8;v4

then..I don't get you well....

¿
Mixed  measure2 by v trat2
/fixed=v trat2 v*trat2
/repeated=v | subject(pat2) covtype(un).
?

Thank you



Gene Maguin escribió:

> Antonio,
>
> There are several ways to analyze your data. I'd assume that trat is not a
> random factor.
>
> One way is manova:
> GLM v1 to v4 by trat.
>
> Another way is mixed (which requires a data restructure vis varstocases). V
> is the restructured v1-v4 and measure is the measure id 1-4.
> Mixed v by measure trat/fixed=measure trat measure*trat/
>    repeated=measure | subject(pat) covtype(un).
>
> The key question, I think, is the structure of the residual covariance
> matrix. If the residual matrix has compount symmetry (covtype=cs) the
> analysis is like a repeated measures. The other extreme is an unstructured
> matrix (covtype=un). The flexibility of mixed is that different covariance
> matrix structures can be tested, which, if a more structured type (fewer
> degrees of freedom) can be justified, a more powerful test results.
>
> Gene Maguin
>
>
>
>>> I've got a sample of 7 patients (I'll get more patients  in the future,
>>>
> maybe about 12 to 15 pats), and I process the sample in 5 deferents
> ways, like 4 treatments .
> I measure 3 different things (interval level, 0 to 100) for each
> sample*treatment (3 independent variables: v1, v2,...)
>
> something like this:
> trat;pat;v1;v2;v3;v4
> 1;1;2;4;0;2
> 1;2;4;10;4;5
> 1;3;5,4;12,4;1,4;11
> 1;4;4;3;2;0
> 1;5;1;1;0;0
> 1;6;4;3;0;4
> 1;7;0;3;0;0
> 2;1;2,9;2,2;5,1;14,5
> 2;2;1;1;23,3;16,5
> 2;3;23,7;17,5;11,34;23,6
> 2;4;3,6;5,4;0,9;21,4
> 2;5;0;0,7;3,4;0,7
> 2;6;;;;
> 2;7;0;1,27;1,27;2,5
> 3;1;2,7;2,7;10,8;13,5
> 3;2;46,8;1,8;30,3;3,7
> 3;3;7,4;9,26;20,4;20,4
> 3;4;15;7;28;7
> 3;5;21,3;8,73;6,67;7,77
> 3;6;19,3;11,7;3,88;3,88
> 3;7;1,9;5,7;0,9;3,8
>
> ¿What analysis should I apply if I want to study relation among
> different treatments and also among the variables (v1,...v4) for each
> treatment?
> I'm not rely interested in the factor 'patient' but I guess I have to
> count it some how.
> At th beginning I thought of a 2 ways MANOVA, one of the factor as
> random effcet (patient) and the other (treatment) as a fixed effect.
>
> Any help will be very much appreciated, I'm lost.
>
> --
>
> Antonio.
> Gracias, Efharisto poli, Grazie, Thank you, Tak,
> Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
> Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
> Go raibh maith agat, Arigato, Dishklenle,
> Dankon, Tashakkur
>
> =====================
> 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
>
>

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

Antonio,

The command would be

Varstocases /make v from v1 to v4/index=measure.

The resulting data structure would be
Trat;pat;v;measure
1;1;2;1
1;1;4;2
1;1;0;3
1;1;2;4
1;2;4;1
1;2;10;2
1;2;4;3
1;2;5;4

By the way, I just was looking more carefully at your data and there's
something odd about records like this one. At first I thought you had data
with semicolon delimited fields but, if so, what are those commas doing in
there.

1;3;5,4;12,4;1,4;11

Gene



'Trat' is not  a randon factor, this are all the treatments we want to
test, there's no more.

when you say to restructure data ¿is like this?:
trat;pat;measure ;V
1;1;2;v1
1;2;4;v1
1;3;5,4;v1
1;4;4;v1
1;5;1;v1
1;6;4;v1
1;7;0;v1
2;1;2,9;v1
2;2;1;v1
2;3;23,7;v1
2;4;3,6;v1
2;5;0;v1
2;6;;v1
2;7;0;v1
3;1;2,7;v1
3;2;46,8;v1
3;3;7,4;v1
3;4;15;v1
3;5;21,3;v1
3;6;19,3;v1
3;7;1,9;v1
1;1;4;v2
1;2;10;v2
1;3;12,4;v2
1;4;3;v2
1;5;1;v2
1;6;3;v2
1;7;3;v2
2;1;2,2;v2
2;2;1;v2
2;3;17,5;v2
2;4;5,4;v2
2;5;0,7;v2
2;6;;v2
2;7;1,27;v2
3;1;2,7;v2
3;2;1,8;v2
3;3;9,26;v2
3;4;7;v2
3;5;8,73;v2
3;6;11,7;v2
3;7;5,7;v2
1;1;0;v3
1;2;4;v3
1;3;1,4;v3
1;4;2;v3
1;5;0;v3
1;6;0;v3
1;7;0;v3
2;1;5,1;v3
2;2;23,3;v3
2;3;11,34;v3
2;4;0,9;v3
2;5;3,4;v3
2;6;;v3
2;7;1,27;v3
3;1;10,8;v3
3;2;30,3;v3
3;3;20,4;v3
3;4;28;v3
3;5;6,67;v3
3;6;3,88;v3
3;7;0,9;v3
1;1;2;v4
1;2;5;v4
1;3;11;v4
1;4;0;v4
1;5;0;v4
1;6;4;v4
1;7;0;v4
2;1;14,5;v4
2;2;16,5;v4
2;3;23,6;v4
2;4;21,4;v4
2;5;0,7;v4
2;6;;v4
2;7;2,5;v4
3;1;13,5;v4
3;2;3,7;v4
3;3;20,4;v4
3;4;7;v4
3;5;7,77;v4
3;6;3,88;v4
3;7;3,8;v4

then..I don't get you well....

¿
Mixed  measure2 by v trat2
/fixed=v trat2 v*trat2
/repeated=v | subject(pat2) covtype(un).
?

Thank you



Gene Maguin escribió:
> Antonio,
>
> There are several ways to analyze your data. I'd assume that trat is not a
> random factor.
>
> One way is manova:
> GLM v1 to v4 by trat.
>
> Another way is mixed (which requires a data restructure vis varstocases).
V

> is the restructured v1-v4 and measure is the measure id 1-4.
> Mixed v by measure trat/fixed=measure trat measure*trat/
>    repeated=measure | subject(pat) covtype(un).
>
> The key question, I think, is the structure of the residual covariance
> matrix. If the residual matrix has compount symmetry (covtype=cs) the
> analysis is like a repeated measures. The other extreme is an unstructured
> matrix (covtype=un). The flexibility of mixed is that different covariance
> matrix structures can be tested, which, if a more structured type (fewer
> degrees of freedom) can be justified, a more powerful test results.
>
> Gene Maguin
>
>
>
>>> I've got a sample of 7 patients (I'll get more patients  in the future,
>>>
> maybe about 12 to 15 pats), and I process the sample in 5 deferents
> ways, like 4 treatments .
> I measure 3 different things (interval level, 0 to 100) for each
> sample*treatment (3 independent variables: v1, v2,...)
>
> something like this:
> trat;pat;v1;v2;v3;v4
> 1;1;2;4;0;2
> 1;2;4;10;4;5
> 1;3;5,4;12,4;1,4;11
> 1;4;4;3;2;0
> 1;5;1;1;0;0
> 1;6;4;3;0;4
> 1;7;0;3;0;0
> 2;1;2,9;2,2;5,1;14,5
> 2;2;1;1;23,3;16,5
> 2;3;23,7;17,5;11,34;23,6
> 2;4;3,6;5,4;0,9;21,4
> 2;5;0;0,7;3,4;0,7
> 2;6;;;;
> 2;7;0;1,27;1,27;2,5
> 3;1;2,7;2,7;10,8;13,5
> 3;2;46,8;1,8;30,3;3,7
> 3;3;7,4;9,26;20,4;20,4
> 3;4;15;7;28;7
> 3;5;21,3;8,73;6,67;7,77
> 3;6;19,3;11,7;3,88;3,88
> 3;7;1,9;5,7;0,9;3,8
>
> ¿What analysis should I apply if I want to study relation among
> different treatments and also among the variables (v1,...v4) for each
> treatment?
> I'm not rely interested in the factor 'patient' but I guess I have to
> count it some how.
> At th beginning I thought of a 2 ways MANOVA, one of the factor as
> random effcet (patient) and the other (treatment) as a fixed effect.
>
> Any help will be very much appreciated, I'm lost.
>
> --
>
> Antonio.
> Gracias, Efharisto poli, Grazie, Thank you, Tak,
> Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
> Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
> Go raibh maith agat, Arigato, Dishklenle,
> Dankon, Tashakkur
>
> =====================
> 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
>
>

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

[antonio ml]

oh no!!, the commas are the decimal instead of the dot!! sorry that's the Spanish way.
So the variables v1, to v4 are continuous with two decimals positions
so the line :

1;3;5,4;12,4;1,4;11

is like this:
v1=5.4
v2=12.4
v3=1.4
v4=11.0
 sorry for the mess!!!

Thank you a lot. I'll try the mixed way.

Antonio




Gene Maguin escribió:

Antonio,

The command would be

Varstocases /make v from v1 to v4/index=measure.

The resulting data structure would be
Trat;pat;v;measure
1;1;2;1
1;1;4;2
1;1;0;3
1;1;2;4
1;2;4;1
1;2;10;2
1;2;4;3
1;2;5;4

By the way, I just was looking more carefully at your data and there's
something odd about records like this one. At first I thought you had data
with semicolon delimited fields but, if so, what are those commas doing in
there.

1;3;5,4;12,4;1,4;11

Gene



'Trat' is not  a randon factor, this are all the treatments we want to
test, there's no more.

when you say to restructure data ¿is like this?:
trat;pat;measure ;V
1;1;2;v1
1;2;4;v1
1;3;5,4;v1
1;4;4;v1
1;5;1;v1
1;6;4;v1
1;7;0;v1
2;1;2,9;v1
2;2;1;v1
2;3;23,7;v1
2;4;3,6;v1
2;5;0;v1
2;6;;v1
2;7;0;v1
3;1;2,7;v1
3;2;46,8;v1
3;3;7,4;v1
3;4;15;v1
3;5;21,3;v1
3;6;19,3;v1
3;7;1,9;v1
1;1;4;v2
1;2;10;v2
1;3;12,4;v2
1;4;3;v2
1;5;1;v2
1;6;3;v2
1;7;3;v2
2;1;2,2;v2
2;2;1;v2
2;3;17,5;v2
2;4;5,4;v2
2;5;0,7;v2
2;6;;v2
2;7;1,27;v2
3;1;2,7;v2
3;2;1,8;v2
3;3;9,26;v2
3;4;7;v2
3;5;8,73;v2
3;6;11,7;v2
3;7;5,7;v2
1;1;0;v3
1;2;4;v3
1;3;1,4;v3
1;4;2;v3
1;5;0;v3
1;6;0;v3
1;7;0;v3
2;1;5,1;v3
2;2;23,3;v3
2;3;11,34;v3
2;4;0,9;v3
2;5;3,4;v3
2;6;;v3
2;7;1,27;v3
3;1;10,8;v3
3;2;30,3;v3
3;3;20,4;v3
3;4;28;v3
3;5;6,67;v3
3;6;3,88;v3
3;7;0,9;v3
1;1;2;v4
1;2;5;v4
1;3;11;v4
1;4;0;v4
1;5;0;v4
1;6;4;v4
1;7;0;v4
2;1;14,5;v4
2;2;16,5;v4
2;3;23,6;v4
2;4;21,4;v4
2;5;0,7;v4
2;6;;v4
2;7;2,5;v4
3;1;13,5;v4
3;2;3,7;v4
3;3;20,4;v4
3;4;7;v4
3;5;7,77;v4
3;6;3,88;v4
3;7;3,8;v4

then..I don't get you well....

¿
Mixed  measure2 by v trat2
/fixed=v trat2 v*trat2
/repeated=v | subject(pat2) covtype(un).
?

Thank you



Gene Maguin escribió:
  

Antonio,

There are several ways to analyze your data. I'd assume that trat is not a
random factor.

One way is manova:
GLM v1 to v4 by trat.

Another way is mixed (which requires a data restructure vis varstocases).
    

V
  

is the restructured v1-v4 and measure is the measure id 1-4.
Mixed v by measure trat/fixed=measure trat measure*trat/
   repeated=measure | subject(pat) covtype(un).

The key question, I think, is the structure of the residual covariance
matrix. If the residual matrix has compount symmetry (covtype=cs) the
analysis is like a repeated measures. The other extreme is an unstructured
matrix (covtype=un). The flexibility of mixed is that different covariance
matrix structures can be tested, which, if a more structured type (fewer
degrees of freedom) can be justified, a more powerful test results.

Gene Maguin



    

I've got a sample of 7 patients (I'll get more patients  in the future,

        

maybe about 12 to 15 pats), and I process the sample in 5 deferents
ways, like 4 treatments .
I measure 3 different things (interval level, 0 to 100) for each
sample*treatment (3 independent variables: v1, v2,...)

something like this:
trat;pat;v1;v2;v3;v4
1;1;2;4;0;2
1;2;4;10;4;5
1;3;5,4;12,4;1,4;11
1;4;4;3;2;0
1;5;1;1;0;0
1;6;4;3;0;4
1;7;0;3;0;0
2;1;2,9;2,2;5,1;14,5
2;2;1;1;23,3;16,5
2;3;23,7;17,5;11,34;23,6
2;4;3,6;5,4;0,9;21,4
2;5;0;0,7;3,4;0,7
2;6;;;;
2;7;0;1,27;1,27;2,5
3;1;2,7;2,7;10,8;13,5
3;2;46,8;1,8;30,3;3,7
3;3;7,4;9,26;20,4;20,4
3;4;15;7;28;7
3;5;21,3;8,73;6,67;7,77
3;6;19,3;11,7;3,88;3,88
3;7;1,9;5,7;0,9;3,8

¿What analysis should I apply if I want to study relation among
different treatments and also among the variables (v1,...v4) for each
treatment?
I'm not rely interested in the factor 'patient' but I guess I have to
count it some how.
At th beginning I thought of a 2 ways MANOVA, one of the factor as
random effcet (patient) and the other (treatment) as a fixed effect.

Any help will be very much appreciated, I'm lost.

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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


    

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

  

[Bruce Weaver]

Gene Maguin wrote

Antonio,

There are several ways to analyze your data. I'd assume that trat is not a
random factor.

One way is manova:
GLM v1 to v4 by trat.

Another way is mixed (which requires a data restructure vis varstocases). V
is the restructured v1-v4 and measure is the measure id 1-4.
Mixed v by measure trat/fixed=measure trat measure*trat/
   repeated=measure | subject(pat) covtype(un).

The key question, I think, is the structure of the residual covariance
matrix. If the residual matrix has compount symmetry (covtype=cs) the
analysis is like a repeated measures. The other extreme is an unstructured
matrix (covtype=un). The flexibility of mixed is that different covariance
matrix structures can be tested, which, if a more structured type (fewer
degrees of freedom) can be justified, a more powerful test results.

Gene Maguin

The OP said:  "I measure 3 different things (interval level, 0 to 100) for each sample*treatment (3 independent variables: v1, v2,...)".  [Somewhere the 3 turned into 4, it seems.]

It's not clear to me whether v1 to v4 represent 4 different variables (e.g., height, weight, resting heart rate, and diastolic blood pressure), or the same variable measured on 4 occasions (or under 4 different conditions, etc).  Gene, does your MIXED syntax assume it's the same variable measured repeatedly?  Thanks for clarifying.

Bruce

[antonio ml]

Yes ok.
v1, v2, v3 and v4 represent 4 different variables (yes 3 turned into 4,
sorry..)

Antonio.




Bruce Weaver escribió:

> Gene Maguin wrote:
>
>> Antonio,
>>
>> There are several ways to analyze your data. I'd assume that trat is not a
>> random factor.
>>
>> One way is manova:
>> GLM v1 to v4 by trat.
>>
>> Another way is mixed (which requires a data restructure vis varstocases).
>> V
>> is the restructured v1-v4 and measure is the measure id 1-4.
>> Mixed v by measure trat/fixed=measure trat measure*trat/
>>    repeated=measure | subject(pat) covtype(un).
>>
>> The key question, I think, is the structure of the residual covariance
>> matrix. If the residual matrix has compount symmetry (covtype=cs) the
>> analysis is like a repeated measures. The other extreme is an unstructured
>> matrix (covtype=un). The flexibility of mixed is that different covariance
>> matrix structures can be tested, which, if a more structured type (fewer
>> degrees of freedom) can be justified, a more powerful test results.
>>
>> Gene Maguin
>>
>>
>>
>
> The OP said:  "I measure 3 different things (interval level, 0 to 100) for
> each sample*treatment (3 independent variables: v1, v2,...)".  [Somewhere
> the 3 turned into 4, it seems.]
>
> It's not clear to me whether v1 to v4 represent 4 different variables (e.g.,
> height, weight, resting heart rate, and diastolic blood pressure), or the
> same variable measured on 4 occasions (or under 4 different conditions,
> etc).  Gene, does your MIXED syntax assume it's the same variable measured
> repeatedly?  Thanks for clarifying.
>
> Bruce
>
>
>
> -----
> --
> Bruce Weaver
> [hidden email]
> http://sites.google.com/a/lakeheadu.ca/bweaver/
> "When all else fails, RTFM."
>
> NOTE:  My Hotmail account is not monitored regularly.
> To send me an e-mail, please use the address shown above.
> --
> View this message in context: http://old.nabble.com/2-ways-MANOVA--or-what--and-how...-tp26791541p26800744.html
> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>
> =====================
> 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
>
>

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

Bruce,

Yes, it does. I assumed that v1 to v4 were the same variable purely on the
fact that Antonio said they all had the same range. Although it turns out
that they are the same, they could have the same range and not be the same.

But suppose they were different variables, with or without different ranges.
Although GLM probably would be what I'd use, can the analysis be done in
mixed. My thought is yes because the residual covariance can be represented
as UNstructured and a measure or treatment by measure interaction would
indicate significant differences in means. If the ranges were different, the
means probably should differ and the measure main effect would be
uninteresting. But a treatment by measure interaction would not necessarily
be uninteresting.

Gene Maguin


>>It's not clear to me whether v1 to v4 represent 4 different variables
(e.g.,
height, weight, resting heart rate, and diastolic blood pressure), or the
same variable measured on 4 occasions (or under 4 different conditions,
etc).  Gene, does your MIXED syntax assume it's the same variable measured
repeatedly?  Thanks for clarifying.

Bruce

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

[antonio ml]

Sure I didn't explain myself well.. sorry. And sorry for long time I took to answer but my whole family had pass the A flu this Christmas!

I'll try to explain my problem again 'cos I still don't get it.


I only have 7 patients (we'll may get some more but not much more), and we measure 9 different variables (independents things like height, weight,..) under 5 treatments (5 different drugs). It's like in 5 different conditions (treatments) we measure the same 9 things to the same 7 patients, what could look like 5 repleted measures (?). The problem is like we only have this 7 patients (!) and, because of the nature of the research is very hard and expensive to find more patients with this characteristics, so we use the same 7 for the 5 experiments and we measure the same things (variables v1 to v9) every time. This will guide our  future research.

The data look like this: (tab separated, decimal separator a dot '.')

caseid    trat    pat    var1    var2    var3    var4    var5    var6    var7    var8    var9
1    1    1    2    4    0    2    0    2    23    15    52
2    1    2    4    10    4    5    0    2    35    5    43
3    1    3    5.4    12.4    1.4    11    0    0    2    0    67.8
4    1    4    4    3    2    0    4    4    10    34    39
5    1    5    1    1    0    0    0    1    5    19    73
6    1    6    4    3    0    4    1    5    14    10    59
7    1    7    0    3    0    0    0    2    5    26    64
8    2    1    2.9    2.2    5.1    14.5    0    2.2    16.6    5    31.17
9    2    2    1    1    23.3    16.5    1.9    1    9.7    1.9    42.7
10    2    3    23.7    17.5    11.34    23.6    0.7    0    0    0    23.16
11    2    4    3.6    5.4    0.9    21.4    1.8    7.1    22.3    18.75    18.75
12    2    5    0    0.7    3.4    0.7    9.6    6.2    4.8    44.5    69.9
13    2    6                                   
14    2    7    0    1.27    1.27    2.5    0.6    7.6    4.43    31.6    50.73
15    3    1    2.7    2.7    10.8    13.5    0    3.6    15.3    2.7    48.7
16    3    2    46.8    1.8    30.3    3.7    0    0    0.9    0    16.5
17    3    3    7.4    9.26    20.4    20.4    0    1.8    11.1    0    29.64
18    3    4    15    7    28    7    20    2    6    3    12
19    3    5    21.3    8.73    6.67    7.77    3.87    0.97    2.9    5.82    41.97
20    3    6    19.3    11.7    3.88    3.88    1    1    6.8    5.9    46.54
21    3    7    1.9    5.7    0.9    3.8    4.7    5.7    7.5    28.3    41.5
22    4    1    4.2    10.86    0    0    0.84    0    0.84    5.06    78.2
23    4    2    31.1    37.9    0    0    0    0    0    0    31
24    4    3    14.8    37.3    0    5    0    0    3    0    34.48
25    4    4    3    20    2    19    1    11    6    6    32
26    4    5    13    45    0    0    0    6    0    0    36
27    4    6    22.1    17.7    0    0.88    0.88    0    0    1.77    56.67
28    4    7    0    4.5    0    1.8    0    0.9    2.7    9.9    80.2
29    5    1    1.74    0.87    13.9    24.35    1.74    0.87    22.6    0.87    33.06
30    5    2    35.13    5.15    12    4.3    14.5    0.85    1.7    1.7    24.67
31    5    3    12.7    4.9    14.6    25.2    0    1    3.9    0    37.7
32    5    4    25.9    6.48    12.95    6.5    19.45    4.6    1.85    0.9    21.37
33    5    5    24.3    16.5    8.7    1.9    13.6    0.97    1.9    2.87    29.26
34    5    6    18.1    12.4    0    3.8    2.86    3.8    5.7    2.86    50.48
35    5    7    0    2.7    0    0.9    0    8    10.7    27.7    50

What we'd like to analyze is differences between treatments for the 9 variables (var1,..vae9), but I guess we have to take in count the patient variable (dependencies). We'd like to conclude things like "treatment 'x' seems to get higher scores in the variables var1, var 2,.... and treatment 'y'  computes higher in var 8 and 9 than the rest of treatments" for example.


Gene, Thank you a lot for your help and comments they are really useful for me, especially all about mixed designs I think there is the clue.
and happy new year!
 


Gene Maguin escribió:

Bruce,

Yes, it does. I assumed that v1 to v4 were the same variable purely on the
fact that Antonio said they all had the same range. Although it turns out
that they are the same, they could have the same range and not be the same.

But suppose they were different variables, with or without different ranges.
Although GLM probably would be what I'd use, can the analysis be done in
mixed. My thought is yes because the residual covariance can be represented
as UNstructured and a measure or treatment by measure interaction would
indicate significant differences in means. If the ranges were different, the
means probably should differ and the measure main effect would be
uninteresting. But a treatment by measure interaction would not necessarily
be uninteresting.

Gene Maguin


  

It's not clear to me whether v1 to v4 represent 4 different variables
      

(e.g.,
height, weight, resting heart rate, and diastolic blood pressure), or the
same variable measured on 4 occasions (or under 4 different conditions,
etc).  Gene, does your MIXED syntax assume it's the same variable measured
repeatedly?  Thanks for clarifying.

Bruce

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

Antonio,

Unfortunately I will not be able to respond in great detail right now, but the answer to the general question as to whether or not MIXED can handle a multivariate normal response is yes. The dependent variables do not have to measure the same construct or have the same range. I have not read the details of your data and research question carefully to state if MIXED is your best choice. If time permits, and you are interested, I can take a closer look at your situation and respond later.

Regardless, I will try to write back to this forum with an example of how to run a multivariate analysis in MIXED soon.

Ryan

amllistas wrote

Sure I didn't explain myself well.. sorry. And sorry for long time I
took to answer but my whole family had pass the A flu this Christmas!

I'll try to explain my problem again 'cos I still don't get it.


I only have 7 patients (we'll may get some more but not much more), and
we measure 9 different variables (independents things like height,
weight,..) under 5 treatments (5 different drugs). It's like in 5
different conditions (treatments) we measure the same 9 things to the
same 7 patients, what could look like 5 repleted measures (?). The
problem is like we only have this 7 patients (!) and, because of the
nature of the research is very hard and expensive to find more patients
with this characteristics, so we use the same 7 for the 5 experiments
and we measure the same things (variables v1 to v9) every time. This
will guide our  future research.

The data look like this: (tab separated, decimal separator a dot '.')

caseid    trat    pat    var1    var2    var3    var4    var5    var6
var7    var8    var9
1    1    1    2    4    0    2    0    2    23    15    52
2    1    2    4    10    4    5    0    2    35    5    43
3    1    3    5.4    12.4    1.4    11    0    0    2    0    67.8
4    1    4    4    3    2    0    4    4    10    34    39
5    1    5    1    1    0    0    0    1    5    19    73
6    1    6    4    3    0    4    1    5    14    10    59
7    1    7    0    3    0    0    0    2    5    26    64
8    2    1    2.9    2.2    5.1    14.5    0    2.2    16.6    5    31.17
9    2    2    1    1    23.3    16.5    1.9    1    9.7    1.9    42.7
10    2    3    23.7    17.5    11.34    23.6    0.7    0    0    0    23.16
11    2    4    3.6    5.4    0.9    21.4    1.8    7.1    22.3
18.75    18.75
12    2    5    0    0.7    3.4    0.7    9.6    6.2    4.8    44.5    69.9
13    2    6
14    2    7    0    1.27    1.27    2.5    0.6    7.6    4.43
31.6    50.73
15    3    1    2.7    2.7    10.8    13.5    0    3.6    15.3    2.7
48.7
16    3    2    46.8    1.8    30.3    3.7    0    0    0.9    0    16.5
17    3    3    7.4    9.26    20.4    20.4    0    1.8    11.1    0
29.64
18    3    4    15    7    28    7    20    2    6    3    12
19    3    5    21.3    8.73    6.67    7.77    3.87    0.97    2.9
5.82    41.97
20    3    6    19.3    11.7    3.88    3.88    1    1    6.8    5.9
46.54
21    3    7    1.9    5.7    0.9    3.8    4.7    5.7    7.5    28.3
41.5
22    4    1    4.2    10.86    0    0    0.84    0    0.84    5.06    78.2
23    4    2    31.1    37.9    0    0    0    0    0    0    31
24    4    3    14.8    37.3    0    5    0    0    3    0    34.48
25    4    4    3    20    2    19    1    11    6    6    32
26    4    5    13    45    0    0    0    6    0    0    36
27    4    6    22.1    17.7    0    0.88    0.88    0    0    1.77    56.67
28    4    7    0    4.5    0    1.8    0    0.9    2.7    9.9    80.2
29    5    1    1.74    0.87    13.9    24.35    1.74    0.87    22.6
0.87    33.06
30    5    2    35.13    5.15    12    4.3    14.5    0.85    1.7
1.7    24.67
31    5    3    12.7    4.9    14.6    25.2    0    1    3.9    0    37.7
32    5    4    25.9    6.48    12.95    6.5    19.45    4.6    1.85
0.9    21.37
33    5    5    24.3    16.5    8.7    1.9    13.6    0.97    1.9
2.87    29.26
34    5    6    18.1    12.4    0    3.8    2.86    3.8    5.7
2.86    50.48
35    5    7    0    2.7    0    0.9    0    8    10.7    27.7    50

What we'd like to analyze is differences between treatments for the 9
variables (var1,..vae9), but I guess we have to take in count the
patient variable (dependencies). We'd like to conclude things like
"/treatment 'x' seems to get higher scores in the variables var1, var
2,.... and treatment 'y'  computes higher in var 8 and 9 than the rest
of treatments/" for example.


Gene, Thank you a lot for your help and comments they are really useful
for me, especially all about mixed designs I think there is the clue.
and happy new year!


--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur




Gene Maguin escribió:
> Bruce,
>
> Yes, it does. I assumed that v1 to v4 were the same variable purely on the
> fact that Antonio said they all had the same range. Although it turns out
> that they are the same, they could have the same range and not be the same.
>
> But suppose they were different variables, with or without different ranges.
> Although GLM probably would be what I'd use, can the analysis be done in
> mixed. My thought is yes because the residual covariance can be represented
> as UNstructured and a measure or treatment by measure interaction would
> indicate significant differences in means. If the ranges were different, the
> means probably should differ and the measure main effect would be
> uninteresting. But a treatment by measure interaction would not necessarily
> be uninteresting.
>
> Gene Maguin
>
>
>
>>> It's not clear to me whether v1 to v4 represent 4 different variables
>>>
> (e.g.,
> height, weight, resting heart rate, and diastolic blood pressure), or the
> same variable measured on 4 occasions (or under 4 different conditions,
> etc).  Gene, does your MIXED syntax assume it's the same variable measured
> repeatedly?  Thanks for clarifying.
>
> Bruce
>
> =====================
> To manage your subscription to SPSSX-L, send a message to
> [e-mail] (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
>
>

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

[MaxJasper]

Running:

 

MIXED
v9 BY treatment
/CRITERIA = CIN(95) MXITER(1000) MXSTEP(50) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
PCONVERGE(0.000001, ABSOLUTE)
/FIXED = treatment | SSTYPE(3)
/METHOD = REML
/PRINT = SOLUTION TESTCOV
/RANDOM INTERCEPT treatment | SUBJECT(patient) COVTYPE(CSR)
/EMMEANS = TABLES(OVERALL) .

Results:

Type III Tests of Fixed Effects(a)

Source	Numerator df	Denominator df	F	Sig.
Intercept	1	6.030	90.409	.000
treatment	4	23.115	4.379	.009

a Dependent Variable: v9.

 

Estimates of Fixed Effects(b)

Parameter	Estimate	Std. Error	df	t	Sig.	95% Confidence Interval
						Upper Bound	Lower Bound
Intercept	35.220000	6.177033	17.262	5.702	.000	22.202668	48.237332
[treatment=1]	21.608571	6.641639	23.035	3.253	.003	7.870464	35.346678
[treatment=2]	5.330439	6.962174	23.247	.766	.452	-9.063451	19.724329
[treatment=3]	-1.384286	6.641639	23.035	-.208	.837	-15.122393	12.353821
[treatment=4]	14.572857	6.641639	23.035	2.194	.039	.834750	28.310964
[treatment=5]	0(a)	0	.	.	.	.	.

a This parameter is set to zero because it is redundant.

b Dependent Variable: v9.

 

Estimates of Covariance Parameters(a)

Parameter			Estimate		Std. Error		Wald Z		Sig.		95% Confidence Interval
											Lower Bound		Upper Bound
Residual			56.369642		233566.456629		.000		1.000		.000000		.
Intercept + treatment [subject = patient]	CSR diagonal	101.690200		175174.842062		.001		1.000		.000000		.
	CSR rho	.036090		636.381380		.000		1.000		-1.000000		1.000000

a Dependent Variable: v9.

 

Variable V9 Conclusion:

Taking into account variations between patients, treatment=1 & 4 have significat (p<0.05) effect on patients relative to treatment=5, and that there is no sig different effect betwen treatment=2,3 and treatement=5.

Above procedure can be repeated for V1.....V9.

 

If V1...V9 are assumed to be dependent, then a Canonical Correlation analysis can be performed using MANOVA (not available in menus).

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Antonio ML
Sent: Monday, January 04, 2010 4:27
To: [hidden email]
Subject: Mixed design? 2 ways MANOVA? or what? and how...

Sure I didn't explain myself well.. sorry. And sorry for long time I took to answer but my whole family had pass the A flu this Christmas!

I'll try to explain my problem again 'cos I still don't get it.


I only have 7 patients (we'll may get some more but not much more), and we measure 9 different variables (independents things like height, weight,..) under 5 treatments (5 different drugs). It's like in 5 different conditions (treatments) we measure the same 9 things to the same 7 patients, what could look like 5 repleted measures (?). The problem is like we only have this 7 patients (!) and, because of the nature of the research is very hard and expensive to find more patients with this characteristics, so we use the same 7 for the 5 experiments and we measure the same things (variables v1 to v9) every time. This will guide our  future research.

The data look like this: (tab separated, decimal separator a dot '.')

caseid    trat    pat    var1    var2    var3    var4    var5    var6    var7    var8    var9
1    1    1    2    4    0    2    0    2    23    15    52
2    1    2    4    10    4    5    0    2    35    5    43
3    1    3    5.4    12.4    1.4    11    0    0    2    0    67.8
4    1    4    4    3    2    0    4    4    10    34    39
5    1    5    1    1    0    0    0    1    5    19    73
6    1    6    4    3    0    4    1    5    14    10    59
7    1    7    0    3    0    0    0    2    5    26    64
8    2    1    2.9    2.2    5.1    14.5    0    2.2    16.6    5    31.17
9    2    2    1    1    23.3    16.5    1.9    1    9.7    1.9    42.7
10    2    3    23.7    17.5    11.34    23.6    0.7    0    0    0    23.16
11    2    4    3.6    5.4    0.9    21.4    1.8    7.1    22.3    18.75    18.75
12    2    5    0    0.7    3.4    0.7    9.6    6.2    4.8    44.5    69.9
13    2    6                                   
14    2    7    0    1.27    1.27    2.5    0.6    7.6    4.43    31.6    50.73
15    3    1    2.7    2.7    10.8    13.5    0    3.6    15.3    2.7    48.7
16    3    2    46.8    1.8    30.3    3.7    0    0    0.9    0    16.5
17    3    3    7.4    9.26    20.4    20.4    0    1.8    11.1    0    29.64
18    3    4    15    7    28    7    20    2    6    3    12
19    3    5    21.3    8.73    6.67    7.77    3.87    0.97    2.9    5.82    41.97
20    3    6    19.3    11.7    3.88    3.88    1    1    6.8    5.9    46.54
21    3    7    1.9    5.7    0.9    3.8    4.7    5.7    7.5    28.3    41.5
22    4    1    4.2    10.86    0    0    0.84    0    0.84    5.06    78.2
23    4    2    31.1    37.9    0    0    0    0    0    0    31
24    4    3    14.8    37.3    0    5    0    0    3    0    34.48
25    4    4    3    20    2    19    1    11    6    6    32
26    4    5    13    45    0    0    0    6    0    0    36
27    4    6    22.1    17.7    0    0.88    0.88    0    0    1.77    56.67
28    4    7    0    4.5    0    1.8    0    0.9    2.7    9.9    80.2
29    5    1    1.74    0.87    13.9    24.35    1.74    0.87    22.6    0.87    33.06
30    5    2    35.13    5.15    12    4.3    14.5    0.85    1.7    1.7    24.67
31    5    3    12.7    4.9    14.6    25.2    0    1    3.9    0    37.7
32    5    4    25.9    6.48    12.95    6.5    19.45    4.6    1.85    0.9    21.37
33    5    5    24.3    16.5    8.7    1.9    13.6    0.97    1.9    2.87    29.26
34    5    6    18.1    12.4    0    3.8    2.86    3.8    5.7    2.86    50.48
35    5    7    0    2.7    0    0.9    0    8    10.7    27.7    50

What we'd like to analyze is differences between treatments for the 9 variables (var1,..vae9), but I guess we have to take in count the patient variable (dependencies). We'd like to conclude things like "treatment 'x' seems to get higher scores in the variables var1, var 2,.... and treatment 'y'  computes higher in var 8 and 9 than the rest of treatments" for example.


Gene, Thank you a lot for your help and comments they are really useful for me, especially all about mixed designs I think there is the clue.
and happy new year!
 


Gene Maguin escribió:

Bruce,

Yes, it does. I assumed that v1 to v4 were the same variable purely on the
fact that Antonio said they all had the same range. Although it turns out
that they are the same, they could have the same range and not be the same.

But suppose they were different variables, with or without different ranges.
Although GLM probably would be what I'd use, can the analysis be done in
mixed. My thought is yes because the residual covariance can be represented
as UNstructured and a measure or treatment by measure interaction would
indicate significant differences in means. If the ranges were different, the
means probably should differ and the measure main effect would be
uninteresting. But a treatment by measure interaction would not necessarily
be uninteresting.

Gene Maguin


  

It's not clear to me whether v1 to v4 represent 4 different variables
      

(e.g.,
height, weight, resting heart rate, and diastolic blood pressure), or the
same variable measured on 4 occasions (or under 4 different conditions,
etc).  Gene, does your MIXED syntax assume it's the same variable measured
repeatedly?  Thanks for clarifying.

Bruce

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

  

[antonio ml]

Thank you very much!. But please one more question.
Why you don't use REPEATED for treatment?
and, why the CSR (compound simetry, with corr) structure matrix? instead of VC (variance components)

I mean:
This
  /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .
instead of this:
 /RANDOM INTERCEPT treatment | SUBJECT(patient) COVTYPE(CSR).

So it would be like this:

MIXED
  var1  BY trat
  /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)  SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
  PCONVERGE(0.000001, ABSOLUTE)
  /FIXED = trat  | SSTYPE(3)
  /METHOD = REML
  /PRINT = SOLUTION TESTCOV
  /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .

I'm really don't understand this deeply so please forgive me if the question is silly.
Thank you again


MaxJasper escribió:

Running:

 

MIXED
v9 BY treatment
/CRITERIA = CIN(95) MXITER(1000) MXSTEP(50) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
PCONVERGE(0.000001, ABSOLUTE)
/FIXED = treatment | SSTYPE(3)
/METHOD = REML
/PRINT = SOLUTION TESTCOV
/RANDOM INTERCEPT treatment | SUBJECT(patient) COVTYPE(CSR)
/EMMEANS = TABLES(OVERALL) .

Results:

Type III Tests of Fixed Effects(a)

Source	Numerator df	Denominator df	F	Sig.
Intercept	1	6.030	90.409	.000
treatment	4	23.115	4.379	.009

a Dependent Variable: v9.

 

Estimates of Fixed Effects(b)

Parameter	Estimate	Std. Error	df	t	Sig.	95% Confidence Interval
						Upper Bound	Lower Bound
Intercept	35.220000	6.177033	17.262	5.702	.000	22.202668	48.237332
[treatment=1]	21.608571	6.641639	23.035	3.253	.003	7.870464	35.346678
[treatment=2]	5.330439	6.962174	23.247	.766	.452	-9.063451	19.724329
[treatment=3]	-1.384286	6.641639	23.035	-.208	.837	-15.122393	12.353821
[treatment=4]	14.572857	6.641639	23.035	2.194	.039	.834750	28.310964
[treatment=5]	0(a)	0	.	.	.	.	.

a This parameter is set to zero because it is redundant.

b Dependent Variable: v9.

 

Estimates of Covariance Parameters(a)

Parameter			Estimate		Std. Error		Wald Z		Sig.		95% Confidence Interval
											Lower Bound		Upper Bound
Residual			56.369642		233566.456629		.000		1.000		.000000		.
Intercept + treatment [subject = patient]	CSR diagonal	101.690200		175174.842062		.001		1.000		.000000		.
	CSR rho	.036090		636.381380		.000		1.000		-1.000000		1.000000

a Dependent Variable: v9.

 

Variable V9 Conclusion:

Taking into account variations between patients, treatment=1 & 4 have significat (p<0.05) effect on patients relative to treatment=5, and that there is no sig different effect betwen treatment=2,3 and treatement=5.

Above procedure can be repeated for V1.....V9.

 

If V1...V9 are assumed to be dependent, then a Canonical Correlation analysis can be performed using MANOVA (not available in menus).

-----Original Message-----
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Antonio ML
Sent: Monday, January 04, 2010 4:27
To: [hidden email]
Subject: Mixed design? 2 ways MANOVA? or what? and how...

Sure I didn't explain myself well.. sorry. And sorry for long time I took to answer but my whole family had pass the A flu this Christmas!

I'll try to explain my problem again 'cos I still don't get it.


I only have 7 patients (we'll may get some more but not much more), and we measure 9 different variables (independents things like height, weight,..) under 5 treatments (5 different drugs). It's like in 5 different conditions (treatments) we measure the same 9 things to the same 7 patients, what could look like 5 repleted measures (?). The problem is like we only have this 7 patients (!) and, because of the nature of the research is very hard and expensive to find more patients with this characteristics, so we use the same 7 for the 5 experiments and we measure the same things (variables v1 to v9) every time. This will guide our  future research.

The data look like this: (tab separated, decimal separator a dot '.')

caseid    trat    pat    var1    var2    var3    var4    var5    var6    var7    var8    var9
1    1    1    2    4    0    2    0    2    23    15    52
2    1    2    4    10    4    5    0    2    35    5    43
3    1    3    5.4    12.4    1.4    11    0    0    2    0    67.8
4    1    4    4    3    2    0    4    4    10    34    39
5    1    5    1    1    0    0    0    1    5    19    73
6    1    6    4    3    0    4    1    5    14    10    59
7    1    7    0    3    0    0    0    2    5    26    64
8    2    1    2.9    2.2    5.1    14.5    0    2.2    16.6    5    31.17
9    2    2    1    1    23.3    16.5    1.9    1    9.7    1.9    42.7
10    2    3    23.7    17.5    11.34    23.6    0.7    0    0    0    23.16
11    2    4    3.6    5.4    0.9    21.4    1.8    7.1    22.3    18.75    18.75
12    2    5    0    0.7    3.4    0.7    9.6    6.2    4.8    44.5    69.9
13    2    6                                   
14    2    7    0    1.27    1.27    2.5    0.6    7.6    4.43    31.6    50.73
15    3    1    2.7    2.7    10.8    13.5    0    3.6    15.3    2.7    48.7
16    3    2    46.8    1.8    30.3    3.7    0    0    0.9    0    16.5
17    3    3    7.4    9.26    20.4    20.4    0    1.8    11.1    0    29.64
18    3    4    15    7    28    7    20    2    6    3    12
19    3    5    21.3    8.73    6.67    7.77    3.87    0.97    2.9    5.82    41.97
20    3    6    19.3    11.7    3.88    3.88    1    1    6.8    5.9    46.54
21    3    7    1.9    5.7    0.9    3.8    4.7    5.7    7.5    28.3    41.5
22    4    1    4.2    10.86    0    0    0.84    0    0.84    5.06    78.2
23    4    2    31.1    37.9    0    0    0    0    0    0    31
24    4    3    14.8    37.3    0    5    0    0    3    0    34.48
25    4    4    3    20    2    19    1    11    6    6    32
26    4    5    13    45    0    0    0    6    0    0    36
27    4    6    22.1    17.7    0    0.88    0.88    0    0    1.77    56.67
28    4    7    0    4.5    0    1.8    0    0.9    2.7    9.9    80.2
29    5    1    1.74    0.87    13.9    24.35    1.74    0.87    22.6    0.87    33.06
30    5    2    35.13    5.15    12    4.3    14.5    0.85    1.7    1.7    24.67
31    5    3    12.7    4.9    14.6    25.2    0    1    3.9    0    37.7
32    5    4    25.9    6.48    12.95    6.5    19.45    4.6    1.85    0.9    21.37
33    5    5    24.3    16.5    8.7    1.9    13.6    0.97    1.9    2.87    29.26
34    5    6    18.1    12.4    0    3.8    2.86    3.8    5.7    2.86    50.48
35    5    7    0    2.7    0    0.9    0    8    10.7    27.7    50

What we'd like to analyze is differences between treatments for the 9 variables (var1,..vae9), but I guess we have to take in count the patient variable (dependencies). We'd like to conclude things like "treatment 'x' seems to get higher scores in the variables var1, var 2,.... and treatment 'y'  computes higher in var 8 and 9 than the rest of treatments" for example.


Gene, Thank you a lot for your help and comments they are really useful for me, especially all about mixed designs I think there is the clue.
and happy new year!
 
Gene Maguin escribió:

Bruce,

Yes, it does. I assumed that v1 to v4 were the same variable purely on the
fact that Antonio said they all had the same range. Although it turns out
that they are the same, they could have the same range and not be the same.

But suppose they were different variables, with or without different ranges.
Although GLM probably would be what I'd use, can the analysis be done in
mixed. My thought is yes because the residual covariance can be represented
as UNstructured and a measure or treatment by measure interaction would
indicate significant differences in means. If the ranges were different, the
means probably should differ and the measure main effect would be
uninteresting. But a treatment by measure interaction would not necessarily
be uninteresting.

Gene Maguin


  

It's not clear to me whether v1 to v4 represent 4 different variables
      

[MaxJasper]

Regarding covariance types:

 

Wald test can be quite unreliable for small samples. So, technically, you should run several models by changing covariance type in SPSS syntax and find the lowest goodness of fit model. Models with random slopes vs. without random slopes, should be compared as follows:

 

p = sig.Chisq( 2RLL(model with random slopes) - 2RLL(model without random slopes), df)

 

2RLL = -2RestrictedLogLikelihood

df = # of parameters dropped = # parameters in model 1 - # parameters of model 2.

 

If p > 0.05 then, you can drop slopes without seriously affecting the goodness of fit of the model.

 

Regarding analyzing your model as a Longitudinal study:

 

If you add TREATMENT as REPEATED, remember you cannot add it to the model again as RANDOM effect. In this model both results are the same.

 

So, my model was similar to this one:

MIXED
v9 BY treatment
/CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
PCONVERGE(0.000001, ABSOLUTE)
/FIXED = treatment | SSTYPE(3)
/METHOD = REML
/PRINT = SOLUTION TESTCOV
/REPEATED = treatment | SUBJECT(patient) COVTYPE(CS)
/EMMEANS = TABLES(treatment) COMPARE REFCAT(FIRST) ADJ(BONFERRONI) .

Estimates of Fixed Effects(b)

Parameter	Estimate	Std. Error	df	t	Sig.	95% Confidence Interval
						Upper Bound	Lower Bound
Intercept	35.220000	6.177033	17.262	5.702	.000	22.202668	48.237332
[treatment=1]	21.608571	6.641639	23.035	3.253	.003	7.870464	35.346678
[treatment=2]	5.330439	6.962174	23.247	.766	.452	-9.063451	19.724329
[treatment=3]	-1.384286	6.641639	23.035	-.208	.837	-15.122393	12.353821
[treatment=4]	14.572857	6.641639	23.035	2.194	.039	.834750	28.310964
[treatment=5]	0(a)	0	.	.	.	.	.

a This parameter is set to zero because it is redundant.

b Dependent Variable: v9.

Again, treatments 1,4 have sig (p<0.05) effect on V9 of patients relative to treatment 5. etc.

 

However, in order to completey understand Linear Mixed Models in SPSS, I urge you to see the 8 examples in the following ref.:

SPSS 15.0 Advanced Statistical Procedures Companion

<A onclick="NewWindow(this.href,'name','560','675','yes');return false;" href="http://norusis.com/img/cover_aspc_v15big.gif">

SPSS 15.0 Advanced Statistical Procedures Companion by Marija J. Norušis ISBN 9780132447126 Prentice Hall Pages 380 To order the book from Prentice Hall, click here. A statistical procedure is not like a sausage: you want to know its contents; you want to know the types of questions it can be used to answer and the types of data for which it is appropriate. The goal of the SPSS 15.0 Advanced Statistical Procedures Companion is to provide you with background information and examples for statistical procedures in the SPSS Advanced and Regression Models modules. It aims to make it less likely that you will succumb to the siren song of melodic statistical procedure names and unleash a disastrous assault on a mutely suffering data file.

From: Antonio ML [mailto:[hidden email]]
To: [hidden email]
Subject: Re: Mixed design? 2 ways MANOVA? or what? and how...

Thank you very much!. But please one more question.
Why you don't use REPEATED for treatment?
and, why the CSR (compound simetry, with corr) structure matrix? instead of VC (variance components)

I mean:
This
  /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .
instead of this:
 /RANDOM INTERCEPT treatment | SUBJECT(patient) COVTYPE(CSR).

So it would be like this:

MIXED
  var1  BY trat
  /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)  SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
  PCONVERGE(0.000001, ABSOLUTE)
  /FIXED = trat  | SSTYPE(3)
  /METHOD = REML
  /PRINT = SOLUTION TESTCOV
  /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .

I'm really don't understand this deeply so please forgive me if the question is silly.
Thank you again

[antonio ml]

I'll take a look to that book.
Thank you very much.

MaxJasper escribió:

> Regarding covariance types:
> Wald test can be quite unreliable for small samples. So, technically,
> you should run several models by changing covariance type in SPSS
> syntax and find the lowest goodness of fit model. Models with random
> slopes vs. without random slopes, should be compared as follows:
> p = sig.Chisq( 2RLL(model with random slopes) - 2RLL(model without
> random slopes), df)
> 2RLL = -2RestrictedLogLikelihood
> df = # of parameters dropped = # parameters in model 1 - # parameters
> of model 2.
> If p > 0.05 then, you can drop slopes without seriously affecting the
> goodness of fit of the model.
> Regarding analyzing your model as a Longitudinal study:
> If you add TREATMENT as REPEATED, remember you cannot add it to the
> model again as RANDOM effect. In this model both results are the same.
> So, my model was similar to this one:
>
> MIXED
> v9 BY treatment
> /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)
> SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
> PCONVERGE(0.000001, ABSOLUTE)
> /FIXED = treatment | SSTYPE(3)
> /METHOD = REML
> /PRINT = SOLUTION TESTCOV
> /REPEATED = treatment | SUBJECT(patient) COVTYPE(CS)
> /EMMEANS = TABLES(treatment) COMPARE REFCAT(FIRST) ADJ(BONFERRONI) .
>
> **
>
> *Estimates of Fixed Effects(b)*
>
> Parameter
>
>
>
> Estimate
>
>
>
> Std. Error
>
>
>
> df
>
>
>
> t
>
>
>
> Sig.
>
>
>
> 95% Confidence Interval
>
> Upper Bound
>
>
>
> Lower Bound
>
> Intercept
>
>
>
> 35.220000
>
>
>
> 6.177033
>
>
>
> 17.262
>
>
>
> 5.702
>
>
>
> .000
>
>
>
> 22.202668
>
>
>
> 48.237332
>
> [treatment=1]
>
>
>
> 21.608571
>
>
>
> 6.641639
>
>
>
> 23.035
>
>
>
> 3.253
>
>
>
> .003
>
>
>
> 7.870464
>
>
>
> 35.346678
>
> [treatment=2]
>
>
>
> 5.330439
>
>
>
> 6.962174
>
>
>
> 23.247
>
>
>
> .766
>
>
>
> .452
>
>
>
> -9.063451
>
>
>
> 19.724329
>
> [treatment=3]
>
>
>
> -1.384286
>
>
>
> 6.641639
>
>
>
> 23.035
>
>
>
> -.208
>
>
>
> .837
>
>
>
> -15.122393
>
>
>
> 12.353821
>
> [treatment=4]
>
>
>
> 14.572857
>
>
>
> 6.641639
>
>
>
> 23.035
>
>
>
> 2.194
>
>
>
> .039
>
>
>
> .834750
>
>
>
> 28.310964
>
> [treatment=5]
>
>
>
> 0(a)
>
>
>
> 0
>
>
>
> .
>
>
>
> .
>
>
>
> .
>
>
>
> .
>
>
>
> .
>
> a This parameter is set to zero because it is redundant.
>
> b Dependent Variable: v9.
>
> Again, treatments 1,4 have sig (p<0.05) effect on V9 of patients
> relative to treatment 5. etc.
>
> However, in order to completey understand Linear Mixed Models in SPSS,
> I urge you to see the 8 examples in the following ref.:
> SPSS 15.0 Advanced Statistical Procedures Companion
>
> Book Cover <http://norusis.com/img/cover_aspc_v15big.gif>     /*SPSS 15.0
> Advanced Statistical Procedures Companion*/
> by Marija J. Norušis
> ISBN 9780132447126
> Prentice Hall
> Pages 380
>
> *To order the book from Prentice Hall, click here.
> <http://vig.prenhall.com/catalog/academic/product/0,1144,0132447126,00.html>*
>
>
> A statistical procedure is not like a sausage: you want to know its
> contents; you want to know the types of questions it can be used to
> answer and the types of data for which it is appropriate. The goal of
> the SPSS 15.0 Advanced Statistical Procedures Companion is to provide
> you with background information and examples for statistical
> procedures in the SPSS Advanced and Regression Models modules. It aims
> to make it less likely that you will succumb to the siren song of
> melodic statistical procedure names and unleash a disastrous assault
> on a mutely suffering data file.
>
>     *From:* Antonio ML [mailto:[hidden email]]
>     *To:* [hidden email]
>     *Subject:* Re: Mixed design? 2 ways MANOVA? or what? and how...
>
>     Thank you very much!. But please one more question.
>     Why you don't use REPEATED for treatment?
>     and, why the CSR (compound simetry, with corr) structure matrix?
>     instead of VC (variance components)
>
>     I mean:
>     This
>     /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .
>     instead of this:
>     /RANDOM INTERCEPT treatment | SUBJECT(patient) COVTYPE(CSR).
>
>     So it would be like this:
>
>     MIXED
>     var1 BY trat
>     /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)
>     SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
>     PCONVERGE(0.000001, ABSOLUTE)
>     /FIXED = trat | SSTYPE(3)
>     /METHOD = REML
>     /PRINT = SOLUTION TESTCOV
>     /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .
>
>     I'm really don't understand this deeply so please forgive me if
>     the question is silly.
>     *Thank you *again
>
>
>>>
>>

--

Antonio.
Gracias, Efharisto poli, Grazie, Thank you, Tak,
Shukran, Hvala, Blagodaria, Bedankt, Danke schön,
Shukriya, Tesekuler, Merci, Spasiba, Obrigato,
Go raibh maith agat, Arigato, Dishklenle,
Dankon, Tashakkur

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

[antonio ml]

Hello List.
    Now I'm wondering how to compute the POWER of a mixed model test....
The think is how to evaluate how good is it or if a would need to sample more (ufff!!!! very expensive and tedious, but if with 4 o 5 more i increase the power a lot i can consider it).

Thank you all in advance.

Finally , my syntax looks is like this:

MIXED
  medida  BY pat trat
  /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)   SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)   PCONVERGE(0.000001, ABSOLUTE)
  /FIXED = trat  | SSTYPE(3)
  /METHOD = REML
  /PRINT = SOLUTION TESTCOV
 /RANDOM INTERCEPT trat  | SUBJECT(pat) COVTYPE(VC)
  /EMMEANS = TABLES(trat) COMPARE ADJ(BONFERRONI)  .

and my data like this (tab separated, and a dot as decimal):

trat    pat    medida
1    1    4
1    2    10
1    3    12.4
1    4    3
1    5    1
1    6    3
1    7    3
2    1    2.2
2    2    1
2    3    17.5
2    4    5.4
2    5    0.7
2    7    1.27
3    1    2.7
3    2    1.8
3    3    9.26
3    4    7
3    5    8.73
3    6    11.7
3    7    5.7
4    1    10.86
4    2    37.9
4    3    37.3
4    4    20
4    5    45
4    6    17.7
4    7    4.5
5    1    0.87
5    2    5.15
5    3    4.9
5    4    6.48
5    5    16.5
5    6    12.4
5    7    2.7


MaxJasper escribió:

Regarding covariance types:

 

Wald test can be quite unreliable for small samples. So, technically, you should run several models by changing covariance type in SPSS syntax and find the lowest goodness of fit model. Models with random slopes vs. without random slopes, should be compared as follows:

 

p = sig.Chisq( 2RLL(model with random slopes) - 2RLL(model without random slopes), df)

 

2RLL = -2RestrictedLogLikelihood

df = # of parameters dropped = # parameters in model 1 - # parameters of model 2.

 

If p > 0.05 then, you can drop slopes without seriously affecting the goodness of fit of the model.

 

Regarding analyzing your model as a Longitudinal study:

 

If you add TREATMENT as REPEATED, remember you cannot add it to the model again as RANDOM effect. In this model both results are the same.

 

So, my model was similar to this one:

MIXED
v9 BY treatment
/CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
PCONVERGE(0.000001, ABSOLUTE)
/FIXED = treatment | SSTYPE(3)
/METHOD = REML
/PRINT = SOLUTION TESTCOV
/REPEATED = treatment | SUBJECT(patient) COVTYPE(CS)
/EMMEANS = TABLES(treatment) COMPARE REFCAT(FIRST) ADJ(BONFERRONI) .

Estimates of Fixed Effects(b)

Parameter	Estimate	Std. Error	df	t	Sig.	95% Confidence Interval
						Upper Bound	Lower Bound
Intercept	35.220000	6.177033	17.262	5.702	.000	22.202668	48.237332
[treatment=1]	21.608571	6.641639	23.035	3.253	.003	7.870464	35.346678
[treatment=2]	5.330439	6.962174	23.247	.766	.452	-9.063451	19.724329
[treatment=3]	-1.384286	6.641639	23.035	-.208	.837	-15.122393	12.353821
[treatment=4]	14.572857	6.641639	23.035	2.194	.039	.834750	28.310964
[treatment=5]	0(a)	0	.	.	.	.	.

a This parameter is set to zero because it is redundant.

b Dependent Variable: v9.

Again, treatments 1,4 have sig (p<0.05) effect on V9 of patients relative to treatment 5. etc.

 

However, in order to completey understand Linear Mixed Models in SPSS, I urge you to see the 8 examples in the following ref.:

SPSS 15.0 Advanced Statistical Procedures Companion

<a moz-do-not-send="true" onclick="NewWindow(this.href,'name','560','675','yes');return false;" href="http://norusis.com/img/cover_aspc_v15big.gif">

SPSS 15.0 Advanced Statistical Procedures Companion by Marija J. Norušis ISBN 9780132447126 Prentice Hall Pages 380 To order the book from Prentice Hall, click here. A statistical procedure is not like a sausage: you want to know its contents; you want to know the types of questions it can be used to answer and the types of data for which it is appropriate. The goal of the SPSS 15.0 Advanced Statistical Procedures Companion is to provide you with background information and examples for statistical procedures in the SPSS Advanced and Regression Models modules. It aims to make it less likely that you will succumb to the siren song of melodic statistical procedure names and unleash a disastrous assault on a mutely suffering data file.

From: Antonio ML [[hidden email]]
To: [hidden email]
Subject: Re: Mixed design? 2 ways MANOVA? or what? and how...

Thank you very much!. But please one more question.
Why you don't use REPEATED for treatment?
and, why the CSR (compound simetry, with corr) structure matrix? instead of VC (variance components)

I mean:
This
  /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .
instead of this:
 /RANDOM INTERCEPT treatment | SUBJECT(patient) COVTYPE(CSR).

So it would be like this:

MIXED
  var1  BY trat
  /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)  SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
  PCONVERGE(0.000001, ABSOLUTE)
  /FIXED = trat  | SSTYPE(3)
  /METHOD = REML
  /PRINT = SOLUTION TESTCOV
  /REPEATED = trat | SUBJECT(pat) COVTYPE(DIAG) .

I'm really don't understand this deeply so please forgive me if the question is silly.
Thank you again

[Bruce Weaver]

amllistas wrote

Hello List.
    Now I'm wondering how to compute the POWER of a mixed model test....
The think is how to evaluate how good is it or if a would need to sample
more (ufff!!!! very expensive and tedious, but if with 4 o 5 more i
increase the power a lot i can consider it).

Thank you all in advance.

Finally , my syntax looks is like this:

MIXED
  medida  BY pat trat
  /CRITERIA = CIN(95) MXITER(100) MXSTEP(5) SCORING(1)
SINGULAR(0.000000000001) HCONVERGE(0, ABSOLUTE) LCONVERGE(0, ABSOLUTE)
PCONVERGE(0.000001, ABSOLUTE)
  /FIXED = trat  | SSTYPE(3)
  /METHOD = REML
  /PRINT = SOLUTION TESTCOV
 /RANDOM INTERCEPT trat  | SUBJECT(pat) COVTYPE(VC)
  /EMMEANS = TABLES(trat) COMPARE ADJ(BONFERRONI)  .

and my data like this (tab separated, and a dot as decimal):

trat    pat    medida
1    1    4
1    2    10

--- snip ---

If you are asking how to compute post-hoc power, I would advise you not to bother, as post-hoc power is just a transformation of the p-value.  I.e., if p < .05, you will find that post-hoc power was adequate, and if p > .05, you'll find it was inadequate.  

If you are asking how to compute an a priori sample size estimate for a multilevel (linear) regression model, take a look at Chapter 8 in Jos Twisk's book, Applied Multilevel Analysis.  It boils down to computing a standard sample size estimate, and then applying a "correction factor", which takes into account the multilevel nature of  the data.  However, there are two possible correction factors.  Twisk describes one as "liberal", and the other as "conservative".  He discusses the relative merits of the two, and the situations in which they might be appropriate.  He concludes:

"In practice, the 'conservative' sample size procedure is mostly used. However, most researchers do not reealise that this procedure leads to an overestimation of the required sample size when the randomisation is performed on the patient level."  (Twisk 2006, pp. 130-131)

p.s. - In your MIXED syntax, I believe you can omit "pat" from "medida  BY pat trat".