SPSSX Discussion

variance structures in MIXED

Classic

List

Threaded

8 messages Options

Kylie

variance structures in MIXED

Hi all,

I am trying to replicate some analyses that were originally done in SAS. I
do not have the code that was used, but do have the analysts writeup of the
analysis method and results. I am trying to confirm if the variance
structure they used is also possible in SPSS.

It is a doubly repeated measures design - multiple time points (23) per
experimental condition (3) per subject. The model includes fixed effects for
time (entered as a factor), condition and the time*condition interaction.
The description of the previous analysis includes this statement regarding
the variances:

"Due to different variances in the outcome across time, variances were
estimated separately for baseline measures (the first three time points),
and then for all measures after."

From the output provided I can see that a variable was created ('period')
that equalled 1 for the observations at the three baseline timepoints, and 2
for all the others. The 'Covariance Parameter Estimates' table from SAS
includes 5 parameters:

Cov Parm Subject Group

Intercept id period 1

Intercept id period 2

time id period 1

time id period 2

AR(1) id

Error

Can anyone suggest how to define this kind of model in SPSS' MIXED
procedure?

Thanks,

Kylie.

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

Reutter, Alex

Re: variance structures in MIXED

What's the structure of your data? Something like:

subject time condition period response
1 1 1 1 ###
1 1 2 1 ###
1 1 3 1 ###
1 2 1 2 ###
1 2 2 2 ###
1 2 3 2 ###
1 3 1 2 ###
...
1 23 1 2 ###
1 23 2 2 ###
1 23 3 2 ###
2 1 1 1 ###
2 1 2 1 ###
2 1 3 1 ###
...

Or perhaps time and condition reversed? Precisely how were the experimental conditions administered (i.e., were subjects assigned a condition and then followed for 23 time periods, then assigned another condition and followed for 23 time periods, then the third condition and followed for the final 23 time periods?)

Alex

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Kylie Lange
Sent: Thursday, July 10, 2008 8:42 PM
To: [hidden email]
Subject: variance structures in MIXED

Hi all,

I am trying to replicate some analyses that were originally done in SAS. I
do not have the code that was used, but do have the analysts writeup of the
analysis method and results. I am trying to confirm if the variance
structure they used is also possible in SPSS.

It is a doubly repeated measures design - multiple time points (23) per
experimental condition (3) per subject. The model includes fixed effects for
time (entered as a factor), condition and the time*condition interaction.
The description of the previous analysis includes this statement regarding
the variances:

"Due to different variances in the outcome across time, variances were
estimated separately for baseline measures (the first three time points),
and then for all measures after."

From the output provided I can see that a variable was created ('period')
that equalled 1 for the observations at the three baseline timepoints, and 2
for all the others. The 'Covariance Parameter Estimates' table from SAS
includes 5 parameters:

Cov Parm Subject Group

Intercept id period 1

Intercept id period 2

time id period 1

time id period 2

AR(1) id

Error

Can anyone suggest how to define this kind of model in SPSS' MIXED
procedure?

Thanks,

Kylie.

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

Kylie

Re: variance structures in MIXED

Hi Alex,

Just to clarify in case I wasn't clear - I'm quite comfortable with running
these kinds of repeated measures mixed models in SPSS. It's just the
additional 'period' difference in variances that I haven't been able to work
out how to define.

Re your specific questions, the data file is indeed as you illustrate
however all rows with time=1,2 or 3 will have period=1. Subjects underwent
the four conditions in random order. During each condition 23 repeated
measures were taken.

Thanks again,
Kylie.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Reutter, Alex
Sent: Friday, 11 July 2008 12:54 pm
To: [hidden email]
Subject: Re: variance structures in MIXED

What's the structure of your data? Something like:

subject time condition period response
1 1 1 1 ###
1 1 2 1 ###
1 1 3 1 ###
1 2 1 2 ###
1 2 2 2 ###
1 2 3 2 ###
1 3 1 2 ###
...
1 23 1 2 ###
1 23 2 2 ###
1 23 3 2 ###
2 1 1 1 ###
2 1 2 1 ###
2 1 3 1 ###
...

Or perhaps time and condition reversed? Precisely how were the experimental
conditions administered (i.e., were subjects assigned a condition and then
followed for 23 time periods, then assigned another condition and followed
for 23 time periods, then the third condition and followed for the final 23
time periods?)

Alex

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Kylie Lange
Sent: Thursday, July 10, 2008 8:42 PM
To: [hidden email]
Subject: variance structures in MIXED

Hi all,

I am trying to replicate some analyses that were originally done in SAS. I
do not have the code that was used, but do have the analysts writeup of the
analysis method and results. I am trying to confirm if the variance
structure they used is also possible in SPSS.

It is a doubly repeated measures design - multiple time points (23) per
experimental condition (3) per subject. The model includes fixed effects for
time (entered as a factor), condition and the time*condition interaction.
The description of the previous analysis includes this statement regarding
the variances:

"Due to different variances in the outcome across time, variances were
estimated separately for baseline measures (the first three time points),
and then for all measures after."

From the output provided I can see that a variable was created ('period')
that equalled 1 for the observations at the three baseline timepoints, and 2
for all the others. The 'Covariance Parameter Estimates' table from SAS
includes 5 parameters:

Cov Parm Subject Group

Intercept id period 1

Intercept id period 2

time id period 1

time id period 2

AR(1) id

Error

Can anyone suggest how to define this kind of model in SPSS' MIXED
procedure?

Thanks,

Kylie.

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

Reutter, Alex

Re: variance structures in MIXED

Just to be sure I understand what kind of covariance setup you've got: are you fitting an AR(1) random effects structure with time and intercept as effects, and and period as a grouping variable to define heterogenous variance estimates for time and intercept for each level of period? MIXED does not explicity support grouping variables, and at the moment I can't think of a way to trick it into doing so.

Alex

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Kylie Lange
Sent: Friday, July 11, 2008 12:17 AM
To: [hidden email]
Subject: Re: variance structures in MIXED

Hi Alex,

Just to clarify in case I wasn't clear - I'm quite comfortable with running
these kinds of repeated measures mixed models in SPSS. It's just the
additional 'period' difference in variances that I haven't been able to work
out how to define.

Re your specific questions, the data file is indeed as you illustrate
however all rows with time=1,2 or 3 will have period=1. Subjects underwent
the four conditions in random order. During each condition 23 repeated
measures were taken.

Thanks again,
Kylie.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Reutter, Alex
Sent: Friday, 11 July 2008 12:54 pm
To: [hidden email]
Subject: Re: variance structures in MIXED

What's the structure of your data? Something like:

subject time condition period response
1 1 1 1 ###
1 1 2 1 ###
1 1 3 1 ###
1 2 1 2 ###
1 2 2 2 ###
1 2 3 2 ###
1 3 1 2 ###
...
1 23 1 2 ###
1 23 2 2 ###
1 23 3 2 ###
2 1 1 1 ###
2 1 2 1 ###
2 1 3 1 ###
...

Or perhaps time and condition reversed? Precisely how were the experimental
conditions administered (i.e., were subjects assigned a condition and then
followed for 23 time periods, then assigned another condition and followed
for 23 time periods, then the third condition and followed for the final 23
time periods?)

Alex

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Kylie Lange
Sent: Thursday, July 10, 2008 8:42 PM
To: [hidden email]
Subject: variance structures in MIXED

Hi all,

I am trying to replicate some analyses that were originally done in SAS. I
do not have the code that was used, but do have the analysts writeup of the
analysis method and results. I am trying to confirm if the variance
structure they used is also possible in SPSS.

It is a doubly repeated measures design - multiple time points (23) per
experimental condition (3) per subject. The model includes fixed effects for
time (entered as a factor), condition and the time*condition interaction.
The description of the previous analysis includes this statement regarding
the variances:

"Due to different variances in the outcome across time, variances were
estimated separately for baseline measures (the first three time points),
and then for all measures after."

From the output provided I can see that a variable was created ('period')
that equalled 1 for the observations at the three baseline timepoints, and 2
for all the others. The 'Covariance Parameter Estimates' table from SAS
includes 5 parameters:

Cov Parm Subject Group

Intercept id period 1

Intercept id period 2

time id period 1

time id period 2

AR(1) id

Error

Can anyone suggest how to define this kind of model in SPSS' MIXED
procedure?

Thanks,

Kylie.

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

Zdaniuk, Bozena-2

proportionality of odds assumption test

Hello, how can I test the proportionality of odds assumption in the Multinomial logistic regression?
I found something called SCORE TEST but I don't know how to request it in SPSS. Thanks so much.
Bozena

Bozena Zdaniuk, Ph.D.
University of Pittsburgh
UCSUR, 6th Fl.
121 University Place
Pittsburgh, PA 15260
Ph.: 412-624-5736
Fax: 412-624-4810
Email: [hidden email]

=====================
To manage your subscription to SPSSX-L, send a message to
[hidden email] (not to SPSSX-L), with no body text except the
command. To leave the list, send the command
SIGNOFF SPSSX-L
For a list of commands to manage subscriptions, send the command
INFO REFCARD

Kylie

Re: variance structures in MIXED

In reply to this post by Reutter, Alex

Hi Alex,

At this stage I don't know for sure. I'm just trying to replicate the SAS model
that I have the short description and results for. The results indicate that the
model includes the 5 random components I listed, but is that not enough to
uniquely identify the model that they used?

I am not familiar with SAS' conventions for MIXED output, but it looks to me
that there is are indeed heterogenous time and intercept variance components for
each level of period. Is this the feature that SPSS can not handle?

Thanks again,
Kylie.

Quoting "Reutter, Alex" <[hidden email]>:

> Just to be sure I understand what kind of covariance setup you've got: are
> you fitting an AR(1) random effects structure with time and intercept as
> effects, and and period as a grouping variable to define heterogenous
> variance estimates for time and intercept for each level of period? MIXED
> does not explicity support grouping variables, and at the moment I can't
> think of a way to trick it into doing so.
>
> Alex
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Kylie Lange
> Sent: Friday, July 11, 2008 12:17 AM
> To: [hidden email]
> Subject: Re: variance structures in MIXED
>
> Hi Alex,
>
> Just to clarify in case I wasn't clear - I'm quite comfortable with running
> these kinds of repeated measures mixed models in SPSS. It's just the
> additional 'period' difference in variances that I haven't been able to work
> out how to define.
>
> Re your specific questions, the data file is indeed as you illustrate
> however all rows with time=1,2 or 3 will have period=1. Subjects underwent
> the four conditions in random order. During each condition 23 repeated
> measures were taken.
>
> Thanks again,
> Kylie.
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Reutter, Alex
> Sent: Friday, 11 July 2008 12:54 pm
> To: [hidden email]
> Subject: Re: variance structures in MIXED
>
> What's the structure of your data? Something like:
>
> subject time condition period response
> 1 1 1 1 ###
> 1 1 2 1 ###
> 1 1 3 1 ###
> 1 2 1 2 ###
> 1 2 2 2 ###
> 1 2 3 2 ###
> 1 3 1 2 ###
> ...
> 1 23 1 2 ###
> 1 23 2 2 ###
> 1 23 3 2 ###
> 2 1 1 1 ###
> 2 1 2 1 ###
> 2 1 3 1 ###
> ...
>
> Or perhaps time and condition reversed? Precisely how were the experimental
> conditions administered (i.e., were subjects assigned a condition and then
> followed for 23 time periods, then assigned another condition and followed
> for 23 time periods, then the third condition and followed for the final 23
> time periods?)
>
> Alex
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Kylie Lange
> Sent: Thursday, July 10, 2008 8:42 PM
> To: [hidden email]
> Subject: variance structures in MIXED
>
> Hi all,
>
>
>
> I am trying to replicate some analyses that were originally done in SAS. I
> do not have the code that was used, but do have the analysts writeup of the
> analysis method and results. I am trying to confirm if the variance
> structure they used is also possible in SPSS.
>
>
>
> It is a doubly repeated measures design - multiple time points (23) per
> experimental condition (3) per subject. The model includes fixed effects for
> time (entered as a factor), condition and the time*condition interaction.
> The description of the previous analysis includes this statement regarding
> the variances:
>
>
>
> "Due to different variances in the outcome across time, variances were
> estimated separately for baseline measures (the first three time points),
> and then for all measures after."
>
>
>
> From the output provided I can see that a variable was created ('period')
> that equalled 1 for the observations at the three baseline timepoints, and 2
> for all the others. The 'Covariance Parameter Estimates' table from SAS
> includes 5 parameters:
>
>
>
> Cov Parm Subject Group
>
>
>
> Intercept id period 1
>
> Intercept id period 2
>
> time id period 1
>
> time id period 2
>
> AR(1) id
>
> Error
>
>
>
> Can anyone suggest how to define this kind of model in SPSS' MIXED
> procedure?
>
>
>
> Thanks,
>
> Kylie.
>
> =====================
> 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
>

Reutter, Alex

Re: variance structures in MIXED

Hi Kylie,

I wasn't 100% sure that you were fitting them as random effects (G matrix) and not repeated measures (R matrix), and so wanted to just repeat what I thought you were fitting so there was less chance of misunderstanding. Thanks.

Yes, that's right; SPSS doesn't currently allow you to specify a grouping variable that would fit the heterogenous time and intercept variance components. Looking back at the columns in your original post:

> Cov Parm Subject Group

I should have twigged to that "Group" column sooner.

Alex

-----Original Message-----
From: Kylie Lange [mailto:[hidden email]]
Sent: Wednesday, July 16, 2008 11:59 PM
To: Reutter, Alex
Cc: [hidden email]
Subject: Re: variance structures in MIXED

Hi Alex,

At this stage I don't know for sure. I'm just trying to replicate the SAS model
that I have the short description and results for. The results indicate that the
model includes the 5 random components I listed, but is that not enough to
uniquely identify the model that they used?

I am not familiar with SAS' conventions for MIXED output, but it looks to me
that there is are indeed heterogenous time and intercept variance components for
each level of period. Is this the feature that SPSS can not handle?

Thanks again,
Kylie.

Quoting "Reutter, Alex" <[hidden email]>:

Kylie

Re: variance structures in MIXED

Hi all,

Following on from this thread, I have managed to get a copy of the SAS code
that was used for the original modelling. If anyone is able to confirm
whether this is able to be replicated in SPSS I would appreciate it.

proc mixed data =d1 covtest method = ml;
class id condition time period;
model logmean = time|condition/ outp=check ;
random intercept time / subject=id group=period;
repeated condition*time / subject=id type=ar(1) ;
lsmeans condition*time / pdiff ;
ods output diffs=temp1 ;
ods exclude diffs ;
run;

Thanks,
Kylie.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Reutter, Alex
Sent: Saturday, 19 July 2008 1:13 am
To: [hidden email]
Subject: Re: variance structures in MIXED

Hi Kylie,

I wasn't 100% sure that you were fitting them as random effects (G matrix)
and not repeated measures (R matrix), and so wanted to just repeat what I
thought you were fitting so there was less chance of misunderstanding.
Thanks.

Yes, that's right; SPSS doesn't currently allow you to specify a grouping
variable that would fit the heterogenous time and intercept variance
components. Looking back at the columns in your original post:

> Cov Parm Subject Group

I should have twigged to that "Group" column sooner.

Alex

-----Original Message-----
From: Kylie Lange [mailto:[hidden email]]
Sent: Wednesday, July 16, 2008 11:59 PM
To: Reutter, Alex
Cc: [hidden email]
Subject: Re: variance structures in MIXED

Hi Alex,

At this stage I don't know for sure. I'm just trying to replicate the SAS
model
that I have the short description and results for. The results indicate that
the
model includes the 5 random components I listed, but is that not enough to
uniquely identify the model that they used?

I am not familiar with SAS' conventions for MIXED output, but it looks to me
that there is are indeed heterogenous time and intercept variance components
for
each level of period. Is this the feature that SPSS can not handle?

Thanks again,
Kylie.

Quoting "Reutter, Alex" <[hidden email]>:

running
> these kinds of repeated measures mixed models in SPSS. It's just the
> additional 'period' difference in variances that I haven't been able to
work

> out how to define.
>
> Re your specific questions, the data file is indeed as you illustrate
> however all rows with time=1,2 or 3 will have period=1. Subjects underwent
> the four conditions in random order. During each condition 23 repeated
> measures were taken.
>
> Thanks again,
> Kylie.
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Reutter, Alex
> Sent: Friday, 11 July 2008 12:54 pm
> To: [hidden email]
> Subject: Re: variance structures in MIXED
>
> What's the structure of your data? Something like:
>
> subject time condition period response
> 1 1 1 1 ###
> 1 1 2 1 ###
> 1 1 3 1 ###
> 1 2 1 2 ###
> 1 2 2 2 ###
> 1 2 3 2 ###
> 1 3 1 2 ###
> ...
> 1 23 1 2 ###
> 1 23 2 2 ###
> 1 23 3 2 ###
> 2 1 1 1 ###
> 2 1 2 1 ###
> 2 1 3 1 ###
> ...
>
> Or perhaps time and condition reversed? Precisely how were the

experimental
> conditions administered (i.e., were subjects assigned a condition and then
> followed for 23 time periods, then assigned another condition and followed
> for 23 time periods, then the third condition and followed for the final
23

> time periods?)
>
> Alex
>
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
> Kylie Lange
> Sent: Thursday, July 10, 2008 8:42 PM
> To: [hidden email]
> Subject: variance structures in MIXED
>
> Hi all,
>
>
>
> I am trying to replicate some analyses that were originally done in SAS. I
> do not have the code that was used, but do have the analysts writeup of

the
> analysis method and results. I am trying to confirm if the variance
> structure they used is also possible in SPSS.
>
>
>
> It is a doubly repeated measures design - multiple time points (23) per
> experimental condition (3) per subject. The model includes fixed effects
for

> time (entered as a factor), condition and the time*condition interaction.
> The description of the previous analysis includes this statement regarding
> the variances:
>
>
>
> "Due to different variances in the outcome across time, variances were
> estimated separately for baseline measures (the first three time points),
> and then for all measures after."
>
>
>
> From the output provided I can see that a variable was created ('period')
> that equalled 1 for the observations at the three baseline timepoints, and

> for all the others. The 'Covariance Parameter Estimates' table from SAS
> includes 5 parameters:
>
>
>
> Cov Parm Subject Group
>
>
>
> Intercept id period 1
>
> Intercept id period 2
>
> time id period 1
>
> time id period 2
>
> AR(1) id
>
> Error
>
>
>
> Can anyone suggest how to define this kind of model in SPSS' MIXED
> procedure?
>
>
>
> Thanks,
>
> Kylie.
>

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