SPSSX Discussion

Mixed Model Missing Data

Classic

List

Threaded

9 messages Options

Salbod

Mixed Model Missing Data

I am new to the mixed model procedure. The output is kind of strange. (That is what happens from years of looking at ANOVAs.) I would appreciate any recommendations that would help me understand SPSS mixed model output? I am trying to model a1c measures (glucose) across five time intervals (appointments): baseline, 3-6mo, 6-9mo, 9-12mo and 12-18mo. A treatment was administered after a baseline was obtained. I want to assess whether the treatment was effective. Missing data is a problem. Out of 118 people, only 37 made all their appointments. Thirty-five percent missed only one appointment and so on. The missing data issue steered me to the mixed model procedure. I’m treating time as a fixed variable, people as a random variable, and use AR(1) as my covariance type. With repeated measures, I would follow the analysis with contrasts. Is this available within mixed model?

Any articles or book chapter recommendations will be most welcome.

TIA

Steve (Pace University, NYC)

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

Re: Mixed Model Missing Data

One very good recommendation is Singer and Willett, Applied longitudinal data analysis. There are other good resources. Another resource is http://stats.idre.ucla.edu/ Another is the MLwin website at University of Bristol (http://www.bristol.ac.uk/cmm/software/mlwin/

Can we assume your model is multilevel model and is something like this?

Mixed glucose with time bcov wcov/fixed=time cov1 cov2/Random intercept | subject(pid) covtype(??).

Where bcov is a person level covariate and wcov is a time point (appointment) covariate.

If this were your model, the covtype would be ‘id’, not ‘ar’.

Or, are you doing a repeated measures model?

Re: contrasts. Emmeans is available and so is Test.

Gene Maguin

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Salbod, Mr. Stephen
Sent: Monday, April 03, 2017 9:43 AM
To: [hidden email]
Subject: Mixed Model Missing Data

Any articles or book chapter recommendations will be most welcome.

TIA

Steve (Pace University, NYC)

Salbod

Re: Mixed Model Missing Data

Hi Gene,

Thanks for getting back to me. I ordered the Singer et al book.

The reason I wanted to work with mixed models is that it might handle my missing data problem. Out of 118 I end up with 37, if I do repeated or Manova.

I have not got the complete dataset yet. There is a comparison group that will soon be available. But, right now I want to understand the Mixed Model procedure.

I’m stuck in ANOVA set. I don’t fully understand what Mixed will offer me nor all the decisions I need to make. Your book recommendation will help there.

Here is syntax from my model.

MIXED glucose BY time

/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,

ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)

/FIXED=time | SSTYPE(3)

/METHOD=REML

/PRINT=DESCRIPTIVES SOLUTION TESTCOV

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

/EMMEANS=TABLES(time) COMPARE REFCAT(FIRST) ADJ(SIDAK).

I selected AR(1) because I assumed the covariance would change as function time. I’m working with SPSS 24, by id you mean identity?

I see you suggested time as a covariate. I put time as fixed because the times were not equal spaced. Is there an advantage to using time as a covariate?

Thank you the push.

--Steve

From: Maguin, Eugene [mailto:[hidden email]]
Sent: Monday, April 03, 2017 10:13 AM
To: Salbod, Mr. Stephen <[hidden email]>; [hidden email]
Subject: RE: Mixed Model Missing Data

Can we assume your model is multilevel model and is something like this?

Mixed glucose with time bcov wcov/fixed=time cov1 cov2/Random intercept | subject(pid) covtype(??).

Where bcov is a person level covariate and wcov is a time point (appointment) covariate.

If this were your model, the covtype would be ‘id’, not ‘ar’.

Or, are you doing a repeated measures model?

Re: contrasts. Emmeans is available and so is Test.

Gene Maguin

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Salbod, Mr. Stephen
Sent: Monday, April 03, 2017 9:43 AM
To: [hidden email]
Subject: Mixed Model Missing Data

Any articles or book chapter recommendations will be most welcome.

TIA

Steve (Pace University, NYC)

Maguin, Eugene

Re: Mixed Model Missing Data

What mixed (and multilevel models, generally) offers is maximal use of available data. Missing data is a serious problem for you, it sounds like. I’d recommend that you look at work by Craig Enders. He’s written a book but he also has several articles that you can find through psychinfo or google. And there are other authors as well.

If you make time a covariate and enter the time since baseline, differences in time from baseline to assessment are accommodated because you are estimating a regression model with actual time as a predictor. Your model treats time as a categorical contrast with assessment 5 as the reference.

Gene Maguin

From: Salbod, Mr. Stephen [mailto:[hidden email]]
Sent: Monday, April 03, 2017 3:42 PM
To: Maguin, Eugene <[hidden email]>; [hidden email]
Subject: RE: Mixed Model Missing Data

Hi Gene,

Thanks for getting back to me. I ordered the Singer et al book.

The reason I wanted to work with mixed models is that it might handle my missing data problem. Out of 118 I end up with 37, if I do repeated or Manova.

I have not got the complete dataset yet. There is a comparison group that will soon be available. But, right now I want to understand the Mixed Model procedure.

I’m stuck in ANOVA set. I don’t fully understand what Mixed will offer me nor all the decisions I need to make. Your book recommendation will help there.

Here is syntax from my model.

MIXED glucose BY time

/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,

ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)

/FIXED=time | SSTYPE(3)

/METHOD=REML

/PRINT=DESCRIPTIVES SOLUTION TESTCOV

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

/EMMEANS=TABLES(time) COMPARE REFCAT(FIRST) ADJ(SIDAK).

I selected AR(1) because I assumed the covariance would change as function time. I’m working with SPSS 24, by id you mean identity?

I see you suggested time as a covariate. I put time as fixed because the times were not equal spaced. Is there an advantage to using time as a covariate?

Thank you the push.

--Steve

From: Maguin, Eugene [[hidden email]]
Sent: Monday, April 03, 2017 10:13 AM
To: Salbod, Mr. Stephen <[hidden email]>; [hidden email]
Subject: RE: Mixed Model Missing Data

Can we assume your model is multilevel model and is something like this?

Mixed glucose with time bcov wcov/fixed=time cov1 cov2/Random intercept | subject(pid) covtype(??).

Where bcov is a person level covariate and wcov is a time point (appointment) covariate.

If this were your model, the covtype would be ‘id’, not ‘ar’.

Or, are you doing a repeated measures model?

Re: contrasts. Emmeans is available and so is Test.

Gene Maguin

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Salbod, Mr. Stephen
Sent: Monday, April 03, 2017 9:43 AM
To: [hidden email]
Subject: Mixed Model Missing Data

Any articles or book chapter recommendations will be most welcome.

TIA

Steve (Pace University, NYC)

Bruce Weaver

Re: Mixed Model Missing Data

Administrator

In reply to this post by Maguin, Eugene

Maguin, Eugene wrote

One very good recommendation is Singer and Willett, Applied longitudinal data analysis. There are other good resources. Another resource is http://stats.idre.ucla.edu/ Another is the MLwin website at University of Bristol (http://www.bristol.ac.uk/cmm/software/mlwin/

Can we assume your model is multilevel model and is something like this?
Mixed glucose with time bcov wcov/fixed=time cov1 cov2/Random intercept | subject(pid) covtype(??).
Where bcov is a person level covariate and wcov is a time point (appointment) covariate.
If this were your model, the covtype would be ‘id’, not ‘ar’.

Or, are you doing a repeated measures model?

Re: contrasts. Emmeans is available and so is Test.

Gene Maguin

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Salbod, Mr. Stephen
Sent: Monday, April 03, 2017 9:43 AM
To: [hidden email]
Subject: Mixed Model Missing Data

I am new to the mixed model procedure. The output is kind of strange. (That is what happens from years of looking at ANOVAs.) I would appreciate any recommendations that would help me understand SPSS mixed model output? I am trying to model a1c measures (glucose) across five time intervals (appointments): baseline, 3-6mo, 6-9mo, 9-12mo and 12-18mo. A treatment was administered after a baseline was obtained. I want to assess whether the treatment was effective. Missing data is a problem. Out of 118 people, only 37 made all their appointments. Thirty-five percent missed only one appointment and so on. The missing data issue steered me to the mixed model procedure. I’m treating time as a fixed variable, people as a random variable, and use AR(1) as my covariance type. With repeated measures, I would follow the analysis with contrasts. Is this available within mixed model?

Any articles or book chapter recommendations will be most welcome.

TIA
Steve (Pace University, NYC)

===================== To manage your subscription to SPSSX-L, send a message to [hidden email]<mailto:[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

--
Bruce Weaver
bweaver@lakeheadu.ca
http://sites.google.com/a/lakeheadu.ca/bweaver/

"When all else fails, RTFM."

PLEASE NOTE THE FOLLOWING:
1. My Hotmail account is not monitored regularly. To send me an e-mail, please use the address shown above.
2. The SPSSX Discussion forum on Nabble is no longer linked to the SPSSX-L listserv administered by UGA (https://listserv.uga.edu/).

Kylie

Re: Mixed Model Missing Data

Also check out Linear Mixed Models - A Practical Guide Using Statistical Software, by West, Welch & Galecki.

SPSS data files and syntax for each chapter are available here:

http://www-personal.umich.edu/~bwest/almmussp.html

From: SPSSX(r) Discussion <[hidden email]> on behalf of Bruce Weaver <[hidden email]>
Sent: Tuesday, 4 April 2017 6:52 AM
To: [hidden email]
Subject: Re: Mixed Model Missing Data

I agree with Gene re Singer & Willett. It's a great general resource, and
IIRC, you can find SPSS syntax for it in the "textbook examples" pages on
the UCLA website.

For books about estimating these types of models using SPSS specifically,
these two by Heck, Thomas & Tabata seem quite good:

http://www2.hawaii.edu/~ltabata/mlm/HTT2013.html
http://www2.hawaii.edu/~ltabata/mlm/HTT2012.html

HTH.

Maguin, Eugene wrote
> One very good recommendation is Singer and Willett, Applied longitudinal
> data analysis. There are other good resources. Another resource is
> http://stats.idre.ucla.edu/ Another is the MLwin website at University of

Statistics - Institute for Digital Research and Education

stats.idre.ucla.edu

The IDRE mission is to support, advance and guide a campus-wide program to position UCLA as a world leader in research and education in computational thi...

> Bristol (http://www.bristol.ac.uk/cmm/software/mlwin/
>
> Can we assume your model is multilevel model and is something like this?
> Mixed glucose with time bcov wcov/fixed=time cov1 cov2/Random intercept |
> subject(pid) covtype(??).
> Where bcov is a person level covariate and wcov is a time point
> (appointment) covariate.
> If this were your model, the covtype would be ‘id’, not ‘ar’.
>
> Or, are you doing a repeated measures model?
>
> Re: contrasts. Emmeans is available and so is Test.
>
> Gene Maguin
>
>
>
> From: SPSSX(r) Discussion [mailto:

> SPSSX-L@.UGA

> ] On Behalf Of Salbod, Mr. Stephen
> Sent: Monday, April 03, 2017 9:43 AM
> To:

> SPSSX-L@.UGA

> Subject: Mixed Model Missing Data
>
>
> I am new to the mixed model procedure. The output is kind of strange.
> (That is what happens from years of looking at ANOVAs.) I would appreciate
> any recommendations that would help me understand SPSS mixed model output?
> I am trying to model a1c measures (glucose) across five time intervals
> (appointments): baseline, 3-6mo, 6-9mo, 9-12mo and 12-18mo. A treatment
> was administered after a baseline was obtained. I want to assess whether
> the treatment was effective. Missing data is a problem. Out of 118 people,
> only 37 made all their appointments. Thirty-five percent missed only one
> appointment and so on. The missing data issue steered me to the mixed
> model procedure. I’m treating time as a fixed variable, people as a
> random variable, and use AR(1) as my covariance type. With repeated
> measures, I would follow the analysis with contrasts. Is this available
> within mixed model?
>
> Any articles or book chapter recommendations will be most welcome.
>
> TIA
> Steve (Pace University, NYC)
>
> ===================== To manage your subscription to SPSSX-L, send a
> message to

> LISTSERV@.UGA

> <mailto:

> LISTSERV@.UGA

> > (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

> LISTSERV@.UGA

> (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
[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://spssx-discussion.1045642.n5.nabble.com/Mixed-Model-Missing-Data-tp5734029p5734036.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

Ryan

Re: Mixed Model Missing Data

In reply to this post by Salbod

Stephen,

I want to call your attention to:

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

The AR1 specification on the RANDOM statement above simplifies to an identity matrix because the intercept term has nothing with which to covary.

If you believe the residuals obtained from observations within "id" further apart in time have an exponential decay in correlations, you should consider using a REPEATED statement with an AR1 structure.

The AR1 structure assumes the time intervals are equal within and between subjects.

Ryan

Sent from my iPhone

On Apr 3, 2017, at 3:42 PM, Salbod, Mr. Stephen <[hidden email]> wrote:

MIXED glucose BY time

/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,

ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)

/FIXED=time | SSTYPE(3)

/METHOD=REML

/PRINT=DESCRIPTIVES SOLUTION TESTCOV

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

/EMMEANS=TABLES(time) COMPARE REFCAT(FIRST) ADJ(SIDAK).

Salbod

Re: Mixed Model Missing Data

Thanks Ryan, I still have problems understanding the language. Identity matrix is clear. I’m not sure what the repeated statement provides. –Steve

PS I’m ‘slowly’ reading Singer et al (2003). To get familiar with the language.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ryan Black
Sent: Friday, April 07, 2017 5:47 PM
To: [hidden email]
Subject: Re: Mixed Model Missing Data

Stephen,

I want to call your attention to:

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

The AR1 specification on the RANDOM statement above simplifies to an identity matrix because the intercept term has nothing with which to covary.

The AR1 structure assumes the time intervals are equal within and between subjects.

Ryan

Sent from my iPhone

On Apr 3, 2017, at 3:42 PM, Salbod, Mr. Stephen <[hidden email]> wrote:

MIXED glucose BY time

/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,

ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)

/FIXED=time | SSTYPE(3)

/METHOD=REML

/PRINT=DESCRIPTIVES SOLUTION TESTCOV

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

/EMMEANS=TABLES(time) COMPARE REFCAT(FIRST) ADJ(SIDAK).

Ryan

Re: Mixed Model Missing Data

Hi Stephen,

The REPEATED statement can be used to account for within-subject correlation due to repeated measurements on the same subject.

Ryan

Sent from my iPhone

On Apr 7, 2017, at 6:25 PM, Salbod, Mr. Stephen <[hidden email]> wrote:

Thanks Ryan, I still have problems understanding the language. Identity matrix is clear. I’m not sure what the repeated statement provides. –Steve

PS I’m ‘slowly’ reading Singer et al (2003). To get familiar with the language.

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Ryan Black
Sent: Friday, April 07, 2017 5:47 PM
To: [hidden email]
Subject: Re: Mixed Model Missing Data

Stephen,

I want to call your attention to:

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

The AR1 specification on the RANDOM statement above simplifies to an identity matrix because the intercept term has nothing with which to covary.

If you believe the residuals obtained from observations within "id" further apart in time have an exponential decay in correlations, you should consider using a REPEATED statement with an AR1 structure.

The AR1 structure assumes the time intervals are equal within and between subjects.

Ryan

Sent from my iPhone

On Apr 3, 2017, at 3:42 PM, Salbod, Mr. Stephen <[hidden email]> wrote:

MIXED glucose BY time

/CRITERIA=CIN(95) MXITER(100) MXSTEP(10) SCORING(1) SINGULAR(0.000000000001) HCONVERGE(0,

ABSOLUTE) LCONVERGE(0, ABSOLUTE) PCONVERGE(0.000001, ABSOLUTE)

/FIXED=time | SSTYPE(3)

/METHOD=REML

/PRINT=DESCRIPTIVES SOLUTION TESTCOV

/RANDOM=INTERCEPT | SUBJECT(id) COVTYPE(AR1)

/EMMEANS=TABLES(time) COMPARE REFCAT(FIRST) ADJ(SIDAK).

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