SPSSX Discussion

need syntax help for GENLINMIXED

Classic

List

Threaded

6 messages Options

Paul Landsbergis

need syntax help for GENLINMIXED

I need some help creating syntax using GENLINMIXED for a mixed model analysis. The syntax reference guide is somewhat helpful, but does not provide enough guidance.

Specifically, my data set is comprised of individuals clustered within schools. Thus, individuals are not completely independent of each other. This is a two-level model. My outcome variable is categorical, either 2 categories or more. Therefore, I would plan to use a multinomial logistic mixed model (for outcomes with more than 2 categories) or a Poisson loglinear mixed model (for an outcome with 2 categories.

The following are some sample variable names of mine:
The variable that defines the group of individuals (school) = schoolID
School level variable = PTR
Individual level independent variable = influence
Dependent (target) variable = health

I constructed basic syntax below using these variable names for a multinomial logistic mixed model:

GENLINMIXED
/DATA_STRUCTURE GROUPING= schoolID
/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE
/TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
/FIXED EFFECTS= influence USE_INTERCEPT=TRUE
/RANDOM EFFECTS=PTR GROUPING= schoolID

Is this correct? For example:
1) am I using the correct /DATA_STRUCTURE subcommand, that is, GROUPING?
2) am I correct to put my individual level independent variable as a FIXED EFFECT?
3) am I correct to put my school level independent variable as a RANDOM EFFECT?
4) am I correct to specify the GROUPING variable schoolID again as a subcommand of /RANDOM EFFECTS?

Once I am sure that this is correct, it should not be a problem to add /BUILD OPTIONS commands. Thanks.

=====================
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: need syntax help for GENLINMIXED

Actually. I t think the genlinmixed command itself is crazily organized compared to mixed or mplus. Anyway, I'm looking at page 743 in the 24 CSR, specifically at the linear mixed model example, which is a three level model. Based on that your data_structure command is incomplete. Let's say you have school and stuid as the school and student ids, respectively. Thus,
data_structure subjects=school*stuid

I'm assuming you want the school intercepts to be random, thus the keyword use_intercept=true gets specified twice. Once where you have it and once in the random statement. Note page 750 examples.
The other thing that gets specified is covariance_type. variance_components is the default and with an intercepts only model, I think (but not sure) the covariance_type subcommand can be omitted. However, you'll need to include it once you include ptr as a random effect.

If I understand your description correctly, PTR is a school level variable. It should be named in the fixed effects list along with influence, the student level variable.

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Paul Landsbergis
Sent: Thursday, March 1, 2018 10:24 AM
To: [hidden email]
Subject: need syntax help for GENLINMIXED

I need some help creating syntax using GENLINMIXED for a mixed model analysis. The syntax reference guide is somewhat helpful, but does not provide enough guidance.

Specifically, my data set is comprised of individuals clustered within schools. Thus, individuals are not completely independent of each other. This is a two-level model. My outcome variable is categorical, either 2 categories or more. Therefore, I would plan to use a multinomial logistic mixed model (for outcomes with more than 2 categories) or a Poisson loglinear mixed model (for an outcome with 2 categories.

The following are some sample variable names of mine:
The variable that defines the group of individuals (school) = schoolID School level variable = PTR Individual level independent variable = influence Dependent (target) variable = health

I constructed basic syntax below using these variable names for a multinomial logistic mixed model:

GENLINMIXED
/DATA_STRUCTURE GROUPING= schoolID
/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
/FIXED EFFECTS= influence USE_INTERCEPT=TRUE
/RANDOM EFFECTS=PTR GROUPING= schoolID

Is this correct? For example:
1) am I using the correct /DATA_STRUCTURE subcommand, that is, GROUPING?
2) am I correct to put my individual level independent variable as a FIXED EFFECT?
3) am I correct to put my school level independent variable as a RANDOM EFFECT?
4) am I correct to specify the GROUPING variable schoolID again as a subcommand of /RANDOM EFFECTS?

Once I am sure that this is correct, it should not be a problem to add /BUILD OPTIONS commands. Thanks.

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

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

Ryan

Re: need syntax help for GENLINMIXED

If there is only one measurement on each student, and students are nested in schools, then the subject variable should be school only.

I should also mention that it is highly unlikely that the procedure allows for correlated random effects, so that is a limitation the OP will have to accept.

Ryan

Sent from my iPhone

> On Mar 2, 2018, at 10:38 AM, Maguin, Eugene <[hidden email]> wrote:
>
> Actually. I t think the genlinmixed command itself is crazily organized compared to mixed or mplus. Anyway, I'm looking at page 743 in the 24 CSR, specifically at the linear mixed model example, which is a three level model. Based on that your data_structure command is incomplete. Let's say you have school and stuid as the school and student ids, respectively. Thus,
> data_structure subjects=school*stuid
>
> I'm assuming you want the school intercepts to be random, thus the keyword use_intercept=true gets specified twice. Once where you have it and once in the random statement. Note page 750 examples.
> The other thing that gets specified is covariance_type. variance_components is the default and with an intercepts only model, I think (but not sure) the covariance_type subcommand can be omitted. However, you'll need to include it once you include ptr as a random effect.
>
> If I understand your description correctly, PTR is a school level variable. It should be named in the fixed effects list along with influence, the student level variable.
>
> Gene Maguin
>
> -----Original Message-----
> From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Paul Landsbergis
> Sent: Thursday, March 1, 2018 10:24 AM
> To: [hidden email]
> Subject: need syntax help for GENLINMIXED
>
> I need some help creating syntax using GENLINMIXED for a mixed model analysis. The syntax reference guide is somewhat helpful, but does not provide enough guidance.
>
> Specifically, my data set is comprised of individuals clustered within schools. Thus, individuals are not completely independent of each other. This is a two-level model. My outcome variable is categorical, either 2 categories or more. Therefore, I would plan to use a multinomial logistic mixed model (for outcomes with more than 2 categories) or a Poisson loglinear mixed model (for an outcome with 2 categories.
>
> The following are some sample variable names of mine:
> The variable that defines the group of individuals (school) = schoolID School level variable = PTR Individual level independent variable = influence Dependent (target) variable = health
>
> I constructed basic syntax below using these variable names for a multinomial logistic mixed model:
>
> GENLINMIXED
> /DATA_STRUCTURE GROUPING= schoolID
> /FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
> /FIXED EFFECTS= influence USE_INTERCEPT=TRUE
> /RANDOM EFFECTS=PTR GROUPING= schoolID
>
> Is this correct? For example:
> 1) am I using the correct /DATA_STRUCTURE subcommand, that is, GROUPING?
> 2) am I correct to put my individual level independent variable as a FIXED EFFECT?
> 3) am I correct to put my school level independent variable as a RANDOM EFFECT?
> 4) am I correct to specify the GROUPING variable schoolID again as a subcommand of /RANDOM EFFECTS?
>
> Once I am sure that this is correct, it should not be a problem to add /BUILD OPTIONS commands. Thanks.
>
> =====================
> 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

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

Paul Landsbergis

Re: [EXTERNAL] RE: need syntax help for GENLINMIXED

In reply to this post by Maguin, Eugene

Dear Dr. Maguin,
        Thank you so much for your advice. I agree that GENLINMIXED is not well organized.
        I have revised my syntax based on your suggestions. Does this look right:

GENLINMIXED
/DATA_STRUCTURE SUBJECTS = schoolID*studentID
/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
/FIXED EFFECTS= influence PTR USE_INTERCEPT=TRUE
/RANDOM EFFECTS=PTR GROUPING= schoolID USE_INTERCEPT=TRUE
    COVARIANCE_TYPE=VARIANCE_COMPONENTS

        I am still a little confused about where all my level 1 (individual level) and level 2 (school level) variables should go. I have more variables at both levels and will include them for appropriate tests and models. Should all my variables be listed under FIXED EFFECTS, and only my school level variables be listed under RANDOM EFFECTS? That seems a bit odd. Thanks again.

Best,
Paul

On 3/2/2018 10:38 AM, Maguin, Eugene wrote:

Actually. I think the genlinmixed command itself is crazily organized compared to mixed or mplus. Anyway, I'm looking at page 743 in the 24 CSR, specifically at the linear mixed model example, which is a three level model. Based on that your data_structure command is incomplete. Let's say you have school and stuid as the school and student ids, respectively. Thus, data_structure subjects=school*stuid

I'm assuming you want the school intercepts to be random, thus the keyword use_intercept=true gets specified twice. Once where you have it and once in the random statement. Note page 750 examples.
The other thing that gets specified is covariance_type. variance_components is the default and with an intercepts only model, I think (but not sure) the covariance_type subcommand can be omitted. However, you'll need to include it once you include ptr as a random effect.

If I understand your description correctly, PTR is a school level variable. It should be named in the fixed effects list along with influence, the student level variable.

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Paul Landsbergis
Sent: Thursday, March 1, 2018 10:24 AM
To: [hidden email]
Subject: need syntax help for GENLINMIXED

I need some help creating syntax using GENLINMIXED for a mixed model analysis. The syntax reference guide is somewhat helpful, but does not provide enough guidance.

Specifically, my data set is comprised of individuals clustered within schools. Thus, individuals are not completely independent of each other. This is a two-level model. My outcome variable is categorical, either 2 categories or more. Therefore, I would plan to use a multinomial logistic mixed model (for outcomes with more than 2 categories) or a Poisson loglinear mixed model (for an outcome with 2 categories.

The following are some sample variable names of mine:
The variable that defines the group of individuals (school) = schoolID School level variable = PTR Individual level independent variable = influence Dependent (target) variable = health

I constructed basic syntax below using these variable names for a multinomial logistic mixed model:

GENLINMIXED
/DATA_STRUCTURE GROUPING= schoolID
/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
/FIXED EFFECTS= influence   USE_INTERCEPT=TRUE
/RANDOM EFFECTS=PTR GROUPING= schoolID

Is this correct? For example:
1) am I using the correct /DATA_STRUCTURE subcommand, that is, GROUPING?
2) am I correct to put my individual level independent variable as a FIXED EFFECT?
3) am I correct to put my school level independent variable as a RANDOM EFFECT?
4) am I correct to specify the GROUPING variable schoolID again as a subcommand of /RANDOM EFFECTS?

Once I am sure that this is correct, it should not be a problem to add /BUILD OPTIONS commands. Thanks.

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

The contents of this email message and any attachments are confidential and are intended solely for the addressee. The information may also be legally privileged. This transmission is sent in trust, for the sole purpose of delivery to the intended recipient. If you have received this transmission in error, any use, reproduction or dissemination of this transmission is strictly prohibited. If you are not the intended recipient, please immediately notify sender by reply email or by (718) 270-HELP. ===================== 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: [EXTERNAL] RE: need syntax help for GENLINMIXED

Yes, that looks correct.

All of your predictors, along with use_intercept=true, should be specified on the fixed_effects command.

But, think about your level 1, student level, model. You name two predictors, influence and ptr, plus the intercept. The level 1 model is health = int.b0 + influence*b1 + ptr*b2.

If either influence or ptr is a school level variable, it is constant across students in each sdhool, thus has no variance across students in each school and cannot be a student level predictor. Let’s say both are student level variables. Your random statement specifies that intercept and ptr are random effects. That means that the regression coefficients for intercept and for ptr are going to vary across schools. The random statement is about the coefficients of variables not the variables themselves. You are saying that the effect of ptr on health is different for each school in your sample. IS THAT what you mean? Let’s say that it is. So now you basically have two new school level dvs: intercept coefficient and ptr coefficient. Both have a variance. They may be allowed to have a covariance. Covariance_type= variance_components does not allow for a covariance, unstructured, for example, does.

I think something that would be maybe helpful is a good multilevel text aimed at data analysis rather than theory AND one that includes spss-based examples. One that somebody bought for our Phd data analysis library is by Ronald Heck, Scott Thomas and Lunn Tabata, Multilevel modeling of categorical outcomes using ibm spss. I think that Bruce Weaver has offered several good suggestions of books. Maybe a search on nabble or the listserv archives.

Gene Maguin

From: Paul Landsbergis <[hidden email]>
Sent: Wednesday, March 7, 2018 8:00 PM
To: Maguin, Eugene <[hidden email]>; [hidden email]
Subject: Re: [EXTERNAL] RE: need syntax help for GENLINMIXED

On 3/2/2018 10:38 AM, Maguin, Eugene wrote:

Actually. I think the genlinmixed command itself is crazily organized compared to mixed or mplus. Anyway, I'm looking at page 743 in the 24 CSR, specifically at the linear mixed model example, which is a three level model. Based on that your data_structure command is incomplete. Let's say you have school and stuid as the school and student ids, respectively. Thus, data_structure subjects=school*stuid

I'm assuming you want the school intercepts to be random, thus the keyword use_intercept=true gets specified twice. Once where you have it and once in the random statement. Note page 750 examples.

The other thing that gets specified is covariance_type. variance_components is the default and with an intercepts only model, I think (but not sure) the covariance_type subcommand can be omitted. However, you'll need to include it once you include ptr as a random effect.

If I understand your description correctly, PTR is a school level variable. It should be named in the fixed effects list along with influence, the student level variable.

Gene Maguin

-----Original Message-----

From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Paul Landsbergis

Sent: Thursday, March 1, 2018 10:24 AM

To: [hidden email]

Subject: need syntax help for GENLINMIXED

I need some help creating syntax using GENLINMIXED for a mixed model analysis. The syntax reference guide is somewhat helpful, but does not provide enough guidance.

Specifically, my data set is comprised of individuals clustered within schools. Thus, individuals are not completely independent of each other. This is a two-level model. My outcome variable is categorical, either 2 categories or more. Therefore, I would plan to use a multinomial logistic mixed model (for outcomes with more than 2 categories) or a Poisson loglinear mixed model (for an outcome with 2 categories.

The following are some sample variable names of mine:

The variable that defines the group of individuals (school) = schoolID School level variable = PTR Individual level independent variable = influence Dependent (target) variable = health

I constructed basic syntax below using these variable names for a multinomial logistic mixed model:

GENLINMIXED

/DATA_STRUCTURE GROUPING= schoolID

/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT

/FIXED EFFECTS= influence   USE_INTERCEPT=TRUE

/RANDOM EFFECTS=PTR GROUPING= schoolID

Is this correct? For example:

1) am I using the correct /DATA_STRUCTURE subcommand, that is, GROUPING?

2) am I correct to put my individual level independent variable as a FIXED EFFECT?

3) am I correct to put my school level independent variable as a RANDOM EFFECT?

4) am I correct to specify the GROUPING variable schoolID again as a subcommand of /RANDOM EFFECTS?

Once I am sure that this is correct, it should not be a problem to add /BUILD OPTIONS commands. Thanks.

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

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

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

Ryan

Re: [EXTERNAL] RE: need syntax help for GENLINMIXED

In reply to this post by Paul Landsbergis

Paul,

1. I posted a message that your subject variable should only be school if you only have one observation per student and students are nested in schools.

2. Regarding your RANDOM statement, School is not a grouping variable, it is your subject variable.

No time to respond further.

Ryan

Sent from my iPhone

On Mar 7, 2018, at 7:59 PM, Paul Landsbergis <[hidden email]> wrote:

Dear Dr. Maguin,
        Thank you so much for your advice. I agree that GENLINMIXED is not well organized.
        I have revised my syntax based on your suggestions. Does this look right:

GENLINMIXED
/DATA_STRUCTURE SUBJECTS = schoolID*studentID
/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
/FIXED EFFECTS= influence PTR USE_INTERCEPT=TRUE
/RANDOM EFFECTS=PTR GROUPING= schoolID USE_INTERCEPT=TRUE
    COVARIANCE_TYPE=VARIANCE_COMPONENTS

        I am still a little confused about where all my level 1 (individual level) and level 2 (school level) variables should go. I have more variables at both levels and will include them for appropriate tests and models. Should all my variables be listed under FIXED EFFECTS, and only my school level variables be listed under RANDOM EFFECTS? That seems a bit odd. Thanks again.

Best,
Paul

On 3/2/2018 10:38 AM, Maguin, Eugene wrote:
Actually. I think the genlinmixed command itself is crazily organized compared to mixed or mplus. Anyway, I'm looking at page 743 in the 24 CSR, specifically at the linear mixed model example, which is a three level model. Based on that your data_structure command is incomplete. Let's say you have school and stuid as the school and student ids, respectively. Thus, data_structure subjects=school*stuid

I'm assuming you want the school intercepts to be random, thus the keyword use_intercept=true gets specified twice. Once where you have it and once in the random statement. Note page 750 examples.
The other thing that gets specified is covariance_type. variance_components is the default and with an intercepts only model, I think (but not sure) the covariance_type subcommand can be omitted. However, you'll need to include it once you include ptr as a random effect.

If I understand your description correctly, PTR is a school level variable. It should be named in the fixed effects list along with influence, the student level variable.

Gene Maguin

-----Original Message-----
From: SPSSX(r) Discussion [[hidden email]] On Behalf Of Paul Landsbergis
Sent: Thursday, March 1, 2018 10:24 AM
To: [hidden email]
Subject: need syntax help for GENLINMIXED

I need some help creating syntax using GENLINMIXED for a mixed model analysis. The syntax reference guide is somewhat helpful, but does not provide enough guidance.

Specifically, my data set is comprised of individuals clustered within schools. Thus, individuals are not completely independent of each other. This is a two-level model. My outcome variable is categorical, either 2 categories or more. Therefore, I would plan to use a multinomial logistic mixed model (for outcomes with more than 2 categories) or a Poisson loglinear mixed model (for an outcome with 2 categories.

The following are some sample variable names of mine:
The variable that defines the group of individuals (school) = schoolID School level variable = PTR Individual level independent variable = influence Dependent (target) variable = health

I constructed basic syntax below using these variable names for a multinomial logistic mixed model:

GENLINMIXED
/DATA_STRUCTURE GROUPING= schoolID
/FIELDS TARGET= health TRIALS=NONE OFFSET=NONE /TARGET_OPTIONS DISTRIBUTION=MULTINOMIAL LINK=LOGIT
/FIXED EFFECTS= influence   USE_INTERCEPT=TRUE
/RANDOM EFFECTS=PTR GROUPING= schoolID

Is this correct? For example:
1) am I using the correct /DATA_STRUCTURE subcommand, that is, GROUPING?
2) am I correct to put my individual level independent variable as a FIXED EFFECT?
3) am I correct to put my school level independent variable as a RANDOM EFFECT?
4) am I correct to specify the GROUPING variable schoolID again as a subcommand of /RANDOM EFFECTS?

Once I am sure that this is correct, it should not be a problem to add /BUILD OPTIONS commands. Thanks.

=====================
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
The contents of this email message and any attachments are confidential and are intended solely for the addressee. The information may also be legally privileged. This transmission is sent in trust, for the sole purpose of delivery to the intended recipient. If you have received this transmission in error, any use, reproduction or dissemination of this transmission is strictly prohibited. If you are not the intended recipient, please immediately notify sender by reply email or by (718) 270-HELP. ===================== 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