SPSSX Discussion

genlinmixed problems and questions

Classic

List

Threaded

5 messages Options

Maguin, Eugene

genlinmixed problems and questions

I’m fitting a growth curve to some data (three waves). In mixed I say

mixed ngrants with wave/fixed wave/print solution/

random intercept wave | subject(rid) covtype(un).

But the data are really counts so better to use a poisson or negbinomial. So:

genlinmixed /fields target=ngrants/

target_options distribution=poisson link=log/

random use_intercept=true effects=wave subjects=rid

covariance_type=diagonal.

I get a NPD message on the final Hessian but I also get this:

Random Effect Covariances

Random Effect Block 1
Random Effect Block	Intercept	wave=0	wave=9	wave=15
Intercept	.226	.000	.000	.000
wave=0	.000	.014	.000	.000
wave=9	.000	.000	.000	.000
wave=15	.000	.000	.000	.142
Covariance Structure: Diagonal Subject Specification: rid

What don’t I understand?

Thanks, Gene Maguin

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

Ryan

Re: genlinmixed problems and questions

Gene:

I’d suggest you start by excluding the random wave effect. You would still be accounting for covariance within subjects (“rid) across waves via the random intercept term.

If you are going to incorporate a random wave effect, consider changing the covariance type to unstructured. This would allow the intercept and slope terms to covary, which is fairly common observed in longitudinal data.

Ryan

On Dec 30, 2019, at 4:54 PM, Maguin, Eugene <[hidden email]> wrote:

I’m fitting a growth curve to some data (three waves). In mixed I say

mixed ngrants with wave/fixed wave/print solution/

   random intercept wave | subject(rid) covtype(un).

But the data are really counts so better to use a poisson or negbinomial. So:

genlinmixed /fields target=ngrants/

   target_options distribution=poisson link=log/

   random use_intercept=true effects=wave subjects=rid

   covariance_type=diagonal.

I get a NPD message on the final Hessian but I also get this:

Random Effect Covariances

Random Effect Block 1

Random Effect Block

Intercept

wave=0

wave=9

wave=15

Intercept

.226

.000

.000

.000

wave=0

.000

.014

.000

.000

wave=9

.000

.000

.000

.000

wave=15

.000

.000

.000

.142

Covariance Structure: Diagonal

Subject Specification: rid

What don’t I understand?

Thanks, Gene Maguin

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

Maguin, Eugene

Re: genlinmixed problems and questions

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ryan Black
Sent: Monday, December 30, 2019 6:31 PM
To: [hidden email]
Subject: Re: genlinmixed problems and questions

Gene:

I’d suggest you start by excluding the random wave effect. You would still be accounting for covariance within subjects (“rid) across waves via the random intercept term.

Ryan

On Dec 30, 2019, at 4:54 PM, Maguin, Eugene <[hidden email]> wrote:

I’m fitting a growth curve to some data (three waves). In mixed I say

mixed ngrants with wave/fixed wave/print solution/

   random intercept wave | subject(rid) covtype(un).

But the data are really counts so better to use a poisson or negbinomial. So:

genlinmixed /fields target=ngrants/

   target_options distribution=poisson link=log/

   random use_intercept=true effects=wave subjects=rid

   covariance_type=diagonal.

I get a NPD message on the final Hessian but I also get this:

Random Effect Covariances

Random Effect Block 1

Random Effect Block

Intercept

wave=0

wave=9

wave=15

Intercept

.226

.000

.000

.000

wave=0

.000

.014

.000

.000

wave=9

.000

.000

.000

.000

wave=15

.000

.000

.000

.142

Covariance Structure: Diagonal

Subject Specification: rid

What don’t I understand?

Thanks, Gene Maguin

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

Maguin, Eugene

Re: genlinmixed problems and questions

In reply to this post by Ryan

Thanks for replying Ryan.

I’ve done that. My initial model included only a random intercept and it estimated. What I described only occurred when I added the wave term. I experimented some more this morning and found the problem. The problem was that wave was “nominal”. When I changed it, variable level, aka “measure” in variable view data window tab, to scale I get what I expected. The only discussion of this “problem” is on page 783 in an oblique, veiled statement. The controlling effect of variable level should be explicitly stated in, at least, the fixed and random subcommands.

Gene Maguin

Gene:

I’d suggest you start by excluding the random wave effect. You would still be accounting for covariance within subjects (“rid) across waves via the random intercept term.

Ryan

On Dec 30, 2019, at 4:54 PM, Maguin, Eugene <[hidden email]> wrote:

I’m fitting a growth curve to some data (three waves). In mixed I say

mixed ngrants with wave/fixed wave/print solution/

   random intercept wave | subject(rid) covtype(un).

But the data are really counts so better to use a poisson or negbinomial. So:

genlinmixed /fields target=ngrants/

   target_options distribution=poisson link=log/

   random use_intercept=true effects=wave subjects=rid

   covariance_type=diagonal.

I get a NPD message on the final Hessian but I also get this:

Random Effect Covariances

Random Effect Block 1

Random Effect Block

Intercept

wave=0

wave=9

wave=15

Intercept

.226

.000

.000

.000

wave=0

.000

.014

.000

.000

wave=9

.000

.000

.000

.000

wave=15

.000

.000

.000

.142

Covariance Structure: Diagonal

Subject Specification: rid

What don’t I understand?

Thanks, Gene Maguin

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

Ryan

Re: genlinmixed problems and questions

In reply to this post by Maguin, Eugene

Hi Gene,

Thanks for sharing with us how you resolved the issue. There would be redundancy if wave were treated as nominal. If you wanted to treat wave as a nominal variable, I would just remove the intercept term. Clearly you don’t so all is well.

Anyway I’m glad you were able to resolve it.

Happy NYE!

Ryan

On Dec 31, 2019, at 8:59 AM, Maguin, Eugene <[hidden email]> wrote:

Thanks for replying Ryan.

I’ve done that. My initial model included only a random intercept and it estimated. What I described only occurred when I added the wave term. I experimented some more this morning and found the problem. The problem was that wave was “nominal”. When I changed it, variable level, aka “measure” in variable view data window tab, to scale I get what I expected. The only discussion of this “problem” is on page 783 in an oblique, veiled statement. The controlling effect of variable level should be explicitly stated in, at least, the fixed and random subcommands.

Gene Maguin

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Ryan Black
Sent: Monday, December 30, 2019 6:31 PM
To: [hidden email]
Subject: Re: genlinmixed problems and questions

Gene:

I’d suggest you start by excluding the random wave effect. You would still be accounting for covariance within subjects (“rid) across waves via the random intercept term.

If you are going to incorporate a random wave effect, consider changing the covariance type to unstructured. This would allow the intercept and slope terms to covary, which is fairly common observed in longitudinal data.

Ryan

On Dec 30, 2019, at 4:54 PM, Maguin, Eugene <[hidden email]> wrote:

I’m fitting a growth curve to some data (three waves). In mixed I say

mixed ngrants with wave/fixed wave/print solution/

   random intercept wave | subject(rid) covtype(un).

But the data are really counts so better to use a poisson or negbinomial. So:

genlinmixed /fields target=ngrants/

   target_options distribution=poisson link=log/

   random use_intercept=true effects=wave subjects=rid

   covariance_type=diagonal.

I get a NPD message on the final Hessian but I also get this:

Random Effect Covariances

Random Effect Block 1

Random Effect Block

Intercept

wave=0

wave=9

wave=15

Intercept

.226

.000

.000

.000

wave=0

.000

.014

.000

.000

wave=9

.000

.000

.000

.000

wave=15

.000

.000

.000

.142

Covariance Structure: Diagonal

Subject Specification: rid

What don’t I understand?

Thanks, Gene Maguin

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

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