Can't get parameter estimates for cluster-level variables in a mixed model

[Dave Voelker]

     Forgive me if this is a duplicate post – I originally submitted it via the archive on Nabble (thinking it was this listserv) and it’s still not appearing many hours after submission.

     I’m having trouble getting parameter estimates for cluster-level variables in a mixed model I’m trying to run.  It’s a two-level design: 16 schools, and within them, nearby stores that sell cigarettes.  The target is dichotomous.  I have some school-level demographic variables (pop. density, pct. race/ethnicity), for each of which I need a single parameter estimate (with sig. & CI).

     The problem is that SPSS is giving me separate estimates for these vars for each of the 16 schools.  I need a single parameter estimate for each school-level var. (e.g., pop. density).  I’ve tried every permutation of var. entry I can think of.  Here’s the last syntax I tried (pasted from the GUI interface):

 

GENLINMIXED

  /DATA_STRUCTURE SUBJECTS=SchoolNum

  /FIELDS TARGET=w2_Q22a TRIALS=NONE OFFSET=NONE

  /TARGET_OPTIONS DISTRIBUTION=BINOMIAL LINK=LOGIT

  /FIXED  EFFECTS=Storetype1 Storetype2 Storetype3 Storetype4 Licensed

PopDens1Mile Prcnt_WhtNH1Mile USE_INTERCEPT=TRUE

  /RANDOM USE_INTERCEPT=TRUE SUBJECTS=SchoolNum COVARIANCE_TYPE=VARIANCE_COMPONENTS SOLUTION=TRUE

  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100

    CONFIDENCE_LEVEL=95 DF_METHOD=RESIDUAL COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0

    SINGULAR=0.000000000001

  /EMMEANS TABLES=Storetype1 CONTRAST=NONE

   /EMMEANS TABLES=Storetype2 CONTRAST=NONE

   /EMMEANS TABLES=Storetype3 CONTRAST=NONE

   /EMMEANS TABLES=Storetype4 CONTRAST=NONE

   /EMMEANS TABLES=Licensed CONTRAST=NONE

  /EMMEANS_OPTIONS SCALE=ORIGINAL PADJUST=LSD.

 

     Any advice would be appreciated!  I tried entering the school-level vars as random effects and got the same result (separate parameter for each school).  As others have observed, SPSS’s implementation of mixed model tests can be confusing, and this is my first shot at one.  Also, it looks like it doesn’t give odds ratios, which is unfortunate.  I’m running SPSS v. 25.

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

[David Marso]

Try registering. That will make your issue show up more quickly.  I am one of
the mods on the UGA list and check it in the am  and pm.  I don't have the
time to check it on the hour.



-----
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email me.
---
"Nolite dare sanctum canibus neque mittatis margaritas vestras ante porcos ne forte conculcent eas pedibus suis."
Cum es damnatorum possederunt porcos iens ut salire off sanguinum cliff in abyssum?"
--
Sent from: http://spssx-discussion.1045642.n5.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

[Dave Voelker]

     I did register, clicked on the confirmation link Nabble emailed me, and
was welcomed on the site ("Hello Dave Voelker"), where it accepted my post.
When it didn't appear and I tried logging back in with the email & PW it had
confirmed a few hours earlier, it said it didn't recognize my email.
Something's wrong at the server end.

-----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
David Marso
Sent: Wednesday, March 7, 2018 6:05 PM
To: [hidden email]
Subject: Re: Can't get parameter estimates for cluster-level variables in a
mixed model

Try registering. That will make your issue show up more quickly.  I am one
of the mods on the UGA list and check it in the am  and pm.  I don't have
the time to check it on the hour.
-----
Please reply to the list and not to my personal email.
Those desiring my consulting or training services please feel free to email
me.
--
Sent from: http://spssx-discussion.1045642.n5.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

[Maguin, Eugene]

What is your data structure? Stores within schools? If so then your data_structure line is wrong. It ought to be

data_structure subjects=schoolnum*stores/

What happens if you make this change?

Gene Maguin

 

 

From: SPSSX(r) Discussion <[hidden email]> On Behalf Of Dave Voelker
Sent: Wednesday, March 7, 2018 7:58 PM
To: [hidden email]
Subject: Can't get parameter estimates for cluster-level variables in a mixed model

 

     Forgive me if this is a duplicate post – I originally submitted it via the archive on Nabble (thinking it was this listserv) and it’s still not appearing many hours after submission.

     I’m having trouble getting parameter estimates for cluster-level variables in a mixed model I’m trying to run.  It’s a two-level design: 16 schools, and within them, nearby stores that sell cigarettes.  The target is dichotomous.  I have some school-level demographic variables (pop. density, pct. race/ethnicity), for each of which I need a single parameter estimate (with sig. & CI).

     The problem is that SPSS is giving me separate estimates for these vars for each of the 16 schools.  I need a single parameter estimate for each school-level var. (e.g., pop. density).  I’ve tried every permutation of var. entry I can think of.  Here’s the last syntax I tried (pasted from the GUI interface):

 

GENLINMIXED

  /DATA_STRUCTURE SUBJECTS=SchoolNum

  /FIELDS TARGET=w2_Q22a TRIALS=NONE OFFSET=NONE

  /TARGET_OPTIONS DISTRIBUTION=BINOMIAL LINK=LOGIT

  /FIXED  EFFECTS=Storetype1 Storetype2 Storetype3 Storetype4 Licensed

PopDens1Mile Prcnt_WhtNH1Mile USE_INTERCEPT=TRUE

  /RANDOM USE_INTERCEPT=TRUE SUBJECTS=SchoolNum COVARIANCE_TYPE=VARIANCE_COMPONENTS SOLUTION=TRUE

  /BUILD_OPTIONS TARGET_CATEGORY_ORDER=ASCENDING INPUTS_CATEGORY_ORDER=ASCENDING MAX_ITERATIONS=100

    CONFIDENCE_LEVEL=95 DF_METHOD=RESIDUAL COVB=MODEL PCONVERGE=0.000001(ABSOLUTE) SCORING=0

    SINGULAR=0.000000000001

  /EMMEANS TABLES=Storetype1 CONTRAST=NONE

   /EMMEANS TABLES=Storetype2 CONTRAST=NONE

   /EMMEANS TABLES=Storetype3 CONTRAST=NONE

   /EMMEANS TABLES=Storetype4 CONTRAST=NONE

   /EMMEANS TABLES=Licensed CONTRAST=NONE

  /EMMEANS_OPTIONS SCALE=ORIGINAL PADJUST=LSD.

 

     Any advice would be appreciated!  I tried entering the school-level vars as random effects and got the same result (separate parameter for each school).  As others have observed, SPSS’s implementation of mixed model tests can be confusing, and this is my first shot at one.  Also, it looks like it doesn’t give odds ratios, which is unfortunate.  I’m running SPSS v. 25.

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

[Dave Voelker]

Maguin, Eugene wrote
> What is your data structure? Stores within schools? If so then your
> data_structure line is wrong. It ought to be
> data_structure subjects=schoolnum*stores/
> What happens if you make this change?
> Gene Maguin

     Right you are, Gene; with that data structure addition, I now have the
single parameter estimates I was looking for.  Thanks very much!
     It looks like Paul L’s GENLINMIXED question here is similar to mine.
Ironically, I have the Heck, Thomas & Tabata book sitting here.  It’s very
helpful for walking one through the SPSS settings for mixed models, as you
say.  I used an example in it as a template for mine, since the research q.
was very similar, but it lacked the data structure element you mentioned.  I
see now why that was needed for my model.  



--
Sent from: http://spssx-discussion.1045642.n5.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