Multilevel binary logistic regression with SPSS 19

[abbottcb]

I am trying to do a multilevel binary logistic regression with SPSS 19 using GENLINMIXED.

 

The output uses the model viewer—something new to me and so far I’m finding it a nuisance.

 

Can I turn the model viewer off and  get output as pivot tables?

 

If not, can I edit the model viewer tables such as increasing the number of decimal places?

 

Has anyone successfully replicated examples 14.2 or 14.3 from Snijders & Bosker,  Multilevel Analysis (1999) using

SPSS 19 GENLINMIXED?

 

--craig

 

 

 

[Ryan]

Craig,

I haven't figured out how to turn off the model viewer. Frankly, I
stopped using GENLINMIXED in part because of this issue. Do you have
access to SAS? If so, consider the GLIMMIX procedure. Not only is the
output much easier to examine, but the actual proc has many more
options and has better estimation methods.

Ryan

On Wed, Sep 21, 2011 at 5:55 PM, Abbott, Craig (NIH/NICHD) [E]
<[hidden email]> wrote:

> I am trying to do a multilevel binary logistic regression with SPSS 19 using
> GENLINMIXED.
>
>
>
> The output uses the model viewer—something new to me and so far I’m finding
> it a nuisance.
>
>
>
> Can I turn the model viewer off and  get output as pivot tables?
>
>
>
> If not, can I edit the model viewer tables such as increasing the number of
> decimal places?
>
>
>
> Has anyone successfully replicated examples 14.2 or 14.3 from Snijders &
> Bosker,  Multilevel Analysis (1999) using
>
> SPSS 19 GENLINMIXED?
>
>
>
> --craig
>
>
>
>
>
>

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

[Alex Reutter]

Hi Craig,

You currently cannot get pivot table output from GENLINMIXED, and the model viewer currently does not allow editing of the tables.

I don't have a copy of Snijders & Bosker.  I see the data is available at: http://stat.gamma.rug.nl/mlbook1.htm, but without the book I'm not sure (a) what variable each column corresponds to, and (b) what are the actual models being fit.

Alex




From:        "Abbott, Craig (NIH/NICHD) [E]" <[hidden email]>
To:        [hidden email]
Date:        09/21/2011 05:00 PM
Subject:        Multilevel binary logistic regression with SPSS 19
Sent by:        "SPSSX(r) Discussion" <[hidden email]>

---

I am trying to do a multilevel binary logistic regression with SPSS 19 using GENLINMIXED.
 
The output uses the model viewer—something new to me and so far I’m finding it a nuisance.
 
Can I turn the model viewer off and  get output as pivot tables?
 
If not, can I edit the model viewer tables such as increasing the number of decimal places?
 
Has anyone successfully replicated examples 14.2 or 14.3 from Snijders & Bosker,  Multilevel Analysis (1999) using
SPSS 19 GENLINMIXED?
 
--craig
 
 
 

[Bruce Weaver]

Hi Ryan.  I've just started using v19, and have not yet tried GENLINMIXED, so I have yet to experience the pleasures of the model viewer.  I hope one can at least use OMS to send desired bits of output to another dataset.  

But...that's not why I'm responding.  I'm interested in knowing the other reason(s) that motivated you to stop using GENLINMIXED.  Thanks.

Bruce

R B wrote

Craig,

I haven't figured out how to turn off the model viewer. Frankly, I
stopped using GENLINMIXED in part because of this issue. Do you have
access to SAS? If so, consider the GLIMMIX procedure. Not only is the
output much easier to examine, but the actual proc has many more
options and has better estimation methods.

Ryan

On Wed, Sep 21, 2011 at 5:55 PM, Abbott, Craig (NIH/NICHD) [E]
<[hidden email]> wrote:
> I am trying to do a multilevel binary logistic regression with SPSS 19 using
> GENLINMIXED.
>
>
>
> The output uses the model viewer—something new to me and so far I’m finding
> it a nuisance.
>
>
>
> Can I turn the model viewer off and  get output as pivot tables?
>
>
>
> If not, can I edit the model viewer tables such as increasing the number of
> decimal places?
>
>
>
> Has anyone successfully replicated examples 14.2 or 14.3 from Snijders &
> Bosker,  Multilevel Analysis (1999) using
>
> SPSS 19 GENLINMIXED?
>
>
>
> --craig
>
>
>
>
>
>

=====================
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]

Hey Bruce,

A couple of reasons:

(1) GENLINMIXED uses pseudo-likelihood (PL) methods to estimate the
model parameters. PL methods have been shown to produced biased fixed
effects estimates, particularly when modeling correlated binary
response data. Instead, one should consider an integral approximation
method or perhaps even a Bayesian approach.

(2) GENLINMIXED does not have what would be equivalent to the /TEST
sub-command in the MIXED procedure. I don't want to have to fit a
model various times to obtain specific contrasts of interest with
respective confidence limits and p-values, especially if it takes
several hours for a model to converge. If I'm employing a random
effects logistic regression, for instance, I want to be able to obtain
odds and risk estimates at each combination at each level of the
categorical predictor variables and at specific values of covariates
with confidence limits. Suppose I want to compare the mean of two
levels of Factor A to the mean of a third level of Factor A? All of
these sorts of contrasts can be easily employed with fairly
straightforward statements using other generalized and/or nonlinear
mixed procedures in other stats packages. Until SPSS catches up, it
just isn't worth it for me.

Having said all of that, for purposes of learning the basics of
generalized linear mixed models or if one wants to run a preliminary
analysis of say a random effects Logistic or Poisson regression, then
I suppose GENLINMIXED could be useful. But, if the stakes are high and
you need a flexible program with top notch estimation methods, you
will have to go elsewhere.

My 2 cents.

Ryan

On Thu, Sep 22, 2011 at 2:53 PM, Bruce Weaver <[hidden email]> wrote:

> Hi Ryan.  I've just started using v19, and have not yet tried GENLINMIXED, so
> I have yet to experience the /pleasures/ of the model viewer.  I hope one
> can at least use OMS to send desired bits of output to another dataset.
>
> But...that's not why I'm responding.  I'm interested in knowing the other
> reason(s) that motivated you to stop using GENLINMIXED.  Thanks.
>
> Bruce
>
>
>
> R B wrote:
>>
>> Craig,
>>
>> I haven't figured out how to turn off the model viewer. Frankly, I
>> stopped using GENLINMIXED in part because of this issue. Do you have
>> access to SAS? If so, consider the GLIMMIX procedure. Not only is the
>> output much easier to examine, but the actual proc has many more
>> options and has better estimation methods.
>>
>> Ryan
>>
>> On Wed, Sep 21, 2011 at 5:55 PM, Abbott, Craig (NIH/NICHD) [E]
>> <[hidden email]> wrote:
>>> I am trying to do a multilevel binary logistic regression with SPSS 19
>>> using
>>> GENLINMIXED.
>>>
>>>
>>>
>>> The output uses the model viewer—something new to me and so far I’m
>>> finding
>>> it a nuisance.
>>>
>>>
>>>
>>> Can I turn the model viewer off and  get output as pivot tables?
>>>
>>>
>>>
>>> If not, can I edit the model viewer tables such as increasing the number
>>> of
>>> decimal places?
>>>
>>>
>>>
>>> Has anyone successfully replicated examples 14.2 or 14.3 from Snijders &
>>> Bosker,  Multilevel Analysis (1999) using
>>>
>>> SPSS 19 GENLINMIXED?
>>>
>>>
>>>
>>> --craig
>>>
>>>
>>>
>>>
>>>
>>>
>>
>> =====================
>> 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
> [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/Multilevel-binary-logistic-regression-with-SPSS-19-tp4828072p4831170.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
>

=====================
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]

Thanks Ryan.  

By the way, I just came across the website for aML, free-ware for multiprocess and multilevel models.  Has anyone heard of it, or used it?

    http://www.applied-ml.com/

Cheers,
Bruce

R B wrote

Hey Bruce,

A couple of reasons:

(1) GENLINMIXED uses pseudo-likelihood (PL) methods to estimate the
model parameters. PL methods have been shown to produced biased fixed
effects estimates, particularly when modeling correlated binary
response data. Instead, one should consider an integral approximation
method or perhaps even a Bayesian approach.

(2) GENLINMIXED does not have what would be equivalent to the /TEST
sub-command in the MIXED procedure. I don't want to have to fit a
model various times to obtain specific contrasts of interest with
respective confidence limits and p-values, especially if it takes
several hours for a model to converge. If I'm employing a random
effects logistic regression, for instance, I want to be able to obtain
odds and risk estimates at each combination at each level of the
categorical predictor variables and at specific values of covariates
with confidence limits. Suppose I want to compare the mean of two
levels of Factor A to the mean of a third level of Factor A? All of
these sorts of contrasts can be easily employed with fairly
straightforward statements using other generalized and/or nonlinear
mixed procedures in other stats packages. Until SPSS catches up, it
just isn't worth it for me.

Having said all of that, for purposes of learning the basics of
generalized linear mixed models or if one wants to run a preliminary
analysis of say a random effects Logistic or Poisson regression, then
I suppose GENLINMIXED could be useful. But, if the stakes are high and
you need a flexible program with top notch estimation methods, you
will have to go elsewhere.

My 2 cents.

Ryan

On Thu, Sep 22, 2011 at 2:53 PM, Bruce Weaver <[hidden email]> wrote:
> Hi Ryan.  I've just started using v19, and have not yet tried GENLINMIXED, so
> I have yet to experience the /pleasures/ of the model viewer.  I hope one
> can at least use OMS to send desired bits of output to another dataset.
>
> But...that's not why I'm responding.  I'm interested in knowing the other
> reason(s) that motivated you to stop using GENLINMIXED.  Thanks.
>
> Bruce
>
>
>
> R B wrote:
>>
>> Craig,
>>
>> I haven't figured out how to turn off the model viewer. Frankly, I
>> stopped using GENLINMIXED in part because of this issue. Do you have
>> access to SAS? If so, consider the GLIMMIX procedure. Not only is the
>> output much easier to examine, but the actual proc has many more
>> options and has better estimation methods.
>>
>> Ryan
>>
>> On Wed, Sep 21, 2011 at 5:55 PM, Abbott, Craig (NIH/NICHD) [E]
>> <[hidden email]> wrote:
>>> I am trying to do a multilevel binary logistic regression with SPSS 19
>>> using
>>> GENLINMIXED.
>>>
>>>
>>>
>>> The output uses the model viewer—something new to me and so far I’m
>>> finding
>>> it a nuisance.
>>>
>>>
>>>
>>> Can I turn the model viewer off and  get output as pivot tables?
>>>
>>>
>>>
>>> If not, can I edit the model viewer tables such as increasing the number
>>> of
>>> decimal places?
>>>
>>>
>>>
>>> Has anyone successfully replicated examples 14.2 or 14.3 from Snijders &
>>> Bosker,  Multilevel Analysis (1999) using
>>>
>>> SPSS 19 GENLINMIXED?
>>>
>>>
>>>
>>> --craig
>>>
>>>
>>>
>>>
>>>
>>>
>>
>> =====================
>> 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
> [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/Multilevel-binary-logistic-regression-with-SPSS-19-tp4828072p4831170.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
>

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

[abbottcb]

Alex,

The data file is also available at http://www.ats.ucla.edu/stat/examples/ma_snijders/default.htm

See the stata file: ch14.dta.

You can open the Stata file using SPSS.  It is an easier way of capturing the data.

I can send you a fuller description of the file and the models in another posting, but briefly cohab is the outcome variable (coded 1 = cohab, 0 = other), reg (region) is the second-level unit and represents 19 geographical regions in Norway, and respnr (respondant number) identifies single cases at the first-level unit.  Respodants range from 35 to 235 per region.  There are 2079 respondants.

--craig