SPSSX Discussion

Account for clusters in binary logistic regression

Classic

List

Threaded

10 messages Options

jess

Account for clusters in binary logistic regression

Hi, I found this forum very helpful but I am very confused about the method of choice for the following problem:

I have a dataset with multilevel/ hierarchical data, containing both data from different hospitals (hospitalID) and surgeons (surgeonID).

In a binary logistic regression model looking at outcome y (yes/no), I want to idenitfy independent predictors (x1-xn) for outcome y, while accounting for hospital and/or surgeon ID.

Which method in SPSS 19 is appropriate to use if I want to run a formal hierarchical binary logitic model?
1) Do I have to go through the complex sampling wizard? What do I have to fill in in the strata, cluster and weight fields in the 'preparing for analyses'? Going to the logistic regression option, how do set indicator variables and do I have to fill in 'factors'?) Is this model now accounting for hospitalID since I used the analyses preparation wizard or do I still have to fill in 'hospitalID' somewhere?

2) Use different 'blocks' in the 'normal' binary logistic regression?

3) Another method listed under regression which might be more appropriate?

Thanks so much,

Jess

Bruce Weaver

Re: Account for clusters in binary logistic regression

Administrator

Hi Jess. Under your option 3, look up GENLINMIXED in the FM. To access the GUI, click Analyze - Mixed Models - Generalized Linear. I've not yet played with GENLINMIXED myself, so cannot offer much advice beyond this. I know some other list members have used it a fair bit, so maybe one of them will jump in if you need more help.

HTH.

jess wrote

Hi, I found this forum very helpful but I am very confused about the method of choice for the following problem:

I have a dataset with multilevel/ hierarchical data, containing both data from different hospitals (hospitalID) and surgeons (surgeonID).

In a binary logistic regression model looking at outcome y (yes/no), I want to idenitfy independent predictors (x1-xn) for outcome y, while accounting for hospital and/or surgeon ID.

Which method in SPSS 19 is appropriate to use if I want to run a formal hierarchical binary logitic model?
1) Do I have to go through the complex sampling wizard? What do I have to fill in in the strata, cluster and weight fields in the 'preparing for analyses'? Going to the logistic regression option, how do set indicator variables and do I have to fill in 'factors'?) Is this model now accounting for hospitalID since I used the analyses preparation wizard or do I still have to fill in 'hospitalID' somewhere?

2) Use different 'blocks' in the 'normal' binary logistic regression?

3) Another method listed under regression which might be more appropriate?

Thanks so much,

Jess

--
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/).

Morris, Ramona (MCSCS)

Automatic reply: Account for clusters in binary logistic regression

I'm sorry I can't reply to you just now, I will be out of the office Monday November 19th.

Thank you
Ramona Morris
R&E Ontario Police College

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

Poes, Matthew Joseph

Re: Account for clusters in binary logistic regression

In reply to this post by jess

You want to use the mixed with log link dv. I believe it's under mixed generalized linear model. HLM is not a sampling issue, it's a nesting issue.

Sent from my iPhone

On Nov 14, 2012, at 5:20 PM, "jess" <[hidden email]> wrote:

> Hi, I found this forum very helpful but I am very confused about the method
> of choice for the following problem:
>
> I have a dataset with multilevel/ hierarchical data, containing both data
> from different hospitals (hospitalID) and surgeons (surgeonID).
>
> In a binary logistic regression model looking at outcome y (yes/no), I want
> to idenitfy independent predictors (x1-xn) for outcome y, while accounting
> for hospital and/or surgeon ID.
>
> *Which method in SPSS 19 is appropriate to use if I want to run a formal
> hierarchical binary logitic model?*
> 1) Do I have to go through the complex sampling wizard? What do I have to
> fill in in the strata, cluster and weight fields in the 'preparing for
> analyses'? Going to the logistic regression option, how do set indicator
> variables and do I have to fill in 'factors'?) Is this model now accounting
> for hospitalID since I used the analyses preparation wizard or do I still
> have to fill in 'hospitalID' somewhere?
>
> 2) Use different 'blocks' in the 'normal' binary logistic regression?
>
> 3) Another method listed under regression which might be more appropriate?
>
> Thanks so much,
>
> Jess
>
>
>
> --
> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201.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

Ryan

Re: Account for clusters in binary logistic regression

In reply to this post by jess

Bruce hit the nail on the head. You should consider a generalized linear mixed model. The correct parameterization of the random effect component depends upon whether surgeons are nested or crossed with hospitals.

Ryan

On Nov 14, 2012, at 3:42 PM, jess <[hidden email]> wrote:

Ryan

Re: Account for clusters in binary logistic regression

In reply to this post by Poes, Matthew Joseph

Matthew,

Unless I've misread the original post, I disagree completely. The
dependent variable is binary. Why are suggesting a log link? And are
you suggesting treating the DV as Gaussian? A more appropriate link
(the canonical link) would be logit, along with a binary/binomial
distribution for the DV. If there are [presumably] multiple
observations per surgeon and/or multiple observations per hospital,
then there is nesting which should be taken into account.

Ryan

On Wed, Nov 14, 2012 at 6:43 PM, Poes, Matthew Joseph
<[hidden email]> wrote:

> You want to use the mixed with log link dv. I believe it's under mixed generalized linear model. HLM is not a sampling issue, it's a nesting issue.
>
> Sent from my iPhone
>
> On Nov 14, 2012, at 5:20 PM, "jess" <[hidden email]> wrote:
>
>> Hi, I found this forum very helpful but I am very confused about the method
>> of choice for the following problem:
>>
>> I have a dataset with multilevel/ hierarchical data, containing both data
>> from different hospitals (hospitalID) and surgeons (surgeonID).
>>
>> In a binary logistic regression model looking at outcome y (yes/no), I want
>> to idenitfy independent predictors (x1-xn) for outcome y, while accounting
>> for hospital and/or surgeon ID.
>>
>> *Which method in SPSS 19 is appropriate to use if I want to run a formal
>> hierarchical binary logitic model?*
>> 1) Do I have to go through the complex sampling wizard? What do I have to
>> fill in in the strata, cluster and weight fields in the 'preparing for
>> analyses'? Going to the logistic regression option, how do set indicator
>> variables and do I have to fill in 'factors'?) Is this model now accounting
>> for hospitalID since I used the analyses preparation wizard or do I still
>> have to fill in 'hospitalID' somewhere?
>>
>> 2) Use different 'blocks' in the 'normal' binary logistic regression?
>>
>> 3) Another method listed under regression which might be more appropriate?
>>
>> Thanks so much,
>>
>> Jess
>>
>>
>>
>> --
>> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201.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

jess

Re: Account for clusters in binary logistic regression

HI Ryan and Metthew,

I am lost in this discussion. I indeed have a binary outcome. And I want to control for certain practice patterns / outcome from surgeons. Indeed, this will make my data hierarchical (nested) I guess.

In stata, there is a command which you can use to perform an official hierarchical model putting CLUSTER(variable) after your logistic command. This is what I want to do...but than with SPSS..

Thanks!

Op 15 nov. 2012, om 11:21 heeft R B [via SPSSX Discussion] het volgende geschreven:

Matthew,

Unless I've misread the original post, I disagree completely. The
dependent variable is binary. Why are suggesting a log link? And are
you suggesting treating the DV as Gaussian? A more appropriate link
(the canonical link) would be logit, along with a binary/binomial
distribution for the DV. If there are [presumably] multiple
observations per surgeon and/or multiple observations per hospital,
then there is nesting which should be taken into account.

Ryan

On Wed, Nov 14, 2012 at 6:43 PM, Poes, Matthew Joseph
<<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=0" target="_top" rel="nofollow" link="external">[hidden email]> wrote:

> You want to use the mixed with log link dv. I believe it's under mixed generalized linear model. HLM is not a sampling issue, it's a nesting issue.
>
> Sent from my iPhone
>
> On Nov 14, 2012, at 5:20 PM, "jess" <<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=1" target="_top" rel="nofollow" link="external">[hidden email]> wrote:
>
>> Hi, I found this forum very helpful but I am very confused about the method
>> of choice for the following problem:
>>
>> I have a dataset with multilevel/ hierarchical data, containing both data
>> from different hospitals (hospitalID) and surgeons (surgeonID).
>>
>> In a binary logistic regression model looking at outcome y (yes/no), I want
>> to idenitfy independent predictors (x1-xn) for outcome y, while accounting
>> for hospital and/or surgeon ID.
>>
>> *Which method in SPSS 19 is appropriate to use if I want to run a formal
>> hierarchical binary logitic model?*
>> 1) Do I have to go through the complex sampling wizard? What do I have to
>> fill in in the strata, cluster and weight fields in the 'preparing for
>> analyses'? Going to the logistic regression option, how do set indicator
>> variables and do I have to fill in 'factors'?) Is this model now accounting
>> for hospitalID since I used the analyses preparation wizard or do I still
>> have to fill in 'hospitalID' somewhere?
>>
>> 2) Use different 'blocks' in the 'normal' binary logistic regression?
>>
>> 3) Another method listed under regression which might be more appropriate?
>>
>> Thanks so much,
>>
>> Jess
>>
>>
>>
>> --
>> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201.html
>> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>>
>> =====================
>> To manage your subscription to SPSSX-L, send a message to
>> <a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=2" target="_top" rel="nofollow" link="external">[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
> <a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=3" target="_top" rel="nofollow" link="external">[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
<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=4" target="_top" rel="nofollow" link="external">[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

If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201p5716227.html

To unsubscribe from Account for clusters in binary logistic regression, click here.
NAML

Ryan

Re: Account for clusters in binary logistic regression

Accounting for characteristics of subjects (e.g., gender, age, years of experience) does not necessitate a hierarchical model of any sort. Whether one should employ a sampling methodology procedure (cluster variable) versus a mixed modeling procedure (random effect) is an interesting issue, but at this point, your design and specific research question(s) are not clear to me.

Perhaps someone else can jump in and help at this point.

Ryan

On Nov 15, 2012, at 1:30 PM, jess <[hidden email]> wrote:

HI Ryan and Metthew,

I am lost in this discussion. I indeed have a binary outcome. And I want to control for certain practice patterns / outcome from surgeons. Indeed, this will make my data hierarchical (nested) I guess.

In stata, there is a command which you can use to perform an official hierarchical model putting CLUSTER(variable) after your logistic command. This is what I want to do...but than with SPSS..

Thanks!

Op 15 nov. 2012, om 11:21 heeft R B [via SPSSX Discussion] het volgende geschreven:

Matthew,

Unless I've misread the original post, I disagree completely. The
dependent variable is binary. Why are suggesting a log link? And are
you suggesting treating the DV as Gaussian? A more appropriate link
(the canonical link) would be logit, along with a binary/binomial
distribution for the DV. If there are [presumably] multiple
observations per surgeon and/or multiple observations per hospital,
then there is nesting which should be taken into account.

Ryan

On Wed, Nov 14, 2012 at 6:43 PM, Poes, Matthew Joseph
<<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=0" target="_top" rel="nofollow" link="external">[hidden email]> wrote:

> You want to use the mixed with log link dv. I believe it's under mixed generalized linear model. HLM is not a sampling issue, it's a nesting issue.
>
> Sent from my iPhone
>
> On Nov 14, 2012, at 5:20 PM, "jess" <<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=1" target="_top" rel="nofollow" link="external">[hidden email]> wrote:
>
>> Hi, I found this forum very helpful but I am very confused about the method
>> of choice for the following problem:
>>
>> I have a dataset with multilevel/ hierarchical data, containing both data
>> from different hospitals (hospitalID) and surgeons (surgeonID).
>>
>> In a binary logistic regression model looking at outcome y (yes/no), I want
>> to idenitfy independent predictors (x1-xn) for outcome y, while accounting
>> for hospital and/or surgeon ID.
>>
>> *Which method in SPSS 19 is appropriate to use if I want to run a formal
>> hierarchical binary logitic model?*
>> 1) Do I have to go through the complex sampling wizard? What do I have to
>> fill in in the strata, cluster and weight fields in the 'preparing for
>> analyses'? Going to the logistic regression option, how do set indicator
>> variables and do I have to fill in 'factors'?) Is this model now accounting
>> for hospitalID since I used the analyses preparation wizard or do I still
>> have to fill in 'hospitalID' somewhere?
>>
>> 2) Use different 'blocks' in the 'normal' binary logistic regression?
>>
>> 3) Another method listed under regression which might be more appropriate?
>>
>> Thanks so much,
>>
>> Jess
>>
>>
>>
>> --
>> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201.html
>> Sent from the SPSSX Discussion mailing list archive at Nabble.com.

>>
>> =====================
>> To manage your subscription to SPSSX-L, send a message to
>> <a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=2" target="_top" rel="nofollow" link="external">[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
> <a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=3" target="_top" rel="nofollow" link="external">[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
<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=4" target="_top" rel="nofollow" link="external">[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

If you reply to this email, your message will be added to the discussion below:
http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201p5716227.html

To unsubscribe from Account for clusters in binary logistic regression, click here.
NAML

View this message in context: Re: Account for clusters in binary logistic regression
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

Kylie

Re: Account for clusters in binary logistic regression

Jess,

Sounds to me like a generalized estimating equations model might be appropriate. I’m assuming that your outcome (y) and the independent predictors (x1-xn) are measured on individual patients, and that the patients are nested within different surgeons who in turn are nested within different hospitals, is that right? You should be able to use a GEE model to handle both the binary outcome and the nesting.

Hope this helps.

Kylie.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of R B
Sent: Friday, 16 November 2012 3:17 PM
To: [hidden email]
Subject: Re: Account for clusters in binary logistic regression

Perhaps someone else can jump in and help at this point.

Ryan

On Nov 15, 2012, at 1:30 PM, jess <[hidden email]> wrote:

HI Ryan and Metthew,

I am lost in this discussion. I indeed have a binary outcome. And I want to control for certain practice patterns / outcome from surgeons. Indeed, this will make my data hierarchical (nested) I guess.

In stata, there is a command which you can use to perform an official hierarchical model putting CLUSTER(variable) after your logistic command. This is what I want to do...but than with SPSS..

Thanks!

Op 15 nov. 2012, om 11:21 heeft R B [via SPSSX Discussion] het volgende geschreven:

Matthew,

Unless I've misread the original post, I disagree completely. The
dependent variable is binary. Why are suggesting a log link? And are
you suggesting treating the DV as Gaussian? A more appropriate link
(the canonical link) would be logit, along with a binary/binomial
distribution for the DV. If there are [presumably] multiple
observations per surgeon and/or multiple observations per hospital,
then there is nesting which should be taken into account.

Ryan

On Wed, Nov 14, 2012 at 6:43 PM, Poes, Matthew Joseph
<<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=0" target="_top" rel="nofollow" link="external">[hidden email]> wrote:

> You want to use the mixed with log link dv. I believe it's under mixed generalized linear model. HLM is not a sampling issue, it's a nesting issue.
>
> Sent from my iPhone
>
> On Nov 14, 2012, at 5:20 PM, "jess" <<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=1" target="_top" rel="nofollow" link="external">[hidden email]> wrote:
>
>> Hi, I found this forum very helpful but I am very confused about the method
>> of choice for the following problem:
>>
>> I have a dataset with multilevel/ hierarchical data, containing both data
>> from different hospitals (hospitalID) and surgeons (surgeonID).
>>
>> In a binary logistic regression model looking at outcome y (yes/no), I want
>> to idenitfy independent predictors (x1-xn) for outcome y, while accounting
>> for hospital and/or surgeon ID.
>>
>> *Which method in SPSS 19 is appropriate to use if I want to run a formal
>> hierarchical binary logitic model?*
>> 1) Do I have to go through the complex sampling wizard? What do I have to
>> fill in in the strata, cluster and weight fields in the 'preparing for
>> analyses'? Going to the logistic regression option, how do set indicator
>> variables and do I have to fill in 'factors'?) Is this model now accounting
>> for hospitalID since I used the analyses preparation wizard or do I still
>> have to fill in 'hospitalID' somewhere?
>>
>> 2) Use different 'blocks' in the 'normal' binary logistic regression?
>>
>> 3) Another method listed under regression which might be more appropriate?
>>
>> Thanks so much,
>>
>> Jess
>>
>>
>>
>> --
>> View this message in context: http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201.html
>> Sent from the SPSSX Discussion mailing list archive at Nabble.com.
>>
>> =====================
>> To manage your subscription to SPSSX-L, send a message to
>> <a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=2" target="_top" rel="nofollow" link="external">[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
> <a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=3" target="_top" rel="nofollow" link="external">[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
<a href="x-msg://23/user/SendEmail.jtp?type=node&node=5716227&i=4" target="_top" rel="nofollow" link="external">[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

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/Account-for-clusters-in-binary-logistic-regression-tp5716201p5716227.html

To unsubscribe from Account for clusters in binary logistic regression, click here.
NAML

View this message in context: Re: Account for clusters in binary logistic regression
Sent from the SPSSX Discussion mailing list archive at Nabble.com.

brucea

(no subject)

In reply to this post by jess

SIGNOFF SPSSX-L