SPSS: random effect in multivariate logistic regression

[vishal]

I am a PhD student here in Germany. Recent days I am dealing with epidemiological data. I would like to perform a multivariate logistic regression adjusting random effect. Is it possible in SPSS? If yes, How? I searched that it is possible in SAS however I am not sure in SPSS.  Please help me. Thank you

Vishal

[Bruce Weaver]

Multivariate means 2 or more dependent (or outcome) variables.  I expect you mean multivariable -- i.e., multiple explanatory variables.  

Have you taken a look at GENLINMIXED?

HTH.

vishal wrote

I am a PhD student here in Germany. Recent days I am dealing with epidemiological data. I would like to perform a multivariate logistic regression adjusting random effect. Is it possible in SPSS? If yes, How? I searched that it is possible in SAS however I am not sure in SPSS.  Please help me. Thank you

Vishal

[Ryan]

The GENLIN procedure is another option.

Ryan

On Mar 17, 2012, at 2:52 PM, Bruce Weaver <[hidden email]> wrote:

> /Multivariate/ means 2 or more dependent (or outcome) variables.  I expect
> you mean /multivariable/ -- i.e., multiple explanatory variables.
>
> Have you taken a look at GENLINMIXED?
>
> HTH.
>
>
> vishal wrote
>>
>> I am a PhD student here in Germany. Recent days I am dealing with
>> epidemiological data. I would like to perform a multivariate logistic
>> regression adjusting random effect. Is it possible in SPSS? If yes, How? I
>> searched that it is possible in SAS however I am not sure in SPSS.  Please
>> help me. Thank you
>>
>> Vishal
>>
>
>
> -----
> --
> 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/SPSS-random-effect-in-multivariate-logistic-regression-tp5573358p5574073.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]

To be clear, I'm stating that it is possible to run a multivariate logistic regression by fitting a multivariate GEE model via GENLIN. If in addition to accounting for a multivariate response, one wanted to account for random effects, then [as suggested by Bruce] the GENLINMIXED procedure should be considered.

 

Ryan

 

On Sat, Mar 17, 2012 at 3:38 PM, <[hidden email]> wrote:

The GENLIN procedure is another option.

Ryan

On Mar 17, 2012, at 2:52 PM, Bruce Weaver <[hidden email]> wrote:

> /Multivariate/ means 2 or more dependent (or outcome) variables.  I expect
> you mean /multivariable/ -- i.e., multiple explanatory variables.
>
> Have you taken a look at GENLINMIXED?
>
> HTH.
>
>
> vishal wrote
>>
>> I am a PhD student here in Germany. Recent days I am dealing with
>> epidemiological data. I would like to perform a multivariate logistic
>> regression adjusting random effect. Is it possible in SPSS? If yes, How? I
>> searched that it is possible in SAS however I am not sure in SPSS.  Please
>> help me. Thank you

>>
>> Vishal
>>
>
>
> -----
> --
> 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/SPSS-random-effect-in-multivariate-logistic-regression-tp5573358p5574073.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

[Ryan]

I was thinking that it might be helpful to show how to fit a multivariate GEE model via GENLIN (syntax BELOW). I can provide details regarding the model fitted below if anybody has a specific question and/or comment.

 

HTH,

 

Ryan

--

 

*Generate Data.
set seed 96754826.
  new file.
 inp pro.
 
comp subject = -99.
comp x = -99.
comp rand_eff = -99.
comp y_indic = -99.
 
leave subject to y_indic.
 
loop subject= 1 to 1000.
  comp x = rv.normal(0,1).
  comp rand_eff = sqrt(.30)*rv.normal(0,1).
 
  loop y_indic = 1 to 2.
  comp eta = 0.50*(y_indic=1) - 0.25*(y_indic=2) + 1.85*(y_indic=1)*x - 1.40*(y_indic=2)*x + rand_eff.
  comp p = 1 / (1+exp(-eta)).
  comp y = rv.bernoulli(p).

 end case.
 end loop.
end loop.
end file.
end inp pro.
exe.
 
delete variables rand_eff eta p.
 
*Fit Multivariate GEE Model.
GENLIN y (REFERENCE=FIRST) BY y_indic (ORDER=ASCENDING) WITH x
  /MODEL y_indic y_indic*x INTERCEPT=NO
 DISTRIBUTION=BINOMIAL LINK=LOGIT
  /REPEATED SUBJECT=subject WITHINSUBJECT=y_indic SORT=YES CORRTYPE=UNSTRUCTURED ADJUSTCORR=YES
  /PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION (EXPONENTIATED).

 

On Sat, Mar 17, 2012 at 4:15 PM, R B <[hidden email]> wrote:

To be clear, I'm stating that it is possible to run a multivariate logistic regression by fitting a multivariate GEE model via GENLIN. If in addition to accounting for a multivariate response, one wanted to account for random effects, then [as suggested by Bruce] the GENLINMIXED procedure should be considered.

 

Ryan

 

On Sat, Mar 17, 2012 at 3:38 PM, <[hidden email]> wrote:

The GENLIN procedure is another option.

Ryan

On Mar 17, 2012, at 2:52 PM, Bruce Weaver <[hidden email]> wrote:

> /Multivariate/ means 2 or more dependent (or outcome) variables.  I expect
> you mean /multivariable/ -- i.e., multiple explanatory variables.
>
> Have you taken a look at GENLINMIXED?
>
> HTH.
>
>
> vishal wrote
>>
>> I am a PhD student here in Germany. Recent days I am dealing with
>> epidemiological data. I would like to perform a multivariate logistic
>> regression adjusting random effect. Is it possible in SPSS? If yes, How? I
>> searched that it is possible in SAS however I am not sure in SPSS.  Please
>> help me. Thank you
>>
>> Vishal
>>
>
>
> -----
> --
> 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/SPSS-random-effect-in-multivariate-logistic-regression-tp5573358p5574073.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

[Matthew Fuller-Tyszkiewicz]

Hi,

I'm away on leave until Wednesday 21st March. I will not have access to my email account while away, and willl endeavour to return your emails shortly after my return.

Cheers,
Matt.

 

[Keli Saporta]

Dear list

i have two IV's  X1 (three levels) by x2 (two levels). I managed to perform interaction contrasts (with the list help) but for on of the  DV's there is only a main effect of x1. So I defined a contrast but it does not work. Any help would be appreciated. Here is the syntax:

 

UNIANOVA Y BY x1 x2

    /METHOD=SSTYPE(3)

  /INTERCEPT=INCLUDE

 /PLOT=PROFILE(x1)

  /PRINT=DESCRIPTIVE

  /CRITERIA=ALPHA(.05)

/LMATRIX = 'x1 =1+2 v x1=3'   x1  1 1 -2

  /DESIGN=   x1 x2 x1*x2.

 

 

kelly

[Hector Maletta]

Vishal,
Generalized Linear Models, through the GENLINMIXED procedure (introduced in
Version 19), provide the capability you mention. In these models, a linear
combination of predictors (the familiar sum of bX) is linked to the
dependent variable via a "link function". Link functions include (among
others) the logistic function, where the probability of an event is EXP(sum
of BX) / [1 + EXP(sum of bX)]. You can have any mixture of nominal, ordinal
and interval-scale predictors, producing fixed or random effects, including
multi-level models where the effects at micro level depend on variables at
higher levels.

There are, however, some problems with that procedure. One of them is that
it precludes the use of fractional weights (they are rounded before using,
so that any weight lower than 0.5 is rounded to zero, and the case
eliminated).

Hope this helps.

Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de
vishal
Enviado el: Saturday, March 17, 2012 07:23
Para: [hidden email]
Asunto: SPSS: random effect in multivariate logistic regression

I am a PhD student here in Germany. Recent days I am dealing with
epidemiological data. I would like to perform a multivariate logistic
regression adjusting random effect. Is it possible in SPSS? If yes, How? I
searched that it is possible in SAS however I am not sure in SPSS.  Please
help me. Thank you

Vishal

-----
vishal suthar
--
View this message in context:
http://spssx-discussion.1045642.n5.nabble.com/SPSS-random-effect-in-multivar
iate-logistic-regression-tp5573358p5573358.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

[vishal]

Thanks for the answers.


Do I need to use Syntax for this GENLINMIXED procedure? If yes than would be nice if you send me the source. I am sending some more detail on my question.Truly speaking I am not aware of this procedure. It would be nice of you if you explain step viz. Below I am sending you more detail on my problem.


My aim is to study prevalence of clinical ketosis (metabolic disease) and its possible predictors in European dairy herds. Therefore I collected 5012 cow data from 507 herds (median n=15 cows/herd) of 9 different European countries. For screening potential plausible predictors (independent factors) of clinical ketosis I used univariate logistic regression (i.e P < 0.20). Therefore I end up with 6 potential predictors.

Dependent and independent variables

Dependent: Clinical Ketosis (0 –No and 1- Yes) (n =201 (4%) having clinical ketosis out of 5012 cows)

Independent factors:

1.       Countries : (9; i.e. 0 to 8)

2.       Subclinical ketosis (No- 0 and Yes-1) (n =1194 (23.8%) having sub clinical ketosis out of 5012 cows)

3.       Season (summer -0 and winter -1)

4.       Milk yield (continuous variable)

5.       Lactation number (1st -0, 2nd- 1 , 3rd - 2 and 4th – 3)

6.    Metritis (0- No and 1 - yes) (n =532 (6.4%) having metritis out of 5012 cows)

I would like to include cow herd as a random effect in this model. Using SPSS > Analysis > Regression > Binary Logistic, I am unable to do that, correct? I am not sure! Please help me in this matter!

Thanking you in advance

Best regard

Vishal  

[David Marso]

GENLINMIXED is part of the Advanced Models option.
I am sure there are menus as well as SYNTAX
If you don't have the Advanced Models option then that may be the reason why you are not aware of it.
You might want to look at the manual and see if you can sort it out (at least poke around a bit)?

--

vishal wrote

Thanks for the answers.


Do I need to use Syntax for this GENLINMIXED procedure? If yes than would be nice if you send me the source. I am sending some more detail on my question.Truly speaking I am not aware of this procedure. It would be nice of you if you explain step viz. Below I am sending you more detail on my problem.


My aim is to study prevalence of clinical ketosis (metabolic disease) and its possible predictors in European dairy herds. Therefore I collected 5012 cow data from 507 herds (median n=15 cows/herd) of 9 different European countries. For screening potential plausible predictors (independent factors) of clinical ketosis I used univariate logistic regression (i.e P < 0.20). Therefore I end up with 6 potential predictors.

Dependent and independent variables

Dependent: Clinical Ketosis (0 –No and 1- Yes) (n =201 (4%) having clinical ketosis out of 5012 cows)

Independent factors:

1.       Countries : (9; i.e. 0 to 8)

2.       Subclinical ketosis (No- 0 and Yes-1) (n =1194 (23.8%) having sub clinical ketosis out of 5012 cows)

3.       Season (summer -0 and winter -1)

4.       Milk yield (continuous variable)

5.       Lactation number (1st -0, 2nd- 1 , 3rd - 2 and 4th – 3)

6.    Metritis (0- No and 1 - yes) (n =532 (6.4%) having metritis out of 5012 cows)

I would like to include cow herd as a random effect in this model. Using SPSS > Analysis > Regression > Binary Logistic, I am unable to do that, correct? I am not sure! Please help me in this matter!

Thanking you in advance

Best regard

Vishal

[David Marso]

FYI:pg 93...
ftp://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/20.0/en/client/Manuals/IBM_SPSS_Advanced_Statistics.pdf

pg 816...
ftp://public.dhe.ibm.com/software/analytics/spss/documentation/statistics/20.0/en/client/Manuals/IBM_SPSS_Statistics_Command_Syntax_Reference.pdf

HTH, David

David Marso wrote

GENLINMIXED is part of the Advanced Models option.
I am sure there are menus as well as SYNTAX
If you don't have the Advanced Models option then that may be the reason why you are not aware of it.
You might want to look at the manual and see if you can sort it out (at least poke around a bit)?

--

vishal wrote

Thanks for the answers.


Do I need to use Syntax for this GENLINMIXED procedure? If yes than would be nice if you send me the source. I am sending some more detail on my question.Truly speaking I am not aware of this procedure. It would be nice of you if you explain step viz. Below I am sending you more detail on my problem.


My aim is to study prevalence of clinical ketosis (metabolic disease) and its possible predictors in European dairy herds. Therefore I collected 5012 cow data from 507 herds (median n=15 cows/herd) of 9 different European countries. For screening potential plausible predictors (independent factors) of clinical ketosis I used univariate logistic regression (i.e P < 0.20). Therefore I end up with 6 potential predictors.

Dependent and independent variables

Dependent: Clinical Ketosis (0 –No and 1- Yes) (n =201 (4%) having clinical ketosis out of 5012 cows)

Independent factors:

1.       Countries : (9; i.e. 0 to 8)

2.       Subclinical ketosis (No- 0 and Yes-1) (n =1194 (23.8%) having sub clinical ketosis out of 5012 cows)

3.       Season (summer -0 and winter -1)

4.       Milk yield (continuous variable)

5.       Lactation number (1st -0, 2nd- 1 , 3rd - 2 and 4th – 3)

6.    Metritis (0- No and 1 - yes) (n =532 (6.4%) having metritis out of 5012 cows)

I would like to include cow herd as a random effect in this model. Using SPSS > Analysis > Regression > Binary Logistic, I am unable to do that, correct? I am not sure! Please help me in this matter!

Thanking you in advance

Best regard

Vishal

[Hector Maletta]

Syntax gives you more options, but you can manage with menu options. Click Analyze-->Generalized Linear Models. You may benefit from consulting the Help system on this subject. What you define as fixed or random effects is your decision.


Hector

-----Mensaje original-----
De: SPSSX(r) Discussion [mailto:[hidden email]] En nombre de vishal
Enviado el: Monday, March 19, 2012 05:51
Para: [hidden email]
Asunto: Re: SPSS: random effect in multivariate logistic regression

Thanks for the answers.


Do I need to use Syntax for this GENLINMIXED procedure? If yes than would be nice if you send me the source. I am sending some more detail on my question.Truly speaking I am not aware of this procedure. It would be nice of you if you explain step viz. Below I am sending you more detail on my problem.


My aim is to study prevalence of clinical ketosis (metabolic disease) and its possible predictors in European dairy herds. Therefore I collected 5012 cow data from 507 herds (median n=15 cows/herd) of 9 different European countries. For screening potential plausible predictors (independent
factors) of clinical ketosis I used univariate logistic regression (i.e P < 0.20). Therefore I end up with 6 potential predictors.

Dependent and independent variables

Dependent: Clinical Ketosis (0 –No and 1- Yes) (n =201 (4%) having clinical ketosis out of 5012 cows)

Independent factors:

1.       Countries : (9; i.e. 0 to 8)

2.       Subclinical ketosis (No- 0 and Yes-1) (n =1194 (23.8%) having sub
clinical ketosis out of 5012 cows)

3.       Season (summer -0 and winter -1)

4.       Milk yield (continuous variable)

5.       Lactation number (1st -0, 2nd- 1 , 3rd - 2 and 4th – 3)

6.    Metritis (0- No and 1 - yes) (n =532 (6.4%) having metritis out of
5012 cows)

I would like to include cow herd as a random effect in this model. Using SPSS > Analysis > Regression > Binary Logistic, I am unable to do that, correct? I am not sure! Please help me in this matter!

Thanking you in advance

Best regard

Vishal

-----
vishal suthar
--
View this message in context: http://spssx-discussion.1045642.n5.nabble.com/SPSS-random-effect-in-multivariate-logistic-regression-tp5573358p5576434.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

[Jennifer S. Randolph]

I am out of the office until Monday, March 26th.  For urgent matters please contact Susan Cronenwett at [hidden email].