SPSS GEE

[Jennifer Lembach]

Greetings,

I am running a GEE model with a multinomial dependent. The output for the parameter estimates is below.

Parameter Estimates    
Parameter       B       Std. Error      95% Wald Confidence Interval    Hypothesis Test
                        Lower   Upper   Wald Chi-Square df      Sig.   
Threshold       [alccompGEE=1.00]       .867    .6483   -.404   2.137   1.787   1       .181   
        [alccompGEE=2.00]       3.142   .8382   1.499   4.784   14.049  1       .000   
        [alccompGEE=3.00]       4.100   1.0535  2.035   6.165   15.149  1       .000   
        friendsuse      2.092   1.0298  .074    4.111   4.129   1       .042   
        (Scale) 1                                                      
Dependent Variable: Alcohol composite

Model: (Threshold), friendsuse                                                 

This is not how I expected it to look. I expected it to be stratified with each of the 3 non-referent categories of the dependent paired with the covariate (friendsuse), as it would be in a multinomial regression model (example of output below).

Parameter Estimates    
Alcohol composite^a      B       Std. Error      Wald    df      Sig.    Exp(B)  95% Confidence Interval for Exp(B)     
                                                        Lower Bound     Upper Bound    
2nd lowest use  Intercept       -.803   1.052   .582    1       .446                           
        friendsuse      5.670   2.437   5.415   1       .020    290.068 2.446   34397.418      
2nd highest use Intercept       -3.529  1.210   8.509   1       .004                           
        friendsuse      10.428  2.606   16.013  1       .000    33790.246       204.452 5584593.566    
Highest use     Intercept       -7.921  1.834   18.652  1       .000                           
        friendsuse      13.769  3.088   19.887  1       .000    954397.923      2246.989        4.054E8
a. The reference category is: Lowest use.                                                      

Does anyone know if there is a way to get the output to look the way I’m expecting or is that not how the output should look?

Thank you.

************************************
Jennifer Lembach
Research Associate
RMC Research Corporation
111 SW Columbia Street, Suite 1200
Portland, OR  97201-5843
503-223-8248
800-788-1887
Fax: [phone]
www.rmccorp.com

P Before printing this document, please think about the environment

[Reutter, Alex]

Hi Jennifer,

 

The GENLIN procedure fits an *ordinal* multinomial model – the response is assumed to be ordinal with a multinomial distribution, so running a GEE on this response is like an extension of PLUM.  Currently there isn’t a GEE equivalent for NOMREG.

 

Alex

 

---

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Jennifer Lembach
Sent: Wednesday, June 03, 2009 4:58 PM
To: [hidden email]
Subject: SPSS GEE

 

Greetings,

 

I am running a GEE model with a multinomial dependent. The output for the parameter estimates is below.

Parameter Estimates    
Parameter       B       Std. Error      95% Wald Confidence Interval    Hypothesis Test
                        Lower   Upper   Wald Chi-Square df      Sig.   
Threshold       [alccompGEE=1.00]       .867    .6483   -.404   2.137   1.787   1       .181   
        [alccompGEE=2.00]       3.142   .8382   1.499   4.784   14.049  1       .000   
        [alccompGEE=3.00]       4.100   1.0535  2.035   6.165   15.149  1       .000   
        friendsuse      2.092   1.0298  .074    4.111   4.129   1       .042   
        (Scale) 1                                                      
Dependent Variable: Alcohol composite

Model: (Threshold), friendsuse                                                 

This is not how I expected it to look. I expected it to be stratified with each of the 3 non-referent categories of the dependent paired with the covariate (friendsuse), as it would be in a multinomial regression model (example of output below).

Parameter Estimates    
Alcohol composite^a      B       Std. Error      Wald    df      Sig.    Exp(B)  95% Confidence Interval for Exp(B)     
                                                        Lower Bound     Upper Bound    
2nd lowest use  Intercept       -.803   1.052   .582    1       .446                           
        friendsuse      5.670   2.437   5.415   1       .020    290.068 2.446   34397.418      
2nd highest use Intercept       -3.529  1.210   8.509   1       .004                           
        friendsuse      10.428  2.606   16.013  1       .000    33790.246       204.452 5584593.566    
Highest use     Intercept       -7.921  1.834   18.652  1       .000                           
        friendsuse      13.769  3.088   19.887  1       .000    954397.923      2246.989        4.054E8
a. The reference category is: Lowest use.                                                      

Does anyone know if there is a way to get the output to look the way I’m expecting or is that not how the output should look?

Thank you.

************************************
Jennifer Lembach
Research Associate
RMC Research Corporation
111 SW Columbia Street, Suite 1200
Portland, OR  97201-5843
503-223-8248
800-788-1887
Fax: [phone]
www.rmccorp.com

P Before printing this document, please think about the environment