Code to run Ordered Logistic Model - using Generalized Estimating Equations (GEE)

[Wesley Vo]

Hi everyone !

Im doing a Master's thesis.
My dataset is a panel data. I just wonder How to run Ordered Logistic Model by using GEE estimating method, because almost of Logistic Models under panel data just using Maximum Likelihood Estimating method to estimates the parameters. Is this strange in the study ? If not strange, So what is the code to run the model ?

Thank for your help

Wesley,

[Joseph Coveney]

You might try something like

ologit <response> <predictors>, vce(cluster <cluster var>)

[Salomat Yuldasheva]

Dear Wesley,

sorry for asking this question heere, but i am new here and i could not find where i can write and post my question? How did you post it? Did you write in your profile where it is stated "your current status"?

[Wesley Vo]

Hi Salomat Yuldasheva

You just go to the link of the Statalist as following: http://www.statalist.org/forums/foru...ussion/general , then you gonna see a rectango icon that is like "+ New Topic " at near top on the left, just click on it in order to post your questions. Just real simple like that .

[Wesley Vo]

Dear Mr. Joseph Coveney ,

The code you wrote above does not suit the panel data fine. Meanwhile my dataset is a panel data. So I cannot use this code to run Ordered Logistic Model by using GEE estimating method.

Thanks for your help anyway. I now still try to find out the code

[Joseph Coveney]

The code you wrote above does not suit the panel data fine. Meanwhile my dataset is a panel data. So I cannot use this code to run Ordered Logistic Model by using GEE estimating method.

Thanks for your help anyway. I now still try to find out the code

I'm not sure what you mean here by not suitable for panel data.

If you have access to the book by M. E. Stokes, Charles S. David and Gary G. Koch, Categorical Data Analysis using the SAS® System, then you can see, in the chapter titled "Generalized Estimating Equations" (Chapter 15), a section titled "Fitting the Proportional Odds Model". In the edition that I have (the second edition, Cary, North Carolina: SAS Institute, 2000), it is Section 15.10. The illustration there uses a dataset ("RESP") with an ordered-categorical outcome measured in each patient in two treatment groups at five visits, one pretreatment (baseline) and again at each of four posttreatment visits. To me, that seems to be what could be called panel data. The dataset was also used in an illustration of fitting binary outcome longitudinal logistic models with GEE (dichotomizing the outcome into two categories) earlier in the same chapter, in Section 15.6 (second edition).

To fit GEE models using SAS, the authors use the GENMOD procedure with the REPEATED option. The authors indicate that, with ordered-categorical data, only the independent working correlation structure is available. (Based upon the "Sample Data and Programs" for the third edition of the book the same is true as of then, too.)

I have fitted the models illustrated in each of the two sections of this GEE chapter. I have attached to this post the do-file, and the log-file in two formats, plain text and SMCL. (The forum software won't permit uploading plain-text ASCII attachments with the .log or .smcl file name suffix, and so I have had to rename them, adding a .txt second suffix to them. After downloading them, rename them to remove the .txt second suffix to view them in Stata.)

In it, I make a link between xtgee . . ., i(subject) family(binomial) link(logit) corr(independent) vce(robust) and logit . . , vce(cluster subject) to establish that the ordinary estimation command with the robust-cluster option produces the same result as the GEE estimation command with the independent working correlation structure and the robust option. By analogy, I believe that the code snippet shown earlier in this thread is equivalent to fitting an ordered logistic model using GEE with an independent working correlation structure, which is the only one available in the two software packages for fitting GEE models to ordered categorical outcome panel data.

[Wesley Vo]

Dear Mr. Joseph Coveney ,

It's cool ! Thanks for your help

[Joseph Coveney]

. . . the book by M. E. Stokes, Charles S. David and Gary G. Koch . . .

Make that ". . . Maura E. Stokes, Charles S. Davis and Gary G. Koch . . .".

[Wesley Vo]

Dear Mr. Joseph Coveney ,

I just would like to ask you one more question by the way.
I would like to compare the prediction performance of the models. I chose the out-of-sample prediction method of Duan (2010). It's convolution calculation technique to the predicted financial-distress probabilities of the out-of-sample firms base on the dynamic logit model. I have read Duan's (2010) convolution calculation technique but I still couldn't know how run it by the Stata. I wonder that Has the Duan's method been interpreted in Stata ?

Thanks for your help,

Wesley