Stepwise

[Liza]

Hello,

I am using the rank ordered logistic regression in my data analysis. I would like to use the stepwise method to select the variables. I am using Stata and the rank ordered logistic regression is not supported by stepwise.
How can I do this?

Thanks,

Liza

[duskstar]

I've honestly never ran into a regression that isnt supported by stepwise. But, looking at the helpfile it appears your right. I don't know much about rologit, so naively, how many variables do you have and is it possible to consider some other kind of (manual?) model selection?

[Liza]

With my data I think that it is not possible to use another kind of regression. My goal is to obtain a model that can predict a final ranking. For this I am using 12 variables.
I am thinking to make this procedure:

Step 1:
To determine the likelihood of the null model and then determine the log likelihood for the model considering only one independent variable at each time. I will obtain 12 models.

Step2:
I will insert in first place the variable that generated the model with the lowest p value. Here I will define a cut off value.

Step 3:
After I Will determine which variable to insert in the second place. I will make this determining new models that consider the variable selected in step2 and one of the remaining variables at each time. I will obtain 11 models. The selection procedure will be the same as in step2.

Do you think I am wrong in doing this?

Thanks

[sg2kilo]

I don't know much about rank ordered log reg. In stepwise-ing regression model i prefer backward model. Testing null model to check its loglikelihood is important.
After checking null model, I usually enter all variable to check the model. Then from p-value which i've got from "enter" method, i've drop variable one by one in order to get the fit model. Which one to drop first? the variable which have the largest p-value (close to 1). Then use "est sto" and "lrtest" to check model significance from previous model or from null model.

(correct me if i'm wrong)