ordinal Logistic regression

[H_G]

Hello,

I have a question and I will be grateful if you can help me,

I am running a regression which I have one dependent variable (ranked from 1 to 6) and several independent variable, which some of them are dummy variables(like gender) and some of them are metric variables like age, and I have some other variables like education which I ranked them .I am using SAS for my research.I used proc logistic for my regression, but it  just considers 3 levels of my dependent variable,(as I mentioned before it has 6 levels).

Would you please help me?

Thanks a lot.

[SteveDenham]

Run a quick cross-tab on your data using PROC FREQ.  While your dependent variable may be allowed to take on values from 1 to 6, it seems that your sample only has three values.

Steve Denham

[H_G]

Thank you for the answer.

actually I am sure that I have data in all 6 levels.still I have the same problem.

[SteveDenham]

Time to share a partial dataset, and the code you are using.  I have no idea whatsoever how PROC LOGISTIC would only see three levels if the data actually had six, aside from 's observation that independent variables are missing values, resulting in the cases not being included.  It is why I recommend a cross tab in PROC FREQ to see what might be happening.

If this turns out to be the case, then I would not trust imputation as a method.  Missing values will be imputed based on similarity of cases, but not based on the response variable, since it takes on unique values for missing independent variables.  As a result, bad fit of the imputed dataset to the dependent variable.  I think you will have to choose from the following: A)reduce the number of independent variables so that cases are not excluded, B) consolidate scores, so that the dependent variable takes on only three levels, C)accept the analysis as is.  I would prefer B, as I believe the results will be much easier to interpret.

Steve Denham

[ballardw · Accepted solution]

Possibly some of your dependent variable values only occur with missing values for (some combination of) the indepent variables. The output should tell if some observations were excluded and this could be the reason.

[H_G]

Yes you are right,

thanks a lot

but do you know how can I solve this problem?

[ballardw]

Only solutions I can think of are to either find values for the missing, possibly by imputation, or remove the offending variables from the model.

[H_G]

Thanks a lot for your help.