Ologit

[Omar Shaher]

I have cross-sectional data for a group of people, the dependent variable is an ordinal variable, and the independent variables include nominal, ordinal, and scale variables.
The model:
Ordinal variable = α + β1 nominal + β2 ordinal + β3 scale + Error term
I want to apply the above model to predict the dependent variable via the independent variables.

1. So in this case, I think I should use the ordered logistic regression. If, not, please do correct me. The code I think will be:

Ologit  Ordinal variable nominal  scale

Then, I will apply the following model for each covariate:

 
 margins, dydx (variablelist)  
 marginsplot

So, could you please correct me if I am on the correct way?

I have a variable that consists of two groups, and I need to examine if there are significant differences between the two groups or not, so in this case, I think I need to create another dummy variable coded 1 for the first group and 2 for the second group.

1. In this case, I think I need to apply the independent sample test, if, not, please do correct me?

1. One of the independent variables is gender, I am very interested in finding whether the component of gender I mean male and female have different predictions on the dependent variable or not, so could you please tell me what is the code if I can do that?
   
   
   Thanks very much in advance

[Richard Williams]

Hi Omar. You may just want to read a short textbook entry on ologit. My own brief online handout is at

https://www3.nd.edu/~rwilliam/xsoc73994/Ologit01.pdf

For a fuller discussion, see

https://methods.sagepub.com/Foundati...ression-models

[Omar Shaher]

Dear Richard Williams,

Thanks very much for the very interesting and informative materials. I didn't want to reply until I have read them. Thanks very much.

I just wanted to confirm with you, please that If I want to use the independent variables to predict the dependent variable which has around 4 to 5 categories that can be ranked, so I can use the following model then:

Ologit Ordinal nominal scale

???

[Richard Williams]

I would probably write it more like

ologit ordinal_depvar i.nominal_variable

For example,

webuse nhanes2f, clear
ologit health i.female

[Omar Shaher]

Dear Richard Williams,

Thanks very much for the answer. Greatly appreciated.

I have one last question, if you don't mind, please.

One of my independent variables is Gender ( i.e., male and female), so can I know which one of them has more effect on the dependent variable, I would suggest creating two variables in this case, the first one is gender one where male coded one and female coded zero, and in the same regression, the second variable will be gender two, where female coded one and male coded zero, and then I will observe the effect?
What do you think? Do you agree?

Many thanks in advance.

[Rich Goldstein]

no, that will not work as the sum will be one for everyone, the same as the constant and you are not allowed to have two constants in the model; just include one of the two terms and if, say, the variable is coded 0 for male and 1 for female, then the sign of the coefficient will, if I understand you correctly, tell you what you want

[Omar Shaher]

Thanks a lot, Rich, it makes sense now.