Regarding ordinal exposure and numrical outcome: using class variable in proc glm?

[xql29]

Dear all.

I have a project regarding bloodvessel size (numerical outcome) in children.

Where the independent variables are height and weight (numerical) and pubertal development (tanner stages; ordered exposure).

Would perform a proc glm to find association between bloodvessel size and height, weight and tanner stage.

Is it valid to use proc glm with tanner stage as a class-variable. Or do I need another method?

Thank you so much in advance.

 

[hlsmith]

Re: Regarding ordinal exposure and numrical outcome: using class variable in proc glm

You should be fine using proc glm. UCLA has great basic level introductions (http://www.ats.ucla.edu/stat/sas/output/sas_glm_output.htm). If I was you, once I got started I may also want to look for collinearity between weight and height (may need to do that piece with proc reg), since they may be explaining the same variability in the dependent variable.

Good luck and let us know if you have any other questions.

P.S., Tanner stages are ordinal groups, so you could add them in the class statement, but I may transform them to numeric values (still placed in class statement) to make sure 1 is your reference group (if that is what you want). Also, you could probably play around with age in the model (if it works with your aims), since it may help explain all of these variables. However, you would also want to examine for interaction between variables by testing multiplication terms (e.g., age*tanner, age*weight, etc.).

 

[xql29]

Re: Regarding ordinal exposure and numrical outcome: using class variable in proc glm

Dear hlsmith

First of all thank you for a fast and clear answer to my question.
I’ve spend some time searching for the right method for ordinal independent variable and numerical dependent outcome, as well as validity of including ordinal categorical exposure and numerical exposure in multiple linear regression using class statement.

As feedback:
The data I’m working with includes a well defined age group. Therefore the difference that we’ve first observed in boys and girls may be explained by different pubertal development stage at this age group.

I’ve used this SAS input: (after assigned dummy variable 1 - 5 for tanner stages)

Proc glm data=dataset;
Class tanner gender;
Model bloodvessel = height tanner height*tanner/ solution clparm;
Run;

Thank you for the reminder of collinearity and interactions.

Best regards.