Testing the assumption of linearity in a multi-level logistic regression analysis

#1
Hello all!
I am working on a multi-level logistic regression (Binary Outcome) model and am trying to finding the best way to check the assumption of linearity for the continuous variables. I am using STATA and have tried lowess graphs using the logit option, Box-Tidwell, and the lincheck methods . All of these options have given somewhat conflicting results. Variables that appear to be linear using the lowess graphs are not considered linear according to Box-Tidwell or the lincheck methods. I know the lowess graphs are smoothed graphs so this maybe explains why they conflict. I am wondering what the most reliable or accepted method for checking linearity using STATA is. I am also familiar with SAS if someone knows a better method using that package.

Thanks
 

Dason

Ambassador to the humans
#2
How are you checking the linearity? Because the linearity assumption is on the logit of the true probability of success.
 

noetsi

Fortran must die
#3
There is no assumption of linearity between the raw data - only between the logit and IV.

I never realized that multilevel analysis could be used with logistic regression before - I thought it required an interval DV. That is useful to know.
 

Dason

Ambassador to the humans
#4
There is no assumption of linearity between the raw data - only between the logit and IV.

I never realized that multilevel analysis could be used with logistic regression before - I thought it required an interval DV. That is useful to know.
You can use it with any type of model you want. Estimation becomes more tedious but it's just a model...
 

noetsi

Fortran must die
#5
My multileval class focused exclusively on interval level DV. Should have realized it was not limited to that (although all the articles were I believe tied to interval level data too). :p