03-19-2012 05:24 PM
Hi to all,
I ran a linear model with proc glm and got a very weired result that I cannot explain what is going on.
I got one categorical variable in the model with 4 levels and I put it in the class statement. Normally I would expect that 3 levels except the reference level would have the estimates. But the output I got resulted in two levels of the variable with nothing estimated.
educ 0 0.0000000 B . . .
educ 1 1.1880336 B 0.99728636 1.19 0.2337
educ 2 -0.9403576 B 1.06206014 -0.89 0.3760
educ 3 0.0000000 B . . .
There's no error message in the log.
Besides, I also tried to create 3 dummy variables and replaced the categorical variable with them. Still, SAS gave no estimate for one of the 3 dummy variables.
I just wonder if there is any statistical trick in the procedure that does not match my conventional thinking. Or there might be some other things I missed or did wrong?
Mostly importantly, do I adjust for this categorical variable sufficiently given this result, since I include this covariate in the model as a confounder?
Thanks so much for the help!
03-19-2012 11:50 PM
I've got some ideas of what the problem might be. One level of the covariate is exactly the same as the level of another predictor in the model, that is, one level of the education was assigned to represent the education level for children<18 while the age variable also has a category for <18yr. Since they provide essentially the same information mathematically, one of them is omitted in the computation.