PROC GLM

[spg]

Can I include dummy variables in proc glm?

I use the following but am not sure if the results should look the way they do.

proc glm data=final;
class income_bracket kiddummy year;
model units= units_cat kiddummy |year;
lsmeans kiddummy year /adjust = tukey;
run;

[Doc_Duke]

yes. That code looks reasonable. However, GLM uses "effect" coding and you are probably thinking of "reference" coding. Check out the documentation for a description of the two.

The various test statistics are the same either way, but I find that it is messier to interpret the coefficients in effect coding. The only way to get reference cell coding is to manually enter the dummy variables as numeric (which makes it harder to get overall effects for classes with more than two levels).