GLM Type III Tests

[Jon Heintz]

I am running a cost analysis, comparing the total cost of treatment 1 to treatment 2. My main question of interest is if there is a significant difference in cost between treatment groups after controlling for var1 and var2. I am running a GLM with a log link and gamma distribution, and the variables are

• Cost: continuous
• Treatment: binary
• Var 1: 9 categories
• Var 2: 6 categories

I'm also clustering by clinician. I would like to test if Var1 and Var2 are significant in of themselves with a type III test. How do I request a type III test in Stata? Here is my code.

glm cost i.treatment i.var1 i.var2, f(gam) l(log) cluster(clinician)

[Leonardo Guizzetti]

Type 3 tests are the default produced, though Stata doesn't use this terminology, and can be read directly from the regression table output (if showing t-statistics) or else can be obtained using -test- or -testparm-. See -help test- for more info.