Thank you Junpeng. Indeed, I have already used the GLM module, but I have trouble understanding something. In the simple case y ~ X with X containing three groups x, y and z. Patsy automatically assigns a group such as the “treatment” group (say x). Then we obtain a coefficient for the other groups (y and z). If these two coefficients are different from zero (HPD does not contain 0 for example, with HPD chosen according to the research domain) then we could say that these two groups are different from x; but how do we know if y is different from z?