There is something in https://www.stata.com/manuals13/tete...ultivalued.pdf that I simply don't follow. This pertains to obtaining ATET estimates for multivalued treatments.

Page 7 onwards, the text says:
...Sometimes, though, we would rather know how much the higher-education degree increases wages among the people who actually have a higher-education degree. To answer that question, we want to examine the ATET rather than the ATE. Here we use the IPWRA estimator to obtain our answer. We specify the control(A) option so that an A-level education is treated as the basis for comparisons. We specify the atet option to obtain ATETs rather than ATEs, and we specify the tlevel(H) option to indicate that we want the ATETs to be calculated for the subset of people who actually receive higher-education degrees.
The results shown are:

Array

But if the ATET has been specifically asked for tlevel(H) and the control level is (A), what do (none vs A) and (0 vs A) represent? Since we asked for tlevel(H), shouldn't we get only 1 row instead of the three rows above?

If I ask for ATET but do not specify tlevel(H), I get 3 sets of cofficients again. I presume that these pertain to the subset of people within each of those groups i.e. ATET for those who were at treatment level none, 0, and H. Is that right?

Thanks for all your help!