Suppose I have a panel dataset with N "individuals" (denoted by the identifier "id") and X*N observations.

I have two periods, one pre-intervention and one post-intervention. I want to keep the coefficient of the interaction id#post.

Suppose I use:

Code:
reg y id post id#post, cluster(id)
In this simple case, the omitted variables are the ones with post==0, which is the intuitive alternative. However, the coefficient vector is too large and contains unnecessary elements.

If I try to use the reghdfe command, for some reason the default is omitting post==1. What bothers me is that even if I try:

Code:
reghdfe y id#io0.post, cluster(id) absorb(post id)
The omitted observations are still those with post==1. That is, they are capturing (PRE - POST) instead of (POST - PRE).

One option is to multiply the whole vector by -1 and be done with it, but I wonder why it isn't omitting post==0? This continues to be the case even if I try something like io0.id#io0.post, it omits post==0 for the first ID, and then for the others, it omits post==1.