Does anyone have any advice on implementing Kroger and Hartmann's (August 25, 2021) interventionist approach from page 31 in their recent paper
xtoaxaca - Extending the Kitagawa-Oaxaca-Blinder Decomposition Approach to Panel Data
. I am trying to apply this to an RCT trial, which should be an obvious use of two types of interventions (treatment and control), but my issue is this:

Specifically, for any particular time and group, an individual only falls in the treatment pile, or the control pile, not both. Thus rendering me zero observations when I try to implement the code on page 31. If I collapse the code over groups, so there is a treatment and control value for each group, for each time period, then I run into the issue of collinearity as I run the model, and it again fails.

Does anyone have a suggestion of a variation of code for a situation such as this? Or any code suggestion for panel data for oaxaca-blinder decomp?

Many thanks.