Different results from unbalanced panel data

[Gofoy]

Dear Liyang:

I would like to ask you some questions about the command "eventstudyinteract". My data is unbalanced panel data, where L* represents dummy variables for post-event relative time, and F represents dummy variables for pre-event relative time. However, during the regression process, I found that there were significant differences in the regression results between "eventstudyinteract L_ F*" and "eventstudyinteract F L_," with the former's results being more consistent with results from other methods. I would like to inquire whether the command "eventstudyinteract" has specific requirements regarding the positioning of relative time dummy variables. Can I use "eventstudyinteract L* F*" as my baseline results?

[lsun20]

Thank you for your interest in the command! Under the hood, the package is calling something along the line of an interacted regression reghdfe outcome rel_time * cohort, absorb(FE) where rel_time is the set of relative time indicators and cohort is the set of indicators for belonging to different treatment groups. Due to unbalanced panel, I suspect there might be cohort that is not present some pre-event or post-event relative time, so this interacted regression has some collinearity. As far as I know, with collinearity, the regressor Stata drops varies with the order the regressor is included. I suspect this is what happened.

[Gofoy]

Thanks a lot!