JWDID estimation equation

[Christiaan de Swardt]

Dear Statalist Comunity,

I am interested in running some difference-in-differences equations with oen treatment group, multiple pre and post time periods, and nonlinear outcome variables. For this purpose, I want to make use of the 'jwdid' command, based on the extended two-way fixed effects developed by Wooldridge (2021: "Two-way fixed effects, the two-way mundlak regression, and difference-in-differences estimators") (2023: Simple approaches to nonlinear difference-in-differences with panel data).

I implement it in Stata with the following command syntax:

jwdid y x1 x2, ivar(id) tvar(wave) gvar(treat) never method(logit) hettype(time) 

. My accompanying estimation equation is attached.


Has anyone worked with this DiD method and/or Stata package before? My question is simply whether, given my Stata syntax and data structure, my estimation equation is correctly specified, or whether I am missing anything important in the Stata syntax or estimation equation?

Thank you in advance for any feedback!

[Jeff Wooldridge]

This looks like a common timing case, correct? You still need to use gvar(cohort) where cohort = 0 for never treated units and equals the time period of the first treatment otherwise. Is this panel data or repeated cross section?

[Christiaan de Swardt]

Thank you for your response! Yes indeed, this is a common case timing. I have an individual panel spanning 8 waves, with treatment starting at wave 6. So, for my treatment group, cohort = 6, and cohort = 0 for all other groups.

Then, I had a quick question related to obtaining heterogeneous effects using postestimation commands. I am interested in differences in effects across rural and urban areas. So, following my main regression, I make use of the following:

estat event, ores(urban == 0) 

and

estat event, ores(urban == 1) 

My question is whether there is a fairly intuitive explanation for how these effects are calculated? It is not the same as running the model separately for just the urban == 0 or just the urban == 1 subsample, right?

[FernandoRios]

The Simplest way to explain this

You make predictions under two scenarios
As observed...and as if never treated (counterfactual)

You estimate the difference, and average for either Urban or Rural areas (your code)