jwdid with staggered adoption and treatment effect heterogeneity

[Sarita Vaswani]

Hi everyone,

I am studying a state-level policy with staggered adoption (panel data: county*industry*year). My main question is not the impact of the policy but whether the policy effect varies with a baseline, pre-determined covariate (Z_c), which is continuous and measured at the county level, and I would like to avoid splitting the sample into low/high groups.

I am considering using jwdid and estimating heterogeneity via interactions of event-time indicators with Z_c.

My questions are:

1. If I use the following code,

jwdid y Z_c , ivar(county_ind) tvar(year) gvar(g_state) ///
    cluster (state)

will I have to use the

estat event , orestriction(Z_c=="")

at every level of Zc to get the event study estimates? And is this even valid to interpret these coefficients as how treatment effect varies with Zc?

2. If the above is the correct way to do this, should I test for differential pre-trends in the same way?

3. I do want test if this relationship varies by industry for which I have three industry levels. Should I run three separate regressions or have 2 covariates?

[FernandoRios]

1) Yes that is how you do it. I think on my last update I also added an "over" option. you could try that
Of course keep in mind that you will have less power (fewer observations)
2) You could do that as well. Its almost like if you are testing the treatment by subgroups
3) Those are (i think) different models. So either one is ok, just need to be clear of what they will be measuring and how they differ from each other

[Sarita Vaswani]

Thank you for your response, Fernando.

I did have a look at the over option, but the Z_c variable is continuous, so I am not sure how well it will work. Although now that I think about it, it will be easier to make it categorical, even if I would like to use "orestriction" because of the lack of observations at every point.