[guneet robin]

I am unable to find a way to get coefficients for control variables in csdid. We only get treatment effects in the standard csdid command. How are other control variables impacting y variable is not known.

[FernandoRios]

Right Csdid produces summary of all of those on the form of the att
it does not show how the att will vary as function of Xs
You can’t recover the coefficients of all models either because there are too many. I I don’t keep track of them

[Till Koeveker]

Hi Fernando,

is it possible to estimate ATTs by groups based on a categorical control variable using jwdid?
Basically, I would like to interact the treatment with a control variable and then recover the treatment effects for all levels of the control variable.

It is possible to do what I am looking for in R using the etwfe command. See section on "Heterogeneous treatment effects" in this vignette:
https://cran.r-project.org/web/packa...tes/etwfe.html

I am wondering if there is a way to do the same thing using jwdid in Stata.

Thanks in advance,
Till

[FernandoRios]

Not at the moment.
I will have to see how exactly things are being done in R. But as of right now, I cant replicate what he presented there.
F

[Yihui Cao]

Hi FernandoRios,

Thanks for sharing the new command. Quite straightforward and useful.

I'm just wondering how you aggregate by 'simple' 'group' and 'event'. Many thanks.

Yihui

[FernandoRios]

If you have the latest version from SSC its as simple as
estat event
estat simple
estat group
F

[Yihui Cao]

Thanks FernandoRios. I understand how to use the command. But I'm wondering what your aggregation rule is by 'simple' 'group' etc. Do you use the weighted scheme? Or do you take the margins?

Many thanks.

Yihui

[FernandoRios]

Its just margins
if you look at the file jwdid_estat.ado, you can see exactly what it does
However, at the end of the day, what margins is doing is similar to the weighting scheme suggested by Callaway and Sant'Anna (2021). In other words, aggregations are weighted averages, where each ATTGT is weighted by the number of observations used for that particular estimate (how many observations treated in year G are observed at time T).
If you use a basic setup, you can see that the results will be identical to CSDID.
HTH
Fernando

[Poyuan Huang]

Hi FernandoRios , just a quick followup a question. I was trying to use the jwdid with the survey data that I want to apply the survey weight in the estimation. I think the manual says it allows for pweight, but when I simply type
jwdid emp [pw = weight], tvar(year) gvar(first_treat) method(poisson)
This gives me the error message "weights not allowed"
Wondering if I did anything wrong here.

Thanks!

[FernandoRios]

Hi Poyuan,
the problem is a bug i accidentally included for trying to a new option.
Soon a new update will be uploaded to SSC to fix that.
Otherwise, you could install it from my repository:

net install jwdid, from("https://friosavila.github.io/stpackages") replace
Thank you for letting me know.
Fernando

[Julia Girard]

Dear FernandoRios ,
I am in a case where I only have treated units in my dataset, they are treated in different years. I do not have never treated units. My understanding is that with the jwdid command, I can only get post-treatment effect estimates, but not pre-treatment effects, meaning I cannot evaluate the presence of pretrends. Is there a way to get those pre-treatment estimates with the etwfe approach ?
Julia

[FernandoRios]

You cant. or not automatically.
Because comparison groups change, jwdid is not equipt to estimate Pre-ATT in those cases.
Your onlyoption is to run separate regressions for each "cohort" and the appropriate "treated groups"
or drop data after treatment of the last treated group, and use it as your never treated.
F

[Julia Girard]

Thank you very much for your answer !

[Md Shoeb]

Hi @FernandoRios. Is it possible to have multiple interactions with the treatment? As of now, I have interacted one variable x1 with gvar and tvar.

jwdid ln_loan, ivar(lender_id) gvar(gvar) tvar(fyear) xtvar(xtvar) xgvar(xgvar) fevar(firm_id) cluster(lender_id) never

I have just multiplied the variable instead of using ## but I am not sure it is the correct way to do it. Anyway, I now want to interact one more dummy x2 to xtvar and xtgvar. If I just multiply it, the other interactions will not come. So is it possible to have multiple interactions ?

And is there a way get robust standard errors for aggregations as well ?

[FernandoRios]

After jwdid and your dep variable you can interact all variables as usual, however you can’t do that with gvar and tvar
those have to remain on the original time-scales. Interactions will make no sense.