Non-staggered treatment and no control group: DID2S

[marilyn ibarra-caton]

I am looking at the impact of covid on labor demand. All the regions in my sample receive the treatment (covid) at the same time. There is no control group.

However, when I run did2s, where the treatment variable is the timing of covid, STATA runs and gives me results (amazing results in fact). In this case, what is the did2s command estimating? what model? As did2s is for staggered treatment and with both treatment groups and control groups. Can I trust the results? If not, what model should I be estimating?

[FernandoRios]

DID2s relies heavily on you providing the correct model specification.
if you say there are no control and other is no staggered treatment, then DID2S wouldn't work.
In fact, as you describe, DID may not be the correct method for what you are doing. However, for the sake of completeness, it would be good if you provide what exactly you typed, and what are the definition of the variables you are using.
F

[marilyn ibarra-caton]

Thanks Fernando for your reply. It is puzzling to me that in theory it should not work but it does. This is what I estimate:

did2s high_low, first_stage(i.state i.week) second_stage(i.post_covid) treatment(post_covid) cluster(firmid)

where:

high_low is the ratio of the number of high skill to low skill job postings at the firm-week-state-level
state is a dummy for the state of the job posting
week is a dummy for the week number of the job posting (goes from week 1 to week 19)
post_covid is a dummy equal to 1 if job ad is in week 9 (March) when covid hits, zero weeks 1-8
firmid is the firm id

[FernandoRios]

Contact the author of did2s though his GitHub
i think what you are seeing here could be a bug because I’m latest iterations the command does not explicitly estimate fixed effects, which could be causing this problem

[Jared Greathouse]

DD without a control group isn't a thing. Plus, I wouldn't use covid as the intervention because everyone was affected by it, so there's no possible control group. Unless you're just going for interrupted time series, then that's different.

[marilyn ibarra-caton]

Thanks Fernando and Jared,

I emailed the author of DID2S and he too recommended interrupted time series.