Is DiD possible for examining effect of a continuous treatment on a continuous outcome among single group dependent sample?

[Brian Young]

Hi Statalist,

I'm new to the forum and a novice Stata user.

I'm studying the effect of a continuous exogenous treatment measure on a small handful of continuous outcomes among a group of hospitals, all of whom are subject to the treatment to some extent.

I have a balanced panel dataset containing 11 years' worth of observations on about 2,200 unique hospitals (an annual observation on each hospital in the five years before the treatment, the treatment year, and the the five years after the treatment).

I'm interested in: (1) whether and the extent to which the treatment had a significant effect on the outcomes of interest, and (2) whether and the extent to which the treatment had a differential effect on the outcomes of interest when the hospitals are broken into two different groups not based on treatment level but a different characteristic that has nothing to do with treatment.

For (1) in particular, would DiD work (or does the fact that there is no control group make this impossible)? Is there a different approach that you might recommend?

Thanks,
Brian

[Jared Greathouse]

No control group=no causal inference, but see this


https://causalinf.substack.com/p/con...-treatment-did

[Brian Young]

Thank you for your reply, Jared. in fact, I’ve read Scott’s blog post. It’s quite illuminating. However, as far as I could tell, the examples he provided all include a control group. In the absence of a control group in my case, I’ve been thinking that an interrupted time series design (using the xtitsa command) as discussed here: http://www.lindenconsulting.org/docu...ITSA_Stata.pdf I'm just not sure if this approach can accommodate a continuous treatment. Thoughts?