Difference in difference with 1 year

[Ann Steele]

Hi! i have 1 year before the treatment and 6 after. Can I still do panel data diff in diff with only 1 year prior?
thanks

[Carlo Lazzaro]

Ann:
instead of bumplng, please take a look at Example 3: DID for panel-data model under -didregress-, Stata .pdf manual. Thanks.

[Ann Steele]

thank you but i already checked it and there's nothing

[Jared Greathouse]

You can, it just won't be of any good quality (generally speaking). 2 by 2 DD is a thing, but it's best used under simple circumstances. I am curious though: why do you have only one year and 6 years after?

[Ann Steele]

it is just a survey that started in 2011...

[Jared Greathouse]

Yeah my advice is honestly to find a different project. What you're asking for is mechanically tractable, but it would immediately raise suspicion if I were reviewing it.

Unless you're prepared to use DD and matching together (presuming you've many comparison units), there's pretty much nothing (or not much) you can do here.

Oh, and there isn't "nothing" under didreg example 3, the pdf literally gives a panel data example of what you want.

[Jeff Wooldridge]

I'm not as pessimistic as Jared. One pre-treatment year is not ideal because one cannot test the parallel trends assumption. But you probably have controls data in 2011 that can help with that. So you can start with the basic two-way fixed effects analysis that assumes a constant effect. Then, you can add to that an interaction of the treatment variable with each post-treatment year dummy and estimate the expanded model by TWFE. You can't do that with xtdidregress. Instead, just use regular xtreg. Here's a linked to a shared Dropbox where I have a paper and Stata code that demonstrates the approach:

DID Dropbox Link