Difference-in-differences without baseline (only midline and endline)

[Filippo Girasoli]

Dear all, good morning,
I am currently dealing with some data issue for an impact evaluation I have to conduct.
I need to perform difference-in-differences regression analysis, but I don't have baseline data.
More specifically, surveys were conducted only at 'midline' (halfway through the intervention) and endline. The midline survey attempted at reconstructing baseline values through retrospective inquiries, but unfortunately that was not the case for all questions (actually, it wasn't for most of them) and sadly not the case for the outcomes I need to evaluate.
Starting from the acknowledgment that endline-on-midline regressions are technically wrong and not meaningful for measuring treatment impacts, I was wondering if you knew about any tool/method to solve this issue and manage to still run the DiD in a meaningful way.
Thanks a lot in advance for your help and availability.

Kind regards
Filippo

[Jared Greathouse]

My short answer is that no pre-period outcome data =no DD analysis. Mathematically, you'd need to have pre-intervention data to impute the counterfactual from your untreated, pre-intervention units.

[Fei Wang]

As #2 said, the dimension of pre/post intervention can not be used. But if lucky enough, you may find another new dimension to conduct a DiD analysis. For example, a policy was implemented on the elderly of region A from year 2015. Then there are three dimensions of the policy: 1. region A vs other regions; 2. elderly vs young; 3. before 2015 vs after 2015. The third dimension fails in your case, but you may still be able to do DiD if you have the first two. Note that DiD simply needs two dimensions, no matter whether the time dimension is included.