staggered vs instantaneous

[kvelleby]

In intro.Rmd, you write that "Staggered adoption does not cause any problem for the DiD-estimator/TWFE if the treatment effect is homogeneous or instantaneous". What does it mean that the treatment effect is instantaneous here? I thought that referred to non-staggered interventions?

[Hax2kon]

I mean that the effect should ideally meanifest in T+1, and thus not spill into later adoptions of treatment.

Example scenario: Unit 1 adopts treatment in T=0, 50% of effect manifest in T=1, and the remaining 50% in T=2. Unit 2 adopts treatment in T=1, 50% of effect manifest in T=2, and the remaining 50% in T=3.

My intuition is that this will bias the ATT?

[Hax2kon]

We cleared this: Effects that manifests over time is not a problem, because the ATT is averaged over the whole period, and the "bias" cancels out across units (as long as effect is homogenous)

Website updated