Hello all,

I'm considering running a generalize/staggered differences-in-differences model to analyze the effect of state-level variation of a particular explanatory variable over time. I also have household-level panel data. My basic equation is:

Y ist = γs + λt + βPst + eist

where P is the explanatory variable of interest.

My potential problem is that all states receive treatment at all values of t. At t=1, each state has some nonzero value for P. The values are all largely unique, and they all change over time. My understanding is that when running a generalized/staggered DiD model to exploit variation in treatment over time, there had to be at least some units in some periods that received no treatment. Is this true? Can anyone provide a citation either way? I have seen one paper claiming that it is perfectly fine to use DiD to exploit variation when all units are treated (but treated differently), but no explanation why.

Thanks