Dear all,

I have a Difference-in-Differences (DiD) setup in which treatment intensity is clearly continuous rather than binary. I also have data on many pre and many post periods.

In a binary treatment intensity setup, the standard procedure for assessing the plausibility of assuming that absent treatment trends of the two groups would have been parallel seems to be to plot trends for the two groups for several pre-treatment periods ("pre trends analysis"). However, with a continuous treatment the number of treatment groups is not two, but is essentially up to as large as the number of units.

I am trying to figure out what is the best practice to examine the parallel trends assumption in this setup. While I see a lot of material on having more than 2 periods, I have not found good material on having more than 2 treatment levels.

Who knows more about this?

Thanks so much, PM