2SLS with staggered diff-in-diff

[Nay Iturbe]

I have a diff-in-diff model where the treatment is staggered and apparently endogenous. I want to use 2SLS to address the endogeneity problem but I am worried that my average treatment effect will be biased since the treatment is staggered. Can I use the next command using 2SLS when having a staggered setup?

 
 ivregress 2sls Y c.time (c.treatment c.time#c.treatment = c.instrument c.time#c.instrument), vce(cluster id)

Is this possible in a staggered setup? Or any suggestions on how to combine 2sls to csdid? Thanks

[Jeff Wooldridge]

Shall I assume there is no exit from treatment during the period? If that's true, what is the instrumental variable for treatment? Does it vary across time? Does the treatment variable above vary across time?

One thing is that you probably don't want c.time#treatment as that only allows the treatment effect to change as a function of a linear trend.

[Nay Iturbe]

There is no exit from the treatment, once the unit is treated it remains treated, but treatment is introduced at different periods of time for each unit. IV for treatment varies across time.