Falsification test for DID

[Alejandro Castillo]

Dear friends,
I have a concern regarding my diff-in-diff regressions on Stata. I am working on a diff-in-diff equation with a database similar to this:

year district_id treated
2005 01 0
2005 02 0
2005 03 0
2005 04 0
2006 01 0
2006 02 0
2006 03 0
2006 04 0
2007 01 0
2007 02 0
2007 03 1
2007 04 1
2008 01 0
2008 02 1
2008 03 1
2008 04 1

I'd like to point out that this is not a panel data, as the individuals are not the same on all years.

In the classical DID analysis I need an interaction between the treatment variable and a time variable which indicates the year when the treatment starts, but as you can see, some individuals can receive the treatment later than the other ones, starting from 2007. I'd appreciate if I can get suggestions as to how to include this multiple period treatment into my regression.

[Rhys Williams]

Hi Alejandro,

I think you are saying that you wish to do staggered difference-in-difference, given that your units receive treatment at different points in time.

Practically, you can implement this as normal and use a two-way fixed effects model such as:
Y = a + bTreatment + cTime + dIndividual + error

However, there has been a lot of theoretical literature in then last few years pointing out that the estimate for b when treatment is staggered is not well identified and may not provide the ATT you're after. I suggest you start by reading the following article and perhaps exploring these papers further.

Best,
Rhys