Dear all,
I have a question about the difference-in-difference (DID) model and its data structure. My data are at firm-year-level level, and the data looks like this:
year window reconstruction firm post adexp performance
1960 -2 1 A 0 0.25 0.23
1961 -1 1 A 0 0.25 0.34
1962 0 1 A 0 0.25 0.23
1963 0 1 A 0 0.25 0.11
1964 1 1 A 1 0.25 0.13
1965 2 1 A 1 0.25 0.56
1967 3 1 A 1 0.25 0.67
1977 -2 1 B 0 0.47 0.34
1978 -1 1 B 0 0.47 0.44
1979 0 1 B 0 0.47 0.45
1980 0 1 B 0 0.47 0.43
1981 0 1 B 1 0.47 0.44
1982 0 1 B 1 0.47 0.47
1967 1 1 B 1 0.47 0.87
My treatment:
I am interested in how the actions the firm does during the reconstruction (change the advertisement expenditure) affect the firm performance. some firms take actions during reconstruction while others were not. I set the year before the reconstruction begun as pre-period (t=-n, -n-1,…-1), year after the reconstruction finished as post-period (t=1,2,…n), and year/years during the reconstruction as t=0

My stata code is:

xi: xtreg performance c.adexp##i.post if (window<0) | (window>0) ,fe cluster(gvkey)

I have two questions:
  1. the time of reconstruction varies across firms. Some firms take 1 year, while some others take 3-5 years. As the treatment in traditional DID model usually happens in 1 year which is different from my case, is it OK for me to use DID?
  2. The actions a firm does during the reconstruction is a continuous variable in my case (e.g. the advertisement expenditure). I take the average adexp during window=0 and extend it to all years of that firm in the data. So adexp is not varies across all years in my case. Does it correct in terms of constructing continuous treatment in DID sample sturcture?
Thank you for your time and help ahead of time!

Kind regards,
Yang