Dear all,
I am referring to the following paper in this post:
INTERACTION TERMS IN POISSON AND LOG LINEAR REGRESSION MODELS - Shang - 2018 - Bulletin of Economic Research - Wiley Online Library
I would like to run a difference-in-difference using Poisson Pseudo-Maximum Likelihood.
Suppose the equation is: E(Y | X, d1, d2) = exp(X*B1 + B2*d1 + B3*d2 + B4*d1*d2)
Suppose d1 and d2 denote the treatment status and timing indicator, respectively. Then the difference in d2 goes first and the difference in d1 goes second (Athey and Imbens, 2006).
Following Ai and Norton (2003), the coefficient on the treated*post cannot be interpreted directly as a meaningful quantity. We must take the discrete double difference, which will correspond to a difference in semi-elasticities.
This will be equal to exp(B2 + B4) - exp(B2). So we would use
However, suppose we run the following command (from ssc):
We will only have B4 from the equation written above. Not B2. How do we proceed to compute the difference in semi-elasticities now?
I am referring to the following paper in this post:
INTERACTION TERMS IN POISSON AND LOG LINEAR REGRESSION MODELS - Shang - 2018 - Bulletin of Economic Research - Wiley Online Library
I would like to run a difference-in-difference using Poisson Pseudo-Maximum Likelihood.
Suppose the equation is: E(Y | X, d1, d2) = exp(X*B1 + B2*d1 + B3*d2 + B4*d1*d2)
Suppose d1 and d2 denote the treatment status and timing indicator, respectively. Then the difference in d2 goes first and the difference in d1 goes second (Athey and Imbens, 2006).
Following Ai and Norton (2003), the coefficient on the treated*post cannot be interpreted directly as a meaningful quantity. We must take the discrete double difference, which will correspond to a difference in semi-elasticities.
This will be equal to exp(B2 + B4) - exp(B2). So we would use
Code:
nlcom (exp(_[bX4]+_[bX2]) - exp(_b[X2])
Code:
ppmlhdfe Y d1*d2, abs(treated time) cluster(id)
Comment