Wald Estimator for Difference-in-difference

[Nicolas Orgeira]

Hi,

I am trying to run a Wald DID on my panel data and have been trying to figure out how to code it. However, in D'Haultfoeuille et de Chaisemartin's paper Fuzzy Difference-in-differences (see website for paper and slides), they mention that Wald-DID for an outcome variable Y is defined as

Wald-DID: DID on Y divided by DID on D, where:

G - dummy variable for control/treatment group
T - dummy variable for time periods
D - treatment (I assume an instrument indicating whether respondents actually received the treatment)

As such, the stata code would then be:
regress outcome_var G##T
regress D G##T

then divide the interaction coefficient in the first regression with the interaction in the second coefficient

In their slides (p.3-4) they also define it as the 2SLS of Y on G, T and D, with D instrumented by GxT, as Duflo used in her 2001 paper Schooling and Labor Market Consequences of School Construction in Indonesia.
The code would then be:

did=GxT
ivregress 2sls outcome_var (did=D G T)

However, I am getting different coefficients (see attached). Has anyone ever used Wald-DID to analyse their data?

Thank you

[Aroni Barkat]

Hey I have been loolking at the same two papers and kind of confused about one thing. Duflo did the DID without a panel dataset right? I mean since she used the cross-sectional survey for individuals with years of education and matched it with area-level data on treatment intensity. But the main data set is not panel, but a cross sectional one. Right?

Sorry I can't provide helpful insights about your question but since I am going through the papers now, I will definitely get back if I get to your level! Thanks!