Difference-in-Difference, with three treatment groups and one control group

[Guest]

Hello everyone,

I am trying to do the analysis with diff-in-diff estimation method and bumped into the question,

I have four groups for estimation. Group 1 - control group, Group 2 - agricultural extension, Group 3 - marketing extension, Group 4 - agricultural and marketing extension services.
All treatment variables are dummy, 1- received the treatment (extension service), 0- did not receive.
I want to see the program (extension service) effect/impact through difference-in-difference method on the "adoption of a new technology" (dummy variable 1-adopted, 0-not adopted) and "adoption intensity" (takes value from 0 to infinity). I have a two period panel data set.

I had only done before the common binary treatment case of a group having the treatment and group having no treatment.
How then I could specify the regression framework for such case?
Do I have to split the full panel into 3 different panels, for each treatment?
Any helpful material. I have been browsing in google but can't find one.


Thank you in advance.

[Clyde Schechter]

Well, it sounds to me like you actually have not three treatment groups and a control but a 2x2 factorial design. There are two treatments: agricultural extension and marketing extension, and each entity in your study receives neither, one or the other, or both. You say you have a two period panel data set, so I assume the two periods are one pre and one post. So first you need a 0/1 variable for agricultural extension and another 0/1 variable for marketing extension. Then your DID analysis would look like:

// PERHAPS ADD COVARIATES AND OPTIONS TO THIS ANALYLSIS
panel_regression_command outcome_variable i.pre_post##i.agricultural##i.marketing, fe

The DID estimator of the agricultural extension alone will be the coefficient of 1.pre_post#1.agricultural. For the marketing extension it will be the coefficient of 1.pre_post#1.marketing. For the combined effect of receiving both treatments, the effect will be estimated by 1.pre_post#1.agricultural + 1.pre_post#1.marketing + 1.pre_post#1.agricultural#1.marketing (which you can evaluate with -lincom-.)

I do not understand your explanation of adoption intensity, so I don't know how it would fit in here.