Difference-in-differences model with more than 2 time periods and changing treatment groups

[Anna John]

Dear Statalist community,

I am looking for help as I am not very experienced with STATA. I would like to conduct a programme impact evaluation by applying a difference-in-difference strategy.

I have a panel dataset for 4 years of household survey data comprising the years 2012-2015. My pre-treatment period is year 2012 – none of the households received treatment in this year.

For the following years (2013, 2014, 2015), all households could choose on a yearly basis whether to take-up treatment or not. If they took up treatment, they are part of the treatment group in the respective year. If they decided not to take up treatment, they belong to the control group in that respective year. (Some households who never got treatment are always in the control group.)

As a result, there are households that receive the treatment at different points in time and control and the treatment group changes each year.

So far I tried it with the following code.

xtset hhid year

xtreg outcome i.treatment##i.year, fe cl(hhid) // where treatment is a dummy equal to 1 if the household received treatment in the respective year

However, the output doesn't give me an average treatment effect for all years. Please advise what modifications to the code are necessary. I was reading through some similar posts in this forum but I am still unsure what kind of code would produce the correct impact estimate.

Thank you very much for your support.
Anna

[Phil Bromiley]

You need to remember that we are not necessarily from your discipline. ATE is readily available in most of the treatment estimators. I'm not sure off hand how to calculate it for a conventional d-i-d -- if you gave us what you wanted to calculate, we could help better. Following the FAQ on asking questions would also help - in addition to code, Stata output, and sample data make it easier for us to help you.