Using xtdidregress for diff-in-diff model

[Federico Stravo]

Hi everyone. I have panel data on US schools from 2008 to 2017. In 2013, a bill was passed and some schools have received benefits from it. Now, I want to estimate the effects of this bill on maths scores with a diff-in-diff model. The dataset looks like this:

year   id  treatment  maths_score  
2008   10      0          2.011
2009   10      0          2.017
2010   10      0          2.435
2011   10      0          2.876
2012   10      0          2.888
2013   10      0          2.921
2014   10      0          3.001
2015   10      0          3.089
2016   10      0          3.110
2017   10      0          3.221
2008   20      1          2.425
2009   20      1          2.552
2010   20      1          2.598
2011   20      1          2.655
2012   20      1          2.701
2013   20      1          3.012
2014   20      1          3.114
2015   20      1          3.302
2016   20      1          3.556
2017   20      1          3.773

And the model is: maths_score = i.id + i.year + beta_t * treatment.
Important: notice that the treatment variable is equal to 1 for all the treated schools even before 2013, the year in which the bill was passed.

Now, I want to estimate the beta_t for each year from 2008 to 2017 and I am trying to use the code

xtdidregress (maths_score b2013i.year#treatment) (treatment), group(id) time(year) aequations allbaselevels vce(cluster id)

to do so, in which I take 2013 as the base variable. The problem is that, if I do this, Stata says that the model is not identified and the treatment variable treat was omitted because of collinearity. I think it is because treatment = 1 for each year from 2008 to 2017, so if I include both the year and the treatment variable, I have a problem of multicollinearity. But if I do

replace treatment = 0 if year < 2013

have solved the multicollinearity problem, but I can no longer compute the beta_t for all the years prior to 2013.

I have also tried

regress maths_score ib2013.year##i.treatment i.id

Which returns an output, but I am not sure whether this is the correct specification.

So, my question is: how can I adapt this model using Stata so that I am able to correctly compute all the beta_t from 2008 and 2017, excluding 2013 which is my base category?

[Jared Greathouse]

Please report what happens when you use this syntax

xtdidregress (y x1 x2...) (treat), group(grpvar1) time(tvar)
/// wildbootstrap

substituting in the xs and ys with your variables.