Dear Stata members,
For my project, I am analysing the effect of a policy change on test scores using panel data. The setting is like this: There was a public contest to determine which schools were going to benefit from the policy. Schools were ranked and the best x schools were approved to be affected by the policy. As the treatment wasn't uniform across schools, I have a panel consisting of each school and its (average) test score over 5 years, the time that took the policy to be fully implemented, with dummies showing when each school was "treated". So far so good.
As the schools that won the contest are likely different from those that lost, I want to compare those schools that just barely won the contest (i.e. those just above the cutoff) vs. those that lost by only a bit (i.e. those just below the cutoff), a setting similar to a Regression Discontinuity Design.
Now my problem is choosing the bandwidth to decide which schools to compare. Obviously I could say that I am going to compare the bottom 10 schools that won the contest vs. the next 10 schools that didn't win, but that isn't very rigurous. I have been trying to use the rdbwselect command in Stata, but this doesn't seem to be exactly what I need, as I just need to find an optimal bandwidth and then run my panel regression (test scores on when the school was "treated"). From my understanding, rdbwselect finds an optimal bandwidth assuming that one is going to regress test scores on the schools' rank.
Any help or advice would be greatly appreciated!!
With kind regards,
Martin
For my project, I am analysing the effect of a policy change on test scores using panel data. The setting is like this: There was a public contest to determine which schools were going to benefit from the policy. Schools were ranked and the best x schools were approved to be affected by the policy. As the treatment wasn't uniform across schools, I have a panel consisting of each school and its (average) test score over 5 years, the time that took the policy to be fully implemented, with dummies showing when each school was "treated". So far so good.
As the schools that won the contest are likely different from those that lost, I want to compare those schools that just barely won the contest (i.e. those just above the cutoff) vs. those that lost by only a bit (i.e. those just below the cutoff), a setting similar to a Regression Discontinuity Design.
Now my problem is choosing the bandwidth to decide which schools to compare. Obviously I could say that I am going to compare the bottom 10 schools that won the contest vs. the next 10 schools that didn't win, but that isn't very rigurous. I have been trying to use the rdbwselect command in Stata, but this doesn't seem to be exactly what I need, as I just need to find an optimal bandwidth and then run my panel regression (test scores on when the school was "treated"). From my understanding, rdbwselect finds an optimal bandwidth assuming that one is going to regress test scores on the schools' rank.
Any help or advice would be greatly appreciated!!
With kind regards,
Martin
Comment