Hi guys,
I'm trying to estimate the impact of a social program on schools' dropout rates. All my variables are at the school level, being the dependent variable the dropout rate in school i and year t. The independent variables include a dummy variable for the treated schools and a set of controls at the school level. I first estimate the TWFE (including year and school FEs) and then the Callaway Sant'Anna approach. Both would ideally return the ATT effect. The problem is that the treatment is not uniform across schools in the treatment group. In some schools 50% of students are exposed to the program, in other schools, 60%, so on and so forth. I do have a variable that measures the % of treated students in each school. Is it ok to include it as a covariate? Or would it be better to estimate the LATE?
Thanks a lot!
I'm trying to estimate the impact of a social program on schools' dropout rates. All my variables are at the school level, being the dependent variable the dropout rate in school i and year t. The independent variables include a dummy variable for the treated schools and a set of controls at the school level. I first estimate the TWFE (including year and school FEs) and then the Callaway Sant'Anna approach. Both would ideally return the ATT effect. The problem is that the treatment is not uniform across schools in the treatment group. In some schools 50% of students are exposed to the program, in other schools, 60%, so on and so forth. I do have a variable that measures the % of treated students in each school. Is it ok to include it as a covariate? Or would it be better to estimate the LATE?
Thanks a lot!
Comment