cqcampos
commented
5 months ago Some background on the intervention follows.
- Students enrolled in second, fifth, and eighth-grade in Fall 2022 are part of the experiment. These are transition years.
- We randomize three school-level treatments, peer quality, school quality, or both. Schools correspond to the students' enrolled school as of Fall 2022.
- The treatment assigned is within school-level size and achievement terciles. Therefore, any experimental analysis conditions on this.
The baseline specification is the following:
$$ Yi = \alpha{b(i)} + \betaP T{Pi} + \betaS T{Si} + \betaB T{Bi} + \gamma' X_i + ei $$ where $\alpha{b(i)}$ are randomization block indicators, $T_{Xi}$ is a treatment indicator for treatment $X$, $X_i$ is a vector of student baseline controls.
To increase precision, we're going to estimate diff-in-diff analogs of this RCT. Let me know if you have any questions about doing that. But let's go with the diff-in-diff analogs for all of the experimental analysis, except balance checks.
On Mercury, work from the infoRCT folder within the lausd project folder. I created a dataset called info_rct_analysis.dta which you can work from. The code for this is in 0_build_infoRCT.do. Make sure to go through it to understand the data structure. I imagine this will be an input to the structural analysis, so it's good for you to know how it was created. Note that to do the diff-in-diff analogs of the RCT, you'll need to work with data from the main build so you need to combine that data with these data I created over the weekend.
The script you will work on should be called 1_analysis_infoRCT.do. It's okay to create different scratch scripts but everything that is final should be incorporated into the 1_analysis_infoRCT.do script and we must ensure it runs.
Please note that the deadline for this task is Friday (1/19) but could perhaps be completed earlier.
Reduced form analysis of the intervention for the following outcomes:
I will add more detail later.
┆Issue is synchronized with this Asana task by Unito