|
Field
Experimental Design (Public)
|
Before
To increase statistical power and to warrant balance across treatment and control units, we will conduct a stratified randomization. After the baseline test, within each district, baseline test scores will be used to create quadruplets of Gram Panchayats with similar academic performance. Thereafter, for each of these strata, two GPs will be selected to participate in the GKA program, while the other two GPs will remain as a “control”. Thus, 49 GPs and their selected schools will be assigned to receiving the program; the remaining 49 GPs and their selected schools will continue with “business-as-usual”. We repeat the above-mentioned randomization procedure ten times, to select the randomization with greatest balance. To this end, we select a vector of covariates – from India’s District Information System for Education (DISE) – that are predictive of baseline scores. Thereafter, we calculate t-statistics for each of the selected variables as well as the baseline score. We do so by estimating regressions of each characteristic on the treatment indicator and strata fixed effects. We then store away the most extreme of these t-statistics, and select the randomization where this value is smallest.
|
After
To increase statistical power and to warrant balance across treatment and control units, we will conduct a stratified randomization. After the baseline test, within each district, baseline test scores will be used to create quadruplets of Gram Panchayats with similar academic performance. Thereafter, for each of these strata, two GPs will be selected to participate in the GKA program, while the other two GPs will remain as a “control”. Thus, 49 GPs and their selected schools will be assigned to receiving the program; the remaining 49 GPs and their selected schools will continue with “business-as-usual”. We repeat the above-mentioned randomization procedure ten times, to select the randomization with greatest balance. To this end, we select a vector of covariates – from India’s District Information System for Education (DISE) – that are predictive of baseline scores. Thereafter, we calculate t-statistics for each of the selected variables as well as the baseline score. We do so by estimating regressions of each characteristic on the treatment indicator and strata fixed effects. We then store away the most extreme of these t-statistics, and select the randomization where this value is smallest.
After selecting the randomization of schools with the greatest balance, we randomized all of the 49 treatment pairs into two arms: one group of GPs with community events (24), and one group of GPs without the community events (25). Both treatment arms continue to get the kits and related training. All pairs of control GPs remained untouched. This randomization for the GP contests took place in July, 2019.
|