Experimental Impact Evaluation Design – Quiz 6

1. Suppose that you are designing an evaluation, and your population of interest consists of households that are grouped into villages. Your design options include stratifying or clustering by village, or neither. Which of the following statements about random assignment are true? (Choose all that apply)

Suppose that the EG program is going to be implemented at the village-level, and you need to conduct random assignment using the EG baseline data. Suppose that the program needs enough SC/ST and non-SC/ST students to run, so you must have at least 10 SC/ST and 10 non-SC/ST students per village. Drop all other villages from the sampling frame.

2. How many SC/ST students are left in your potential evaluation sample?

3. How many villages are left in your evaluation sample?

Now assign 50% of remaining villages to the treatment group and 50% to the control group. Within each village, randomly sample 10 SC/ST students and 10 non-SC/ST students.

4. Conduct a balance test comparing the fraction of students who are SC/ST in your treatment sample versus your control sample. What is the difference between groups? Round to the nearest 0.01

5. Which of the following statements about this random allocation is true? (Choose all that apply)

6. Given this evaluation design, what is the minimal detectable effect size for treatment effects on test scores (in standard deviations) among SC/ST students, at 80% power and 5% significance? Assume that intra-village correlation in SC/ST endline test score will be equivalent to intra-village correlation in SC/ST baseline test scores in these villages (the ones that had at least 10 SC/ST students & 10 non-SC/ST students). Ignore any power gains from controlling for baseline scores or other covariates. Round your answer to the nearest 0.01 SD.