Question 1: Additional information needed. Answered as requested in Community updates, but repeated here.

“The programmatic threshold is given.
However, we still need to specify the following:
- Delta, ie the value upon or below which we consider that the calculated coverage is truly different from the threshold
- Direction
Annex B1 suggests that it is conservative to use direction below, ie we are interested in pinpointing districts with coverage lower than our threshold
- Alpha, the classical type 1 error. I will assume this is 5%
- Beta, the classical type II error, which is directly associated with power (=1- Beta). I will choose a power of 90%, which can be lowered to 80% if needed to reduce sample size.”

Question 0. Sample size for a survey over 10 provinces instead of 100 districts, assuming all other parameters remain the same as previously provided.

Given this information, what would be the overall sample size? N

- Using the data provided here, Guidance from Annex B1 and sample size calculator v2, we have:

A=10; B=264; C=4; D=5; E=1.2; m=10.

- The overall sample size is NA= AXBXC=10x264x4=10 560 interviews to be completed.

How many households would have to be visited?

NB=AXBXCXDXE=10x264x4x5*1.176=62 118 households

Question 2.

Calculating a sample size for the classification of the 10 provinces

Data provided:

Threshold: 90%
Direction: below
Delta: 10%
Power=90%
Beta: 1-power=10%
Alpha: 5%

Entering these as inputs in the sample size calculator v2, we get:

Number of interviews to be completed: 4 480
Total number of households: 26 353

Question 3. What decision would you make regarding the survey scope and objectives? What is your rationale for this decision?

Opting for classification instead of estimation in the 10 districts, the sample size goes from 10 560 to 4 480, which represents nearly 60% reduction in sample size. This will subsequently lead to a substantial reduction of costs. Moreover, the data will be used to estimate national coverage. Although this estimation will yield a less precise coverage compared to an estimation using the sample size of 10560, the reduction in cost is noteworthy. Therefore, I will opt for conducting a classification in the 10 provinces.