Q1. Given this information, what would be the overall sample size?

Ans1: 105,600 completed interview in all 100 districts

Q2. How many households would have to be visited?

Ans 2. 621,177 households

Q3.What do you think of this estimated sample size?

Ans 3. This is very big sample size which will yield a good coverage estimate (with the low margin of error at 5%) at the granular level (district). Time and cost may be high but this is a first steps as Harmonica requested very granular level coverage estimate at the district level.

Q4. How feasible will it be to conduct this survey?

Ans 4. This survey will require a lot of resources at district level therefore cost will be very high. Large number stratum will escalate the cost of the survey. We can reduce the stratum to province level or relax the precision level from 5% to 10% to decrease the ESS. The sample size at district level needs to be adjusted to accommodate population density as per district.

Q5. What are the trade-offs in terms of time, money, and quality of survey implementation?

Ans 5. Cluster sampling preferred over random sampling for the cost consideration. But the number of stratum impact the survey cost. Each stratum needs to have its own cluster sampling. A large number of strata may have serious cost and time implication on survey result and may not increase overall (aggregated at province or national level) precision and may need high sampling and data collection efforts.

We can reduce the strata to reduce the cost and duration of the survey we can also consider or relax the precision level from 5% to 7% or 10%.

We can also consider the grouping the districts (instead of 100 district, group them in 35 districts) with based on the social and spatial characteristic of the population. This way we will still achieve the same result with great granularity

If we want to reduce the number of teams we can increase the number of households sampled per cluster (m). This will result in less number of clusters per stratum.