Method: manual calculation

Target population:+

children 12-23 months of age

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

Estimation of overall sample size needed if simple random sampling was used:

Using Table B-1, with precision for 95% CI and expected coverage of 85%, the effective sample size (ESS) is 265.

I have also tried to calculate the EES using the formula but the estimate sample size is much lower than the one derived from the Table B-1

n=((1.96)^2*p(1-p))/e^2 with p=85% and e=0.05 >> n=(3.8416*0.85*0.15)/(0.05^2)=195.918 approximately 196 children which is much lower than 265 children from the table.

DEFF=1+(m-1)*ICC=3.664 approximately equal 4

The number 265 needs to be inflated to consider clustering (cluster effect).

The design effect is the multiplier factor (4) ICC=1/3

* Ncs=100 (districts)*265 (EES)*4 (DEFF)=106000

The total completed interviews needed is 106000 to have the same precision as the estimate obtained from a simple random sampling.

*Using Ncs and the D (number of households to find eligible child) and (the inflation factor) the number of households to be visited can be estimated.

Nhh to visit=106000x5x1.18=625400

*The target number of households to visit in each stratum is 625400/100=6254

Alternatively, it is 265x4x5x1.18=6254 households to be visited in each stratum (district).

This can be done but logistically challenging if the infrastructure is not well implemented (example from district to district or from clusters to clusters.

*Number of clusters per stratum would be (265x4)/10=106

2) How many households would have to be visited?

*Total households to visit per cluster

5x1.18x10=59 households

The total number of clusters in the survey is 100x106=10600 clusters in the survey

3) What do you think of this estimated sample size?

The DEFF is quite high that we are using a large number of clusters.

The estimated sample size appears to be large and as sample size increases the 95% CI become narrow and the standard errors decrease.

Precise estimates can be obtained from the survey depending on the available financial and human resources.

4) How feasible will it be to conduct this survey?

Depending on the available resources, and good planning at the national and sub-national levels, this survey could be conducted to get precise estimates of the vaccine coverage. Furthermore, districts could be classified based on the estimated coverage.

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

Comparing the cost of conducting the survey using simple random sampling vs and cluster survey sampling will provide some useful information to guide the implementation.

Cluster survey is feasible (logistically and in term of saving money and obtaining high quality data).

The variance and the cost can be minimised by selecting few individuals per cluster.