# Stage 3: Likelihood survey and estimation of coverage

Once the prior mode has been finalised and its shape parameters entered into the Bayes calculator, a recommended sample size will be generated. This figure is the recommended minimum number of acutely malnourished children which need to be found during the likelihood survey to achieve the desired level of confidence in the posterior, or the overall coverage estimate.

This section outlines how to plan and carry out the likelihood survey, how to generate the final coverage estimate and how to analyse and present the data gathered on the barriers to access.

# 1. Planning and undertaking the likelihood survey:

The likelihood survey will usually use a two stage sampling procedure: Selection of villages to sample and in-community sampling.

A. Selection of villages to sample:

Based on the sample size generated by the Bayes calculator (i.e. the number of cases of acute malnutrition that need to be identified), the following formula can be used to calculate the number of villages or communities to visit (n=recommended sample size):

For example, if:
• Target sample size (n) = 41
• Average village population (all ages) = 600
• Prevalence of SAM = 1%
• Percentage of children aged between 6 and 59 months = 20%

The minimum number of villages to be sampled would be:

With this information, the surveyors need to select the villages they need to sample in a random way. There are two methods which can be employed to do this:

• Centric Systematic Area Sampling method: this involves taking a spatially stratified sample of the health district being investigated by drawing a grid of equally sized squares (‘quadrats’) over the area to be sampled and sampling the community or communities located closest to the centre of each square. This method requires a large printed map of the health district marked with every known village or community. More information is available here.
• Spatially stratified sampling method: if no map is available containing all known villages or communities, this method can be used to select the villages. All villages or communities need to be listed by health centre catchment area. More information is available here.

With the finalised list of villages to sample, the survey leader can formulate a sampling plan with the logistics team. The Team composition sheet (Tools) can be used to allocate responsibilities for each day of the sampling.

B. In-community case finding:

Once in the communities, survey teams can use two in-community case finding techniques to locate cases in each village:

• Active and adaptive case finding
• Door-to-door case finding

These are the same techniques which are used in Stage 2 to test the hypothesis. Here is a guide on how to carry out each.

The findings of the interviews with carers of acutely malnourished children and results of the sampling should be recorded on forms. Templates for these are available in Tools.

# 2. Actions on completion of case finding:

The survey team should complete the in-community sampling in ALL of the villages and communities which they set out to sample at the start of the likelihood survey; even if they reach the required sample size (n) before completing sampling in all villages.

If, during the in-community sampling, the team finds MORE than the required sample size, then this is fine.

However it they do not succeed in finding the required sample size then they have a number of options:

• They can randomly select more villages to sample until they reach the required sample size (as many more as they deem necessary).
• They can change the coverage estimator from Point to Period coverage (see below).
• They can adjust the Precision on the Bayes calculator (see below).

# 3. Completing the conjugate analysis:

When analysed, the interview forms from interviews with carers of acutely malnourished children will provide surveyors with the following information:

• Number of current cases attending the program (Covered cases)
• Number of current cases NOT not attending the programme (Non-covered cases)
• Number of recovering attending the programme (children who are no longer acutely malnourished but who are still attending the programme because they have not yet met programme discharge criteria - Recovering cases)
• Number of current cases (both in the programme and not in the programme)

With this information coverage, the survey team can calculate the numerator and denominator for the Single coverage estimator. The single coverage estimator is a new estimator (designed to replace the formerly used Point and Period coverage estimators).

More information is available here about the Single coverage estimator. The Single coverage estimator can also be used during SQUEAC assessments which are assessing the coverage of MAM. This article explains more.

The CMN has also developed a calculator to help surveyors calculate the Single coverage numerator and denominator. More information is available here.

Surveyors can then go back to the Bayes calculator (having previously entered the shape parameters for the Prior), adjust the numerator and denominator scales, click "Use survey data" and generate the posterior (see Image 1).

Image 1: Screen shot of Bayes calculator

# 4. Additional analysis of data from likelihood survey:

The completed interview forms from carers of children NOT covered by the programme, will indicate the main reason why the carer had not enrolled his or her child.

The reasons stated should be tallied and presented in a chart. For example:

Figure 2: Example of chart summarising reasons for non-attendance identified during likelihood survey.

Team composition sheet

Case finding procedure

Questionnaire for covered cases

Questionnaire for non-covered cases

Active case finding data collection form