Hi, Number of children found during small area surveys should not be added to the number of children found during wide area survey because it will give biased results. Small area survey is mainly done to test the hypothesis as per finding of stage 1. But in wide area survey, sampling is done by representing whole study area. Thanks

Sanjay is right; during the small area survey (step 2) you « purposely » chose villages with low coverage an « purposely » chose villages with high coverage and this to test homogeneity/heterogeneity of coverage across your area. Therefore, using this data will introduce a bias. In your case, because the investigation is already done, you can do nothing and you cannot use the small area data. However, in some cases (e.g. very small surveyed area, very low prevalence) you may have used the small area data IF the same village(s) was/were again selected during the random sampling of step 3. In this case, data from the randomly selected village(s) can be used (since, of course, you will not go out again and do active case finding in the same village). In the reality, unless obliged to do so because of the specific conditions mentioned above, you will not use the “same” village even if they are sampled during random sampling (just pick up the next village on your sampling list) because the more villages you visit, the more change you have to identify uncovered children and treat them. Hope this helps. Lio

Combining data from small-area and wide-area surveys is not a proper thing to do. There are a couple of reasons why this is so: (1) In a Bayesian analysis (typical of a wide-area survey in SQUEAC) we will have already used the findings from the small-area survey(s) to inform the prior. In the conjugate analysis we combine the prior and likelihood. This means that we already use the results from the small-area survey(s) in the analysis of the wide-area survey data. To use the data from the small-area survey(s) in the likelihood would be double-counting. This increases the nominal sample size without increasing the effective sample size. The result of the wide-area survey would appear to have better precision (i.e. a narrower 95% CI) than is warranted. (2) Small-area surveys investigate hypotheses about coverage. The sample that is taken is purposively biased to test a hypothesis. The small-area samples will be taken (mostly) from areas / populations that are "special" in some way. If you mix this sample with a wide-area sample then you risk introducing a bias to the overall coverage estimate. The problem is that the small-area survey samples are not intended to be representative of the wider-area (and is unlikely to be so). This is Sanjay's point. Issues (1) and (2) can combine to give an inaccurate (biased) and spuriously precise estimate. Even if the bias from (2) is not very large the effect of (1) could be that the 95% CI does not include the true value for coverage. There is, as Lio points out, a special case. This is when the wide-area sample selects communities already visited by the small-area sample. We need not sample these villages again. To do so may introduce an upwards bias as we would hope that all of the cases found in the small-area survey had been referred and some or all had attended the CMAM program. I am not happy with this approach (i.e. including the small-area sample) as it introduces double counting (see (1) above) so I would usually sample an alternate village. If using a grid (CSAS) sample or list-based systematic sample I would take the next nearest village that had not already been sampled. This is Lio's point. I hope this helps.

Thank you all for the information. It clarifies my concerns