The simple answer is “Yes”. There are a couple of options: (1) Continue as you describe (i.e. select HH, apply the mortality questionnaire, if children between 6 and 59 months then do the anthropometry) until you have collected the sample you need. This has the advantage of simplicity. It also increases the mortality survey sample size (this is a good thing as these surveys often lack precision due to high design effects if mortality is clustered). (2) Stop the mortality survey and continue the anthropometry survey selecting only HHs with eligible children until you have you sample size. This might be a little less work but is a little more complicated to do in the field. It looks like you understand the following point already ... I think it worth repeating ... be careful of doing the mortality and anthropometry surveys together. If you only do the mortality in households with children aged between 6 and 59 months (I have seen this) then your survey will underestimate mortality by not selecting HHs in which all children have died and by limiting the sample to HHs with adults of reproductive age (i.e. exclude the elderly). I hope this helps.

Dear Mark, Will it cause increased bias if we continue to select households only with eligible 6-59 months children for anthropometric assessment? Will this HH be selected randomly or purposively? As per SMART methodology, replacement of households or children are not recommended though adequate number of children or HHs are not found in any cluster. Generally, we calculate sample size for Anthropometry and Mortality using SMART and highest sample size is taken as the sample size for both mortality and anthropometry survey and required number of clusters are estimated on the basis of number of households can be visited each day. each day after arrival in assigned cluster, defined number of HH is selected randomly and data is collected only from those identified households. Replacement of households or children is not done.

I may be a little confused . . . I do not think it will add any bias (other than what is already there in the SMART method) to the anthropometric survey because you only measure children in HHs with eligible children anyway. This means that you effectively skip the HHs without eligible children. It seems to me that you are doing something other than a typical SMART survey. In a typical SMART survey (two-stage cluster sample with PPS in the first stage and proximity sampling in the second stage) the only random selection is in selecting the starting point for the PPS sample and the starting household in each cluster. Subsequent HHs are usually selected by a proximity rule. All you would need to do is continue what you were already doing (e.g. take the nearest HH to the right and repeat) until you have your quota. You appear to have a different design to this. The advantage of the method outlined above (known as “modified EPI”) is that it does not run into the problems you are facing now (see below about quota sampling). How do you sample HHs randomly? That is quite a difficult thing to do. A typical SMART survey uses a quota sample. You stop sampling in a cluster when you meet your quota. The main reason to stop sampling before you meet your quota is if the village is too small to hold the quota. I think it a mistake to define quotas in terms of HH numbers. The quotas should be people sampled (these are the denominators in our estimators). For anthropometry this is children aged between 6 and 59 months. For mortality this is something like all people living in the sampled HHs during the recall period. It can be useful to use a HHs / day calculation for planning purposes but it is a mistake to use them to define sampling quotas unless you are interested in household level data such as WASH indicators or household dietary diversity. I do not know why you do not use replacement HHs. This is standard practice for quota sampling.