You can do stratified random sampling only if you have a list of all the Basic Sampling Units (BSU) - i.e. households (or children < 24 months) - in the survey area. By the way, if you do have this list, you can just do a survey using simple random sampling - just randomly select a sample using this list and do your survey (I presume you only need an overall estimate for the survey areas as you are thinking about stratification only because of high design effect?) In any case, note that stratification refers to putting the survey population into mutually exclusive and exhaustive groups (called stratum). You will still need to use random or cluster sampling method to select samples from different strata. If you do not have the list of all the BSU in the survey area you will still need to use cluster sampling method to select sample from each strata. Thus, you will still need to account for design effect. Would it be possible to know what your analysis plan is and what computer software you are planning to use for the analysis? Thank you.

The total general population am working with is about 12,000 according to census data and focus is on 6-23months. I am trying to see if cluster sampling can be totally avoided since according to literature it is a minimum of 25 clusters and an inclusion of a design effect which increases sample size (main challenge). My thoughts are…If you have total number of household and total number of population from census data, is it possible to segregate the population into two groups (strata) to calculate sample size using simple random sampling (segregation so as to take care of the differences). Following this, villages are randomly sampled to fit into the two groups based on PPS method. Households in selected areas(villages) are enumerated to get the exact number and following this, simple random sampling is applied inorder to determine which households to be visited. Is this possible? The software available for analysis is called STATISTICA.

You would still have design effects because you would very likely use a cluster sample within each stratum. You would have a "stratified-cluster" sample. You may have some sample size savings but (I guess) this will not be translated into cost savings as you'd still have a lot of clusters to sample if you wanted useful precision on within-stratum estimates (see below). You would have many small clusters. This is good for precision but (in my experience) a lot of the costs of a survey are in getting to the clusters rather than sampling within clusters. The cluster sampling method has some advantages. It does not require complete lists of households or individuals (in you case the list will be out of date in a single day as children are born or reach their second birthday). It is also (usually) the cheapest method per individual sampled. If you believe that there are two populations (i.e. with greatly differing IYCF practises) then you really need to sample these separately. One overall figure for both will not be of much value (unless they are similar). If you do not have a large sample from each population then you will have poor precision for the stratum-specific estimates. I don't think you will get a cost saving. I think you may have increased costs. You will need to use specialist survey software (e.g. SUDAAN, SPSS Complex Samples, STATA svy* commands, CSAMPLE in EpiInfo). I am not familiar with STATISTICA but a quick web search suggests that this software does not have built-in facilities for handling complex samples. If your pastoralists are transhumant then PPS cluster sampling will be a poor choice since troupe populations and locations are unstable. The PSM method would probably be more appropriate. You can reduce design effects by having many small clusters but this increases costs. You can also reduce design effects by having a more representative (i.e. than proximity sampling) within-cluster sampling method such as map-segment-sample. BTW ... You will need to sample the 0-6 month age group for the exclusive breastfeeding indicator. I hope this is of some use.

Thanks for this. Has given me some good insights and direction.