Now that we've sampled our clusters, we're going to talk about how to sample households using what I'm calling the gold standard method. This is basically the method that is used in the Demographic and Health Surveys, the Multiple Indicator Cluster Surveys, and other rigorous household surveys. How do we sample households in a cluster? Again, if we're using census enumeration areas, the first thing you want to do is to obtain base maps. Once you have sampled your clusters, obtain base maps for the sampled clusters from the National Statistics Office. If you are not using census enumeration areas, you are going to need to create your own maps from scratch, which is one of the advantages of using EAs. Then once you have the base maps from the National Statistics Office, you will need to go to each cluster to update those maps. Again, those maps are from the most recent census. Even if that census was only a year ago, we have sometimes found important changes in the cluster. It's important for somebody to go to the cluster to update that map, first of all, and then to make a list of all the households in the cluster. That is your sampling frame. Generally, we send teams of two to do this. One of the members of that team is a mapper, a cartographer, the other member of the team is an enumerator. The cartographer is responsible for creating the map, the enumerator is responsible for listing the households, but they work together. They move through the cluster together. If you're using experienced mappers and enumerators, it requires approximately one day per cluster. A little bit later in this module, we're going to actually talk about how to do the mapping. But here, once you have done the mapping, once you have your list of households in the cluster, you can then sample households from that list using systematic random sampling, just as you did for clusters. In terms of how many households to sample per cluster, usually, we recommend sampling 30 households per cluster or less to avoid excessive clustering or design effects. Basically, if you sample a lot of households within a cluster, you will have more clustering, essentially less information that is collected from the cluster, and your variance is going to be larger. This is an example of a hand-drawn sketch map which you will get from some national institutes of statistic. You can see they have divided the cluster into different blocks which often correspond to different neighborhoods, they've marked off major landmarks, and then each structure is marked with a rectangle that is numbered. The numbering isn't haphazard. They start from a point and then move in a clockwise direction around the cluster. The reason for this then, part of the reason, is that when you develop your sampling frame and you do your systematic sampling, you list the households according to these structure numbers, and then you can make sure that your sample of households is dispersed throughout the cluster instead of all being clustered in one particular part of the cluster. This is an example of a map drawn using GPS, which you may also get from some national institutes of statistics if they are working with the appropriate technology. Here, again, you can see major landmarks that are featured, roads. Then each structure is denoted by a square, and next to the square is the structure number. This shows an example of a household listing. We've locked out the actual names of the household heads here so as to preserve confidentiality, but, essentially, each row here corresponds to a household. You can see data in second column from the left, the sequential numbering of households. Then in the third column from the left, you can see the structure numbers. You can see that many structures have more than one household in them. These are often compounds where you may have multiple households living in a compound. We also collect information about the address, or if there's no address, a description of where the structure is located, whether it's a residential structure, the sequential number of the household within the structure, and the name of the household head, and whether the structure is occupied, whether there's somebody living there. This is what sampling households in an Excel worksheet looks like. Again, as for sampling clusters, Excel is a really useful tool to automate sampling households. It's quite similar to sampling the clusters. Here, each row corresponds to a cluster. You can see that we've listed the number of enumerated households, the number of households that we want a sample, and then we have the sampling interval and the random starting point. The systematic sampling works just like it does for clusters, except here it is applied to households. Columns J onwards lists the number of sampled households.
