If no sampling frame exists for your population of interest, best practice is to conduct a listing. If the sampling strategy has multiple stages (e.g. clustered sample designs), multiple listings may be needed.
Start at the lowest sampling unit for which you have full information. For example, if you have a list of all the villages (or other administrative unit), you can sample villages, and then within each sampled village conduct a listing of households, from which you can then sample. How much cost this will add depends on what scope geographic area you need to cover, and how the enumerators are getting around.
Example of household listing in Malawi: the research team was provided with a list of all 500-some sectors (level below district, above village) in the country from the national statistics office. We then sampled 100 sectors. We then requested a list of all villages for each of those 100 sectors. Survey teams visited each village, and listed all the households using a simple form we designed. They started by visiting the village head, who then assigned a couple of kids to show them the boundaries of the village in different directions, and they worked their way back inwards. Each enumerator wrote a number in chalk on the door of each household they listed to ensure that the same household was not visited twice. Each enumerator had a different starting number for their household list (e.g. Enumerator A started with 100, numbered the HHs he visited 101, 102, 103, etc) to avoid duplicating numbers. If no one was at the HH, they typically collected information from the neighbor (as we weren’t asking for anything complicated or sensitive). Once all households had been listed (one enumerator assigned to quality control went around towards the end of the day to make sure no HHs were left unnumbered) the supervisor collected all the lists, and aggregated in the evening. Using a randomly generated start number, and a skip number based on the desired sample size, he then marked the sampled households. We wanted 30 households per village, so for example if the total sample size was 90, the skip number would be 3. Given a random start number of 56, they would then select HH 59, 62, 65, 68, etc. All of this would be much simpler if done via CAPI, household IDs could be automatically assigned and each case sent to the server for easy aggregation, then actual sampling code run to automate selection. But all the above was done by paper.
Back to Parent
This article is part of the topic Preparing for Data Collection
Please add here related articles, including a brief description and link.