Ghana - Ghana Living Standards Survey : 1987-1988, First round

The methodology that was used reflects the purpose of the survey. To balance the desire for a large, representative sample with the expense of a long, detailed survey instrument, a sample size of 3,200 households was selected. The households were to be chosen in such a manner that each household had an equal probability of being selected. At the same time, the logistics of locating the households and conducting all interviews within a specific time frame required that the households be grouped into "workloads" of 16 households each. A final concern was that all three of the country's ecological zones (coastal, forest and savannah), and each of urban, semi-urban and rural areas (population greater than 5000, 1500 to 5000, and less than 1500, respectively) form the same proportion in the sample as they do in the national population.

To achieve the three objectives simultaneously, a stratified selection process was used. For the 1984 Census, all of Ghana was divided into approximately 13,000 enumeration areas (EAs). From this list it was determined what proportion of the 200 GLSS workloads should be selected from each of the nine zone/urban categories. Two hundred sampling areas were then selected from the enumeration areas in the sub-divided list. For each enumeration area, the probability of being selected was proportional to the number of households contained in that area.

After the 200 sampling areas were selected, households in those areas were enumerated in 1987. Therefore it was possible to take into account changes in the number of households and preserve the self-weighting nature of the sample. The 200 workloads were assigned among the 200 sampling areas with probability equal to the number of households in that area in 1987 divided by the number of households in that area in 1984 and multiplied by the total number of households in 1984 divided by the total number of households in 1987. That is, sampling areas that had greater than average increases in size had a greater than one chance of being selected. Thus, each sampling area was assigned zero, one, two, or even three workloads of sixteen households. The households (sixteen selected and four replacement for each workload) were then chosen randomly from the household list for each sampling area. The resulting list is 3200 households and 800 replacement households in something less than 200 sampling areas (specifically 178 in 1987-88 and 170 in 1988-89). Each group of 16, 32 or 48 households within a sampling area is referred to as a cluster in the GLSS data sets and in this document.


Weights are not computed. This is because there were not enough informations on how the weights were imputed against the variables.