Ghana - Ghana Living Standard Survey 5: 2005, With Non-Farm Household Enterprise Module

Sampling Frame and Units
As in all probability sample surveys, it is important that each sampling unit in the surveyed population have a known, non-zero probability of selection. To achieve this, there has to be an appropriate list, or sampling frame of the primary sampling units (PSUs).The universe defined for the GLSS 5 survey is the population living within private households in Ghana. The institutional population (such as schools, hospitals etc), which represents a very small percentage in the 2000 Population and Housing Census (PHC), is excluded from the frame for the GLSS 5 survey.

The Ghana Statistical Service (GSS) maintains a complete list of census EAs, together with their respective population and number of households as well as maps, with well defined boundaries, of the EAs. . This information was used as the sampling frame for the GLSS 5. Specifically, the EAs were defined as the primary sampling units (PSUs), while the households within each EA constituted the secondary sampling units (SSUs).

Stratification
In order to take advantage of possible gains in precision and reliability of the survey estimates from stratification, the EAs were first stratified into the ten administrative regions. Within each region, the EAs were further sub-divided according to their rural and urban areas of location. The EAs were also classified according to ecological zones and inclusion of Accra (GAMA) so that the survey results could be presented according to the three ecological zones, namely 1) Coastal, 2) Forest, and 3) Northern Savannah, and for Accra.

Sample size and allocation
The number and allocation of sample EAs for the GLSS 5 depends on the type of estimates to be obtained from the survey and the corresponding precision required. It was decided to sample a total sample of 8000 households nationwide .

To ensure adequate numbers of complete interviews that will allow for reliable estimates at the various domains of interest, the GLSS 5 sample was designed to ensure that at least 400 households were selected from each region.

A two-stage stratified random sampling design was adopted. Initially, a total sample of 550 EAs was considered at the first stage of sampling, followed by a fixed take of 15 households per EA. The distribution of the selected EAs into the ten regions or strata was based on proportionate allocation using the population.

Under this sampling scheme, it was observed that the 400 households minimum requirement per region could be achieved in all the regions but not the Upper West Region. The proportionate allocation formula assigned only 17 EAs out of the 550 EAs


Under this sampling scheme, it was observed that the 400 households minimum requirement per region could be achieved in all the regions but not the Upper West Region. The proportionate allocation formula assigned only 17 EAs out of the 550 EAs nationwide and selecting 15 households per EA would have yielded only 255 households for the region. In order to surmount this problem, two options were considered: retaining the 17 EAs in the Upper West Region and increasing the number of selected households per EA from 15 to about 25, or increasing the number of selected EAs in the region from 17 to 27 and retaining the second stage sample of 15 households per EA.

The second option was adopted in view of the fact that it was more likely to provide smaller sampling errors for the separate domains of analysis. Based on this, the number of EAs in Upper East and the Upper West were adjusted from 27 and 17 to 40 and 34 respectively, bringing the total number of EAs to 580 and the number of households to 8,700.


At the end of the survey, 8,687 households were successfully interviewed representing a 99.85 percent response rate


The GLSS 5 is not a self-weighting sample design because disproportionately larger samples from regions with smaller populations were drawn. Therefore each sample household did not have the same chance of selection into the sample. Hence, weights were computed to reflect the different probabilities of selection in order to obtain the true contribution of each selected EA in the sample based on the first and second stage probabilities of selection.

The household weight variable is called WEIGHT which is attached to each section data
A grossing up factor (637.89) was used for the agriculture sections instead of the weight variable.