Comprehensive Food Security and Vulnerability Analysis Assessment 2008

The CFSVA sampling strategy aimed at providing sufficiently precise estimates of several key food security indicators for all rural regions, as well as Urban Accra, and all other urban areas together in one domain.

As a CFSVA aims to provide estimates of many different indicators, no single indicator can guide sample size requirements. Therefore, when calculating minimal sample size, an assumed prevalence of 50% was used, this yields the largest sample sized for a required precision. A design effect of 2 was assumed (food security indicators typically used in similar CFSVAs usually have design effects ranging from near 1 to 4). 95% confidence is always used, with intervals of +/- 8%. Following the standard sample size calculation for estimating prevalences, this yields a sample of approximately 260 households per domain. Where possible, larger samples were taken to increase precision.

Due to the time and cost limitations of drawing a new accurate sample, it was decided to 'piggy-back' on the existing sample already drawn for the on-going DHS survey. The DHS sample is a 2-stage cluster sample, with the following sample distribution:

Ghana DHS 2008 cluster numbers by region (urban and rural)
Region Urban Rural Total
Western 15 24 39
Central 13 21 34
Gt. Accra 53 7 60
Volta 10 25 35
Eastern 16 27 43
Ashanti 36 31 67
Brong Ahafo 16 22 38
Northern 11 27 38
Upper East 5 23 28
Upper West 7 23 30
Total 182 230 412

The clusters used throughout the country are 'Enumeration Areas', or EAs. At the time of the CFSVA, Ghana Statistical Services had already recently drawn up this sample, complete with household listings for each selected EA (cluster). Rather than draw a new sample, a sub-sample of this was taken for the CFSVA. In this sub-sample, all rural clusters selected for the DHS were maintained in the CFSVA sample, and a sub-set of urban clusters were randomly selected from the DHS sample in each of the regions for inclusion in the CFSVA (thus maintaining the PPS selection of clusters). This resulted in a non-self-weighting sample, so probability weights were used in analysis to account for this.

Once the clusters were selected this way, it was decided to select 12 households per cluster to allow for sufficient total sample size per domain, while allowing for enumerator teams (consisting of 1 team leader and 4 enumerators) to complete, on average, one cluster per day. This also yielded at least 260 households per strata (or just under, and with the exception of rural Accra). As GSS has previously conducted a complete listing for all EAs in the DHS, they were able to randomly select 12 households per cluster (with 3 additional replacement households if one or more of the 12 were unavailable). The enumerator teams were supplied with maps of the EAs and the locations of the households and the names of the household heads.

This resulted in the final sample, by domain:

Domain Clusters Households (planned) Households (actually sampled) Number of these that were replacement households
Western Rural 24 288 288 38
Central Rural 21 252 252 37
Gt. Accra Rural 7 84 84 19
Volta rural 25 300 299 56
Eastern rural 27 324 324 55
Ashanti rural 31 372 372 59
Brong Ahafo rural 22 264 264 38
Northern rural 27 324 324 43
Upper East rural 23 276 276 27
Upper West rural 23 276 276 17
Urban Accra 42 504 504 129
Urban Other 49 588 588 103
Total 321 3852 3851 621

It should be noted here that Rural Accra has well below the goal of 260 households for that strata. Due to the very small rural population in that Region, and the fact that the DHS only had 7 rural clusters selected, it was decided that a sample yielding very low precision would be acceptable for that region.

In each cluster, an attempt to give the households an advance notice was made wherever possible, particularly in urban areas. Additionally, enumerators were instructed to make multiple re-visits (within the same day of visit to the EA) in order to try to capture the selected households. However, as can be noted from the table above, there was a high number of replacement households (overall, 16% of the sample) particularly in urban areas (21% of urban households and 14% of rural households). This could result in some bias if the absent/unreachable households were different than the randomly selected replacements.

Additional geographic reporting strata included urban/rural, and ecological zone. A probability weights were used in the analysis to compensate for the unequal selection probabilities throughout the country, the sample was representative for any geographic division, and the sample size allowed for sufficient precision within these alternative stratifications.

For nutrition indicators of children under 5, it was determined that the sample size would be too low to yield sufficiently precise estimates. Therefore, it was decided to aggregate the nutrition estimates at the zonal level and urban/rural for the majority of the analyses. In all households selected, all children under 5 were selected to be weighted and measured, as well as all women between 15 to 49 years old (pregnant women were not weighed and measured). Further discussion of the nutrition sample is presented in the nutrition data annex. For the community questionnaire, one questionnaire was administered in each EA (cluster).