Comprehensive Food Security and Vulnerability Analysis 2004

Sampling of communities and households was done using the latest data of number of households by colline provided by ISTEEBU. A systematic random sample of collines was chosen, with their probability of being chosen proportional to the number of households in the colline. Once this was done, one sous-colline was chosen at random within the colline. Population data was not available at the sous-colline level; however, sous-collines within a colline are generally similar in size, so the simple random selection of one sous-colline in each selected colline was estimated not to have a significant biasing effect on the sample.

Within each selected sous-colline, enumerators conducted one community key informant interview and 10 household interviews. Households were randomly selected from a list of all households in the sous colline. When the household members were not present, the household was revisited later in the day. If no one was available, a replacement household was chosen at random from the list of households in the community.

For the maternal and child health and nutrition sections of the household survey, only women of reproductive age (15 to 49 years) and their children were eligible for inclusion in the sample. If there was more than one eligible woman in a household, only one was chosen at random, and all her children 0-59 months of age were included in the survey. Although this method may produce slightly biased results, the resources were not available to include all eligible women and children in a household. Additionally, the purpose of the study was not to produce precise national or sub-national estimates of the nutritional status of children, but rather to produce estimates that would provide information on the utilization aspect of household food security.

This method of household and sous-colline sampling produces a self-weighting sample, which facilitates analysis in that results can be produced nationally, provincially, by natural region, and where sample size allows, by commune. The overall sample size gives sufficient numbers in all natural zones, provinces, and many communes.

Although commune level results give the most information for programming within a province, providing accurate estimates for all the communes in the country (approx. 120) would require an extremely large overall sample size. This sampling methodology provides 10 to 70 household interviews per commune. In some communes, there were sufficient household interviews conducted to produce commune level estimates. However, where the sample size per commune was less than 40, groups of communes were clustered together to produce cluster-level estimates. Despite the clustering, the sample size per commune/commune cluster is still too small to produce statistically representative results and thus the findings should be interpreted with caution and should be used as general and comparative estimates rather than precise figures.