<< Previous    Next >>

Box 1.  Calculating demographic and health indicators for livestock production systems

To demonstrate how assigning geographic coordinates to survey data can enhance their use, we have obtained the geo-referenced 1998 Kenya Demographic and Health Survey data (NCPD, 1999) and calculated a number of population-based indicators for the three mixed-system categories in Kenya. Similar statistics could be calculated for the production system categories in WA.

     National household surveys like the DHS are based on a probability sample designed to provide valid estimates at the national and state or regional levels. They do not provide data for maps at the sampling level, for example health indicators by household, since there are too few observations to derive statistically valid estimates. However, geo-referencing surveys, that is assigning a latitude and longitude to a sampling point, can be used to:

•    Integrate internationally standardised surveys such as the DHS across countries for regional assessments. For example, this was done for the West Africa Spatial Analysis Prototype, a pilot effort to geo-reference DHS household clusters for 12 countries in WA  
  

•     Aggregate sampling points to new units of analysis, so long as corrections are made for differences in the probability of selection and a sufficient number of sample points are selected for each new unit of analysis.

 A DHS is a national sample survey designed to provide information on fertility, family planning and health. A core component of the survey involves interviewing a randomly selected group of women in the reproductive age group (15–49 years). Typically, the DHS sample is selected in two stages. First, a stratified random sample of enumeration areas (EAs) is chosen with probability of selection proportional to population size for each region and urban/rural area within the region. Second, a complete household listing is carried out in each EA from which a number of households are chosen at random. The number of households chosen is inversely proportional to the population size of the EA. In the Kenya DHS, each sampled EA or cluster includes between 7 and 22 households. The Kenya DHS used Global Positioning System (GPS) units to geo-reference the centroids of 271 survey clusters (4,310 households), about half of the 530 clusters surveyed nationally. They are depicted in Map B1 in an overlay with the map of livestock production systems. The sample design excluded approximately 4% of Kenya’s population in the arid regions of the country with no survey clusters being covered in the following seven districts: Garissa, Mandera and Wajir in Northeastern Province, Isiolo and Marsabit in Eastern Province, and Samburu and Turkana in Rift Valley Province (NCPD, 1999).
     The DHS collects information on important dimensions of human well-being, including housing characteristics, households assets, household-member characteristics, high-risk births and family planning, early childhood mortality, child nutrition and school enrolment. Though the DHS does not collect any information on household consumption or income, recent research has demonstrated the value of a household-assets index that can be used as a proxy measure for socio-economic status in the absence of income or consumption data (Gwatkin et al., 2000).
     The household-assets or wealth index is based on detailed information about household ownership and access to a variety of consumer goods and services. For each household, a score has been calculated using weights generated through principle components analysis and assuming a specific distribution of asset scores. All households are then grouped into quintiles based on these asset scores. This approach has been used to re-analyse DHS data for 44 countries and examine the linkages between socio-economic differences and health and nutrition outcomes (more detail can be found at www.worldbank.org/poverty/health/data/).  
     We calculated an average asset score for each cluster in Kenya to show the spatial distribution of households in the two poorest quintiles across production systems (Map B2). Generally, households in the western part of the country have fewer assets, meaning a higher proportion of households within each cluster are classified within the poorest quintiles. It should be borne in mind that this is just a visual examination of the raw data (similar to the visual examination of the relationship between two variables in a scatterplot) and that the cluster selection is based on a population-based rather than a geographic sample.  
     Table B1 shows the results of the aggregated DHS for each production system category. The calculated mean is based on appropriate sampling weights and includes a corresponding standard error and the number of unweighted cases used for the calculation. Because only half of the Kenya clusters were geo-referenced, and the DHS uses a population-based sample, we could only produce statistics for the more populated mixed productions systems. The other system categories contained too few clusters to produce statistically significant results.
     In general, where the means are not too close to be within the margin of error, the temperate/tropical highland category does better on the selected variables than the other two categories. For most of the selected variables, the mixed rainfed humid/subhumid category ranks last. These systems can be found in areas bordering Lake Victoria and the  Indian Ocean.

<< Previous    Next >>