Adding soil sampling to household surveys: Information for sample design from pilot data

Large sample surveys with households, or individuals within households, as the basic sampled units, are important sources of information on variables related to household income, economic activity, food security and nutritional status. In many circumstances the advantages of supplementing these surveys with sampling of the soil from fields or other land units which the households cultivate may seem obvious, as a source of information on the quality of the soil on which households depend, and potential limitations on their food security such as soil pH or nutrient status. However, it is not certain that household surveys, designed to examine social and economic variables, will be efficient for collecting soil information, or will provide adequate estimates of soil property means at scales of interest. Additional sampling might be necessary, so an attendant question is whether this is feasible. In this paper we use data on soil pH and soil carbon inferred by spectral measurements on soil specimens collected from land cultivated by households in Uganda and Ethiopia to estimate variance components for these properties, and from these the standard errors for mean values at District (Uganda) or Zone (Ethiopia) level by household surveys with different designs. Similar calculations were done for direct measurement of soil carbon and soil pH from a spatial sample in Malawi from which variograms were used to infer the variance components corresponding to the levels of a household survey. The results allow the calculation of sample sizes at different levels of the design, required to allow estimates of particular quantities to be obtained with specified precision. The numbers of sampled enumeration areas required to obtain estimates of district or zone-level means with the arbitrary specified precision were large, but the feasibility of such sampling must be judged for a particular application, and the precision appropriate for that. The presented method makes that possible. • We used data on soil sampled in pilot household surveys of Ethiopia and Uganda. • We computed standard errors for estimates of regional means from contrasting designs. • We did a comparable analysis using data from a spatial coverage survey of Malawi. • Precise mean soil pH or organic carbon for districts or zones required large samples. • The method can evaluate sample requirements for other variables and objectives.