Using nonprobability samples in design-based estimation of spatial means of soil properties

摘要

To estimate spatial means of soil properties of a region from data collected by convenience or purposive sampling discussions on the validity of the results can be avoided by collecting additional data by probability sampling. The nonprobability and probability sample can be combined by interpolating the data at the nonprobability sample to the probability sample points, and to use these interpolated values as an auxiliary variable in the design-based regression estimator. The regression estimator is approximately unbiased even if the measurements at the nonprobability sample are severely biased. The gain in precision compared to the π estimator that uses the probability sample only, depends on the correlation between the target variable and interpolated auxiliary variable at the probability sample points. This correlation on his turn is partly determined by the quality of the interpolation. Factors that govern this quality are the density of the nonprobability sample, its configuration and the interpolation method.