Estimating sampling bias by using paired samples

Abstract

A method for investigating the possibility of bias between sampling methods has been devised. It is effected by collecting paired samples from a number of different sampling targets. Paired samples are created by executing a standard sampling protocol and a sampling method under validation in a randomised fashion at each target. After chemical analysis of the two samples, the difference between a pair of results is an estimate of the bias for that target. Repeating the procedure for many distinct targets improves the precision of the estimate. If the analyte concentration range of the targets is small, the mean of the difference estimates is taken as the overall bias estimate. The mean bias can be tested for significance, if required, by a one-sample test. If the concentration range is wide, potential biases can be considered as functions of analyte concentration. Weighted functional relationship estimation is appropriate for this purpose.