A.6.3 Sample Representativeness and Bias
A representative sample of bulk material is one that has the same properties as the bulk material contained in the population. In the case of site investigations, the sample may be obtained through discrete, composite, or ISM sampling and the population is the mass of contaminant in the environmental medium throughout the DU. A representative bulk material sample has all particulates (e.g., size, shape, concentrations) in the same proportion as the particulates present in the DU. When a representative sample is collected from the DU, we are more likely to obtain an unbiased estimate of the DU mean.
Sampling bias (nonrepresentativeness) is introduced due to composition (e.g., size, shape) and distribution (e.g., segregation, grouping, and spatial patterns) of particulates present in the DU. The distribution and composition of particulates in a sampling DU affect the representativeness of the collected sample. For example, denser particles might have settled at the bottom (e.g., stockpiles), or new contaminants might have settled onto the surface of the DU. Contaminant particulates might be heavily deposited in one part of the DU, resulting in a subarea with elevated concentrations (e.g., spill area or training target location) or clumps with concentrated mass of contaminant (e.g., pesticides in a farm field).
It should be noted that contrary to the common belief, the use of structured sampling patterns such as the serpentine pattern (Figure A-21) and systematic random within grid pattern (Figure A-22) may not yield an unbiased estimate of DU mean unless the mass/particulates are evenly, uniformly, or homogenously distributed throughout the DU. In the absence of any knowledge about uneven and/or segregated distribution of particles in the DU, the simple random sampling (incremental or discrete) approach is the best approach to collect representative samples from heterogeneous DUs (from the perspective of sample bias). These issues are illustrated by performing simulation experiments on hypothetical heterogeneous target DU considered in this appendix.