# Hyperlink 2. Nonrepresentative Data, Sampling Errors, and Decision Making

Consider an experiment where it is known that the true mean concentration of a large soil mass is 12.3 mg/kg. Assume an analysis was performed on a 1 g aliquot from a sample jar and the result was 12.3 mg/kg. The actual mass of contaminant present in the 1 g analytical subsample is 0.0123 mg. Now imagine that a particular 1 g analytical subsample from the same sample jar captured a highly concentrated “nugget” made of organic carbon (red dot). The organic carbon nugget contains 0.1000 mg of contaminant mass in addition to the baseline 0.0123 mg. The contaminant mass in that 1 g subsample would be 0.1123 mg. When analyzed, the reported sample result would be 0.1123 mg/g, or 112.3 mg/kg, an order of magnitude different from the first result. The value of 112.3 mg/kg would be an erroneous result (a sampling error) caused by within-sample heterogeneity. It is not an analytical error because the analysis correctly measured the 0.1123 mg of contaminant present in that 1 g of soil. It is a subsampling error because the subsample did not contain the contaminant in the same proportions as the large soil mass.

Another factor is the concept of sample support is illustrated graphically in Figure H2-1. Imagine it was possible to extract and analyze 1 kg instead of 1 g of soil. Imagine also that the nugget was present in the kilogram of soil being analyzed. The baseline amount of contaminant present in the 1 kg analytical sample is 12.3 mg. Adding the mass of the 0.1000 mg nugget gives 12.4 mg, so the reported concentration is 12.4 mg/kg, quite close to the true concentration of 12.3 mg/kg. This scenario illustrates how strongly analytical subsample support affects reported concentrations. A nugget can cause much larger sampling errors when analytical subsamples are small.

Figure H2-1. Illustration of the relative effect of a concentrated nugget on a 1 g sample vs. a 1 kg sample.