“Selection bias” is used to describe systematic inaccuracy that creeps into statistical studies when samples are not selected randomly, a common situation in mining studies. Prime examples include selection of samples focused on higher grades for check assays, screened metallics or a second fire assay for gold. In cases of poor homogeneity, the second set of assays will tend to report lower grades. Although the problem is pervasive, it is poorly understood or, worse, not even recognized. Good projects can be stalled by the false belief that the mineral resources are smaller than they actually are. This presentation aims to make the problem better recognized so that project development decisions can be improved. Following discussion of a commonly-used example of selection bias, a study of air force pilots, the presentation examines causes of selection bias in minerals projects, with a focus on gold projects where a few flecks of gold, literally barely parts per million, causes large variation in sub-samples. Three case studies of assay data from different mining projects highlight various aspects of how selection bias arises, and how it can easily lead to poor project development decisions.
