Average relative difference alone won’t tell you if your duplicate data are fit-for-purpose. For a structured assessment, use two complementary tools: the Thompson-Howarth plot and Half Absolute Relative Difference.
The Thompson-Howarth plot shows the mean vs. absolute difference for each pair, revealing the expected heteroscedastic pattern in geochemical data. Outliers above a 2× median HARD envelope flag problematic pairs, while a shift in scatter can indicate a detection limit or preparation heterogeneity issue. HARD, calculated as $|A – B| / (A + B) \times 100\%$, gives a robust, easy-to-interpret metric — a median <5% is typically acceptable for base metals at grade, with near-LOQ values excluded.
Key practice: Assess precision element-by-element. A single summary statistic can hide failure. Gold at 0.1 g/t and copper at 2000 ppm do not behave the same under duplication and require separate evaluation.
Save this post for your next Q/AQC review.
Leave a Reply