Use of Trimmed Duplicates Derived from Laboratory Data To Estimate Standard Deviation

Improved standard deviation estimates from possibly biased duplicate measurements can be derived from appropriately trimmed plots of standard deviation estimates using pairs of replicates vs the quantiles of a half-normal distribution. Simulated studies show that these estimates exhibit generally lower mean-squared errors and biases than do more standard robust estimators of location—¾ times the interquartile range and 3/2 times the mean absolute deviation from the median.