bias
/ BY-us /
Picture an archer who always lands a hand's width to the left of the bullseye, shot after shot. Their aim is consistent, but consistently off to one side. Bias is that lopsidedness: a tendency for measurements to land on one particular side of the truth — too high, or too low — on average, rather than scattering evenly around it.
Bias is the difference between the average (expected) value of many measurements and the true value. It is the visible footprint of systematic error: if a method has a constant or proportional flaw, repeating it many times and averaging will not give the truth but a value shifted off to one side, and that shift is the bias. Because it is an average effect, you detect bias by analysing a sample of known value — a certified reference material — and seeing whether your mean systematically lands above or below it.
Bias matters because it is the enemy of accuracy that hides behind good precision: a biased method can be wonderfully reproducible and still consistently wrong. The honest caveat is that some bias can be measured and corrected (subtract a known offset, apply a recovery factor), but unrecognised bias is dangerous precisely because the data look clean and trustworthy. Good practice spends real effort hunting for bias that the numbers themselves will never reveal.
A method run twenty times on a reference material certified at 100.0 mg/L gives a mean of 96.5 mg/L. The scatter is small, but the mean sits 3.5 below the certified value every time — a bias of −3.5 mg/L pointing to a systematic loss somewhere in the procedure.
A tight mean that still sits to one side of the truth.
Bias is to systematic error what scatter is to random error: bias is the net directional shift that systematic errors produce in the long-run average.