The Same Three Temperatures Give Three Different Coefficients of Variation

Near a mean of zero the ratio stops meaning anything

Dividing by the mean is fine while the mean is comfortably away from zero and hopeless as it approaches. A series of monthly profits averaging close to break-even produces a coefficient of variation in the hundreds or thousands of per cent, and the number swings wildly on a change that barely moves the underlying data. Worse, a mean that is negative flips the sign of the whole thing unless the absolute value is taken, as it is here. When the mean can plausibly cross zero, the coefficient of variation is the wrong summary and the standard deviation on its own is the right one.

Where it earns its place: comparing consistency

Two machines fill jars, one at 50 grams with a standard deviation of 2 and one at 500 grams with a standard deviation of 15. The second looks far worse in absolute terms and is in fact more consistent: 4 % against 3 %. That comparison is the whole reason the measure exists, and it works on anything with a genuine zero — masses, lengths, counts, durations, prices. A laboratory reporting the repeatability of an assay, a factory comparing lines that produce different products, and an athlete comparing the steadiness of a sprint time against that of a throw distance are all doing the same thing with the same ratio.

Frequently asked questions

Which scales have a real zero?

The test is whether "twice as much" means anything. Twice 4 kilograms is 8 kilograms and the sentence is true, so mass has a real zero; twice 10 degrees Celsius is not 20 degrees in any physical sense, so Celsius does not. Lengths, masses, durations, counts, prices and absolute temperatures pass; Celsius, Fahrenheit, calendar years, pH, and most rating scales from one to five do not. On a scale that fails the test, use the standard deviation and say what units it is in.

Is a lower coefficient of variation always better?

For a manufacturing process, yes — the whole point is that jars come out the same. For an investment it is a different question, because the ratio can be lowered either by reducing the spread or by raising the mean, and those are not the same decision. A fund can improve its coefficient of variation by taking on more of a return it never earns, and one with a tiny spread around a mean barely above zero will score badly while losing almost nothing. Read it as a description of consistency relative to size, not as a ranking.