Standard deviation tells you the spread of one data set in its own units. But suppose you want to compare the consistency of two sets that are measured differently, or that simply have very different averages — say monthly rainfall against daily temperature, or the run-scoring of a batsman who averages $80$ against one who averages $25$. A raw standard deviation cannot settle that comparison fairly, because a larger mean naturally tends to come with a larger spread.
The fix is to express the spread as a percentage of the mean. This is the coefficient of variation (C.V.), a unit-free, relative measure of dispersion:
Because the units cancel in the ratio $\dfrac{\sigma}{\bar{x}}$, the C.V. is a pure number (a percentage) and can compare any two distributions directly. The rule is simple:
The data set with the greater C.V. is the more variable (less consistent); the one with the smaller C.V. is the more uniform or reliable. The comparison below shows two players with the same average but different consistency:
| Player | Mean $\bar{x}$ | S.D. $\sigma$ | C.V. $=\dfrac{\sigma}{\bar{x}}\times100$ | Verdict |
|---|---|---|---|---|
| A | $50$ | $5$ | $10\%$ | more consistent |
| B | $50$ | $15$ | $30\%$ | more variable |
When two sets happen to have the same mean, the comparison reduces to comparing standard deviations directly — the means cancel and a smaller $\sigma$ already means greater consistency.
Deeper Insight — why "relative" spread beats "absolute" spread: The coefficient of variation exists because dispersion only means something in proportion to the scale you are working at. A standard deviation of two kilograms is enormous for the weights of medicines but trivial for the weights of adults; the same number tells opposite stories depending on the average it is set against. By dividing by the mean we strip out that scale, leaving a clean statement of "how much does the data wobble relative to its own typical size?" This is exactly why C.V. is the right tool for comparing consistency across players, factories, crops or share prices that operate at different levels. Two cautions worth carrying forward: the C.V. is undefined when $\bar{x}=0$ and becomes unstable when the mean is close to zero, and it is only meaningful for data on a ratio scale with a true zero — comparing the C.V. of temperatures in Celsius would be nonsense, since the choice of zero is arbitrary. Used within those limits, the C.V. turns the vague word "consistent" into a number you can defend.