CONSISTENCY AND THE COEFFICIENT OF VARIATION

Imagine two players who both average twenty points over their last twelve games. Player A scores between eighteen and twenty-two almost every night — dull, dependable, almost boring. Player B alternates wildly between six and thirty-four. Their averages are identical to the decimal; their profiles could not be more different.

If you only look at the average, these two players appear interchangeable. They are not. Player A is a known quantity you can build around. Player B is a coin flip wearing the same jersey. Any honest analysis has to separate them, because the risk you take on when you rely on each of them is completely different.

This is the single most important idea in performance analytics that casual observers miss: the average tells you the centre of a player's distribution, but says nothing about its width. Two distributions can share a centre and still overlap only partially. Width is where predictability lives.

The coefficient of variation (CV) is a clean way to capture that width. It is the standard deviation of a player's recent games divided by their average. In plain terms: how spread out are the results, relative to the typical result?

Dividing by the average is the clever part. A raw standard deviation of five points means something very different for a player who averages eight than for one who averages twenty-eight. By scaling the spread against the mean, CV becomes dimensionless — it lets us compare consistency fairly across players of different volume, and even across different stat types entirely.

A low CV means tight, repeatable output — the metronome. A high CV means wide swings — the coin flip. Because it is scale-free, we can meaningfully ask whether a receiver's yardage is more or less consistent than a point guard's assists, a comparison that a raw standard deviation could never support.

Raw CV is analytically tidy but a little awkward to read, because lower is better and the numbers are small decimals. Most people find that counter-intuitive. So we convert it into a reliability percentage where higher is better, which is how the figure appears throughout the product.

A player described as 'ninety percent reliable' is one whose recent games cluster tightly around their projection. A player at sixty percent is one whose output regularly strays far from the middle. The percentage is simply a more human-friendly restatement of the same underlying spread that CV measures.

It is worth stressing what reliability is not. It is not a measure of ability. A modest role player can be extremely reliable at a small number, while a genuine star can be brilliant on average yet wildly erratic game to game. Both facts are useful — they just answer different questions. Reliability answers 'how confident can I be in the projection?', not 'how good is this player?'

Once you internalise consistency, projections stop being single numbers and start being ranges. A projection of twenty points at ninety percent reliability describes a narrow band you can lean on. The same twenty at sixty percent describes a wide fog of possibilities that happens to be centred on twenty.

This is why we never present a projection without its reliability alongside it. Stripping the two apart invites exactly the mistake we are trying to prevent — treating a shaky estimate and a solid one as if they carried the same weight simply because their midpoints match.

Consistency is most valuable when paired with the projection and the matchup. A reliable player in a favourable matchup is the clearest kind of signal analytics can offer. A reliable player in a brutal matchup is a caution flag, because even dependable production can be suppressed by a strong opponent.

And an erratic player, however talented, always carries more uncertainty than the headline number suggests. That is not a reason to dismiss them — variance cuts both ways, and their ceiling is often what makes them exciting — but it is a reason to size your confidence accordingly. Knowing which kind of player you are looking at is half the battle.