Correlation Is Not a Cause in Sports

Two numbers that rise together are correlated. A cause is a harder claim: changing one would change the other, in a direction you can name, for a reason that is not some third fact hiding underneath. Sports talk makes the jump constantly, because the tables are full of numbers that move together for boring mechanical reasons.

Pace creates fake skill

A team that runs more plays will, all else equal, record more rebounds, more fouls, more shots, and more points. A ranking of raw rebounds per game will therefore flatter fast teams. They may also be good rebounders. The raw total cannot separate the two. A rate per possession, or per hundred plays, removes the pace before you compliment the skill. Until you remove it, you are complimenting the tempo.

The same trap sits inside individual counting stats. Players who stay on the floor longer accumulate more of everything. Minutes are an opportunity. Opportunity is not the same as the rate at which the player does the thing. Quote the rate when the question is skill, and quote the minutes when the question is how large a role they had. A checklist before you trust a stat starts by forcing that denominator into the open.

Selection disguises itself as a discovery

Teams that win more games also tend to have better point differentials. That correlation is real and partly mechanical: winning is related to outscoring. It does not mean the win total caused the margin, or that a team can add wins without changing the scoring. Point differential versus the record keeps the two summaries apart so the correlation is not asked to be a lever.

A darker version selects on the outcome and then admires a trait. Look only at champions and you will find habits champions happen to have, including habits that losers also have and that you did not write down. You selected the end of the story and then read the middle as a recipe. Recipes need the failures in the denominator.

A third factor is the usual engine

Height correlates with rebounds. It also correlates with position, and position correlates with the job of standing near the basket. Ban a tall player from the paint in your mind, and the rebound correlation shrinks. The third factor was the role. Age, coaching, and who the team plays next are other third factors. Injuries and lineup changes are third factors with a date attached. A correlation observed on a healthy roster is not a cause you can carry into an injured week.

None of this says correlations are useless. They are a place to start looking for a mechanism. The mechanism has to be stated in words that could be wrong. Outscoring teams tend to win is almost a definition. Playing faster creates more raw counting stats is a mechanism. This player is winning because of a lucky number on the jersey is not.

Do not rent the correlation to a bet

A correlation is not a price. It does not tell an adult in a legal market what a game is worth, and it tells a minor nothing about a wager because a minor should not have one. The useful action is to rewrite the claim with the third factor included, or to drop the claim. Regression to the mean is the companion caution when the correlation was computed on an extreme week you noticed because it was extreme.

Frequently asked questions

If two stats move together, can I treat one as the reason for the other?

Not yet. Check pace, minutes, role, and selection first. A third factor often produces both numbers.

Are per-game totals a fair way to rank players?

Not when playing time differs. A rate plus a note about minutes is the fairer pair. The total alone rewards whoever stayed on the floor.

Does a strong correlation justify a wager?

No. Correlation is a description of how numbers have moved. It is not a cause, a price, or a recommendation.