← All articles

Every analytics revolution is a per-possession revolution

Every major team sport has now gone through some version of the same analytical revolution, on roughly the same timeline, against roughly the same resistance. The shorthand for it differs by sport — per-100 possessions in basketball, per-drive or per-play in football, per-60 in hockey, per-plate-appearance and per-batter-faced in baseball, per-90 in soccer — but underneath the sport-specific vocabulary the move is identical. The league reorganized its statistics from counts per game to rates per opportunity. The reorganization changed everything that followed.

It's worth recognizing the pattern explicitly because the next sport to do this — and the next stat in any of these sports to undergo the shift — will follow the same playbook. The per-game era of any sport is the era when the league has not yet fully understood that the denominator matters. The per-opportunity era is the era when the league understands the denominator and treats it as a first-class variable. Every transition between those two eras looks the same.

What the denominator change actually does

Counting stats per game inherit two confounds. The first is opportunity. A team that plays fast generates more possessions per game than a team that plays slow, and their per-game totals reflect both their true efficiency and their pace. The second is participation. A player who plays more minutes generates more per-game events than a player who plays fewer, and the per-game totals reflect both their true rate and their court time. Either confound, by itself, can dominate the comparison. Together they make per-game stats almost uninterpretable for cross-team or cross-player comparison.

Switching to rate stats — per possession, per drive, per plate appearance, per ninety minutes — strips both confounds. A team's per-possession offensive efficiency is independent of how fast they played; a player's per-minute scoring rate is independent of how many minutes they played. The resulting numbers don't change what was true about the team or player; they change how legible the truth is in the column.

Every analytics revolution in every sport has, at its core, been a series of arguments about which denominator to use and which numerator to count. The denominator choice is almost always the more important of the two. Per-game stats are wrong in the same way in every sport. Rate stats are right in the same way in every sport. The order in which the leagues figured this out is the order they reached intellectual maturity in their analytical era.

Baseball got there first

Baseball's analytical revolution started earliest — Bill James's first Baseball Abstract in 1977 — for two structural reasons. The sport's atomic events already had natural denominators baked in. A batter's opportunity to hit is a plate appearance, an at-bat, or a time on base. A pitcher's opportunity to record an out is a batter faced. Those denominators were already in the box score; the rate stats just hadn't been prioritized.

The transition from batting average and pitcher wins to on-base percentage, slugging, OPS, FIP, wOBA, and eventually WAR was the gradual replacement of the wrong-denominator and wrong-numerator versions of each question with better-denominated, better-counted versions. The change was bitterly resisted by the baseball establishment for thirty years, then absorbed completely. By 2010 every front office had adopted the new framework. The broadcasts took another decade to catch up.

Basketball was second

Basketball's revolution arrived in the early 2000s, anchored in Dean Oliver's Basketball on Paper and the slow public adoption of per-100-possessions framing. The argument was the same as baseball's. A team that scored 110 points per game in a 105-possession league wasn't necessarily a better offense than a team that scored 100 points per game in a 90-possession league. The per-game totals confounded efficiency and pace. Per-possession totals separated them.

The shift from per-game to per-100 in basketball analytics was accompanied by the same set of adjustments baseball had been through twenty years earlier. Player evaluation moved toward per-minute rates. Team comparisons moved toward offensive and defensive ratings rather than points scored and points allowed. The four-factor framework — effective field goal percentage, turnover rate, offensive rebound rate, free-throw rate — was the rate-stat reformulation of everything the box score had been telling the league for forty years in the wrong denominator.

Hockey, football, and soccer followed

Hockey got serious about per-60 rates in the late 2000s and early 2010s through the work of public analysts who rebuilt the case from first principles. A player who averages eighteen minutes a game accumulates fewer events than a player who averages twenty-three minutes a game, and per-game totals had been treating that difference as if it were a difference in skill. Per-60 stats fixed the problem. Corsi and Fenwick — the precursors to expected goals in hockey — were natively per-60 from the start, because by the time anyone was building those stats publicly, the analytical community had already absorbed the lesson from baseball and basketball.

Football's version arrived through Expected Points Added per drive and per play, and the rise of public models that evaluated decisions and players in rate-per-opportunity terms rather than total yards or total touchdowns. The single most important conceptual shift in football analytics over the past twenty years is the move from "yards per game" framing to "expected points per drive" framing. The change-of-possession unit, the drive, became the denominator that mattered. Pace of play and opportunity count stopped polluting the headline numbers.

Soccer's revolution centered on per-90 stats, beginning in the late 2000s with the public-analytics community that built around StatsBomb-style data. A striker's goals per ninety minutes is the rate that matters; a striker's goals per appearance is contaminated by how much they actually played. Once per-90 became the standard frame for player evaluation, the rest of the soccer analytics movement — xG, xA, expected threat, pressing and recovery rates — followed in the same denominator, because there was no other reasonable denominator left.

The resistance always sounds the same

Every league's transition to rate stats encountered the same set of objections, almost word for word. The new stats were "alphabet soup" that "obscured the game." The old stats were "what fans care about." The numbers "didn't pass the eye test." The new framework "overcomplicated" something that had been "fine for a hundred years."

The objections, considered structurally, were arguments for keeping a known-broken denominator because the new denominator required relearning. There was no version of the argument in which the per-game framing was actually more informative than the per-opportunity framing. The per-game framing was just older. The people defending it were defending the version of the game that had matched the era of statistics they had learned, not a claim that the older statistics carried more information than the newer ones.

What hasn't shifted yet

Pockets of every sport remain stubbornly per-game. The NFL still organizes its passing leaderboards by total yards rather than yards per drive. NBA broadcasts still lead with points-per-game leaders rather than points-per-75-possessions leaders, the per-game-feel version of a true rate stat. MLB still publishes pitcher win totals on the standings page. EPL still shows possession percentages on the live score graphic as if they were a verdict. Each of those is an old denominator outliving its accurate-but-better replacement, in a sport whose analytical community has already moved on.

The leaderboards will catch up eventually. They always have. The lag between when a stat is publicly known to be inferior and when it stops being the leaderboard's default is usually about a generation. The next generation of sports broadcasts will lead with rate stats the current ones don't even compute on screen. Then they'll move on to the next denominator argument, which will be about the opportunity inside the opportunity — which possessions actually counted, which drives were garbage time, which plate appearances came against the worst arm in the bullpen.

The lesson for the next sport

The next sport to undergo this transition — and the next stat in any current sport to be reformulated — will follow the same pattern. The new rate version will be proposed by a small group of public analysts. It will be ridiculed for a decade. It will quietly become the standard in front offices first, then in better broadcasts, then in mainstream broadcasts. The transition will look, in retrospect, like an obvious and overdue correction. It will not, while it's happening, look obvious to most of the people defending the older numbers.

The shortcut, for any fan who wants to skip ahead of the curve, is to ask one question whenever you see a sports statistic: what's the denominator, and is it the right one. If the denominator is "games played" or "the whole season," it's probably the wrong one. If it's "the opportunity the player or team actually had," it's probably the right one. The rest of the analytics argument is mostly an elaborate version of that single question, asked sport by sport, stat by stat, decade by decade.