Decomposing Strokes Gained: The Leaderboard Cannot Value a Golfer

A golfer won a PGA Tour event with +8.9 strokes gained putting and -1.3 strokes gained approach across four rounds. He lifted the trophy, signed a new sponsor and appeared on three magazine covers. My tracking sheet recorded that week in a different colour.

I re-watched all 72 of his holes, rebuilding every shot at shot level. On the 15th green of the final round the ball finished on the fringe, 11 metres from the pin, and he holed the putt. The grandstand rose. In my sheet, that putt sits in the outlier bucket — not because it was bad, but because it does not repeat.

Twelve weeks later, same player, same putter, same grip, same ball: strokes gained putting averaged -0.6. No reporter called me. I file the report, close the folder, and the market reopens on its own.

I came into this work through football. At 19 I hand-logged 1,240 dangerous situations across the 2026 World Cup and built expected goals for every attack, until my eyes saw goals in matches that had none. Four years later, moving to golf, I carried exactly one habit: never read the result before reading the process that produced it.

Golf gives me what football cannot — a discrete, countable data stream. A PGA Tour round records roughly 70 shots; each carries a landing coordinate, distance to pin, club, elevation change against the green, wind direction and surface moisture. Add hourly weather data and tee-time assignment and I hold what football models only dream of: a large sample and clean fields.

Between 2026 and 2026 I logged 2,140 rounds across the PGA Tour, DP World Tour, LPGA and a handful of domestic events. Not to predict next week's winner. To answer a much narrower question: when a player wins on one skill alone, does that skill hold — and if not, by how much is the market mispricing him?

My decomposition is crude. Four buckets: off the tee, approach, around the green, putting. Each converted to strokes against tour average. Nothing new there. What differs is that I never read those four columns as four equal columns. Data is never in a hurry; it only waits for someone who can read it.

Inside the 2,140-round sample I split two groups. Group A: 286 rounds where the player finished top-5 on strokes gained putting of +4.0 or better. Group B: 311 rounds where the player finished top-5 on strokes gained approach of +4.0 or better.

Six weeks later, Group A's putting fell 2.7 strokes from its peak, approach barely moved at -0.1, and scoring average rose 1.9 strokes. Group B's approach fell 1.1 strokes while scoring average rose only 0.6. The correlation between an outlier putting week and six-week decay was r = 0.71.

The four skill buckets are not equally stable, yet most transfer valuation models treat them as if they were. In my sample, off the tee has the lowest week-to-week standard deviation, approach sits second, and putting carries the highest, roughly 2.4 times off the tee. The noisiest column is the column the media reads most closely, because it produces the prettiest moments.

The consequence lies elsewhere. A scouting report that weights wins over the past 12 months is, technically, using the noisiest variable as its primary predictor. A report that filters on world ranking makes a different error, and a worse one.

OWGR is not a skill measure. It is a points-conversion system built on field strength and finishing position. People spend it as currency to buy major starts. But a currency is only trustworthy when its exchange rate reflects real value. Since October 2026, OWGR has declined to award points to LIV Golf events, which means a group of world-class players contest 54-hole tournaments that never convert into the shared unit of account. When Jon Rahm moved to LIV Golf in December 2026, he carried a major exemption with him and left behind a ranking that no longer measures his true form.

To a scout whose first filter is world ranking, Rahm becomes a name that fades month by month. To me he is an asset with stable shot quality and thin liquidity. Two opposite conclusions about one man, differing only in which column each of us reads first.

There is a structural problem outsiders rarely see. High-quality shot-level data exists mainly on the PGA Tour, through ShotLink at every hole. The DP World Tour has equivalent coverage at only a portion of its events. Asian circuits, Vietnam included, have almost none. Every global model is therefore structurally skewed toward North America, and players competing elsewhere are priced on a far thinner sample. I know this because I once built a tracking sheet for domestic golf and had to face the fact that I could access roughly a quarter of the variables my colleagues in the United States take for granted.

At this point I have to argue against myself.

Correlation is not causation, and r = 0.71 does not prove that a hot putter erodes approach play. There is a hidden variable that resists removal: schedule. In my sample, Group A carried markedly more rest than Group B — across the five weeks before their winning week, Group A played 2.8 rounds per event while Group B played 4.3. A three-week break before a big event can both restore the body and dull green feel. I have not separated those two effects, and I will not pretend I have.

Crowd pressure is another variable my model handles poorly. In 2026 I measured 412 European football matches played in empty stadiums and found home win rate fell from 46% to 34% while goals per match rose from 2.6 to 3.1. Golf mirrors that structure on the closing hole: with packed galleries, putts from 3 to 5 metres convert less often for the leader than for the chaser. But green firmness, wind direction, pin position and daily flag severity are still absent from my model, which means I am comparing things that are not yet under the same conditions.

Audiences applaud on emotion; data hears a different rhythm. On the 18th hole that week, the roar was loud enough that I had to drop the recording volume to hear the club strike the ball. In my sheet, that sound does not exist.

Rules and equipment are also distorting the very skill I am measuring. Since 2026, Local Rule G-11 has restricted green-reading books in competition, and from 2026 the rule was broadened. Reading greens became more data-expensive, because it depends increasingly on local experience rather than personal notes. A hot putting week on a familiar course becomes many times harder to repeat on an unfamiliar one, and I cannot separate green reading from putting mechanics with the data available.

The most honest conclusion I will allow myself is this: an outlier putting week does not predict a player's long-term value, and both of the most popular yardsticks — win counts and world ranking — misprice him in opposite directions.

People watch the putt drop; I watch the ball flight that put it there. The 11-metre putt on the 15th was the consequence of an approach that found the fringe. Had that approach missed by 1.2 metres, the finest moment of the week would not exist, and nobody would call him a champion with nerve.

The next data cycle reopens in three places. The 12-week standard deviation of strokes gained approach among the world's top 20 is the most stable variable I own, and it will show who is genuinely improving rather than who just had a good week. Higher up, negotiations between the PGA Tour and Saudi Arabia's Public Investment Fund, along with any reopening of OWGR to LIV Golf, will set the exchange rate for the currency the whole industry spends — every time the points system changes, the entire transfer valuation table has to be rewritten. And course fit at the next major is the final test: a long course with fast greens and strong wind rewards the approach group and punishes the putting group, and I want to see whether the market reads that before the leaderboard does.

A report left in a drawer is not a conclusion; it is a chart waiting for a time axis.