The Empty Cell in Golf Data: ShotLink, OWGR, and the Cost of What Cannot Be Measured

The Empty Cell in Golf Data: ShotLink, OWGR, and the Cost of What Cannot Be Measured

Three in the Morning, and a Column Full of N/A

Three in the morning. On screen is a spreadsheet with forty-two columns and one thousand one hundred seventy-six rows. Column eleven is headed SG: Approach. Every cell in that column carries the same value: N/A.

Not because the players struck the ball poorly. Not because I built the formula wrong. It is because for the entire duration of that tournament, nobody recorded the position of the ball after the second shot. No coordinates, no distance to the hole, no lie type, no grass type. Only the total strokes at the end of the day, retyped from a blurry scanned sheet.

I stared at that column for about twenty minutes, and then did something I still consider the single best data decision of my life: I did not interpret it. I closed the spreadsheet, shut the machine down, and went to sleep.

The next morning someone called to ask what I had read from that pile of numbers. I said I had read one thing — this event has no proximity data. The reply was the sentence I have heard at least forty times in thirteen years in this trade: "Just estimate it. Who's going to check?"

That is why I am writing this. Numbers do not lie. But reputation whispers into the ear of the person who never reads the table.

An empty data column, in the eyes of a careful practitioner, is a system fault in need of repair. In the eyes of the market, it is a blank space in need of filling. And the market always fills blank spaces faster than technicians repair faults. Golf has lived off that blank space for nearly a century. This piece walks through four layers of the problem: the measurement infrastructure, the ranking systems, the habit of replacing data with inference, and a concrete contingency plan for Vietnamese golf.

Golf's Measurement Architecture: Three Tiers, One With Eyes

Golf is the most unevenly measured professional sport among the major ones. Not because anyone is lazy — because the cost of a single measurement in golf is far higher than in football, basketball, or tennis.

In football, one fixed camera in the stand plus ball-tracking software can generate data for all twenty-two players simultaneously, every second. In golf, each group holds only three or four players, but they scatter across sixty hectares, moving along non-repeating routes, and each man hits only about seventy-five shots across five and a half hours. To record every ball position, you need hardware or humans following every group, in every group, all day long.

The PGA Tour can do this. ShotLink, powered by CDW, was deployed from the early 2000s and has covered the tour's full schedule since the 2026 season, logging every shot by every player through on-site positioning and a large volunteer corps. Each season the system processes hundreds of thousands of shots, each tagged with distance, lie, distance to hole, and outcome.

From that foundation, in 2026 Mark Broadie, a professor at Columbia Business School, published the Strokes Gained framework, and in 2026 systematised it in the book Every Shot Counts. It was the sport's conceptual turning point: instead of counting outcomes, people began measuring contribution.

But the mechanism needs stating plainly, because many people use Strokes Gained without knowing what it stands on. Strokes Gained is fundamentally a conditional subtraction. To subtract, you need a baseline: the expected strokes to hole out from distance X, on lie type Y, at course Z, on a specific putting surface. That baseline is not an assumption. It is built from millions of real shots with real coordinates.

Put another way: Strokes Gained is not a statistic. It is a proprietary database wearing the clothes of a statistic.

No coordinates, no baseline. No baseline, no Strokes Gained. The entire advanced metric stack of golf — SG: Off the Tee, SG: Approach, SG: Around the Green, SG: Putting — rests on exactly one thing: somebody standing there with a measuring device, pressing a button the moment the ball stops.

The second tier consists of tours with tracking that is real but uneven. The DP World Tour has data at most events, though group-level coverage is patchy and often truncated outside the leaders. The LPGA Tour runs its own tracking with a different level of granularity. The Korn Ferry Tour and PGA Tour Champions share the ShotLink backbone. This tier has data, but it has holes in it.

The third tier — and the largest by number of players on the planet — has nothing at all. The Asian Tour, the Japan Golf Tour, the Korean Tour, the China Tour, the VGA Tour, most national amateur championships and every club-level team event sit here. The only data that exists is the scorecard: strokes per hole, par, yardage. If the organiser is diligent you get fairways and greens in regulation — but GIR is usually judged by eye by a volunteer standing at the edge of the green, and nobody audits the volunteer.

The proportion matters more than it appears. Tens of thousands of professionals and elite amateurs compete regularly worldwide. Of those, the number with complete weekly shot data does not reach a few thousand. Everyone else plays, wins, loses, and vanishes from every comparison table except the scoreboard.

An Empty Cell Is Not Neutral: It Is a Subsidy

This is the part I want to give the most space to, because it is where most commentary on sports data goes astray.

People often describe an empty cell as "untapped territory", as "opportunity", as "potential". Wrong. An empty cell is a subsidy, and that subsidy is paid to whoever controls the storytelling.

Follow the pricing mechanism. When a scout or sponsor wants to evaluate a player with no shot data, they are forced onto proxy variables. The four most common, in order of influence:

First, finishing position. But finishing position is a brutally compressed variable: it merges four distinct skills, plus weather, plus luck on the greens, plus the quality of your draw, into a single number. A player who finishes eighth with a strokes-gained figure of minus four and a player who finishes eighth at plus two receive the same cheque — and, more importantly, the same market valuation.

Second, nationality and tour affiliation. Professional golf runs on regional pipelines. A player performing well on the Asian Tour is priced at Asian Tour levels. A mediocre PGA Tour player is priced at PGA Tour levels. The gap between those two pricing floors is larger than the true skill gap between the two men, and no statistic can correct it, because no statistic compares them.

Third, the club, the academy, or the agent. This is a proxy for a quality signal when no other signal exists. It works well enough to perpetuate itself, and badly enough to discard the wrong people.

Fourth, the narrative. A player with a good story gets paid more than a player with good numbers. Not because the market is foolish — because the market only pays for what it can see, and the story is always visible.

Now the important question. When was the last time a Vietnamese player was priced by the international market using data rather than narrative?

The answer: when they entered a ranking system with an audit trail. The World Amateur Golf Ranking, WAGR, launched in 2026 and co-administered by the R&A and the USGA, processes amateur results worldwide under a single formula. It is the only data pipeline in the lower tiers of world golf that runs reasonably smoothly, because it requires exactly one thing: tournament results, reported to standard.

Nguyen Anh Minh is the clearest example I have tracked over several years. This Vietnamese amateur carries a WAGR listing, and it was that ranking position — not any newspaper article — that opened doors for him at international events and in the eyes of American college recruiters. When he played well at an Asian event, WAGR recorded it. When he played well at a domestic event outside WAGR recognition, the system recorded nothing at all — and to the market, that round barely existed.

Here is the point I want to drive home: a round that is not entered into an audited system is not a bad round. It is a round that did not happen.

And this is where the subsidy appears. The organiser of a domestic event saves the cost of recording data. That saving does not evaporate. It is converted into an advantage for the players who already have data, already have names, already have agents. Nobody does anything wrong. A cost simply goes unpaid, and the invoice is sent to the best player nobody has heard of.

OWGR: A Data Machine That Blocked Its Own Input

If the Asian Tour illustrates missing infrastructure, OWGR illustrates a different kind of deficit — a missing shared definition. It is more serious, because it happened at the very top of the sport.

OWGR was founded in 2026 and runs on points derived from finishing position, field strength, and a minimum divisor across a rolling two-year window. Technically it is a pure outcome index, not a skill index. It does not measure how well a player strikes the ball. It measures where a player finished, in a field whose quality is estimated.

That was acceptable for forty years, until LIV Golf arrived.

LIV Golf launched in 2026 with a different format: fifty-four holes instead of seventy-two, no cut, a largely invitational and locked field, and a shotgun start. LIV applied for OWGR points in July 2026. On 10 October 2026, OWGR rejected the application.

The stated reasons were technical: a fifty-four-hole format is not equivalent to seventy-two; the absence of a cut removes a filtering mechanism; an invitation-based, locked field provides no open pathway for outside players. On principle, these are valid arguments. A ranking system has to compare apples with apples.

But look at the data outcome, not the reasoning.

Over the following two seasons, a group of dozens of major-calibre players — including major champions and former top-10 players — competed continuously without accumulating ranking points. Their points column stood still on the official system while they kept winning tournaments. Their ranking position slid not because they played worse, but because the denominator of the division was running backwards.

This is precisely the N/A column I stared at three in the morning, only at a different scale. In my spreadsheet, the column was empty for want of hardware. In OWGR, the column was empty for want of a shared definition. Both produce the same thing: a table that looks complete and is hollow at exactly the most important rows.

And both trigger the same reaction: the market builds its own measurement system. LIV created its own season-long individual title — Dustin Johnson took the first in 2026, Talor Gooch in 2026, Jon Rahm in 2026. Sponsors learned to read two parallel rankings. Fans learned to choose which one to believe.

Here I want to say something that will probably cost me a few followers. Creating a parallel measurement system is not a solution. It is a symptom. When both sides control their own data and neither accepts the other's, the sport loses the one thing that makes comparison meaningful: a shared definition.

What golf needed in that situation was not two better rankings. It was an independent measurement body with access to the raw data of both systems, operating under a public protocol. In medicine, that is called an independent clinical trial. In aviation, it is called an independent accident investigation authority. In golf, it does not exist.

Machine Learning Will Not Rescue a Course Nobody Measured

After every piece I write about data gaps, the same question arrives: why not use a model to fill the blanks?

I have tried. Many times. And I learned something I regard as the first principle of this trade.

In statistics, missing data comes in two types. The first is missing at random — lost without correlation to anything in the system. The second is missing systematically, meaning the probability of a cell being absent depends on the very thing that cell was supposed to measure.

Golf's third-tier data is missing the second way, and in the worst possible form.

Look at the mechanism. An event has shot data when it has enough money. An event has money when it has a major sponsor. A major sponsor arrives when there are famous players. Famous players arrive when the event has enough money. The loop locks itself, and it locks precisely where the players who most need measuring are the least likely to be measured.

Now imagine filling those blanks with a model. Any imputation algorithm will learn from the observable portion — meaning from PGA Tour players, on American courses, at high green speeds, on Bermuda and bentgrass. The imputed values will carry that entire bias. A Vietnamese player competing on a course with a different grass, slower greens and strong seasonal sea wind will be assigned a strokes-gained figure calibrated to a course half a world away.

And worst of all: the figure will look entirely reasonable. It will have two decimal places. It will sit in a bordered table. It will look right.

I walked into exactly that trap once, in a different sport, and it taught me more than any statistics course.

In 2026, still an International Communication student in Binh Duong, I spent three months building an xG model in Excel to analyse twenty-six rounds of V.League. The result produced a piece that was mocked fairly heavily: Quang Nam won the title with average possession of only forty-eight per cent — lowest among the top five — but with a shot conversion rate among the highest in the league, around seventeen point five per cent. Three months later Quang Nam were crowned, and the piece passed two thousand shares.

What I did not mention in that article is that my model worked because I had event data: shot location, angle, phase type. Without those, the model would have said nothing at all.

Three years later, in 2026, Becamex Binh Duong hired me as a data assistant, and COVID closed every stand. When I split the data by crowds and no crowds across forty-two matches, the home win rate fell from forty-nine per cent in the 2026 season to thirty-eight per cent. The coaching staff wanted to keep the home and away game plans unchanged. I objected, presented the comparison table, and proposed shifting to proactive defending away from home. The team won four of the next five.

Empty stadiums in 2026 made me ask: does home advantage come from the ground or from the crowd? The data has an answer.

I tell these two stories because they are two faces of the same lesson, and that lesson applies directly to golf. In 2026, the model worked because the input was sufficient. In 2026, the model worked because I split the context instead of pooling it. At the event with the all-N/A approach column, the model did not work because there was nothing to run.

I hate uncertainty. But 2026 taught me that one unforeseen variable can be stronger than any algorithm.

The Counterintuitive Angle: More Data Is Not Automatically the Answer

By now the conclusion seems obvious: golf lacks data, so collect more. I do not entirely agree, and this is the section for practitioners.

First, the ShotLink corpus is itself a biased sample. It covers almost all of the PGA Tour — meaning one very specific course archetype: irrigated fairways, fast greens, Bermuda or bentgrass, American design. When someone builds a "course fit" model on that foundation and applies it to a Scottish links or a tropical Asian course, the model is extrapolating outside its training range. Correlation is not causation. A player's strong approach play in Phoenix does not prove strong approach play on a windy coastal course.

Second, the industry misprices putting. Broadie's decomposition shows that at professional level, putting contributes roughly fifteen per cent of the scoring difference between players, while the approach category accounts for the largest share. Yet judging by how broadcasters and sponsors behave, you would think putting were half of it. The cause is structural: the putt is the easiest shot to film, easiest to commentate, easiest to turn into a moment. And putting is the least stable skill over time. A hot three-round putting streak is routinely extrapolated linearly into a good season. That is the most common analytical error in golf.

Third, the biggest risk is not missing data. The biggest risk is data that looks complete. A spreadsheet with visible blanks makes a careful analyst careful. A spreadsheet already filled by imputation makes him confident. Between those two risks, I take the first every time.

Fourth — and this is the point I think golf analytics has under-exploited — the strongest signal in golf may be the one nobody wants to tabulate: recovery behaviour after a bad hole. Golf has a term for it, bounce back, the ability to regain a stroke immediately after losing one. It exists inside PGA Tour data but almost never appears in player valuation discussions. A player with an average strokes-gained profile and a high bounce-back rate is a fundamentally different asset from one with an average profile and a low bounce-back rate. I have never seen a transfer model that takes that variable seriously.

Here I also want to say plainly how we read metrics: most current valuation models overrate young players' potential based on early-career strokes-gained figures, and underrate unmeasurable factors such as dressing-room chemistry and weekend pressure tolerance. It is the same error at both ends — trusting what is measured, ignoring what is not, and calling that objectivity.

I once wrote about a similar case ahead of a major tournament, when a national team was rated highly on possession share. I was not being clever. I simply did not believe the myth.

Plan B: A Minimum Measurement Infrastructure for Vietnamese Golf

Having stated the problem, I owe you the plan. This is my working principle: any analysis without a failure point and an alternative path is just criticism.

A full ShotLink installation costs more than any Southeast Asian tour can currently afford. But ShotLink is not the only starting point. There is a minimum dataset a domestic professional tour can collect at very low cost.

The minimum set has five fields per shot: driving distance, lie type after the first shot, distance to hole on the next shot, club used, and outcome. Those five fields are enough to build a crude Strokes Gained without absolute coordinates. No expensive GPS required. It needs a phone application and two volunteers per group.

Operationally, the real cost sits in three places. The first is labour: a four-round event with thirty groups needs roughly two hundred forty volunteer shifts per round. That is not a financial barrier, it is an organisational one, and it is solvable through relationships with local golf clubs and universities with sports-science students.

The second is definition standardisation. This is the genuine failure point. If each event defines GIR differently, if each volunteer judges fairway hits differently, the data cannot be pooled. Data without a shared definition is a dialect, not a language. This is precisely the OWGR lesson: two parallel measurement systems that cannot agree on definitions cannot compare anything.

The third is publication. Data collected and kept in an internal machine creates no valuation value. Once the data is published openly and stably across at least two seasons, it begins to be ingested by international analytics platforms. At that point a Vietnamese player with a good strokes-gained figure becomes visible to people who have never set foot on that course.

On baselines, one misconception needs clearing. Many assume the baseline must be the PGA Tour baseline to carry value. It must not. A baseline only has to be correct for its own ecosystem. A baseline built from three seasons of VGA Tour data, with slower greens and a different monsoon wind, will produce meaningful figures inside that ecosystem — and precisely because it differs, it has comparative value. The one thing to avoid is applying the American baseline to Vietnamese data and calling it an international standard.

The failure point to anticipate: if after two seasons only three events have joined, the sample is too small to build distance-banded baselines, and the whole system must be downgraded to a two-factor version — putt counts and greens hit per hole. The two-factor version still beats a bare scorecard. No option is all-or-nothing.

On the human factor, I want to add a short passage I usually leave out of technical reports, though I should not. Players in the no-data tier carry the psychology of the unseen. They play events nobody counts, win events nobody records, and tell themselves the door is closed on merit. In most cases I have examined, the door was closed because nobody knocked. A measurement system does not create good players. It only makes good players findable. Those are different things, and they differ in that one is development and the other is access.

What to Watch in the Next Round

Looking at the whole picture, the signal I am tracking is not any player's statistic. It is an infrastructure event: the first Asian event to publish audited hole-by-hole shot data and keep it public across an entire season.

When that happens, it will change how everything downstream is read. Scouts will finally have grounds to price Southeast Asian players by metric rather than by nationality. Federations will have grounds to select national teams by contribution rather than by standing. And players performing well in obscurity will have one row in some table somewhere, instead of an empty cell.

I do not predict. I read the data and accept the consequences.

As for the blank cell in my spreadsheet from that night, it is still there, in an archived file. I keep it for a reason. If someone ever asks me why a small event should pay for a data recording system, I will open the file and point at column eleven. That blank is more honest than any number I could have invented to fill it. And if you run a tournament, ask yourself this: if this season ended and all your event's data were deleted, how many players in your field would vanish from the sport's history? If the answer is all of them, the problem is not the players.