Not Enough Data: The Hardest Discipline in Golf Analysis

On the 18th green, the ball rolled from about eleven metres, caught the edge of the cup, spun half a turn, and dropped. The gallery erupted as if a derby had just produced a ninetieth-minute goal. In the press room, the story had already been written: his putter was back. I opened my laptop and checked ShotLink. That week he had gained 4.3 strokes putting. But when I stretched the window out to twenty weeks, the contribution fell to 0.2 strokes per round, barely different from the field average. The "comeback" the whole gallery had just sung into being existed only inside seventy-two holes. I have followed golf long enough to know that most wrong calls in this sport do not come from bad data. They come from using good data on too small a sample. One holed putt is a single event; a career is tens of thousands of events. The distance between those two things is where every analytical error is born. In 2026, the PGA Tour began recording ShotLink on nearly every shot of every round. In 2026, Mark Broadie, a professor at Columbia Business School, published the Strokes Gained method; by his 2026 book "Every Shot Counts," it had become the common language of golf. Before that, we judged by scoring average, by greens in regulation, and by eye. After that, we could split a round into four skill groups: Off the Tee, Approach, Around the Green, and Putting. That data revolution freed the analyst, but it also handed everyone a licence to conclude. Television broadcasts, podcasts, and what I still call the second gallery — social media — all had numbers in hand. In 2026, I realised the second gallery has no seats but it does have real people. And when everyone has numbers, everyone thinks they have the truth. The problem lies in sample size. A golf tournament is seventy-two holes, usually across four days. A PGA Tour season runs twenty to twenty-five events. Each week, a golfer makes only about seventy to eighty shots, divided among four skill groups. That means each group has only fifteen to twenty observations. With a sample that small, the standard deviation is large enough that a good week may be pure noise, and a bad week too. The irony is that not every skill group is equally noisy. In the data that Data Golf and the research community have accumulated over more than a decade, Approach (SG: Approach) is the most stable and the most predictive. Off the Tee (SG: Off the Tee) is also relatively trustworthy. But Putting (SG: Putting) is the most volatile variable of all: a golfer can gain three strokes putting one week and lose three the next without any change in technique. In other words, the skill audiences love most — the decisive putt on the 18th green — is the one that says least about the long term. We are drawn to what is easiest to see, while the truth sits in what is harder to see. An approach from 180 metres into a firm green draws no roar, but it repeats far more reliably. A useful distinction: there are descriptive stats and predictive stats. Descriptive stats tell you what happened; predictive stats tell you what is likely to repeat. A single-round scoring average is descriptive. SG: Approach sustained over months is predictive. When a broadcast blends the two, it turns a past event into a promise about the future, and that is where the error begins. This is where Broadie overturned a golfing cliché. "Drive for show, putt for dough" sounds entirely reasonable, but at the professional level he showed that approach play and driving explain most of the scoring gap between elite golfers. Putting matters more for amateurs, where raw skill gaps are still wide. At the top, everyone putts well; what separates them is the ability to hit the ball close. Based on my experience covering tournaments, the most common mistake in coverage is assigning cause to a hot putting week. People call it "finding form," "regaining belief," when in reality it is usually statistical variance. A long putt holed is skill plus luck; four good putting rounds in a week can be luck alone. To see the opposite, the truly large sample sits elsewhere. Scottie Scheffler led the PGA Tour in Strokes Gained: Total in 2026 and collected a set of titles including the Masters and Olympic gold. Rory McIlroy completed the career Grand Slam at the 2026 Masters, a milestone that only appears after more than fifteen years of data. There, the sample is large enough to conclude: not one week, but an entire career. Then there is the question of course fit. Every course has a technical portrait: wide or narrow fairways, firm or soft greens, altitude above sea level, grass type, and prevailing wind direction. A golfer may thrive on a wide, long course yet struggle on a narrow one that demands precision. But a player's personal history with a course is usually only three or four appearances, far too few to call a sample. We still call it "course experience" and turn it into a predictive criterion, when statistically it is nearly silent. There is another trap: the technique-transition period. When a golfer changes coach or alters the swing motion, the body needs twelve to twenty months to stabilise. During that window, competitive results are nearly meaningless as a measure of ability. Yet the leaderboard still appears every week, and someone still concludes that he is "finished." That judgment is right in the numbers and wrong in the timing. The same logic applies to team events. The Ryder Cup or Presidents Cup gives each golfer only a few matches, usually three to five across three days. With a sample that small, a missed putt on the 17th hole can decide an entire week, an entire legend, an entire media fate. A team is not led by tactics alone, but by people calling each other's names — and at the Ryder Cup, that call grows loud enough to hide the fact that the sample is too small to say anything certain. I learned this lesson in a year without spectators. In 2026, when the pandemic closed every stadium, the PGA Tour returned at Colonial in June to empty fairways. The wind I recorded that year still blows in me whenever the course is empty. With no roar to deceive the senses, I was forced to listen through the numbers, and I realised that many things I once believed were only the echo of a crowd. And I realised this is true of my own craft. When I write about a golfer, I always have to choose: tell the story that is easy to hear, or the story that fits the sample. The second is shared less often. But it is the only thing still standing once the season closes. Now it is signing season. LIV Golf, sponsorship deals, equipment contracts, and the argument over the Official World Golf Ranking. In the rhythm of the transfer market, everyone watches the clock; I listen for departing footsteps. Every report of a golfer moving to LIV comes with a grand narrative: about money, about loyalty, about the future of the sport. But strip away the storytelling and mostly what remains are a few verifiable facts — contract value, length, and release clause. The rest is noise. In signing season, the only way to filter the noise is to rank rumours by evidence. A signed contract on paper is fact. A negotiation confirmed by two independent sources is possibility. A post from an anonymous account is noise. Readers deserve to know which rung of that ladder they are standing on. This is where I think of the hardest discipline in analysis: saying "not enough data." In a serious analytical system, when the input is empty, the only correct conclusion is "insufficient information, cannot assess." Not a bold call, not a prediction that pleases the audience, but a blank column left blank. But the sports media industry does not reward silence. It rewards conclusions. A report willing to say "I do not know yet" is treated as weak; a report willing to assert after seventy-two holes gets shared. That reward mechanism creates a paradox: the more data we have, the more confident we become, and the more confidently wrong. I once saw a full eight-dimension analysis, with cells for technique, form, tournament, governance, rules, risk, public narrative, and industry value chain. Every cell read "insufficient information." At first I thought it was a failure. Then I realised it might be the most honest document of the week: it refused to fill the gaps with speculation. A blank table is more useful than a wrong conclusion neatly presented. A lack of data has never been the biggest problem. We have had ShotLink since 2026, Strokes Gained since 2026, systems measuring every putt. The problem is that we have learned to use data to confirm the story we want to tell, rather than to let data refute it. That is overfitting in thought: fitting a model to a small sample, then believing the model will hold on a large one. An empty course, and the wind still keeps time for the ball. Next time a report declares that a golfer "is back" after one Sunday afternoon, I will open the twenty-week window before I open my praise. And if that window is empty, I will leave it empty, because holding a blank column is sometimes the only way not to lie.