The Empty Cell in the Strokes Gained Table: When Golf Data Goes Silent

At nine in the evening I opened a Strokes Gained table and found the SG: Approach column completely empty. The shot-tracking system at the course had lost signal for the final nine holes of the second round; everything else was intact, enough for the table to look normal. Four people were in the room that night. Three read that empty cell the same way: “I don't see a problem.”

The most expensive mistake in golf analysis is not a wrong number. It is an empty cell read as a zero. A null field and a field holding the value zero differ in kind, and only one of them is a safe signal. Confusing the two does not produce a small margin of error; it produces a fully formed wrong conclusion, delivered with a tidy appearance.

Numbers do not lie. An empty cell, however, says nothing at all, and silence is always filled in by whoever is reading.

Golf has the most clearly stratified data infrastructure of any sport. On the PGA Tour, ShotLink records every shot — distance, lie, ball position — and from that source Strokes Gained became the standard language: SG: Off the Tee, SG: Approach, SG: Around the Green, SG: Putting. Each metric measures a player's advantage over the tour average from the same distance and lie. That is why SG: Approach is regarded as the metric most strongly correlated with scoring.

The picture thins out fast once you leave the elite tier. The DP World Tour and the Asian Tour track fully at only a share of events. Domestic tours, including the Vietnam Golf Association's schedule, mostly publish scorecards. No tracking means no SG: Around the Green. No wind and elevation data means course-fit analysis loses one term of the equation entirely.

Even where data is richest, coverage has limits. Four rounds generate roughly 280 shots, but nearly half of those are putts, while approach shots number only about forty. A weekend of brilliant putting is a sample of sixty to seventy putts. That is far too small to support any long-term conclusion.

Three misreadings of the empty cell recur in the reports that reach me.

The commonest is assigning zero to a blank. A player with no SG: Putting figure in the table gets filed under “no putting problem.” In reality he belongs to “not yet measured.” Those two sentences lead to two different decisions before the final round.

Another form is applying PGA Tour baselines to every course. Tour-average SG: Approach is built on fast greens, thick American-standard rough and mostly low elevation. A course in Dong Nai or Da Nang has different grass, humidity and green speed. Grading a shot there against the old baseline produces systematic bias, not random noise.

More dangerous still is turning four rounds into a trend. Rory McIlroy led the 2026 US Open at Pinehurst until the last two holes, then missed short putts at the 16th and 18th and lost to Bryson DeChambeau by a single stroke. Read only that week's SG: Putting and you can write a story about a putting crisis. Read the whole season and you find a player whose putting sat near tour average, with the failure located in back-nine pressure rather than technique. The same data, two opposite conclusions, separated by sample size.

The same holds for points systems. A player who misses the cut after 36 holes collects zero prize money, zero OWGR points and an “MC” line in his record. That zero is real, but it explains no cause. It states only that the tournament ended early for him. To learn why, you need hole-by-hole shot data — which most scorecards do not supply.

Then there is the system layer. The FedExCup grants the standings leader a starting-strokes head start at the season finale. In 2026, Scottie Scheffler entered the Tour Championship at -10 and Rory McIlroy still won. If the input standings are wrong by a single week, the whole forecast model downstream collapses — not because the algorithm is weak, but because the input was skewed.

At the equipment layer, the USGA and R&A ball reform announced on 14 March 2026 applies to elite competition from January 2028 and to recreational play from 2030. Every projection of its effect rests on flight data gathered under current conditions. That is an extrapolation problem, and extrapolation always requires stated assumptions.

Numbers do not lie. Reputation, though, whispers into the ear of anyone who does not read the table.

The counterintuitive point sits here: the biggest problem in golf analytics is not a model that predicts badly. It is the habit of filling empty cells with assumptions and then presenting those assumptions as findings. A report with three measured metrics and two missing ones is usually read as a report with five metrics, two of which are “fine.” Player reputation fills the rest: famous names default to fine, quiet names default to troubled. The correlation between reputation and long-run putting ability is far weaker than broadcast copy assumes.

Another blind spot belongs to how the industry prices potential. Golf's data models rate driving distance and youth very highly, while what decides final-round victories tends to be short-game stability and pressure handling — qualities that only emerge across several seasons. And when young players are pushed into packed schedules at sixteen or seventeen, bodies that have not finished developing absorb an adult's shot volume. Back and wrist injuries in that group are not accidents; they are the output of a model optimising the wrong variable.

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

For Vietnamese golf, the data gap is an opportunity, not a source of shame. A scorecard archive covering ten years still yields scoring average by hole, birdie rate on par-5s and closing-three-hole performance — enough to separate steady players from explosive ones. The task is not to wait for ShotLink. The task is to state clearly which metrics remain unmeasured, and never to let a blank sit on a table without a note.

If every published statistics table carried its own missing-data rate alongside it, how many sports narratives would have to be written from scratch?