Historic Golf Courses in the United States, Plotted by Coordinate

Filter the plot to the 1880s and 1890s and it collapses almost entirely into one corner. 5 of the 14 entries sit in the northeast (Yonkers, Southampton, Brookline, Newport, and the Pennsylvania outlier at Foxburg), and all 5 are inside a radius that a nineteenth-century train timetable would have made a comfortable day's travel.

That is the founding pattern in a single image. American golf was not adopted across a country; it was adopted by a small group of people who could reach each other. The clubs that wrote the rules in 1894 were the clubs whose members already met, and the plot shows why that was possible.

The one point that breaks the pattern is also the earliest by more than a century. South Carolina Golf Club plots at Charleston, SC, in 1786, eight hundred miles south of everything else on the grid and ninety-eight years before the next entry. Charleston was a major Atlantic port trading directly with Britain, so golf reached it first and by sea. Nothing spread inland from it, which is exactly what an isolated point on a plot looks like.

Move the decade filter forward and the distribution inverts. Everything from 1933 onward is southern, western or midwestern (Augusta, Pebble Beach, Kiawah, two Michigan resorts), and none of it is in the founding cluster. Golf's second American expansion followed climate and tourism rather than society, and it went where the season was long or the scenery sold. The two eras barely overlap geographically.

Two things about it are deliberate and worth stating, because both are places where a map can quietly mislead.

Each point is a locality, not a property. A golf course occupies somewhere between eighty and two hundred acres; a single dot cannot mean more than "this town". The coordinates here are the town, so the plot answers "where in the country" and refuses to answer "where on the ground".

No coastline is drawn. The frame carries its longitude and latitude rulings instead of a national outline. An outline reconstructed by hand would be wrong in ways a reader could see, and a misshapen country would undermine the accurate thing the plot is actually for. A graticule is honest about being a plot sheet.

The filters compose. Choosing an architect narrows the decade options to the decades that architect worked in, and any option that would return nothing is greyed out rather than left selectable, so a dead end is visible before it is chosen rather than after. Below 700 pixels the grid gives way to the list it already contains, because a plot crowded past legibility is worse than a table.

The plot has a second axis that is easy to miss, because it is not on the map. Of the fourteen places recorded here, seven were laid out between 1884 and 1893, half the record inside ten years. One entry sits a century earlier at Charleston in 1786, then nothing until the 1880s. After 1893 the record thins to 1921 and 1933, and picks up again only from 1966.

That shape belongs to this record and not to the country: fourteen documented places are a selection, not a census, and a national count would look different. What the shape does say is where the disputes come from. When half a set of foundings falls inside a single decade, the competing claims to be "first" are separated by a season or two rather than a generation, and the tie is broken by which test a reader applies rather than by the dates themselves.

Present condition is the other unmapped dimension. Seven of the fourteen are in continuous use, four operate as resorts, one is private, one was restored after being abandoned, and one is gone with nothing of the ground left. Age and survival are close to unrelated in this set: the oldest entry is the one that vanished.

The plot is embeddable. Paste the snippet below and it loads the same figure, with the same data, from this site, so it stays current as records are added rather than freezing a screenshot of today.

The snippet is two parts and both matter. The frame is the figure; the four lines after it listen for the height the figure reports and apply it, because no single fixed height fits. Below about 768 pixels the plot becomes a list and needs roughly 536 pixels of room, while at 1100 pixels wide it needs 859. The listener checks the sender's origin before trusting a message.

If your publishing system strips inline scripts, paste the frame alone: the height="760" attribute then applies and the figure is simply not fitted exactly. Nothing breaks either way.

Attribution is built into the frame rather than requested in a paragraph nobody reads: the embedded figure carries a link back to this page.