Where did Britain put its gallows? Hundreds of hills, woods and fields on the Ordnance Survey map still carry the word. I measured 340 of them for height, roads, towns, views and boundaries, and compared each with fifty ordinary places of the same kind nearby.
Every county once had a place where people were put to death in public, and many of those places kept the name long after the gallows had rotted away. The Ordnance Survey's open gazetteer of Great Britain still lists 296 names built on Gallows or Gallow, from plain Gallows Hill to Gallowstree Common and Gallows Knowe, and 44 built on Gibbet. Most are hills: 67% of them.
It is often said that these places were chosen with care: on a rise, beside a main road, just outside the town but in full view of it, and perhaps on a boundary, so the dead stood at the edge of the community that had cast them out. I wanted to know how much of that the map itself can confirm.
So I took every Gallows and Gibbet name and measured five things: how high it stands above the ground around it, how far it is from a town, whether a person in that town could see it, how far it is from a main road and from a Roman road, and how close it lies to a boundary. Each name was then compared with fifty ordinary names of the same kind (hills with hills, woods with woods) drawn at random from the same 50 km square, and with fifty random points of land.
The short answer: gallows names sit close to towns and very close to roads, much more than ordinary names do. They are on hills, but on low, modest hills rather than the most commanding ones. More of them are in sight of the nearest town than chance would give, but that is mostly because they are nearer. And the boundary story, at least at the resolution I can test, is weak.
This is a companion to Bloody Place Names, which asked whether Battle, Slaughter and Gallows names mark real battles. This study leaves battles aside and asks only where the execution places stood.
Each dot is a named feature on the OS map, coloured by the word it contains. Hangman names (a weaker class) and Hanging names (which usually mean a steep slope) can be switched on for comparison. The ringed marks are four gallows that are items in Wikidata, and the small green squares are village stocks and pillories mapped in OpenStreetMap. Tap any mark for its measurements.
The map has two halves. In England the names are mostly Gallows Hill (42 of the 79 on the map) and Gibbet Hill, spread across the lowland counties. In Scotland the usual form is Gallow Hill (85 of 103), thickest in Aberdeenshire (47 names), Dumfries and Galloway (27) and Moray (18). Scotland has 184 of the names, England 154 and Wales only two. I cannot tell from the map why Wales has so few; it may be a matter of language, since the OS records many Welsh names in Welsh.
The clearest result is distance to a town. I measured from each name to the nearest place the OS calls a town or city. Of the Gallows and Gibbet names, 31.5% lie within 3 km of one. Of the matched ordinary names, only 15.1% do, and in 2,000 replicate draws of matched names, none came close (chance below 0.001). The median distance is 5.0 km, against 6.7 km for matched names.
But they are rarely in the settlement itself. Only 5.3% lie within 500 m of the centre of a village or town. Village stocks and pillories, the punishments that were meant to be in the middle of things, make the contrast plain: the 89 mapped in OpenStreetMap sit a median of 266 m from a village centre, and 66% are within 500 m. The gallows names sit a median 1.7 km away.
The second strong result is roads. 29.1% of Gallows and Gibbet names lie within 500 m of a modern A-road, against 16.6% of matched names. Because A-roads run out of towns, I repeated the test with controls that are also matched on distance to the nearest town: the matched share rises only to 19.0%, and the gap remains (chance below 0.001). Gibbet names are the most roadside of all: a median of 542 m from an A-road, and 41% within 500 m.
Modern A-roads are not medieval roads, but many follow old main routes. The Roman roads in the Itiner-e atlas are a fixed, older network, though thin outside England. In England, 21.4% of the names lie within 1 km of a Roman road, against 12.8% of matched names (chance 0.001).
Gallows names really are on raised ground. Measured against the mean height of the ground within 1 km, 72% of them stand higher than their surroundings, against 47% of random points of land. But that is what you would expect of any named hill, and against other named features of the same kind the picture flips. Taking only the hills, a Gallows hill rises a median of 7 m above the ground around it; matched named hills rise 15 m. The Gallows hills are also lower in absolute terms: a median 87 m above sea level, against 166 m.
The gap comes mostly from Scotland, where gallows names (hills and others) rise a median 7 m above the ground around them against 15 m for matched names (chance below 0.001). In England, where named hills are lower anyway, gallows names are no less prominent than their neighbours (4.6 m against 3.4 m, chance 0.115).
The gallows went on the first rise outside town, not on the highest hill in the district.
That fits the town result. A low hill close to a town and beside its road is a different kind of place from a high summit, and it is the kind of place the gallows names favour. Some still stand well above their surroundings: the highest in my set is a Gallow Hill in Angus, 65.4 m above the ground within 1 km.
For each name within 10 km of a town, I traced a line of sight from 2 m above the town's centre point to 3 m above the name's point, across the terrain model, allowing for the curve of the earth. 36% of the Gallows and Gibbet names are in sight of their nearest town, against 29% of matched names (chance 0.009). Nearer things are easier to see, so I repeated the test against controls at similar distances from a town: they reach 31%, and the gap narrows (chance 0.023). From the nearest village within 5 km, 56% are in sight, against 46% of distance-matched controls.
So the gallows names are a little more visible than ordinary places, and most of that edge comes from being close. Being in sight was a by-product of being on a low rise near town, or the other way round; the map cannot tell which was the aim.
The boundary idea is the hardest to test, because I have no map of medieval parish boundaries. I used three stand-ins. First, the edges between Domesday hundreds, drawn halfway between neighbouring Domesday places recorded in different hundreds, which only works for England and only roughly. Second, the position of each name between its two nearest Domesday places (1 means exactly halfway, 0 means on top of one). Third, modern district and county boundaries, which in places still follow older lines.
The signal is weak. 25% of names lie within 500 m of a hundred edge, against 19% of matched names: in the expected direction, but a gap that luck produces in 10% of draws. Against distance-matched controls the chance is 0.033, and with so many measures tested, one or two values of that size are expected by chance. The median position between neighbouring Domesday places is 0.71, against 0.66 for matched names. Modern district boundaries show nothing at all (12% within 1 km, against 15%).
That does not mean gallows were not on boundaries. A boundary drawn as a straight line between two village points can be a kilometre or more from the real one, and that blur would hide a real pattern. It means only that this map, at this resolution, cannot show it.
Hangman names are few (37) and behave like ordinary names on almost every measure: they are no nearer towns or roads than matched names. Some may well be execution places; others, such as the coastal hills Great Hangman and Little Hangman in Devon, may have nothing to do with hanging. I keep them apart and draw no conclusion from them.
Hanging names are a useful test of the method. In English place names "hanging" often describes ground on a steep slope, as in a hanging wood. If that is right, Hanging names should be steep, and they are: a median slope of 3.6 degrees, against 2.2 for matched names, with 34% on slopes over 5 degrees against 17%. They are not nearer towns or roads. Gallows and Gibbet names show the opposite: only 8.5% are on slopes over 5 degrees, against 13.7% of matched names.
Only four gallows or gibbets in Britain are items in Wikidata with a location, and OpenStreetMap maps just two (both among the four). They are far too few for statistics, but they show the same habits. The measurements below come from the same code as everything else; the descriptions link to Wikipedia.
Two more of the best-known execution places in England are not in the data as points, so I leave them off the map: Tyburn, London's main place of execution for centuries, near today's Marble Arch, and the Knavesmire, York's place of execution, now the racecourse.
I searched all 267,744 OS Open Names features of type populated place, landform, landcover and hydrography for names containing "gallow" (leaving out Galloway), "gibbet", "hangman" and, as a contrast, "hanging" at the start of a word. That gives 296 Gallows, 44 Gibbet, 37 Hangman and 194 Hanging names. A match means only that the letters are there. OS Open Names leaves out field names and most minor names, so many former gallows sites are missing, and some features (a Gallows Hill and a Gallowhill Wood beside it) are two names for one place. Counting names within 1 km of each other only once leaves 306 places; the main results hold (see the table below).
For each name, I drew 50 ordinary OS names of the same local type (hill, wood, valley and so on) from the same 50 km grid square, falling back to 100 km or the whole country when fewer than 20 were available; names with grim words were excluded from the pool. I also drew 50 random points of land in the same square. Each test compares the real names with 2,000 replicate sets, each taking one control at random per name. Chance values are two-sided. The distance-matched controls are the 10 of the 50 closest to the name in distance to the nearest town.
Ground height comes from AWS Terrain Tiles at zoom 10 (cells of roughly 100 m in Britain), read with bilinear interpolation. Height above surroundings is the height at the point minus the mean of 64 samples on rings of 250, 500, 750 and 1,000 m. Slope uses samples 100 m either side. Towns are OS Open Names places of type City or Town, and villages also include type Village; distances are to their label points, not their edges. Lines of sight run from 2 m above the town point to 3 m above the name, sampling the ground every 45 m with earth curvature and standard refraction, as in Within Sight; the model has no trees or buildings. Roads are OpenStreetMap A-roads (modern) and the Itiner-e Roman roads, corrected for a datum shift as in How Roman Is Your Road. Boundaries are modern ONS district and county lines (shared edges only, so the coast does not count) and Voronoi edges between Domesday places in different hundreds or counties, for points within 3 km of a Domesday place in England.
The OS point for a hill or wood is a label point, not the gallows site; the gallows may have stood a few hundred metres away. Today's towns are not medieval towns, and today's A-roads are not medieval roads, though many follow them. Many gallows belonged to manors or baronies rather than towns. The Domesday hundred edges are crude. I ran 26 measures on several groups of names, so a single chance value near 0.03 should be read with caution; the strong results (towns and roads) have chance values of 0.001 or less and hold in every version of the test.