Ireland has tens of thousands of ringforts, long known as fairy forts that it was unlucky to disturb. I set every one in the national record against other earthworks of similar build to see whether the ringforts survived better. At first sight they did. But most of that advantage comes from how sites are classed, and the fairies are hard to find in the evidence.
A ringfort is a round farmstead of the early Middle Ages: a circular bank and ditch, or a drystone wall, around a family's houses and animals. In Irish they are the ráth and the lios, or the caiseal and cathair when built of stone, and those words run through thousands of place names.
They are also fairy forts. Folklore across Ireland held that the aos sí, the people of the mounds, lived in them, and that ploughing, levelling or cutting a thorn bush in one brought bad luck. People often say that this belief is why so many ringforts survive. It is a lovely idea, and it can be tested, at least in part.
The National Monuments Service keeps a record of every known archaeological site in the Republic: 151,308 entries. 31,431 of them are ringforts, 21% of the whole record. For many, the record notes how the site was drawn on the first Ordnance Survey six-inch maps of the 1830s and 1840s, how it appears on the latest edition, and what the surveyors found on the ground. That lets me ask a simple question. Did ringforts survive better than other earthworks of similar build and age that carried less of the folklore?
Each dot is a monument. The colour shows what its notes say about it now: still visible, damaged, or no visible trace left. Switch to the hexagons to see how thick on the ground the ringforts are, or where most have gone. Use the buttons to add the other earthworks I compare them with, or search for a townland.
The ringforts cover the country, but not evenly. Across the Republic there are about 0.45 per square kilometre, one for every 2.2 km². Sligo is the densest county, with 1.25 per km²; Dublin the sparsest, with 0.08. Cork has the most, 3,863. In the busiest hexagons there are more than three ringforts in every square kilometre, and 106 of the 813 hexagons have at least one per km².
I use two measures of survival, and neither is perfect.
On the maps. The first Ordnance Survey six-inch maps of Ireland were surveyed in the 1830s and 1840s, and they drew a great many forts. The record says, for each site, how it was drawn on that first edition and on the latest edition. Where the latest edition gives a date, the median is 1921 (half fall between 1916 and 1927). So, for every earthwork drawn around 1840, I can ask: was it still drawn about eighty years later? That covers the age of the Famine, emigration and the great shift from tillage to grazing.
In the notes. For ringforts, enclosures and souterrains, the record holds the surveyors' descriptions, most of them written for the county inventories from the 1980s onwards and revised since. I read every note with a set of rules that pick out phrases such as "levelled", "no visible surface trace", "destroyed", "removed", "not visible at ground level" or "visible only as a cropmark". Notes that mention only partial damage, such as a bank levelled at one side, count as damaged, not gone.
A check against a fresh random sample of 60 notes, each read in full, found the rules agreed 54 times out of 60. In 4 cases the rules called a site gone when part of it survives, and in 2 they missed a site that has gone, so the errors lean slightly towards overcounting loss. 3,021 ringfort records (10%) and 2,359 enclosure records only hold a placeholder saying the description is not yet online. I leave those out. They are not spread evenly: in Clare, 60% of ringfort records are placeholders.
If fairy belief protected ringforts, they should have survived better than earthworks that are similar in size and build but carried less of the lore. The record offers several: plain enclosures (round or oval banks the survey could not firmly call ringforts), barrows (prehistoric burial mounds, often ringed by a bank and ditch), moated sites (the square, ditched homesteads of Anglo-Norman settlers), mounds and earthworks of uncertain type.
On this measure ringforts did well. Of the 11,555 drawn around 1840, 88.6% were still drawn on the latest edition. Enclosures managed 70.1% and the miscellaneous earthworks 61.5%. But barrows, which are not fairy forts, did just as well, at 88.0%, and moated sites were close behind at 80.6%. Allowing for county, height and slope, ringforts come out 2.4 percentage points ahead of barrows and 1.9 ahead of moated sites, but 19.8 ahead of enclosures.
Ringforts kept their place on the map far better than plain enclosures. But barrows kept theirs just as well.
The notes tell a similar story with bigger numbers. Of the ringforts drawn around 1840 that have notes, 16.3% now have no visible trace. For enclosures drawn at the same time the figure is 42.4%. Adjusted for county, height and slope the gap barely moves: 16.4% against 40.2%. Counting partial damage too, 38% of these ringforts are damaged or gone, against 60% of enclosures.
That looks like a fairy premium of about 23.8 percentage points. I do not think it is, or not much of one, and the reason is in the chart. Look at the unclassified ringforts: 52% of them have gone, more than enclosures. The record's classes depend on what survives. A site with a good bank, ditch and entrance can be called a ringfort; one levelled before anyone looked closely often cannot, and becomes an "enclosure" or an "unclassified" ringfort. So the word "ringfort" in the record already means, in part, "survived well enough to be recognised". Comparing ringforts with enclosures builds survival into the comparison.
The map measure is touched by the same problem. Unclassified ringforts drawn around 1840 were still on the latest map in only 49.5% of cases, against 91.6% for raths. The fairest test I can make is to stop letting survival decide what counts as a ringfort, and pool ringforts and enclosures together as round earthworks. Of the 14,858 drawn around 1840, 84.5% were still drawn on the latest edition. Adjusted for county, height and slope that is 84.5%, against 86.3% for barrows and 86.6% for moated sites.
Count every round earthwork drawn in 1840, not just the ones later called ringforts, and they kept their place on the map no better than barrows or Anglo-Norman moated sites.
Land quality matters more than lore. I have no soil map here, so I use height and slope from a terrain model as rough stand-ins: lower, flatter land is easier to plough, drain and improve. Ringforts on low ground were much more often lost than those on higher ground. In the notes model, each extra 100 metres of height cuts the odds of a site having gone by about a quarter (odds ratio 0.75). Slope, at this scale, makes little difference (odds ratio 0.98 per degree).
The ringfort and enclosure lines stay far apart at every height, which fits the idea that the gap is mostly about classification. The ringfort line falls from 19% below 50 metres to 8% at 200 metres and over.
If fear of the fairies kept the plough away, you might expect the surveyors to hear about it. They rarely wrote it down. I searched all 47,024 notes for fairies, the sí, the púca, superstition, bad luck, ghosts and the word liosachán (a little fort, glossed in one note as a fairy fort). Only 52 notes match, 28 of them for ringforts: about 9.9 in every 10,000 ringfort notes. That says more about what an archaeological inventory records than about what people believed. But it means the record cannot show belief protecting any particular site.
The handful that do mention the fairies are a mixed lot, and several describe sites that went anyway.
Of the 28 ringforts whose notes mention such beliefs, 10 now have no visible trace. That is too few to compare, and a story may be more likely to be written down when a site has been destroyed.
Two other clues point at local attention. First, townland names. 1,799 of the 25,628 townlands holding these earthworks begin with Lis, Lios, Rath, Raheen, Caher or Cahir, or contain the word Rath, from the Irish words for a fort. Second, the 1840 map: some forts were written on it with a word, usually "Fort", or a name such as Caher, Lisheen or Cahermore; others were simply drawn.
Ringforts in fort-named townlands have gone slightly less often: 19.0% against 21.0% (adjusted, 19.2% against 21.0%). Forts named on the 1840 map look much safer at first: 12.6% of ringforts and enclosures named then have gone, against 24.2% of unnamed ones. But named sites are mostly ringforts (91%) and sit in counties with less loss. Among ringforts alone, adjusted for county, height and slope, the difference vanishes: 16.3% named against 16.2% unnamed.
In every one of the 16 counties with enough data, a smaller share of ringforts than of enclosures has gone. The median gap is 22.9 points, from 2.0 in Leitrim to 45.8 in Mayo. On the map measure, ringforts beat the other earthworks in 16 of 17 counties.
The county figures for ringforts alone vary hugely: from 2.8% gone in Leitrim to 36% in Longford. Some of that is real, but some will be how each county's inventory team wrote its notes. I would not rank counties on it.
Britain has its own small folk, and some of them are on the Ordnance Survey map. In OS Open Names I count 112 named features containing the words fairy, Puck, pixie, Hob or boggart: 70 fairy names, 21 Hob, 9 Puck, 8 pixie and 4 boggart. There are 20 fairy hillocks, and 20 of them are in Scotland. These are names on a modern map, not a survey of belief, and some Puck and Hob names may come from other words or surnames.
The honest answer is: not in any way the national record can see. Ringforts drawn around 1840 kept their place on the map far better than plain enclosures and earthworks, and their notes record far less loss. But barrows, which are not the forts of the fairy stories, survived on the maps just as well. Much of the gap with enclosures comes from how sites are classed, and when ringforts and enclosures are counted together the advantage over barrows and moated sites disappears (-1.8 and -2.1 percentage points). Any small edge the best ringforts have could come from their being bigger, better built, or simply useful, as easily as from fear of the sí.
None of this means the belief did nothing. Some people clearly did leave forts alone, and say so. But on the evidence of the national record, what protected Ireland's ringforts was mostly their number, their height above the best land, and luck. The fairies may have saved some forts, one field at a time. They have left very few fingerprints.
Data. The full Sites and Monuments Record for the Republic of Ireland (151,308 records in 470 classes), from the National Monuments Service's open data service. I keep 68,000 earthworks: ringforts (raths, cashels and unclassified, 31,431), enclosures (16,621), barrows of all kinds (3,901), moated sites (1,329), earthworks (2,009), mounds (1,219), souterrains (4,853) and fulachta fia (6,637). Notes exist only for ringforts, enclosures and souterrains.
Map survival. A site counts as drawn around 1840 when the record's first-edition field describes a depiction (hachured, indicated, outlined, embanked enclosure or a written name) and as still drawn when the latest-edition field does too. Fields that are blank or "NULL" are unknown and left out. The notes often mention the first edition too, but they mention it far more often for sites that have since gone, so I ignore those mentions (3,080 sites). For 16,177 ringforts the first-edition field is blank, and coverage varies by county: the field is filled for fewer than one in ten ringforts in Cork, Carlow, Cavan, Kerry and Laois, so those counties barely enter the map test.
Notes. Rules based on regular expressions, applied sentence by sentence, mark a note "gone" when it says the monument as a whole has no visible surface trace, has been levelled, destroyed or removed, is not visible at ground level, or is known only from aerial images or a cropmark; "damaged" when it mentions partial levelling or removal; otherwise "visible". The check on 60 randomly drawn notes is in the data folder.
Statistics. Shares with Wilson 95% intervals. Adjusted shares come from logistic regressions with county fixed effects (counties with fewer than 30 sites or no variation left out), elevation and slope, averaged over the sample as if every site belonged to each group. Elevation and slope come from AWS Terrain Tiles at zoom 10. Hexagons are drawn on the Irish Transverse Mercator grid.
Limits. The record is a survey of what was noticed and recorded, not a census of what was built. Survival is a proxy, and the classes partly depend on survival. Association is not cause: ringforts differ from the other earthworks in size, date, build and use. The notes for some counties are mostly placeholders. Northern Ireland has its own record and is not included.