Between 1760 and 1900 the coasts of Britain and Ireland were ringed with new lights. I counted the wrecks recorded at each spot before and after its light was first lit.
Lighthouses were one of the great public works of the eighteenth and nineteenth centuries. Between 1760 and 1900 new lights went up on rocks, headlands and sandbanks all round Britain and Ireland, some of them on reefs that took years to build on. The obvious question is whether they worked. Once a light was lit, did the wrecks at that spot stop?
I cannot count the wrecks that never happened. But Wikipedia's volunteers have built something close to a register of the ones that did: month-by-month lists of shipwrecks, compiled largely from Lloyd's List and the newspapers of the day. Most entries say where the ship was lost: "wrecked on the Gunfleet Sand", "driven ashore at Flamborough Head". So for each lighthouse I counted how often the record names its rock, headland or sandbank, year by year, and compared the years before its first light with the years after.
The answer is less tidy than I expected. Wrecks named at a site rose by about 37% in the five years before it got its light. Lights were lit where ships had just been lost. After the light, mentions were about 12% lower than in the earlier, quieter years, but the 95% interval runs from 31% lower to 12% higher. That is consistent with a modest fall, and also with none.
The record is not a steady instrument. The chart below shows how many entries I could parse for each year. Before 1750 most yearly pages redirect to decade pages I did not have, so the study starts in 1750. In the 1760s there are about 200 entries a year. In the 1850s there are 2,409, and the busiest year before the First World War, 1880, has 3,844. Then the monthly pages stop. From 1891 there is one page per year, and from 1895 to 1913 the lists average 368 entries a year.
That matters more than anything else here. A place can look safer simply because fewer wrecks were written down. So I never compare raw counts across years. In the simple comparison I divide each site's mentions by the number of entries in British and Irish waters that year. In the models every calendar year gets its own baseline, shared by lit sites and comparison sites, which soaks up changes in how much the record holds. I stop in 1913, before war changes what the lists are about.
Of the 193,562 entries from 1750 to 1913, 104,832 name a place in Britain, Ireland, the Isle of Man or the Channel Islands. Together they mention my 179 sites 8,936 times.
The map shows every site I used. Drag the slider, or press play, to watch the coast light up. Each circle is a hazard, sized by how often the record names it. Blue circles were named less often in the ten years after their first light than in the ten years before, as a share of that year's wrecks in home waters. Orange circles were named more often. Dashed green circles are places that never got a lighthouse in these years, such as the Manacles and the Runnelstone off Cornwall. Small gold dots are the other lights in the Wikipedia lists and Wikidata, many of them harbour lights.
Blue and orange are almost evenly matched. Of the 87 test sites named at all in those twenty years, 44 were named less often after the light and 43 more often.
The simplest test adds up all 128 test sites. In the ten years before first lighting they were named in 551 wreck entries. In the ten years after, 450. Allowing for the size of the record each year, the rate after lighting was 0.76 times the rate before. Over twenty years either side it was 0.96 times.
For a yardstick I gave the never-lit places the same dates, one copy for each lit site. Their ratio was 1.03. So over ten years the lit places did fall relative to places that got no light. Over twenty years most of that gap closes.
One site carries a lot of the ten-year fall. The Gunfleet, a sandbank off the Essex coast, got its lighthouse in 1850. It was named in 171 wreck entries in the ten years before and 55 in the ten years after. Souter, on the Durham coast by Whitburn, went from 59 to 32 either side of 1871. But for every Gunfleet there is a Happisburgh, where the light of 1791 was followed by more mentions of the Haisborough Sands, not fewer: 6 before, 19 after.
Adding up before and after hides the shape. The chart follows the test sites in five-year steps, against comparison places over the same calendar years: the never-lit hazards, and lights that had not yet been lit. Every value is relative to the five years just before the light.
The striking thing is on the left. In the five years before a light was lit, its site was named 37% more often than in the fifteen years before that (95% interval 14% to 64%). It fits a simple story: a bad run of wrecks made the case for a light, and a few years later the light was lit.
After lighting, the line falls back. But it falls to about where it was before the run-up, not far below it. Measured against the years 6 to 20 before lighting, mentions after lighting were 12% lower, with a 95% interval from 31% lower to 12% higher.
The wrecks came first, then the light. After the light, wrecks at the spot went back to normal, and perhaps a little below it.
This is why a plain before-and-after comparison misleads. Compare the twenty years after with the twenty years before, run-up included, and you get a fall of 21% (95% interval 37% lower to 0.4% higher). Much of that is the run-up fading, which would probably have happened with no light at all. Statisticians call it regression to the mean.
Choose any site to see its own record. The orange line marks the first lighting, and the shaded blocks are the ten years either side used in the simple comparison. Most sites are named only a handful of times a decade, so a single site tells you little on its own. That is why I pool them. Try the Gunfleet, Souter, Flamborough Head, or the Manacles, which never got a light.
I tried the estimate many ways. The chart shows the after-lighting change measured against the years 6 to 20 before, so the run-up is set aside each time, plus the plain version and two placebos at the bottom.
Using only never-lit places as the comparison gives −3%. Using only lights not yet lit gives −10%. Dropping the sites whose first-lit dates I corrected by hand gives −12%. Leaving out any one test site moves the estimate between ×0.86 and ×0.98. None of these intervals rules out no change.
Two splits differ more. Lights first lit from 1841 to 1900 show −27%, against +19% for those lit from 1760 to 1840. Headlands and sands show −19%, rocks and islands −1%. I would not lean on either. The later lights were lit when the record was fullest, and I tried several splits, so one of them looking stronger is no surprise.
Then the placebos. If I pretend each light was lit ten years early, the method finds a "rise" of +26%, because the fake after-period contains the real run-up. If I move the fake date fifteen years early and stop before the run-up, the placebo is +12% (95% interval −7% to +35%).
Last, I dealt the real lighting years out at random among the test sites 200 times and re-estimated each time. 38% of the shuffled estimates were as low as the real one or lower. A real effect this size would be hard to tell from luck in data this sparse.
Put together, the record does not show lighthouses ending wrecks at the spots they guarded. It shows lights arriving after a bad run, and wrecks there then settling at or a little below their earlier level. That is not the same as saying lights did nothing. A light can save ships that would have been lost somewhere else along the coast, or make voyages faster and trade heavier. Wreck lists by place cannot see those gains.
Wrecks. I parsed every entry in English Wikipedia's "List of shipwrecks" pages from 1700 to 1945 (236,326 entries), taking the date from the section it sits in, the ship's name, its flag and the description as plain text. A check of 20 random entries against the page source found 0 errors in date, ship, flag or description, and 1 cosmetic leftover (an empty bracket where a map reference was removed).
Lighthouses. I combined the Wikipedia lists of lighthouses in England, Wales, Scotland, Ireland, the Isle of Man and the Channel Islands with Wikidata. The first-lit year is the list's year, then Wikidata's service entry, then its inception. Where a list gives two years (first light and current tower) I use the first. A check of 10 random list rows found 0 mismatches.
Sites. I wrote out by hand which hazard each light guards and the names a wreck description uses for it. The full list, with every pattern, is in aliases.py in the data release so anyone can audit it. I checked the British names against the Ordnance Survey's gazetteer for same-named places elsewhere. Test sites are lights first lit from 1760 to 1900 that are not excluded: 128 of them, first lit from 1776 to 1900. Comparison places are 13 never-lit hazards and lights not yet lit.
Models. For each test site I built a comparison of its years from 20 before to 20 after lighting with the comparison places over the same years. I stacked these and fitted a Poisson model with site and year effects inside each comparison, clustering standard errors by site. The main estimate gives the five-year run-up its own term, so the after-lighting change is measured against years 6 to 20 before.