The Rigged Demo That Made Color Blindness a Safety Test

On 13 October 1876, the general director of the Swedish State Railways sat in a darkened hall in Uppsala and picked a signal. He chose white. At the far end of the room, a conductor in full uniform held up a red lantern. A second conductor, at the other end, answered with green.

Nobody spoke for a while. Then the physiologist Frithiof Holmgren said that if an accident ever came from color blindness, there was no doubt where the responsibility lay. Within three days the railway ordered its doctors to learn his screening method.

The demonstration did what Holmgren wanted. It was also rigged. The story of how it was rigged, and why the crash that prompted it can't be pinned on color blindness alone, is one of the best case studies we have in how signal colors fail.

A green tablecloth that looked like blood

Color blindness was described long before anyone tested for it. In 1794 John Dalton, then in his late twenties and living in Manchester, described his own color vision to the city's Literary and Philosophical Society. It was the first scientific account of the condition, which later became known as Daltonism.

His comparisons were homely. A green woollen tablecloth looked to him like a "dull, dark, brownish red." A laurel leaf matched a stick of red sealing wax. He suspected the jelly in his eye, the vitreous humour, was tinted blue and filtering what he saw, and he asked for his eyes to be examined after he died.

The doctor Joseph Ransome did it on 28 July 1844, the day after Dalton's death. The vitreous humour was "perfectly colourless." The theory was wrong. In 1995 a DNA analysis of preserved tissue, published in Science, found Dalton lacked the gene for the medium-wavelength (green) receptor. He had deuteranopia, a red-green deficiency.

The National Eye Institute puts the rate at about 1 in 12 men. That is common enough to matter the moment a job depends on telling one light from another. Railways were the first industry where it did.

The night two express trains met

Just after 1 a.m. on 15 November 1875, two night expresses collided on a single-track line near the Lagerlunda estate, nine kilometers west of Linköping, in southern Sweden. Nine people died. They included the engineer of the northbound train, a fireman and an oiler.

The trains were Sweden's most prestigious. By 1875 standards they were luxurious, with gas lighting and reclining seats in first class. The Swedish paper Ny Illustrerad Tidning sent a young art student to sketch the wreck. His name was Carl Larsson, and he later became one of Sweden's best-known painters.

The crash itself came out of a scheduling scramble. The two trains normally passed each other at Linköping. The northbound was running 51 minutes late in heavy snow, so the traffic director moved the meeting point to the small station at Bankeberg. He sent the order at 00:56, only minutes before the trains arrived. Nobody told the northbound engineer, who had last stopped at Mjölby Junction. Regulations didn't require it.

At Bankeberg the stationmaster, Uno Björkelund, tried to stop the northbound train with his hand lantern, first green, then red. The train slowed almost to a halt, then accelerated away toward the oncoming express.

Three colors, one lantern

Here is where color enters the story. Night signals in 1875 had three meanings. A red light, or any light moved up and down, meant stop. Green, or a light moved slowly side to side, meant caution. A white light swung in a circle meant all clear. There was no signal for "back up."

A lineman named Oskar Johansson, who should have been at the gates, was running after the train with a lantern. The oiler Larsson, riding in the baggage car behind the tender, looked back and said a man was running with the all clear. He called it to the engineer three times.

The Cambridge and Chalmers researchers J.D. Mollon and L.R. Cavonius reconstructed the night from trial records and railway archives. They name Larsson as the strongest candidate for a color-deficient observer, though he may simply have misread a swinging lantern. They also point to a woman who lived beside the track, Hanna Reit. She watched from her kitchen window and saw a still green light, then another light waving up and down that she made out to be red. She expected the train to back up.

That is the whole problem in miniature. The same lights, seen by people with different eyes and different assumptions, produced opposite conclusions.

What the court never heard

You would expect the trial to turn on color blindness. It didn't. Mollon and Cavonius read the handwritten transcript and confirmed that color deficiency isn't mentioned anywhere in it.

The court acquitted the traffic director and the conductor. It convicted Björkelund, the stationmaster, who was dismissed and sentenced to six months in prison. He also owed the railway 125,683 kronor, at a time when his monthly salary was 1,200. The authors conclude he was made a scapegoat, since he had no more broken a regulation than the traffic director had.

Their own list of causes is long. Snow had delayed the train. The northbound engine's slide valves may have been leaking. The regulations didn't require telling an engineer about a changed meeting place. A single semaphore couldn't tell a driver the difference between "enter and continue" and "enter and stop." The signal for "back up" didn't exist. Their verdict: the accident "cannot be attributed to color blindness alone."

Nobody could test the oiler or the engineer, because both died in the crash. Holmgren's claim was a hypothesis from the start, and he made it ten months after the crash, in a letter to the State Railroad Board dated 25 September 1876.

Holmgren's wool

Holmgren already had a tool. In 1874 he had devised a test in which a person matches skeins of colored wool to samples in a box. Light and dark shades are mixed in to confuse the person being tested.

In July 1876 he presented the method to a medical congress in Gothenburg. The assembled Nordic physicians resolved that the incidence of color deficiency among railway staff needed investigating.

The State Railways' management was unconvinced. Their reply, as Holmgren quoted it, was that if color blindness really existed, it couldn't be among employees, because they would surely have noticed. Engineers and conductors rise through the ranks, so they had "amply proved their ability to distinguish signals."

Holmgren won permission to test all 266 employees of the Uppsala-Gävle line, which was privately run. He found 13 who were color blind, or 4.8 percent. They included a stationmaster, an engineer, two conductors and two linemen. Color-blind men were already doing safety-critical work, and the railway didn't know.

The rigged lanterns

Even that didn't persuade the state railway. So Holmgren built a stage.

He picked two color-blind conductors, one a protanope and one a deuteranope, and put them in uniform at opposite ends of a darkened hall. He gave them signal lanterns. The director general chose white, the first conductor showed red, the second answered green. The room went quiet.

The trick, which Mollon and Cavonius documented, was in the glass. One lantern held three red filters of different densities. The other held three green ones. The conductors were used to judging colors by brightness, so the varying brightness of each light fooled them. The demonstration, in the authors' words, "was rigged, although not by bribing the conductors."

Lanterns matching that description survive in the Medicinhistoriska museet in Uppsala, with a dusty label saying they were Holmgren's. The deception was mentioned in a footnote of an 1898 obituary and stayed largely unknown in the literature until Mollon and Cavonius wrote it up.

It worked. The railway ordered doctors to learn the method. F.C. Donders traveled from the Netherlands to Uppsala to consult Holmgren on behalf of the Dutch railway. The Smithsonian arranged an English translation of Holmgren's monograph in 1877. In Boston, B. Joy Jeffries wrote in 1878 that Holmgren had left American railroads "no excuse" for not testing their employees.

Why brightness beat hue

The trick exposes a physical fact designers still trip over. For a viewer with a red-green deficiency, the difference between a red and a green light often isn't a difference of hue at all. It's a difference in luminance, or none.

Mollon and Cavonius measured a surviving signal lantern and calculated how its lights would look to dichromats. For a protanope the red signal came out about seven times dimmer than the green (4.52 against 30.6, in arbitrary units). For a deuteranope the two were close, 47.79 against 54.45, so brightness gave little to go on.

Distance made it worse. At 600 feet, which the authors take as a conservative figure for the key sightline, a four-inch lantern subtends roughly two minutes of arc. At that size even a normal eye loses much of its blue sensitivity, and a dichromat's eye becomes effectively monochrome. The authors conclude that red, green and clear lights would not have differed in hue for a dichromat looking back at Bankeberg from that spot.

Up close it was different. The "white" of an 1875 oil lamp was very yellowish, and both protans and deutans could tell it from red and green at short range.

If you see color typically, you may be sure two lights are different. About one man in twelve in your audience may not be.

What this means for your interface

A railway lantern and a dashboard badge share one problem. Red for error and green for success are the oldest pairing in signalling, and the history of that convention is tangled enough to be worth reading. The pairing fails exactly where Holmgren's audience failed.

A few habits follow from Lagerlunda.

• Never let hue carry a message alone. Pair color with a shape, an icon, a label or a position. The 1875 rules did add a second channel, since a light moved up and down meant stop, but a lantern carried by a running man can mimic other signals. The authors raise that possibility for Johansson's lantern.
• Check luminance, not just hue. Run your status colors through a contrast checker and make sure the pair differs in lightness by a clear margin. Our guide to a colorblind-friendly UI palette walks through building one.
• Test small and far. A status dot in a dense table is Lagerlunda's two-arc-minute lantern. Preview it with the color blindness simulator, and don't trust a large swatch to predict how a small one reads.
• Look at the system, not just the sign. The investigators found no single failure. Missing signals, unstated rules and a stressed operator all lined up. Good accessibility practice works the same way, with redundancy at several layers.

Brightness has traps of its own. Saturated colors can look brighter than they measure, and the number of cone types in your eye shapes what you see at all.

Holmgren got railways to test their staff with a plausible hypothesis and a stage-managed demonstration. The evidence that color blindness caused Lagerlunda is thin. The evidence that color alone is a fragile way to say "stop" is solid, and it is still the lesson worth keeping.

Sources

• J.D. Mollon and L.R. Cavonius, "The Lagerlunda Collision and the Introduction of Color Vision Testing," Survey of Ophthalmology 57(2), 2012
• Chalmers Research record for the same paper, with abstract and DOI
• Science and Industry Museum, Manchester: John Dalton and colour blindness
• Science Museum Group: Holmgren's coloured wool test for colour blindness
• National Eye Institute: Color blindness