The Straight Dope (Cecil Adams, 1986) writes:
When the first primitive railroad signaling devices were developed in the 1830s and 1840s, red meant “stop,” green meant “caution,” and clear (i.e., white) meant “go.” This system had several defects. One obvious problem was the fact that the white signal could easily be confused with an ordinary white light. What was worse, however, was the fact that the system wasn’t fail-safe. This was tragically demonstrated sometime around 1914. The red lens supposedly fell out of a signal so that it erroneously showed a white indication. This caused a train to sail through the “stop” signal, resulting in a disastrous crash. The railroads subsequently decided to drop white and make green “go” [...]
This story is repeated in a few other places on the Web; e.g. TodayIFoundOut cites the Straight Dope story and embellishes it with an exact date and a second train:
The choice of a white light for go turned out to cause a lot of problems. For instance, in an incident in 1914 a red lens fell out of its holder leaving the white light behind it exposed. This ended with a train running a “stop” signal and crashing into another train.
GreenLightOptics omits the details, shifts the timeline as far back as the "mid-1800s," implies that there might have been multiple such accidents, and incidentally confuses the engineer with the conductor:
The mid-1800s in England saw many accidents on the railways. Because of this, a standardized color system was put into place. The rails would now use red to indicate danger, white to indicate safety, and green to indicate “proceed with caution” [...] To indicate danger or proceed with caution, a colored lens was placed in front of the light or lamp. But, those lenses could shift and fall out of place, leaving a white light blazing. Conductors would think that meant go, and would continue onward causing an accident. It is said this is when railways chose to adopt green for go [...]
This is a beautiful story, but it shows all the classic signs of an urban legend that gets vaguer and vaguer the farther back you trace it. I eyeballed Wikipedia's List of rail accidents (1910–1919) in vain — but there were so many recorded railway accidents that I'm sure I could miss the right one even if I searched every decade back to the "mid-1800s."
Did such a railroad accident ever actually occur? If so, when and where?