Even a 20mm autocannon round only has a ~50% chance of destroying a plane on an engine hit, as opposed to the ~100% assumption used for the illustration (I think the flak round statistics were per-fragment-hit and not per-shell)! Weird how few people were suspicious about a 100% loss rate for engine hits to a two-engine plane.
Speaking of which, there is no source for the example data, which is noted as “hypothetical” in the report. The hypothetical data set was for a 1000-plane raid on a single objective, which implies a strategic bombing mission context, and thus that the planes were 4-engine strategic bombers like the B-17 instead of the 2-engine bomber in the illustration (this is a nitpick, but I thought it was neat). There is also no exact hit location data, with hits being aggregated by category.
It’s still a nice-looking illustrations though, and I don’t think the contributor did anything wrong here. It’s just a reminder to look more closely at anything that looks too good.
The famous survivorship bias image is a “loose reconstruction” of methods used on a hypothetical dataset
File:Survivorship-bias.svg—Wikimedia Commons
A later reprinting of the “unillustrated report” referenced:
A_Reprint_Plane_Vulnerability.pdf
Even a 20mm autocannon round only has a ~50% chance of destroying a plane on an engine hit, as opposed to the ~100% assumption used for the illustration (I think the flak round statistics were per-fragment-hit and not per-shell)! Weird how few people were suspicious about a 100% loss rate for engine hits to a two-engine plane.
Speaking of which, there is no source for the example data, which is noted as “hypothetical” in the report. The hypothetical data set was for a 1000-plane raid on a single objective, which implies a strategic bombing mission context, and thus that the planes were 4-engine strategic bombers like the B-17 instead of the 2-engine bomber in the illustration (this is a nitpick, but I thought it was neat). There is also no exact hit location data, with hits being aggregated by category.
It’s still a nice-looking illustrations though, and I don’t think the contributor did anything wrong here. It’s just a reminder to look more closely at anything that looks too good.
Inspired by this tweet:
Liron Shapira on X: “A plot showing zero planes returned after being shot in various large regions doesn’t prove which areas are best to reinforce since we don’t know the distribution of shots, but does prove that someone made a fake exaggeration of how much survivorship bias dominates other effects. https://t.co/7UGoyzv2zn” / X
Fighter pilots do in fact aim for specific parts of a plane with MG and autocannons, while flak rounds are shot from the ground.