Space Shuttle Challenger
Is it the data, the tool, or the editorial choices that are made in analyzing the data that cause good people to make bad decisions based on poor data visualizations?
For instance, on the morning of Jan 28, 1986, the Space Shuttle Challenger broke apart 73 seconds into its flight. The disintegration of the shuttle was caused by the failure of an O-ring seal in its right solid rocket booster. The seal failure allowed a “flare” to weaken the structure of the external fuel tank and aerodynamic forces promptly broke apart the vehicle.
The rubber O-ring failed because the launch took place on a particularly cold morning and the cold caused the seal to become brittle and rigid. They needn’t have launched on that day. So why did the team decide to launch that morning knowing full well that it was going to be colder than it had been on any other launch day?
As Tufte notes in “Visual Explanations” pp.39-53, each launch generates “incident” data: what part or parts failed on that day and what was the launch environment. The team made a single editorial mistake when deciding to launch: the team chose to chart only the launch data where O-ring failures occurred:
The fact that the rubber rings were failing primarily in colder temperatures was apparent only when you plotted all the data–especially data from launches where the O-rings didn’t fail:
So, one poor editorial decision and seven people died. No one in the process wanted that to happen. There is probably no software-only way to protect against this kind of mistake. But that won’t stop us from looking for one.
