This is an area of data science that the public is less familiar with. This example involves small data, simulations, and 18 years old crowdsourcing.
It's an attempt to explain the cause of the TWA Flight 800 that exploded near New York on July 17, 1996. I raised the possibility that a potential cause for the Malaysia Airlines flight that went missing last week, was being hit by a missile (accidental or not). Likewise, many people still believe that TWA 800 was destroyed by a missile.
However, the numerous investigations strongly debunk this myth. Here I summarize some of the data science concepts used to debunk the missile theory in the TWA flight 800 tragedy.
Traces of explosives were found, later attributed to military investigators who might have accenditally contaminated the wreckage AFTER the investigation started. But it was clearly ruled out (based on fractures and burns on human remains) that the fireballs were not caused by a bomb or a missile warhead.
Intervews with 258 witnesses scattered in the area, were recorded. Witnesses answered questions about the flight path, sound, time between sound and light streak in the sky (including light and sound intensity) and other questions. That's the crowdsourcing part, and it was crucial in ruling out the missile theory, as explained in the next paragraph. Note that this is data science, involving data recording, using the right metrics (that is, asking the right questions), and checking whether the answers are consistent across witnesses. Some criticized the FBI for asking questions that contained the keyword "missile", as this can create a bias in the answers.
Monte Carlo simulations, a data science technique, were used. After missile visibility tests were conducted in April 2000, the NTSB determined that if witnesses had observed a missile attack they would have seen: (1) a light from the burning missile motor ascending very rapidly and steeply for about 8 seconds; (2) the light disappearing for up to 7 seconds; (3) upon the missile striking the aircraft and igniting the CWT another light, moving considerably more slowly and more laterally than the first, for about 30 seconds; (4) this light descending while simultaneously developing into a fireball falling toward the ocean. None of the witness documents described such a scenario.
Note that there's a lot of variance in witnesses perception of what happened (based on location and personal factors). There's also a lot of variance in how a "missile hitting a plane" scenario could appear, based on where it is launched from, the size/payload of the missile and other factors. Reconciliating (or proving in this case that witness reports and missile simulations are totally incompatible no matter how you slice the various factors), is a challenging data science task.
Anyway, after reading the article about the investigations, I am convinced that it was not caused by a missile.
Do you still believe in the missile theory? Maybe you believe these analyses are wrong, the truth is hidden? What about the recent Malaysia Airlines crash? Could it be a missile?