Concorde Crash, July 25 2000
From the prediction registry:
Note: the correct time for the crash was 14:44 GMT, so the figures here are incorrect -- the formal graph should be shifted about 15 mins earlier.
Prediction, Reinhilde Nelson: July 25th, 2000 at 15:02 GMT. The Concorde crash upon takeoff from Paris will engage more attention than usual because the Concord is a high-profile plane, beautiful and symbolic of the romance of flying. It is the first crash of a Concorde, which has been considered one of the safest planes in the world since its introduction in 1969. This was a chartered flight with mostly German tourists on their way to a cruise. All 100 passengers and 9 crew were killed, as were four people on the ground.
The primary analysis will be of the half hour surrounding the crash and a four hour aftermath, with the formal hypothesis test based on the half hour period, assessed at one-second resolution, expectation high. The 4 hour period is an exploration addressing the question whether developing awareness of the news may reflected in the data. Secondary analysis will determine whether European eggs show greater reaction.
Analytical results: The outcome of the analysis is a positive deviation, with chisquare 1830.2, on 1800 df, for a p-value of 0.304. The first figure shows the half hour surrounding the crash, beginning at 14:47, UTC.
Arguably, the prediction might have specified the half hour following the crash. If so, the result would have been highly significant, with chisquare 1959.6 on 1800 df, for a p-value of 0.0047. The next figure shows The half hour beginning at 15:02, UTC. It is followed by a figure that shows the next four hours of what we may regard as the aftermath, when people are hearing the news of the crash.
Reinhilde, who made the prediction, also suggested that a fair presentation would show the data for a few hours prior to the crash, expecting it to look essentially random (flat trend). But the figure shows periods with rather strong trends, some of which are nearly as impressive as the segment immediately after the crash, begining at hour 4 in this graph. This indicates, as does the comparison of the first two figures above, and as we have seen in previous explorations, the important role played by the specific formulation of a prediction. A slightly different point of incursion or a different length of time might yield a very different outcome.