From msnbc.com and NBC News
From the Globe and Mail Wednesday morning
The official geological agencies provided the details:
Tuesday, January 12, 2010 at 21:53:09 UTC
The GCP event was set for an 8 hour period beginning at 20:00 UTC, a little less than 2 hours befor the 7.0 temblor. The result is Chisquare 28307.741 on 28800 df, with p = 0.980 and Z = -2.060.
By the end of Jan 13, a day after the quake, it is still chaotic. The damage is immense, and by now there are some rough guesses being made about the loss of life, with suggestions of many 10's of thousands, possibly more than 100,000 people dead.
Our formal prediction includes 8 hours of time, in keeping with the majority of previous earthquake events, but given the magnitude of the event, it seems worthwhile to look at the context. In the next figure, 48 hours are shown, with the main temblor near the middle. The early part is shown in response to requests relating to unsettled feelings interpreted as possibly precognitive. As it happens, this part shows normal looking variation. This is the case also in the last part of the figure, which shows the 13th (UTC time) when the extent of the disaster was becoming more clear.
We take a different view of the data around the quake, plotting the average deviation of the eggs over time. The next figure shows the smoothed average (using a 2 hour window). The data show a large spike at the time of the main temblor. It has odds of about 100 to 1, assuming all parameters were prespecified. This is an exploratory analysis, so should not be interpreted as definitive, but the spike does have exquisite timing.
To provide longer context, pursuant to a suggestion by Dick Bierman, we look at 9 days, with the quake approximately centered. Bierman's CIRT model postulates time symmetry following the implications of mathematical physics in which the past and future are equally represented. Thus we might predict the anomalous structure in GCP data to show symmetry around the nominal event. The quake's main temblor is marked. (The current database includes several eggs that had not reported data at the time of the original analysis.)
It is important to keep in mind that we have only a tiny statistical effect, so that it is always hard to distinguish signal from noise. This means that every "success" might be largely driven by chance, and every "null" might include a real signal overwhelmed by noise. In the long run, a real effect can be identified only by patiently accumulating replications of similar analyses.