Based on an analysis of 2.5 months of GCP data, I conclude that a statistical anomaly in this database was associated with the date, time and general location of the terrorist attacks of September 11, 2001. Of more pragmatic importance, there is evidence that the statistical anomalies began to appear about two hours before the actual events unfolded. With a much larger RNG network, distributed uniformly around the world, and with the application of more sophisticated analyses, the GCP may prove to be useful in predicting future events with similar massive impact.

Data analysis procedure

from each day's raw datafile, calculate empirical mean and sd per egg

exclude raw egg values < 65 or > 135 (i.e., |z| > 4.9, probably bad data)

use resulting mean & sd to calculate z score per egg, per day

calculate z-squared per egg

sum up all z-squares across eggs, per day, keeping track of the number of eggs

create 5-minute consolidations of the per-second data, as sums of z-squares

analyze data using 6 hour sliding window

calculate z score equivalent for the resulting chi-squares & df

calculate odds associated with the z scores

 
plot results, serve with sprinkle of nutmeg, and enjoy

Figure 1. z scores of 5-minute summaries, across all eggs, for June 30 - September 18, 2001.
 

Figure 2. Histogram of z scores in Figure 1 (in black) and expected normal curve (in red). This shows that overall the eggs are well behaved, with a slight positive skew. The theoretical mean for the z score is 0; the observed mean is z = -0.0002. The theoretical standard deviation is 1, the actual sd = 0.9972. The Stouffer Z for the observed mean shift is z = -0.031.
 

Figure 5. This shows observed z scores (in blue, 6 hour sliding window, September 3 to 13) vs. similar scores (in pink) using pseudorandomly generated data. This confirms that the analytical method is not introducing artifacts into the results. The "0" on the x-axis indicates the beginning of a day boundary.
 

Figure 6. This shows the odds associated with the z scores in Figure 5, using a log scale.
 

Figure 7. This shows the dates associated with maximum z scores obtained after applying sliding windows ranging from 5 minutes to 12 hours in length, in 5 minute increments (a total of 144 windows), to all data from July 15 - September 16. The date where the majority (60%) of the maximum z scores occurs is September 11. This confirms that the spike observed on September 11 is not especially sensitive to the choice of specific window lengths. The second largest cluster of maximum z scores occurs on August 4, which is also the second largest spike in Figure 3.
 

Figure 8. This shows the number of times that the date of September 11 appears in the top 10 maximum z scores, after applying the 144 different window lengths. We see that the majority of maximum z scores appears in window lengths ranging from 5.2 to 6.8 hours. This suggests that the length of the "event" on September 11 may be in this range.
 

Figure 9. This shows the time of day associated with sliding windows that resulted in maximum z scores observed on September 11. The clustering of times suggests that the "event" may have had three especially meaningful moments: around 5 AM, 6:30 AM, and 8 -10 AM.
 

Figure 10. This shows a composite z score for September 11 across all 36 available eggs, using a 6-hour sliding window. The peak is at 9:10 AM. This plot is essentially a high-resolution version of the z scores shown in Figure 5, centered on September 11.
 

Table 1. This lists the egg numbers and their associated locations in terms of state or region, country, continent (roughly), and hemisphere.
 

Table 2. Analysis of 6-hour windowed Stouffer Z scores associated with 9:10 AM, September 11, by combining all eggs within each hemisphere. This suggests that a larger effect "occurred" in the Western hemisphere.
 

Table 3. Analysis of 6-hour windowed Stouffer Z scores at 9:10 AM, by country, possibly suggesting that the primary "effect" took place in Northern America.
 

Table 4. Analysis of North American eggs, suggesting that the primary "effect" was in the East coast.
 

Table 5. Summary: Through analysis of 144 sliding windows, from 5 minutes to 12 hours, in 5 minute increments, we find that over a period of 2.5 months, one date is primarily associated with a statistical anomaly: September 11, 2001. On this date, the time range appearing most often is 6 AM to 10 AM, with a peak at 9:10 AM. The Western hemisphere appears to be responsible for the majority of the statistical anomaly, especially North America, and in particular the East Coast.