Dr. Nelson:
Since the GCP prediction registry had been at a static point during your
absence, I thought it was a perfect moment for me to carry out an
independent analysis of the effect size by event category to compare
against those results found by Dean Radin and yourself. What I did was
the following:
1.) I gathered the formal results from all of the 137 events that were
registered in Table 2 of the "Results" section as of May 1, 2003. At
that time, there was the entry in Table 2 that was marked "Various
Unfinished Analyses." The 4 events that were under this entry were
excluded from the analysis, so there were actually 133 events total.
2.) With the given Chi-Square and degrees of freedom, I calculated a
z-score for each event using the relation z = Sqrt(2*chi-square) -
Sqrt(2*df - 1)
3.) I place each event into 1 of 13 general categories based on my own
subjective impressions of the event from reading its Prediction
Registry note. Even though there were several events that I thought
could belong in more than one category, I decided to place those events
in the category that seemed most fitting, so there was no overlap.
4.) A Stouffer's Z was calculated for all the events in each category
using the total number of events listed in each category as N, and
the results were plotted as in the first attached graph.
The second graph is a further exploration of event effect size, in which
I included several suitable events listed in the "Non-formal
Explorations" section along with the formal prediction results just to
see what else these events might be able to tell us about event effect
size. An event was deemed "suitable" if it used all of the EGGs in the
network, reported Chi-Square and probability statistics, and had some
form of predicted time period for the event. 11 (2 of which are context
analyses of one event) of the 24 events currently listed seemed to fit
this criteria.
The events added were:
1.) Indian Elections
2.) Geomagnetics, all EGGs
3.) Brahms Deutsch Requiem
4.) New York Stock Exchange, Market Plunge
5.) Death of Larissa Vilenskaya, 1-hour event pred.
6.) Death of Larissa Vilenskaya, 3-hour aftermath
7.) 100 Nations Agree on Kyoto
8.) Destruction of Buddhas of Bamiyan, full day
9.) Funeral of Jennifer Sandy
10.) Super Bowl 2003, seconds resol.
11.) Super Bowl 2003, 15-min. resol.
These events were also analyzed using the same process described in
Steps 2 - 4. The second graph shows these readjusted results, with
combined formal and non-formal exploration results.
The third graph shows the effect size with distance for the data from
Sept 11, 2001, using the 15-minute summary data. I grouped the EGGs
according to location, then obtained a Stouffer's Z for each using the
z-scores in the group.
Just as found by you and Dean Radin, the event category with the largest
effect size is "Celebration," and the smallest being political events.
The non-formal exploration results changed the findings slightly
overall, reinforcing some while weakening others. The strongest effects
occurred in North America on Sept. 11, just as found by Radin. Overall,
these results seemed to replicate the results found by Radin and
yourself. I think they point to some interesting questions on EGG
response with distance and event intensity.
Regards,
Bryan Williams, UNM