Global Consciousness Reanalysed

[Robin]

The Global Consciousness Project has appeared in this section a couple of times. The idea of it is that there are a number of hardware random number generators (called ‘EGGS’ or ‘REGS’) around the world generating 200 binary numbers every second and these are fed back daily to a main server. Their hypothesis is that these generators will return unexpected results during times of significant global events (such as the Papal visit to Israel, Tsunami in SE Asia and so on). Their claim is that this hypothesis is borne out by the results and that the probability of getting the results they saw were about 50,000 to one and that this can be seen as fairly good evidence for some kind of connection between consciousnesses. If you are new to it you can catch up in: http://noosphere.princeton.edu/story.html. If you are interested you might also want to visit the old “Global Consciousness Project” thread in this section.

In the last discussion one of my observations was that the results they claim are derived using a number of inconsistent statistical methods and time intervals. You can see the results page at http://noosphere.princeton.edu/results.html. I was interested enough to download their data and reanalyze using just one statistical method and time grouping.

I used the 1 second intervals and the Stouffer Z Meanshift method (described on their “recipes” page). What I found was that the results using a consistent method and 1 second intervals the results were more like 1400 to one and even less using 1 minute, 10 minute and 15 minute intervals.

One in 1400 is very different to the one in 50,000, but nevertheless it is still statistically significant, but it is interesting to note that there is no trend up until around event 50 (shown on the graph below as a blue spike). Prior to that time the GCP people analysed using mostly 15 minute intervals, after that time they used mostly 1 second intervals. So there is no detectable deviation from expected random results while the data is being analysed using a method or interval different to the one in the original analysis.

This, to me, raises the suspicion that some sort of ‘cherrypicking’ has gone on in relation to the start and finish times of the events, or the events selected for inclusion. In addition some of the early events have very sparse data and no ‘EGGS’ reporting for most of the time, yet these are included and add significantly to the overall result. Some events are analysed in 15 minute intervals when they don’t even last 15 minutes.

Anyway this is the graph of the results I found, against the results that the GCP claim with the GCP event number on the x axis. It should be compared with the graph on the GCP results page cited earlier:

 

[CFLarsen]

This, to me, raises the suspicion that some sort of ‘cherrypicking’ has gone on in relation to the start and finish times of the events, or the events selected for inclusion.

You bet they cherry-pick.

They did an analysis of Earth Day 2001 (which is claimed to show GC), but not any analysis of Earth Day 2002, 2003, or 2004.

They change the timeframe of the predictions to create positive results:

During the big tsunami on Dec 26, 2004, they chose to begin half an hour before the main quake, using an 8 hour period.

But during the Indian Ocean Quake Mar 28 2005, they chose to begin a little more than an hour before the quake, using an 8 hour period.

They don't even always use an 8 hour period:

During the Earthquake in Afghanistan on March 25, 2002, they use a 4 hour period, starting half an hour before the event.

During the Western India Quake, on January 26 2001, they use a 30 minute period, starting 15 minutes before.

All the same event: An earthquake. Yet, they change everything: The period length and the starting time.

You bet they cherry-pick.