AAAS2006.pres.ppt

advertisement
Anomalous Anticipatory Responses
In Networked Random Data
Roger Nelson
Princeton, New Jersey
Frontiers of Time: Reverse Causation -- Experiment and Theory
AAAS Symposium, University of San Diego, June 2006
Global Consciousness Project
http://noosphere.princeton.edu
Global Consciousness Project
(aka The EGG Project)
The People: An international collaboration of
100 Scientists, Engineers, Researchers
The Tools: REG technology, Field applications,
Internet communication, Canonical statistics
The Question: Is there evidence for Non-random
Structure where there should be none?
Random Event Generator – REG
Reverse Current in Diode: White Noise
Electron Tunneling – A Quantum Process
Sample Resulting Voltage, Record 200-Bit Sums
It is like flipping 200 coins and counting the heads
Trial Scores: 100 ± 7.071
Plotted as a sequence, 1 trial per sec
Binomial Distribution of Data
Compared to Theoretical Normal
100 is expected mean
A World Spanning Network
Yellow dots are host sites for Eggs
http://noosphere.princeton.edu
Internet Transfer to Data Archive in Princeton
Here are data plotted as sequences of 15-minute
block means, for a whole day, from 48 eggs
We begin to see what’s happening
If we plot the Cumulative Deviations
If we average the cumulative deviations
Across REGs we may see a meaningful trend
Expected
Trend is
Level
Random
Walk
Cumulative deviation is a
Graphical tool to detect change
Process control engineering
A Replication Series
Of Formal Tests
The Hypothesis:
Global Events Correlate with
Structure in the Random Data
Test Procedure:
Pre-defined events,
Standardized Analysis
Bottom Line:
Composite Statistical Yield
Current Result: Formal Database, 7.5 Years
204 Rigorously Defined Global Events
Odds: About 1 part in 300,000
9/11
Now we proceed to new questions
First, how good are the data?
 Equipment: Research quality Design, Materials,
Shielding, XOR, Calibration standards
 Errors and Corrections: Electrical supply failure,
component failure. Rare but identifiable
 Empirical vs Theoretical: Mean is theoretical, but
tiny differences in Variance (expected)
 Normalization: All data standardized; empirical
parameters facilitate comparison and interpretation
Identify and exclude “Bad Trials” <55 or >145
Identify and exclude device failures, “Rotten Eggs”
Identify Individual “Rotten Egg”
Calculate Empirical Variance for Individual Eggs
REG device failure
Effect of “Rotten Eggs” on the Full Network
REG device failure
Fully vetted, normalized data
Theoretical vs Empirical Distribution
(We also assess pseudorandom clone data,
and use resampling and permutation analyses)
Note: These are (0,1)
Normal Z-scores
The Diffs are TINY
Negative difference
Means that formal
Tests are conservative
Three Independent statistics
The netvar is Mean(zz). It measures the
average pair correlation of the regs:
<zz> = <z[i]*z[k]>
where i & k are different regs and z is
trials for one second.
The devvar is Var(z) the variance across regs
Calculated for each second.
The covar is Var(zz). It represents the
variance of the reg pair products:
{ z[i]*z[k] - <zz>}^2
Suggestions of precursor effects
Sept 11 2001 Terror Attacks
Stouffer Z across REGs per second
Cumulative sum of deviations from expectation
Variance across REGs per second
Cumulative sum of deviations from expectation
Attacks
Attacks
Attacks
Attacks
Moderately persuasive suggestion
that trend may begin before event
Strong and precise indication that
change begins 4 hours before event
And very recently, the Indonesian earthquake
on May 27 this year also seems to show
evidence of a precursor response
To go further we need a better database
Suggestive single cases but low S/N ratio
Need replication in multiple samples
“Impulse” events are sharply defined
E.g. crashes, bombs, earthquakes
Subset of formal series: 51 impulse events
Epoch average for covar and devvar may
Depart from expectation prior to T=0
Covar
The suggestion
of early shift is
clearest in covar
Devvar
Netvar
51 Impulse events, Covar epoch average
Deviation may begin ~ 2 hours before T=0
Approx Slope
Impulse events vary -- We need consistency
Earthquakes are a precisely defined,
Prolific subset of impulse events
They show similar responses
Impulse events shown as Red, Earthquakes as Blue trace
Netvar
Covar
Earthquakes: Important to People,
Numerous, Accurately Located,
Rigorously Scaled, Precisely Timed
All Earthquakes, Richter 6 or More
Select those on Land with People and Eggs
Eggs shown as
orange spots
Selected regions outlined in orange
Included quakes shown as grey dots
Controls shown
as blue dots
In the Earthquake database, the covar
measure appears to be the most useful
of our three independent statistics
For quakes R>6 (grey dots) the covar measure
Responds before and after the primary temblor
Before
-8 hrs
Mostly
Negative
After
Mostly +8 hrs
Positive
Average location of quakes in grid square marked as a colored point
Size is cum Z-score; Red: positive; Blue: negative; Green: no calc, less than 2 quakes
Strong covar response in populated
Land areas where we have eggs
North America and Eurasia
Symmetrical, Significant Z-scores Pre & post
Null covar response in unpopulated
Regions (ocean) and areas where we have few eggs
Control: Quakes in the Oceans
All Z-scores less than 0.5
Major earthquakes in populated areas
Compared with quakes in the oceans
Covar measure, epoch average
Cum Dev T=0 ± 30 hours
North America and Eurasia
Significant structure around T=0
Scale of departure ~ 80 units
Ocean Quakes
No structure around T=0
Scale of departure ~ 40 units
Closer look: T=0 +/- 10 hours
North America Europe and Asia
Unpopulated Ocean regions
Significant structure around T=0
Scale of departure > 50 units
No structure around T=0
Scale of departure ~ 20 units
Data split: T=0 ± 8 Hrs
North American vs Eurasian Quakes
Similar structure, independent subsets
The case for an anticipatory response
Magnified central portion
T=0 ± 50 hr
Raw data
T=0
3-Hour
Gaussian
smooth
Estimating significance:
The drop between T-7 Hrs and T=0
Corresponds to a Z score of 4.6 
After Bonferroni correction
Compare slope with 3  envelope
Same data as a cumulative deviation
Many questions remain, e.g.,
Fatal quakes should be test case.
Subset with N > 5 fatalities and R > 5
The picture is less clear.
CAUTIONARY NOTES
The effects we see are very small, buried in a sea
of noise. Is “signal” an appropriate term?
Statistical and correlational measures. Need to
understand inconsistencies.
Fundamental questions remain unanswered.
(e.g., effects of N of eggs, Distance, Time).
Selectivity of analyses needs balance of independent
perspectives and replication.
We invite efforts to confirm or deny these indications.
POSSIBILITIES
The GCP database of networked random events is
unique. No other resource like it exists.
Opportunity for useful questions and answers.
Probably holds surprises.
Fundamental questions that should be asked are
known (e. g., N of eggs, Distance, Time).
A couple of years of supported analytical research
would break new ground.
GCP Homepage
http://noosphere.princeton.edu
Special Links
Status
Day Sum
Results
Extract
Complementary
Perspectives
Web Design
Rick Berger
The following are extras. Some are
explanatory, some provide additional info.
An example of new perspectives:
Is there evidence of periodicity?
The generalized short answer is no.
But formal events may show FFT spikes
Fourier Spectra and Event Echoes
Dec 26 2004 Tsunami vs Pseudo Data
Analysis by William Treurniet
The pre-event frame shows a substantial peak (black trace)
Compared with the pseudorandom data (right panel).
And check out post-event frame 3 (pale bluegreen).
EGG Network Response (Quakes on Land)
Cumulative Deviation of Covariance
Primary Temblor +/– 30 Hours
Control Data: Oceans &
Low Population Zones
North America and Eurasia
Note: This is an early figure with somewhat different
Circumscription and hence a different N of quakes.
Epoch or Signal Averaging
A tool for revealing structure
In repeated low S/N ratio events
Graphical presentation: Cumulative Deviation
Used in Statistical Process Control Engineering
Example, Raw data
Dev from Expectation
Begin Cum
Dev from
Expectation
Raw data and Gaussian smoothed data
Raw
Quakes on land
T=0 ± 30 hours
3 Hour
Largest spikes are near T=0
1 Hour
The crossover is exactly at T=0
The minimum is -3 sigma and
The maximum is +3 sigma
Cumulative deviation of covar for unpopulated
regions (ocean) and areas where we have fewer eggs
South America
Nippon, East Asia
Control: Quakes in the Oceans
No trends, and
No structure
Related to T=0
Range is
1/2 to 1/3 of
Land quakes
A very early suggestion that the
REG data might show evidence of
Precursor response to major events
-5 minutes
T=0
+5
Cumulative Deviation
From Expectation
95% confidence
Expectation
Assassination of Prime Minister Rabin, 1995
Download