September 11 2001: Exploratory and Contextual Analyses

Roger Nelson, Director, GCP

On September 11, 2001, beginning at about 8:45 in the morning, a series of terrorist attacks destroyed the twin towers of the World Trade Center and severely damaged the Pentagon. The disaster is so great that in New York we have as yet, two days later, only guesses about how many thousands of people perished when the WTC towers collapsed. Commercial airliners were hijacked and flown directly into the three buildings. The first crashed into the North tower at 8:45, and about 18 minutes later the second airliner hit the South tower. At about 9:40, a third airliner crashed into the Pentagon. At about 9:58, the South tower collapsed, followed by the North tower at 10:28.

The following material shows the behavior of the Global Consciousness Project's network of 37 REG devices called "eggs" placed around the world as they responded during various periods of time surrounding September 11. A book chapter gives a compact summary. These eggs generate random data continuously and send it for archiving and analysis to a dedicated server in Princeton, New Jersey, USA. We analyse the data to determine whether the normally random array of values shows structure correlated with global events. This page shows a wide range of exploratory analyses that provide context for the formal hypothesis testing related to the events on September 11. A number of people have done supplementary and complementary analyses, as well as direct replications. Links to these are provided below. An especially interesting effort was undertaken by Bryan Williams, who used data in 15-minute blocks, to compare with the seconds resolution used in the formal analyses. To the extent his results are similar, this provides some response to the question whether a general, external influence is at work, as opposed to an "experimenter effect" operating via fortuitous (albeit anomalous) selection of the analysis specifications.

The underlying motivation for this work is to discover whether there is evidence for an anomalous interaction driving the eggs to non-random behavior. In a metaphoric sense, we are looking for evidence of a developing global consciousness that might react to events with deep meaning. The whole world reeled in disbelief and horror as the news of the terrorist attack and the unspeakable tragedy unfolded. Our analyses show that the EGG network registered an unmistakable and profound response.

Introduction

I want to acknowledge that I like the notion of Global Consciousness, but that this idea is really an aesthetic speculation. I don't think we have real grounds to claim that the statistics and graphs representing the data prove the existence of a global consciousness. On the other hand, we do have strong evidence of anomalous structure in what should be random data, and clear correlations of these unexplained departures from expectation with well-defined events that are of special importance to people. The events share a common feature, namely, that they engage our attention, and draw us into a common focus.

This is a report of what we see in the data recorded on September 11, 2001 and the surrounding period. It is the best description we can give of measurements and effects that are essentially mysterious. We do not know how the correlations that arise between electronic random event generators and human concerns come to be, and yet, the results of our analyses over the past three years repeatedly indicate such correlations. We cannot explain the presence of stark patterns in data that should be random, nor do we have any way of divining their ultimate meaning, yet there appears to be an important message here. When we ask why the disaster in New York and Washington and Pennsylvania should appear to be responsible for a strong signal in our world-wide network of instruments designed to generate random noise, there is no obvious answer. When we look carefully and discover that the eggs might reflect our shock and dismay even before our minds and hearts express it, we confront a still deeper mystery. This network, which we designed as a metaphoric EEG for the planet, responded as if it were measuring reactions on a planetary scale. We do not know if there is such a thing as a global consciousness, but if there is, it was moved by the events of September 11, 2001. It appears that the coherence and intensity of our common reaction created a sustained pulse of order in the random flow of numbers from our instruments. These patterns where there should be none look like reflections of our concentrated focus, as the riveting events drew us from our individual concerns and melded us into an extraordinary coherence. Maybe we became, briefly, a global consciousness.

These are pictures drawn from the data, with brief descriptions of the exact procedures for those who want to know the details. We use statistical and mathematical tools to visualize the structures that appear, and graphs to display them. I think these are transparent images, pictures worth a thousand words.

Formal: Deviation of Means

The Global Consciousness Project has a standard protocol for testing the hypothesis that great events in the world may affect the eggs in a way that can be detected by statistical analysis. The data from all the eggs are first combined in a single (Stouffer) Z-score representing all eggs for each second. These Z-scores are squared and summed to yield a Chisquare for the whole period. The departure from expectation of this Chisquare reflects a correlated response across the eggs. Our prediction is for a positive deviation, which would indicate a tendency toward increased deviations from theoretical expectation. More information about the analysis procedures is given in the Methodology section of the GCP Web site.

The first formal prediction for September 11 is essentially the same as that made for the terrorist bombing of American Embassies in Africa in August 1998. That specified a period beginning a few minutes before the bombing and included an aftermath period of three hours. Following that model, I specified a period beginning 10 minutes before the first plane crashed into the WTC tower, and ending four hours after, thus defining a similar aftermath period.

To visualize the data, we plot the cumulative deviation of the Chisquare sequence from chance expectation. If there is no effect, such a plot will show a random walk (sometimes called a "drunkard's walk") around the horizontal line of expectation. That is, the trace will wander up and down but will have no clear trend.

The graph of data from the formal prediction for September 11 shows a fluctuating deviation throughout the moments of the five major events, during which ever-increasing numbers of people around the world are hearing the news and watching in stunned disbelief. Times of the major events are marked by boxes on the line of zero deviation. The uncertain fluctuation of the EGG data continues for almost half an hour after the fall of the second WTC tower. Then, at about 11:00, the cumulative deviation takes on a strong trend that continues through the aftermath period and ultimately exceeds the significance criterion. There were 37 eggs reporting on September 11, and over the 4 hours and 10 minutes of the prediction period, their accumlated Chisquare was 15332 on 15000 degrees of freedom. The final probability for the formal hypothesis test was 0.028, which is equivalent to an odds ratio of 35 to one against chance.

Formal graph: 
<!--
<img src=

Explorations of Context

The formal test indicates a nominally significant departure from expectation but it is not especially persuasive by itself, given the enormity of the event. Moreover, the outcome clearly is dependent on a fortuitous specification of the timing in the formal hypothesis.

It is therefore helpful to examine the larger context by looking at the behavior of the eggs before and after September 11. We find that while there is nothing unusual in the data from preceding days, the opposite is true following the attacks. During most of the 11th, 12th, and 13th there is a strong trend indicating correlated behavior among the eggs. This persistent deviation from random behavior appears to begin a little before the first crash, and it continues well beyond the time specified in the formal prediction. As it happens, an aftermath of a few hours could not capture what appears in context to be a long-lasting aberration in the normally random flow of data. The persistent departure from expectation continues for more than two full days.

The next figure shows the period from September 7th to 13th, and the time of the attacks on September 11 is marked with a black rectangle. You can see that shortly before the terrorist attack, the wandering line takes on a strong trend representing a persistent departure from what is expected of random data. A small probability envelope inserted at that point provides a scale to indicate the extraordinary increase in non-random deviation. The slope of the graph beginning just before the the first WTC tower was hit and continuing for over two days, to noon on the 13th, is extreme. An informal estimate for the probability can be made from its slope, and lies between 0.003 and 0.0003 (suggesting an odds ratio on the order of 1000 to 1). If we extrapolate the anomalous trend, it begins at about 04:00 (08:00 GMT), several hours before the first World Trade Center tower was hit, and the total length of this persistent trend is about 56 hours. The mean number of days between segments with a slope like this, continuing for so long, is on the order of 2300 days, which is consistent with the 1000 to 1 odds ratio suggested by the slope.

Context graph:
Terrorist Attacks, September 7-13 2001

Odds Against Chance

The multi-day perspective places the formal specification in a larger context, and we should also look at finer details. The next figure shows the raw odds against chance, second-by-second, for the squared Stouffer Z-scores (Chisquares) for September 11. The maximum odds ratio, shown as a spike in the center of the graph, is equivalent to a Z-score of 4.81, and occurs at 10:12:47, EDT, not long after the first World Trade Center tower collapsed. A Z-score this large would appear by chance only once in about a million seconds (roughly two weeks). Thus it is not terribly unusual to find such a spike in our large database, but it is thought-provoking that one does occur within the brief time-span of the attacks, about an hour and 45 minutes. The ratio of this period to the mean time between spikes of this magnitude is 1/192, so the odds of this being just a chance occurence is nearly 200 to 1. To see more clearly the contributions to the Chi-square deviations, the figure includes a display of the odds ratios calculated for the same data, but with the Stouffer Z-scores passed through a 1-hour smoothing window. The graph is rescaled, to enlarge the detail.

Raw
second-by-second odds ratio: September 11 2001

Omega Complexity

Brain EEG data from large numbers of electrodes can be reduced to a smaller number of dimensions for summary measures. One of these is "Omega Complexity" which yields three parameters called Sigma, Phi, and Omega, corresponding roughly to amplitude, frequency, and complexity in the data. This measure was applied to the data from September 11, and two types of figures made. The first shows a 3-dimensional plot that shows the development over time of the combination of all three parameters. The actual data are plotted in red, and pseudo-random data for the same day are in green.

Context graph:
Omega complexity, Sept. 11 2001

The same data were plotted more simply in the next figure. The Sigma parameter, which appears to drive the deviation most powerfully, is plotted as a cumulative deviation from the mean. For comparison, there is again a corresponding plot using the pseudorandom clone data.

Context graph:
Cumdev Omega, Sept. 11 2001

Formal: Variance of the Egg data

The next figure shows the cumulative deviation of a measure of the variability of scores (variance) among the 37 eggs over the course of the day of September 11. It was generated as a test of Dean Radin's prediction that the variance would show strong fluctuations: "I'd predict something like ripples of high and low variance, as the emotional shocks continue to reverberate for days and weeks." Although this was only a partial specification it is effectively a prediction that the variance around the time of the disaster should deviate from expectation. I added the necessary specifications for a formal prediction.

The variance measure shows a normal fluctuation around the horizontal line of expectation until about three or four hours before the attack, and then a steep and persistent rise indicating a great excess of variance, continuing until about 11:00. Shortly thereafter, a long period begins during which the data show an equally precipitous decrease of variance. It is difficult to make a direct calculation of probability for this figure, but the extreme excursion in Dean Radin's similar analysis reaches a level of more than three sigma, which corresponds to odds of less than 1 in 1000. An additional analysis using permutation of the data to determine how often such an extreme excursion occurs in randomly ordered sequences, to compare with the original temporal sequence leads to an estimate of p = 0.0048, based on 10,000 iterations. (A more recent computation using the mean of variance as the standard rather than theory yields p = 0.0009.)

A more conservative estimate is included in the formal database. It is based on assessing the fast rise and the fast fall of the variance measure surrounding the period of the attacks. The probability for each was calculated by extrapolation of the probability envelope as far (as many seconds) as would be needed to achieve the extreme rise or fall by chance, compared to the much shorter envelope that covers the time of the actual rise or fall. The ratios of these times were divided by the square root of 2 to compensate for the fact that a cumulative deviation trace reaching the terminal significance level at some prior time during the cumulation is twice as likely as the terminal probability itself. The resulting estimate is p = 0.096.

For a visual indication of the likelihood that this is merely a random fluctuation, a comparison can be made with pseudo-data generated for September 11, 2001, and plotted in the same format. In contrast to the real data, there are no long-sustained periods of strong deviation in the algorithmically generated data. While we intend to do a more rigorous test using resampling procedures, this comparison with the pseudo-data indicates that the variance measure is unusual around the time of the attacks.

In this figure, the times on the X-axis are Eastern Daylight Time, allowing a direct assessment of the timing of the strong deviations. We note also that the distinctive shape of the graph is suggestive of a classic "head and shoulders" graph seen in stock market analysis. As in the first figure showing the cumulative deviation of the Chisquare, there is an indication that the effects registered for this horrendous event might have begun several hours prior to the first attack. Again, the pseudo data are used for a direct comparison. More on this topic, in the context of exploratory analyses, can be found on the extended analysis page.

Terrorist Attacks,
September
11 2001

Formal: Silent Prayer

Since the horrible event, innumerable calls for prayer have been made. On the 14th of September there was a special emphasis on such collective spiritual moments, including major organized periods of silence in Europe and America. Doug Mast made a specific formal prediction for a deviation of the Chisquare "over the time periods 1000 to 1003 GMT, corresponding to a European organized mourning and the time period 1200 to 1203 EDT (1600 to 1603 GMT) corresponding to the beginning of the Washington service and many organized mourning events in the Eastern US." Here is the resulting graph.

Doug Mast Pred: 
Silent Prayer, September 14 2001

The picture is very compelling, I think, although it does not confirm the formal prediction. Instead, the trend shows a marginally significant decrease in the deviations of the egg data. The Chisquare is 150.68 on 180 degrees of freedom, with probability 0.9455. The trend is steadily opposite to the usual (and specified) direction, but somehow it looks right -- symbolic of the moment's contrast to the preceding days.

Although there was no formal prediction made for a reaction of the EGG network to the event, we do not want to forget the heroic sacrifice that brought down the fourth hijacked plane. Exploratory analyses of the time prior to the crash itself suggest a poignant correlation ­ an extraordinary rise in the cumulative deviation followed by a precipitous fall.

Exploratory Work by Independent Analysts

Dean Radin produced a variety of analyses of the September 11 events. Some samples are presented here, with more in the extended analysis page. A paper with a detailed analysis of many aspects of the data gives special attention to the location of the eggs. Dean's treatment of the low-level data is different from the GCP's standard approach. Instead of a composite (Stouffer) Z across eggs, he calculates the Z-score per egg and sums the squared Z-scores and degrees of freedom across eggs. This responds to the variability among the eggs while the standard analysis responds to correlation among the eggs. This approach also mandates empirical instead of theoretical variance for the Z-score calculations. Dean uses sliding window smoothing or moving averages of the data across time. This can make interpretation difficult because the results depend very heavily on the choice of parameters such as the window width and centering. Because Dean also tries several sets of parameters to "optimize" the presentation, there is a form of data selection, so any probability or odds ratio that appears in the figures is very much an overstatement. It is, moreover, very difficult to compensate with the usual Bonferroni adjustment for multiple analysis because of the uncertain number of analyses he does. Dean believes that work of this kind is legitimate in the context of good evidence from properly designed studies. I present the analyses here with the caveat that they have no evidentiary value, although as complements to formal analysis, they may lead to useful questions. I should add that Dean says everything he tried showed unexpected structure in the data from September 11.

The following graph shows results for a 6-hour sliding window covering Sept 6 - 13. Dean identifies the peak value in this graph as occurring at 9:10 AM, Sept 11. However, the algorithm that he used for the sliding window averages the data for the six hours preceding the plotted point. Thus, in terms of the original, unsmoothed data, the "peak" weight of the averaging actually occurs three hours earlier, at 06:10. (The "19's" on the x-axis indicate 19:00 on each day.) In any case, the drop between this peak and the equally strong negative peak about 7 hours later is extreme, corresponding to 6.5 sigma (odds against chance of 29 billion to 1 if this were calculated for an a priori prediction). Dean observes that "such large changes will eventually occur by chance, of course, but this particular change happened during an unprecedented event, suggesting that this `spike' and `rebound' were not coincidental." A permutation analysis shows that the likelihood of getting a 6.5 sigma drop in Z-scores (based on a 6-hour sliding window) in one day, and within 8 hours or less (as observed) is p = 0.002.

Radin window_z.emp.jpg

The next figure shows the 2-tailed probabilities associated with the smoothed Z-score as odds ratios. There is an extraordinary spike near the time of the attacks, driven by large deviations that precede the first plane crashing into the World Trade Center towers; its weighted center is at 06:10, corresponding to the peak in the Z scores. The second spike occurs roughly seven hours later, with the weighted center at about 1:00 pm. The "0's" in the x-axis shows the start of each day.

Terrorist Attacks, Dean Radin 2

To help assure that there was no mistake in the processing, this same figure was recreated using algorithmically generated pseudo-random data instead of the real data generated by the truly random eggs located in countries all around the world. This figure speaks for itself.

Terrorist Attacks, Dean Radin 3

Peter Bancel has taken another independent perspective, focusing on the temporal development or autocorrelation of the data to gain perspective on possible linkage of the eggs' output over time. He explains "Basically, what I do is do an autocorrelation of the sec-by-sec Stouffer z's over regs, using Fourier techniques. The resulting autocorrelations are then normalized to the Sqrt (of the number of data pts-the lag). This gives a distribution of "z" values that should be very closely N(0,1) distributed. I then visualize the result by taking the cumulative sum, much as we do for a classic reg experiment.

"The large rise in the autocor 8-12 figure can be understood as coming from a large excusion in the cumulative deviation of the z-scores (sec-by-sec Stouffer's z - not the z^2) which occurs from 9:50 to 11:50. This positive excursion as an isolated data set has a two-tailed-pvalue of 2 x 10-4 (z=3.71). So it's strong and it lasts for 1.9 hours. Placed in the context of a 24 hours data window I guess a Bonferroni correction would put the pval at 2.5 x 10-3."

Peter Bancel, Autocorr 8 to 12 on 010911

A much more refined analysis of autocorrelation shows the relative predictability of the device variance over lags up to four hours, second by second, passing each autocorrelation window over the full 24 hour EDT day. The next figure shows the data from Sept 11 in red, compared with 60 surrounding days of August and September. The latter show a cloud of essentially random traces, nearly all of which remain within a 5% probability envelope. In stark contrast, the Sept 11 autocorrelation is consistently large over the first hour of lags, continuing at a lower but still significant level for the second hour, after which the cumulative deviation line becomes essentially flat again. Besides the obvious difference from the comparison traces, an indication of the likelihood is given by the fact that the cumulative trace penetrates a one in a million envelope -- but see below.

Peter Bancel, Autocorr Var 4 hrs of Lag on 010911

Further analysis using a more appropriate asymmetrical distribution suggests a less extreme probability of 0.0005. This is shown in the next figure. In any case, there is little question that these data show structure where there should be none.

Peter Bancel, Autocorr Var using Asymmetric envelopes

A third set of independent analyses has been generated by Richard Shoup, who has examined the data from the beginning of July to the end of October to see empirically how unusual September 11 is. A paper describing his survey, available at the Boundary Institute contains interesting perspectives. An analysis of the increase in variance on each day over the four-month period, for example, shows that the largest peak by far occurs on September 11. He uses smoothing windows of various widths and reports that they all peak on the llth. He calculates odds of 131,000 to 1 for the largest spike, produced with a one-hour smoothing window.

Shoup correlations ofcumvar, 010911

In addition, Shoup looks at correlations and trends in variance. Here is an example, where 32 of the 37 eggs (those with no missings) are divided several ways into two groups of 16, and the cumulative deviation of the sum of squared Z-scores plotted separately. The parallel trends beginning at about 08:00, GMT, suggest substantial inter-egg correlation. A similar analysis on Aug 11 and on Sept 10 shows no similar pattern.

Shoup correlations ofcumvar, 010911

Discussion and Interpretation

Using exploratory methods to visualize structure in the data, it appears from various perspectives that there is a concentration of strong deviations around the major events of September 11. A far more comprehensive survey of such explorations, including analyses by Dean Radin and Peter Bancel, is available in an extended analysis page. Some focused work addresses still other aspects of the GCP data. One important question, both theoretically and practically, is whether distance makes a difference. Dean Radin has done a careful analysis of results by location that is detailed in the context of validity checks on the procedures, and studies of the timing of the egg network response. Another independent analysis, by Peter Bancel, looks at autocorrelation of the Stouffer Z-scores. This perspective assesses the correlation across time, which is independent of correlations across the eggs. A strongly skeptical look is provided by Ed May and James Spottiswoode in their critical analysis of selected items from the formal and exploratory work by Roger Nelson and Dean Radin. (downloadable pdf file, about 55 K.)

To begin a discussion of these remarkable results, I must acknowledge that the very best we can do is to report the data honestly and completely, because we do not have a theoretical understanding of the sort that must underlie robust interpretations. Of course we try hard to understand, and we are asked to explain the results. It is important to identify the answers we give as speculative and provisional, but having said that, I would like to describe a speculation that appeals especially to my aesthetic self, to my right brain. You can find more general discussion of alternatives and cautions elsewhere on the website.

One way to think of these startling correlations is to accept the possibility that the instruments have captured the reaction of a global consciousness beginning to form. The network was built to do just that: to see whether we could gather evidence of a communal, shared mind in which we are participants even if we don't know it.

Groups of people, including the group that is the whole world, have a place in consciousness space, and under special circumstances they ­ or we ­ become a new presence. Based on evidence that both individuals and groups manifest something we can tentatively call a consciousness field, we hypothesized that there could be a global consciousness capable of the same thing. Pursuing the speculation, it would seem that the new, integrated mind is just beginning to be active, paying attention only to events that inspire strong coherence of attention and feeling. Perhaps the best image is an infant slowly developing awareness, but already capable of strong emotions in response to the comfort of cuddling or to the discomfort of pain.

The hypothesis we set out to test is that the REG devices we use may respond to the concerted effect of large numbers of people turning their attention in one direction, becoming absorbed in the same focus. The terrible events of September 11 were a powerful magnet for our shared attention, and more than any event in the recent memory of the world they evoked the extraordinary emotions of horror and fear and commiseration and dismay.

The EGG network reacted in a powerful and evocative way. While there certainly are sensible alternative explanations, this is not a mistake or a misreading. It can be interpreted as a clear, if indirect, confirmation of the hypothesis that the eggs' behavior is affected by global events and our reactions to them. More important than any scientific question, however, is the question of meaning. What shall we learn, and what should we do in the face of compelling evidence that there may be such a thing as global consciousness? In fact this is not a new question. The results from this scientific study are an apparent manifestation of the ancient idea that we are all interconnected, and that what we think and feel has effects on others, everywhere in the world. The implication of the GCP/EGG data reflecting our shock and dismay is in some sense quite obvious. It says that even insensate electronic random generators can see the effects of hatred born of pain and despair. It means that the earth cannot support us in comfort as things now are. It urges a new understanding that we must learn to accept each other and help and support each other, everywhere in the world, if we are to live in peace on this beautiful earth.

More

Many other analyses and graphs have been generated, and some show certain details and perspectives that may interest you. The extended analysis page has most of the figures shown here, but in the context of the developing analysis program over the first few days following the tragedy. The data are treated in with different procedures by Dean Radin and Peter Bancel in a series of explorations seeking more insight. They require very careful interpretation so are not included in this primary summary of the September 11 results. There is also an Interpretations page in the works, and one that simply presents the flood of messages from people all over the world who are involved in the GCP/EGG project. For more details about the project itself, you can go to the GCP home page where you will find links to all aspects.


GCP Home