GCP FAQ

A number of aspects of the Global Consciousness Project (abbreviated GCP and sometimes also called the EGG project) are unfamiliar to a general audience, including people who are seriously interested in the questions we are addressing. We also get critical and skeptical questions that help to identify scientifically challenging aspects of our findings. Specific queries are answered directly, but it may be of value to assemble the most salient of these in an FAQ.  Here is an example from an email we received:

I checked over your site just now, and wonder just what's being represented here. REG means random number generator? It creates random numbers? Then you look over those numbers to check for randomness? What do up and down on the chart mean? Boy, it looks fascinating, but I'm quite fuzzy about what I'm seeing.

This and other questions treated in this FAQ are raised with some frequency. Some example emails with my brief responses may be helpful as a summary of issues that are treated more completely here. Occasionally I give an email interview that summarizes things concisely. Sometimes they come in batches, as in Spring 2007: an online version of a major German newspaper, Die Zeit, an electronic journal associated with What the Bleep, called the Global Intelligencer, and the Italian edition of Marie Claire magazine.

There are detailed explanations of all aspects of the project which can be found by following links from the GCP menu, especially under About and Procedures. For a brief account of some of the most unfamiliar aspects of the measurement and analysis techniques, the following FAQ-style descriptions should be helpful. For more specific information, go to the measurement page. There are a variety of immediately informative links on the media information page. If you have other questions, or cannot find an explanation of some aspect of the project, please feel free to email questions or comments to Roger Nelson, the Project Director.

Frequently Asked Questions

I have been trying to really understand your data on your web sight for the last 30 minutes and frankly it's a little confusing. ...I'm not of a scientific bent. Please explain to me simply what the graphs mean.

The EGG Story includes a step by step explanation. Briefly, the expectation for the graphs is a random walk that has a horizontal trend, but lots of random fluctuations. We sometimes see large deviations from that expectation (the line takes on a slope) that are correlated with big events in the world. At this point we do not know whether this is only a response to the news getting out, or whether it is something more fundamental. There are some cases where the apparent reaction is more consistent with the latter—those cases make me think maybe there is a global consciousness that is deeper than mere news recognition. It may be useful also to look at a discussion of interpretation of these graphs. It is important to keep in mind that an individual event and its analysis cannot provide a complete answer. See the next FAQ entry.

I viewed the information under the Primary Results heading, Table 2 Results, for individual events. Most of the variances expressed for the 250 events [at the time] are fairly small, however, the last event, Obama getting the Democratic nomination, shows a 2.172 variation. Of all the events happening in the world why would this one have such a big variation? I am trying to understand this. Can you help me?

It is a good question, for which we have a partial answer. The signal to noise ratio is very low in this experiment, which means that in some cases a true signal may be buried in the noise, and hence not seen. It also means that in some cases there may be nothing but noise fluctuations, but these may masquerade as signal. The typical case will be a mixture. Unfortunately, there is nothing that allows us to distinguish deviations that are just random fluctuation from those that are driven by consciousness of the event. Nothing, that is, but patient accumulation of instances that are similar, and hence can be combined. In short, the results for single events cannot be reliably interpreted—even though it may look like it. The average effect size is about 0.3 to 0.5 standard deviations, and this means that we need at least a dozen or more events to get a reliable answer to the question whether there is a significant effect. The apparent response to the Obama nomination is probably partly effect and partly noise—the two sources of variation just happen in this case to come together to make a strong appearance.

We are gathering a large number of people to meditate and pray for peace in the world. Can you measure the power of our efforts to effect social change?

We need to explain the GCP instrument in this context. To begin, the events on which we usually focus are global in scope, and typically engage the emotions of millions of people. Beyond this, the tools we use are statistical, and the data from the network are normally distributed random numbers in which we see effects as very tiny statistical deviations from expectation. The signal to noise ratio is so small that the average effect size is equivalent to 1/3 sigma. This means that single events can't identify effects in a reliable way. We need to patiently replicate the hypothesis testing many times—at least 20 or 30 events must be assembled as a composite to get an interpretable statistic if the true effect size is average. If the type of event has a larger average effect the required number will be smaller, but under the best of circumstances, we need many repetitions of similar events. We might wish it were otherwise, but our system isn't sensitive enough to measure immediate effects of intention and meditation. A better option is our human sensitivity, the direct perceptions of heart and mind, which are the finest tools we have for this purpose.

Has an experiment event ever been staged for test results specifically for GCP? Has a global link up ever happened staging the same the event repeatedly 20-30 times or days in a row?

Although our policy or protocol is simply to monitor what happens, and not to use the eggs locally or generally for intentional experiments. We did do one "experiment" where we sponsored and helped publicize an event. It was called "Just a Minute" and asked people to take just one minute of time at noon January 1 2000 to meditate or think about a bright future. You can see the description and results at http://www.global-mind.org/jam.html. From my perspective, there are many repetitions of events or types of events, such as New Years, religious holidays, giant organized meditations, and unfortunately also terrorist attacks, and natural disasters. So we have plenty of replications, and indeed we see that the faint signal which otherwise is buried in statistical noise does rise out of that noisy background to make a persuasive statistical bottom line.

I looked up Global Consciousness Project on Wikipedia, and it appears that the editors there don't think much of the GCP.

Yes, we have noted that the Wikipedia article on the GCP changed substantially sometime in 2008 or 2009 from a previously reasonable state. We believe this is the work of a couple of non-neutral editors (evidently passionate skeptics), who revert efforts to correct errors of fact or attempts to create balance. In the discussion page, I made some comments about the situation, which produced an invitation to say what errors I see in the article, and a promise to try to fix them. Here is our discussion of skeptical bias in the Wikipedia GCP article. This is the best we can do until Wikipedia takes care of itself via its slow but sure process of signal replacing noise.

What are the data made of, how are they generated, and what do we measure?

The data come from electronic random sources called random event generators, abbreviated as REG or sometimes RNG, for random number generator, which produce a steady stream of unpredictable bits. They are the equivalent of high-speed electronic coin flippers. We simply record the actual value of 200 coin flips every second at each device, expecting it will be about 100 (50/50 chance for a bit to be 1 or 0). The resulting count is a varying quantity, of course, depending on chance fluctuations. We record these trial counts continuously at each node in the network, every second, every day over months and years. Thus, we have data to examine for changes correlated with events in the world. From the statistics, especially the variation in the mean or average count, we can determine the relative randomicity of the datastream, and thereby identify unexpected structure that might be correlated with global events of importance to humans.

I don't understand the graphs of results. Are you doing time-series analysis?

No, we are not plotting the raw data in an ordinary time series, so a slope does not mean the scores are getting larger and larger (like temperature increasing during the spring). Instead, a slope indicates a tendency for each score to be slightly but consistently different from expectation. We are interested in whether there is a persistent departure of the whole network's data from the theoretical or empirical expectation. The graphs used most often show the history of accumulating deviations of an average score across the eggs. This is a type of graph used in control engineering, in which it is easy to see changes in the behavior of data. A description of the calculations may help. The statistic we most often use is the squared Stouffer's Z, which we call the network variance. This is calculated as the sum of the Z-scores across eggs, normalized by the square root of their number, then squared. The result is a Chisquare with one degree of freedom (DF). The graphs are the accumulating sum of Chisquare minus the DF over the period of an event. But the statistic that is important is the total accumulation of deviation, or the end-point of the graph. That tells us how strong or significant the departure from expectation is for that event. The egg story article, in the section on how it works, gives some background. I wrote the article many years ago, but we continue to do the same thing.

Why doesn't the graph of results for the same event in two different years look the same?

The data from the GCP have a low signal to noise ratio and one important implication is that single events cannot be interpreted reliably. There is a caveat at the end of each detailed analysis (in the formal series) that says:

It is important to keep in mind that we have only a tiny statistical effect, so that it is always hard to distinguish signal from noise. This means that every success might be largely driven by chance, and every null might include a real signal overwhelmed by noise. In the long run, a real effect can be identified only by patiently accumulating replications of similar analyses.

This is exactly the situation, without exception, even for monster events. The world is complex and large, and there will be numerous events and circumstances competing for a limited array of global attention, so to speak. Thus, we cannot interpret the result this year (this applies to every year) as a literal picture of the effects of meditation, or Valentine's day, or Sun eruptions, or the turmoil in the Middle East. Only when we have dozens of events can we combine them to give reliable interpretations. We cannot take one event as proof of global consciousness, nor can we conclude that nothing happened today. A graph may show interesting trends or uninspiring flatlines, but neither will support our creative explanations and interpretations. We are literalists, and we think they must, but the data are just too noisy. A metaphor may help see the problem: think of a fish in the sea, breaching the surface amidst large and small waves. Any ripples caused by that fish will be very hard to see. That's where we are, and it is a small miracle (aided by good statistics) that we see anything at all.

What does the actual measurement look like and what does it mean?

We look for anomalous structure or order in what is normally expected to be random data. The most common measure of such structure is based on differences from theoretical predictions, in particular, deviations from the mean or average that is expected for the trial scores. If there is a persistent tendency for the data to differ from expectation, this will show up in statistical measures. The idea that we can use the scale of such variations in the data as a measure of some aspect of consciousness is derived from three decades of laboratory research indicating that conscious intention can affect the randomness of REG devices in controlled experiments. What the GCP/EGG project is doing is a direct extension of such work. A more extended discussion of how we do the measurement is available.

How do you jump from there to global consciousness measuring the effect of organized meditations or major public events around the world?

In field studies with REGs we have found consistent deviations from expected random data sequences taken in situations where groups become integrated. During deeply engaging meetings, concerts, rituals, etc., the data tend to show slightly greater order, and we are able to predict this deviation with small but significant success. In the GCP case, exactly the same thing is done—we predict a detectable ordering (in the form of slight meanshifts, or changes in the variability of the mean) in otherwise random data during world-class events that are likely to engage the attention of large numbers of us around the globe. The continuous data streams registered by the EGG network have a well-defined statistical character, namely, random expectation, and we simply look at the empirical statistics to see whether our predictions of structure are supported by the data. That is, we predict differences from the random expectation that are correlated with the events, and use standard statistical tests to see whether the predicted structure exists in the data.

What do you mean by consciousness in general? That is a pretty slippery term.

Consciousness seems to result from coalescing connections among the elements of brain and mind. Consciousness is created when coherence develops in an otherwise chaotic, random flux of subtle chemistry and faint electrical signals. Ordering influences may be subtle external agents and operators, and they also may be internal, self-organizing principles. The essence is order, pattern, structure, and, ultimately, meaning. The metaphor can be extended in most any domain. Consciousness can be small and simple, like what we would imagine for mice, birds, snails, bacteria. It can also be stretched mightily, to help think about forests, oceans, flocks, herds—and groups of people. And, of course, it can be extended to the world, where we can apply the metaphor on multiple levels, ranging from crowds and cultures to gaia herself. In human terms, consciousness is usually associated with being awake and aware, possibly even self-reflective. Because we are here looking at a broader set of possibilities, our usage necessarily implies also the unconscious and subconscious aspects of the organized activity that defines the mental world.

I can't figure if someone actually predicted the Concorde crash or if you are getting at something else.

What we do is to predict that if there is a powerfully engaging event in the world—like the Concorde crash—the focused attention of large numbers of people will produce a departure from expected behavior in our network of REG detectors. We speculatively imagine a consciousness field that becomes relatively coherent and structured when these occasional global events occur. Continuing the speculation, we suggest that the information in such a field can somehow be absorbed by the REG devices, which then show patterns where none should exist.

What does the graph that accompanies most of the analyses show?

Our most frequent graphical display for the results shows a cumulative deviation of trial scores from expectation. Each raw score is processed (normalized and squared) to create a well-defined quantity called a Chisquare which can readily be combined and compared. The deviations of these values from their expectation should average to zero in truly random data, but if there is structure in the data, the deviations may accumulate in a consistent positive or negative trend. The length and steadiness of such trends, and more importantly, their correlation with events which have been identified in formal predictions, are indicators of anomalous structure. The graphs typically include a smooth curve showing the criterion for statistical significance. Data traces that show a steady deviation that progresses outside this envelope are unlikely to be just chance fluctuations.

How can the correlations be explained? What physical mechanisms could allow the anomalous effects of focused world consciousness?

No fully satisfactory explanation is available, but some speculations in physics, healing research, and parapsychology touch upon the same issues. Basically, we have evidence for an influence or correlation that does not depend on ordinary physical mechanisms. Most likely, we will need some extensions of standard models that will include consciousness and its special attributes, including non-locality, true creativity, and meaning.

If I understand correctly, the notion of global consciousness suggests that if many people are thinking about the same thing at the same time it can influence other events. What I find hard to grasp, however, is the process or mechanism by which shared consciousness (e.g., 9/11) could influence something inanimate like a random number generator. What am I not realizing?

It isn't what you miss, but what we don't know. The empirical case is good, but theoretical modeling is weak and speculative. The best bets are quantum mechanical entanglemant (operating in a quasi-macroscopic realm) and active information a conceptual structure in David Bohm's physics.

My own model is that consciousness or mind is the source or seat of a nonlocal, active information field (this is not a standard, well defined physical construct). Such fields interact, usually with random phase relationship and no detectable product. When some or many consciousness (information) fields are driven in common, or for whatever reason become coherent and resonant, they interact in phase, and create a new, highly structured information field. The REG has an informational aspect (entropy) and a completely undetermined future, and I speculate, with Bohm, that it manifests a need for information which allows or guides the actualization of the active information sourced in human, group, or global consciousness.

What about disturbance in the power grid, or extraordinary levels of cell phone usage, or other EM fields? Might these be an explanation for the deviations in a case like September 11 2001?

Such influences would have a geographical concentration. In this example, they would center on New York and Washington, of course, but the eggs are distributed around the world. Their average distance from New York is more than 4000 miles (∼6400 Km). More important, the design of the research-grade instruments we use includes both physical shielding and a logic stage that excludes first-order biasing from electromagnetic or other physical causes. Thus we are forced to look elsewhere for the source of the induced structure.

Where are your control data? How do we know the statistical tests using theoretical expectations are legitimate?

We can do several kinds of control trials, including matched analysis with a time offset in the database, or matched analysis using a pseudorandom clone database. However, the most general control analysis is achieved by comparisons with the empirical distributions of the test statistics. These provide a rigorous control background and confirm the analytical results for the formal series of hypothesis tests.

Have you picked a random event as a control and done an analysis of randomness based on that? If so, where might I find the results?

There is a full description of the statistical characterization of the data on the GCP website under the scientific work set of links. You can start with EGG data archive. Some relevant points are discussed elsewhere in this FAQ. It is possible to pick a control random event as you suggest, but that is not a satisfactory way to address the implied question. We use the more powerful techniques of random resampling, permutation analysis, and full database statistical characterization. For some purposes, e.g., to establish statistical independence of measures, we use simulations and modeling. The brief answer to your question is that the normalized data do show expected values across the full database in all moments of the appropriate statistical distributions. The same is true of the individual physical random sources and their composite, and it is true for each of the measures we use in the hypothesis testing. This is the background. Against that background, the replicated hypothesis tests in the formal series show departures from expectation, with a composite Z-score of about 4.5 (May 2007).

How often do large variances/deviations occur which do not coincide with major world events? Or put another way, If only predictions are focused on, isn't it possible that there are major variances which naturally occur...at random and yet are not captured or reported?

We can describe the expected distribution of variances that naturally occur...at random. This is just the statistical background against which we compare the variances observed in the formal hypothesis tests. Another way to look at this is to consider calibration or control tests. These show a range of variation, including some surprising deviations and spikes. For example, we can calculate exactly how often we should see a 200-bit trial sum of, say, 134 when 100 is expected. (less than once in a million trials). In truly random data, we should see such examples, of course, and we do. The question then becomes one of the frequency of big deviations correlated with specified events. Our finding is that such correlations happen more often than expected. On the other hand, we can't determine how many large deviations occur in response to major world events which have not been the subject of hypothesis test predictions. It would make sense that there are some, assuming that the events or people's reactions are the source of effects on the EGG network, as seems to be indicated by our analysis of the formal event sequence.

[Another question with similar intent:] How many times have the eggs spiked without any type of global event happening? Where can I find this data?

The data are demonstrably random in the grand ensemble. This is described in some detail in the section dealing with the GCP Data. If you are familiar with randomness and random data distributions, you will know that fluctuation, including extreme deviations (spiking) are inherent, expected aspects of the distribution. Thus, the answer to your first question is, literally, the expected number of times.

Another way to address your question is by considering the formal scientific protocol we follow, which pre-specifies the period of time and the corresponding data in which we hypothesize there will be a departure from expectation, versus the alternative implied by your question: Look for spikes, and then try to find out whether there was a global event. The latter protocol fails as good science because it has unknown degrees of freedom. Moreover, in our large and very complex world, you probably can find something to construe as a global event almost any time—but there will be no way to establish a correlation with data fluctuations except by testing a pre-defined, properly constrained hypothesis.

The September 11 graphs suggest a precursor effect, as has been seen in a few prior cases. Could this be used as a warning?

The best guess is we cannot use the EGG data for such practical applications. One major reason is the statistical nature of our measures. Nobody has yet come up with anything more direct, and this means that there will be, by definition, both false positives and negatives. Moreover, the effect size is so tiny that we almost always require repeated measures, or measures over a long time to detect any anomalies. To see precursors we have to look back across that time from a post facto perspective. Unique point events have little chance of being seen, at least by our current methods.

Could the GCP correlations be due to more local consciousness effects? For example, all of us who are following the GCP are also somewhat synchronized through the Internet and we spend much time close to the Eggs.

The source of anomalous effects probably has several dimensions, including the experimenters, and geographically local groups as well as dispersed interest groups. The good news is that the GCP database offers some potential for answering this question. The issues are subtle, but there are sensible criteria such as correlation of effect size with numbers of people potentially involved, and timing of knowledge about the event, that may help distinguish between experimenter effects and local versus general influences. The bad news is that we still have only a tiny statistical effect, so that it is always hard to distinguish signal from noise. This means that every success might be largely driven by chance, and every null might include a real signal overwhelmed by noise. In the long run, the signal can be seen, of course, but to separate the signal into its components (global consciousness, experimenter effects, local group influence, and so on) will require a still longer run and well-thought-out analysis strategies. There is more discussion of the sources of effect.

Is it possible for us to contaminate the data by focusing on the dot? In other words, are those of us who are in closer physical proximity to the Eggs capable of exerting undue influence over the Egg?

It's a good question, for which we have only partial answers. The GCP dot shows a composite across the whole network which is an average correlation of pairs of eggs. The effects are generally very small compared with noise—it's a classic low signal-to-noise situation. The S/N ratio is so small we can't reliably detect effects of even major events if we look at only one case. We need a dozen or two events of a similar kind to get reliable statistical estimates. The implication is that individual personal influences on the system will be too small to detect, although they may be contributing to the overall global consciousness. Because the GCP dot's composite is an average correlation of pairs of eggs, a local influence (close proximity) actually doesn't quite mean what it seems to in terms of our measure. That is, if you influence your egg, and I influence mine, that won't affect their correlation—unless we both exert the same, well-synchronized influence. If you think about it, when that is the case it is pretty much the same thing we mean when we talk about the global consciousness operationally. We say that large numbers of people are engaged by the event, they have the same response and share thoughts and emotions. I think this shared response may reach a level of coherence that is capable of supporting an information (or consciousness) field that becomes the source of the effects we measure.

So, can I tell when some major event is breaking by watching the color dot, or reading the graph that goes with it? How should I interpret the colors?

The simplest way for me to address interpretation is to say those displays are artistic, aesthetically expressive but not suitable for scientific interpretation. The GCP's formal program of analysis gives evidence of structure appearing in the random data correlated with mass consciousness. But the evidence is cumulative over many tests showing tiny effects, and it becomes reliably interpretable only by the patient accumulation of months of data. Those instantaneous displays are pretty, but in the context of scientific interpretation, they are not currently useful for data analysis. The language about what colors mean is perhaps too suggestive. What can be said is that, over the long run, the increased variance shown as red dot colors would suggest greater coherence. But in fact the instantaneous, momentary color is as likely to be statistical noise as otherwise.

I am aware that many people would like to have a direct reading thermometer to monitor our social temperature, but the GCP instrument is not that. It is more like an EEG measuring evoked responses in the brain, where many measurements have to be averaged to see meaningful patterns. What the GCP can do is show that there is a small but ultimately important manifestation of our human consciousness when it becomes synchronized or coherent. But this demonstration depends on examining many tests of a well-formed experimental hypothesis, and it requires processing data accumulated over months and years.

What about the experimenter effect? What is really the source of the anomalous effects?

Some people among the professionals who pay attention to psi research think the experimenter effect is a major contributor to anomalous effects, even in something like the GCP network. This would be the closest thing to a model relevant to your question, whether a powerful personal/individual experience would show up. My opinion is that the weight of interpretable evidence does not support that model—but, there is a caveat of some importance. I think that as in experimental physics, the nature of the question helps to determine the answer, as particles versus waves depend on the question asked.

In psi research this may be a more potent (albeit still poorly defined) question: some of us think that an experiment can be aimed or focused in ways that are unexpected in the usual scientific perspective. For example, the FieldREG experiments I did while at PEAR were more-or-less local in the sense the focus was on a group in the same room or space. But the same technology and analytical tools were applied to progressively wider ranging targets. When Prime Minister Rabin was assassinated in Israel, I was in Germany, but I had a continuous running REG in Princeton. This historically momentous tragedy seemed to me to have global impact and implication, and I set a hypothesis (which effectively assigned a question) that the very non-local REG in Princeton would show a reaction at that time.

I think that we will ultimately discover three or four major sources of the effects, with the nominal source being most prominent, the experimenter effect second, and then a couple of other sources, including the form/nature of the question, and finally, an anthropomorphized universe or cosmos that pushes us to stay on our toes, to recognize and understand that in a complex world no simple answers will do for questions about the nature of consciousness.

On close examination, there are several remarkable assumptions or speculative leaps in this process, but the result is an instructive confirmation of the hypothesis, a remarkable departure from expectation in the REG trace, precisely focused on the time of the murder. The point is, of course, that this effect was comprised of or driven by:

1. the nominal source: shock and compassion
2. the experimenter expectation: REGs see information fields
3. the nature of the question: nonlocal space, local time
4. and probably the Coyote: the Universe's wake-up call

How do you make the leap that the deviations from randomness are related to world events or consciousness? After all, when you find a deviation you can check the news and ALWAYS find some world event that is taking place, because world events happen every day. There are never days without world events anymore, so it seems that there is a possibility that this is just a coincidence.

The leap we make is only to ask the question. The answer seems to be yes, there are correlations. With regard to your concern that we can always find a special event to fit the data, we fully agree. However, we do our experimental work the other way around from what you have inferred. First we make a prediction that some identified event will have an effect, then we assess the data to see the actual outcome. Though some people suggest that we should do so, we never find a deviation [and then] check the news, because you are right—it will always be possible to find some event that we might imagine was the cause. The GCP methodology is prediction-based. Before the data are examined, a prediction is registered, with all necessary analysis specifications, and only then do we perform the analysis that allows us to quantify the correlation and assign it a probability against chance.

What do skeptics say? What do they offer that may help understand what is going on?

We have had a few useful skeptical commentaries. Here are a references or links to thoughtful expressions of skeptical views. The FAQ item following this introduces an extended discussion of issues raised by Ed May.

For an extended discussion of problems in the quality of skeptical attention to the general area of psi research I recommend articles by Brian Josephson, linked from his home page, and a broad assortment of useful articles on the Skeptical Investigations website.

Dr. Jeff Scargle wrote a critical commentary to accompany two articles treating the GCP response to September 11 2001, one by Roger Nelson and one by Dean Radin, for the Journal of Scientific Exploration, Vol 16, No 4, pp. 571-578 (2002). There is a letter of response to Scargle's article in JSE Vol 17, No 2, by Roger Nelson.

More recently, Prof. Gerard 't Hooft has created a website that he calls parabet in which he provides a CHALLENGE for the BELIEVERS in the PARANORMAL. Though he does not mention the GCP, I have had correspondence with him and this is his intended focus. He expresses his conviction by making a bet that

If the experiment is repeated exactly in accordance with conditions to be formulated by me, and if the statistics is subsequently handled with mathematical precision, I am prepared to bet for EUR 1000,- that the signal will be less than 5 sigma.

In another, related page, 't Hooft describes a few scenarios that he believes might explain away anomalies seen in parapsychological research. Though again there is no specific reference (and most of the examples are not relevant) this provides some indication of his reasoning and his skepticism with regard to the GCP results. A response to the concern on which 't Hooft bases the formulation of his bet is available.

What concerns about protocol and procedural issues have been expressed by critics? How might the statistically significant results be explained away?

Dr. Edwin May has made a number of critical commentaries, and he posted several specific concerns to the Parapsychology Research Forum – a serious discussion of research issues (method & theory), prf at jiscmail.ac.uk on 9 Aug 2001. (The PRF may be defunct now.) I took the opportunity to respond in some detail to the points, and because they are germane here, the exchange is reproduced in the style of the FAQ as a series of questions and answers.

There has never been a stable analysis time and to invent one post-hoc is problematic.

All the analysis times are determined before data are examined—none of the formal analyses have any post hoc aspect.

We have hardly addressed what the hypothesis actually is. For example is [the prediction for] the time of the event when hardly anyone but the locals know about it? Or is it a fuzzy time later when more and more people awaken and read their papers? If so when?

The GCP website addresses this issue in detail under Prediction Procedures. It is also addressed in an article forthcoming in the Journal of Parapsychology. (That article is now published.) Briefly, we can specify both a Time of the event and an Aftermath prediction, though in most cases we pick one of these options. In some cases we have a time period that is clearly defined by the event; sometimes we have a point event (say, an explosion) and we must arbitrarily specify a period surrounding the moment. The predictions are made and registered before examining the data; they are not made by inspection and they are not conditioned by the results.

[The overall database is claimed to show a 5 Sigma departure from expectation.] I doubt it when corrected for [selection] problems and in some cases multiple analyses.

We don't do multiple analyses, and do not have a need for corrections. There is one exception to this, namely the Y2K analysis Dean Radin did. He said that he tried several variations before choosing the best representation. That case is included in the formal database, with a Bonferonni correction of a factor of 10. (The detailed presentation discusses the correction. The event is excluded in our formal parametric analyses.) In addition, I did a number of exploratory analyses with different algorithms but a similar fundamental hypothesis—variance reduction at midnight, and I found that most of them showed evidence of structure focused on midnight. Check the Results page, and the explorations that are linked from the Y2K analysis in the formal results table.

How do you choose which events to count and which not? The GCP has consistently ignored many major events of high spiritual value with 1 to 20 x 10^6 people in India. e.g., India was left out of the analyses all together, as far as I know, even though they account for about 1/6 of the world's population.

In fact, India figures prominently in several formal predictions: India, Train Crash, 990801, Typhoon, India, 2 Hours, 991029, Typhoon, India, 24 Hours, 991029-30, Just A Minute, 1 Min Epoch, 20000101, Kumbh Mela, India, 20010124 and in several more exploratory analyses, including the elections in October 1999. The GCP does not pretend to assess all worthwhile/reasonable/major events, but we definitely have not ignored India.

Pongala is another example. Each year at a specific time and date 1+ Mill women only gather and light a fire in honor of Bhadrakali at a specified instant. Surely if the GCP folk think that a meeting of 100 odd (very odd indeed :-) ) people at the PA in NY can whack the GCP, then surely Pongala must, and [so] must also many other events that are nearly continuously [happening] that we do not know of. Maybe some folk might think that this is why the generators appear random in the first place!

I don't know anything about Pongala. I am, on the other hand, willing to accept predictions made by others, as you can see in the Results table, in the column that tells who made the predictions. The Bhadrakali fire ceremony sounds much like the Kumbh Mela in India for which we did make a prediction in January, 2001. As the methodology and procedures section on the website seeks to make clear, we intend to test predictions that are sensible in terms of the defined hypothesis. We do not include the meeting of 100 odd people at the PA in NY in that category—that is a spurious example; indeed, it is a straw man.

So, DAT [Decision Augmentation Theory—which proposes that apparent effects on REGs can be explained as precognition of opportune times to push the button to extract non-random subsets of data from an ongoing sequence] can and probably does enter in the following circumstances:

1. [Selecting the] Window of analysis

The window of analysis is specified in the prediction registry, without knowledge of the data.

2. [Deciding] which events to keep and/or ignore

All predicted events are kept, none are ignored.

3. [Deciding] which time zones to keep.

I used 24 time zones in my first New Years analysis, 36 zones beginning with the Y2K analysis, and since then. I may be slow on the uptake, but I am honest.

4. [Deciding] which analysis to do.

Again, this is prespecified in the prediction registry.

Yes, why not DAT? I think something like an experimenter-psi-based selection of events that seem to be good prospects, plus good guesses about the parameters of analysis, etc., are likely to be part of the picture. Maybe something like precognition could be a factor. But DAT is not the only possible source of putative effects that can and probably do enter. My personal litany on potential sources includes four: The nominal source (e.g., global consciousness), the experimenter(s), the environmental matrix in which the question is asked and answered, and the cosmic trickster (which I'll tell you about over beers).

More important, our ongoing program of deeper analysis and modeling has been producing useful results in recent years. In particular, we now know that structure in the data is is present in 2 orthogonal measures and three independent correlations either between them or with other parameters. This provides material for empirical modeling, and the calculated results are inconsistent with DAT, and supportive of field-like models. See a Response to a letter from Helmut Schmidt in the Journal of Scientific Exploration, Vol. 23, No. 4, 2009

Radin used a third order statistic on the variance and computed a sliding median (rather than the sliding average) because the above worked and variants did not. In addition it only worked with integer time zones leaving significant others out of the 37 zones. Including just them and keeping Dean's exploratory analysis made the effect go away.

Reviewing the Results page and the links from the Y2K events yields an extensive exploration of these issues. I'll just make two remarks here: Dean made a prediction about reduced variance in good faith, and proceeded in good faith to use the 24 timezones most people believe pretty much cover the territory. The 37 timezone analysis is post hoc. Moreover, in my own explorations addressing the same general hypothesis, I found much the same results with 24 or 36 timezones. Indeed there are still more timezones than 36 or 37, including at least two which are shifted not 1/2 hour, but 7+ minutes. However, the right way to do an analysis is to follow the plan prescribed in the experimental design, without additional conditions that are discovered after the fact and after the results have been examined. Such explorations can be very informative, and they can help in prospective designs, but they cannot be a part of formal hypothesis testing.