Document Type: 
Document Number (FOIA) /ESDN (CREST): 
Release Decision: 
Original Classification: 
Document Page Count: 
Document Creation Date: 
November 4, 2016
Document Release Date: 
June 11, 1998
Sequence Number: 
Case Number: 
Publication Date: 
February 3, 1993
Content Type: 
PDF icon CIA-RDP96-00787R000300230001-5.pdf8.13 MB
Ap SeQpF-c Rel a 2000/08/10 : CIA-RDP96-00787R0003W2-30000D5O1O Phenomenological Research and Analysis Edwin C. May, Ph.D., Wanda L. W. Luke, and Nevin D. Lantz, Ph.D. AEIWWAF~_ ME 0 Science Applications International Corporation An Employee-Owned Company Contract MDA908-91-C-0037 (Client Private) Submitted by: Science Applications International Corporation Cognitive Sciences Laboratory 1010 El Camino Real, Suite 33330P..O.. Box 1412, Menlo oPPark, CA94025 ? (415) 325-8292 Apfs~~i?~E~1nCJ{y/~pf' ri'tSvi'~r"`~l'LYP~F7?`Vt7ia?~at7 ~tYi9t~ son ApgX jJ1FF"6e se 2000/08/10 : CIA-RDP96-00787R000300230001-5 TABLE OF CONTENTS LIST OF FIGURES ..................................................................iii LIST OF TABLES ....................................................................iv I OBJECTIVE ................................................................ 1 II BACKGROUND ............................................................ 3 1. Historical Perspective ..................................................... 3 2. Current Program ........................................................ 4 EXECUTIVE SUMMARY .................................................... 5 1. Target Dependencies ..................................................... 5 2. Enhancing AC with Binary lkrgets .......................................... 6 3. AC in Lucid Dreams ..................................................... 7 4. Magnetoencephalograph .................................................. 8 5. Enhancing AC of Binary Targets ........................................... 9 IV TARGET DEPENDENCIES ................................................. 11 1. Objective .............................................................. 11 2. Introduction ........................................................... 11 3. Approach ............................................................. 13 4. Hypotheses ............................................................ 25 5. Results and Discussion .................................................. 25 V ENHANCING DETECTION OF AC OF BINARY TARGETS .................... 31 1. Objective .............................................................. 31 2. Background ............................................................ 31 3. Approach ............................................................. 32 4. Results and Discussion .................................................. 36 VI MA GNETOENCEPHALOGRAPH ........................................... 39 1. Introduction ........................................................... 39 2. Approach ..................:.......................................... 40 3. Results ................................................................ 46 4. Discussion ............................................................. 46 5. Suggested Research ..................................................... 47 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-51 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report VII, ENHANCING DETECTION OF AC WITH BINARY ENCODING ............... 49 1. Objective .............................................................. 49 2. Background ............................................................ 49 3. Approach ............................................................. 49 4. Results ................................................................ 52 5. Discussions and Conclusions .............................................. 54 VIII SUBCONTRACTS .......................................................... 57 1. Edinburgh University .................................................... 57 2. Psychophysical Research Laboratories (PRL) ............................... 57 3. The Lucidity Institute ................................................... 58 IX OTHER ACTIVITY ........................................................ 61 1. Correlations between AC and Geomagnetic Activity .......................... 61 2. Assessment of Theoretical Constructs ...................................... 62 3. Anomalous Perturbation ................................................. 63 4. Fuzzy Set Analysis ...................................................... 63 5. Empirical Tlaining Overview ............................................. 65 6. A Potential New Training Method ......................................... 68 X GLOSSARY ............................................................... 71 REFERENCES ..................................................................... 73 APPENDIX A: Target Elements for the Fuzzy Set Representation of AC Targets .............. 77 APPENDIX B: The Ganzfeld Novice: Four Predictors of Initial ESP Performance ............ 79 APPENDIX C: Impact of the Sender in Ganzfeld Communication: Meta-Analysis and Power Estimates ...................................... 81 APPENDIX D: Effects of the Sender on Anomalous Communication in the Ganzfeld .......... 83 APPENDIX E: A Preliminary Study of Anomalous Perception During Lucid Dreaming ........ 85 APPENDIX F: Possible Effects of Geomagnetic Fluctuations on the Timing of Epileptic Seizures ............................................ 87 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 ii Ap@W9yf4,?M6, jse 2000/08/10: CIA-RDP96-00787R000300230001-5 LIST OF FIGURES 1. City with a Mosque ............................................................. 14 2. Green Intensity Distribution for the City Target (Macrol-pixel 3,3) .................... 15 3. City with Mosque (I AS I = 1.88 bits) .............................................. 15 4. Pacific Islands (I AS I = 1.45 bits) ................................................. 16 5. Zener Target Cards (Average I AS I = 0.15 bits) ..................................... 16 6. Cluster Diagram for Dynamic Targets ............................................. 17 7. Cluster Diagram for Static Targets ................................................ 18 8. Target and Response with a post hoc Rating of 7 .................................... 22 9. Target and Response with a post hoc Rating of 4 .................................... 23 10. Target and Response with a post hoc Rating of 1 .................................... 24 11. Correlation of Post Hoc Score with Static Target AS ................................. 27 12. Correlation of Post Hoc Score with Dynamic Target AS ............................... 28 13. Two-tailed SA Decision Graph ................................................... 33 14. Operating Characteristic Function-1 Tail ......................................... 34 15. Operating Characteristic Function-2 Tail ......................................... 35 16. Sequence of Events for Stimuli Generation ......................................... 41 17. Phase Calculation for a Single Stimulus ............................................ 43 18. A Two-by-Five, Error Correcting Block Code ....................................... 52 19. Correlation: Levels of Visual Complexi with Post Hoc Ratin and Blind Ranking for Post Hoc Scores Greater than Three (i.e., evidence for AC) ........................ 64 20. Correlation: Levels of Visual Complexity with Post Hoc Ratings and Blind Rankings for all Post Hoc Scores .......................................................... 65 21. Experimental Paradigm for Raining .............................................. 70 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5iii Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report LIST OF TABLES 1. Effect Size as a Function of Target Type ........................................... 12 2. Experiment Conditions ......................................................... 19 3. 0-7 Point post hoc Assessment Scale ............................................... 21 4. Effect Sizes ................................................................... 25 5. ANOVA Results ............................................................... 26 6. Receiver 7 .................................................................... 36 7. Receiver 83 ................................................................... 37 8. Receiver 531 .................................................................. 37 9. Hypothesis Testing for Each Receiver ............................................. 46 10. Attributes for Thn Target Packs ................................................... 50 11. Statistics for the Sum-of-Ranks ................................................... 53 12. Statistics for First Place Ranks ................................................... 53 13. Receiver First Place Ranks ...................................................... 54 14. Analyst First Place Ranks ....................................................... 54 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5w Ap fug g FiRsl W Rse 2000/08/10 : CIA-RDP96-00787R000300230001-5 1. OBJECTIVE The objective of this document is to provide a technical final report on tasks 6.2, "Basic Research," 6.3, "Applied Research," and 6.4, as listed in the 1991 Statement of Work. This report covers the time peri- od from 4 February 1991 to 30 June 1992, and includes all subtasks.* Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 1 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report at Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-j ApMERIfp> fig, rfse 2000/08/10 : CIA-RDP96-00787R000300230001-5 II. BACKGROUND With regard to this final report, anomalous mental phenomena (AMP) can be divided into two broad categories:* ? Anomalous Cognition (AC): A form of information transfer in which all known sensorial stimuli are absent. ? Anomalous Perturbation (AP): A form of interaction with matter in which all known physical mecha- nisms are absent. For the purpose of this document, we define research that is primarily directed at understanding the nature of AMP (e.g., signal transmission, neurophysiology, etc.) as basic. Research that is primarily directed at improving the quality of output (e.g., analysis techniques, choice of target material, etc.) as applied. Basic and applied research domains are broad and are highly interactive and mutually support- ive. Understanding the technical details of AC phenomena, for example, will improve its application potential, and likewise, being sensitive to the restrictions of a real-world problem may provide insight into underlying mechanisms. 1. Historical Perspective Serious government research of AMP began in 1973 when a modest effort began at SRI International in Menlo Park, California, to determine if AMP could be verified and to assess the degree to which AMP could be applied in practical situations. In fiscal year 1986, SRI International conducted the first coordinated, long-term examination of AC and AP phenomena. This program had three major objectives: ? Provide incontrovertible evidence for the existence of AC and AP. ? Determine the physiological and physical basis for AC and AP ? Determine the degree to which AC data could be applied in practical situations. The results and conclusions from this program were as follows: ? The first objective was partially met. An information transfer anomaly (i.e., AC) exists that could not be explained by inappropriate protocols, incorrect analyses, or fraud; however, there was insufficient evidence to conclude if AP existed. ? Significant progress was made in meeting the second objective. For example, (1) The central nervous system (i.e., the brain) of individuals with known AC ability appeared to re- spond to isolated AC stimuli. ? A definition of terms may be found in the Glossary in Section X on page 71. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 3 Appproved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report (2) Tvo physical models were constructed. One, called Decision Augmentation Theory, suggests a possible physical transfer mechanism for AC data. The other is a speculative physical model for the type of information that is sensed by AC. Under Ithe same research program, different physical systems were examined for their susceptibility to putative AP effects. They included single-cell algae, single alpha particles, and electronic devices such as random number generators and piezoelectric strain gauges. However, in these carefully controlled experiments, some with experienced AP subjects, no evidence of AP was observed. 2. Current Program Beginning in February 1991, the sponsor initiated a comprehensive, 18-month investigation of AMP at Science Applications International Corporation. The primary thrusts of this effort were to: ? Prepare a comprehensive, integrated, 5-year research plan. ? Conduct basic and applied research of AMP. This finial report provides a comprehensive technical review of this program and suggests possible paths of inquiry for the future. Major sections within this report are as follows: ? Primary experiments carried out under this program. ? Theoretical and analytical problems. ? Results from three subcontractors and their final reports. In the following section, we provide an executive summary of this 18-month effort. The executive sum- mary is designed for the non-technical reader; however, the technical and statistical details can be found in the body of the report, which begins with Section IV on page 11. References to the statement of work are given at the beginning of each section and elsewhere, where appropriate. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-4 Ap m Fig, Rse 2000/08/10: CIA-RDP96-00787R000300230001-5 III. EXECUTIVE SUMMARY During the course of this 18-month contract, we conducted five experiments that were designed to ad- dress specific issues of applied and basic research of AMP Additionally, we conducted a variety of other investigations that did not require further experimentation. As an example of the latter, we applied fuzzy set theory to the data from one of the experiments. In this section, we provide a non-technical summary of the five experiments. Details on all tasks may be found in the body of the report. A well-designed experiment provides valuable information regardless of the particular outcome. In our experimental effort during this contract, three studies produced positive outcomes and two did not. All, however, provided useful guidelines for a follow-on effort. 1. Target Dependencies 1.1 Abstract The purpose of this experiment was to determine if the quality of AC depends upon an intrinsic target property, which is called the change of entropy (i.e., the amount of information contained in visual tar- get material). This was examined for two different target types, photographs and short video clips. A second objective was to determine if the quality of AC depends upon a sender (i.e., a person who is isolated from the receiver but who is focusing upon the target material). The experimental results indicate that the quality of AC does not require a sender to know about, or to focus his or her attention on, the target. Most importantly, we found a strong correlation between the quality of the AC and the change of entropy in a target: That is, the more information determined by information theory contained in the target, the better the AC. Should this result replicate in other ex- periments, it may be the first indication of an independent physical variable that is fundamental to AC. If so, this information can be used to vastly improve many other types of AC experiments. 1.2 Approach Each of five receivers, who had previously demonstrated an AC ability, contributed 40 trials each. All receivers worked alone from their homes and, at a prearranged time, conducted an AC trial for a target that was located no less than 500 km away. The target was either a photograph from the National Geographic magazine or a short clip from a video movie. For half of the trials, the experimenter acted as a sender, and for all trials, the receivers were unaware of the target type or if there was a sender. After receiving the responses by facsimile machine, the experimenter mailed each receiver the target as feed- back. Standard statistical procedures were use to determine whether there were differences in AC quality among these various conditions. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-55 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 1.3 Results Three of the five receivers independently demonstrated significant evidence for AC. AC of photographs was statically significant while AC of videos was not, but the difference between them was not large enough to demonstrate a meaningful preference for the photographs. Using the combined data for all receivers, we found a significant correlation, as measured by a subjective past hoc technique, between the change of target entropy (i.e., its information content) and the quality of the response. We did not observe statistical evidence of AC when the targets were video clips, unlike previous re- search'by Honorton. We speculate that the receivers, who had rarely been exposed to this type of target, were unable to discriminate AC from their internal, non-AC experiences. We expand this point in the body of the report. 1.4 Conclusions A sender is not an intrinsic requirement of AC; however, a sender might contribute to a conducive envi- ronment for AC. In addition, it is unknown whether a sender facilitates the reception of specific target elements. We are currently examining this last point with one of our subcontractors. The total change of target entropy may be fundamental to the functioning of AC. This may shed impor- tant light on the question, "What is being sensed by AC?" A number of replications are required, how- ever, before we can be certain. 2. Enhancing Detection of AC of Binary Targets 2.1 Abstract It is often thought that targets in AC experiments are much too complicated. Frequently they consist of photographs of complex scenes such as a city near a mountain. This complexity makes it difficult to quantitatively analyze the information contained in the target and the response. To eliminate this prob- lem, researchers can use binary targets (e.g., red/black cards, 0/1), which are completely defined and can be analyzed by simple statistics. Earlier experiments have used sophisticated mathematical procedures to enhance the detection of AC of binary targets. The purpose of this experiment was to replicate these earlier experiments. In this experiment, one individual who had demonstrated an ability to use AC successfully when the targets are single binary bits, continued to show his ability. In this case, however, we applied statistical enhancement techniques from information theory to improve the scoring rate. We need to identify a more robust statistical technique to improve the overall efficiency because the receiver was required to "guess" over 21,000 times to reach 100 definite decisions about the binary targets. Of the 100 decisions, he or she was correct 76 times. 2.2 Approach Each of three receivers contributed 100 AC trials in a computer-driven, binary AC experiment. One receiver had demonstrated consistent AC ability in similar experiments, whereas the other two receiv- ers were inexperienced in binary target AC. It is beyond the scope of this summary to describe the math- ematical technique since it may be found in detail in the body of the report. Simply stated, a sophisti- cated procedure called sequential analysis was used to provide many redundant responses to a single Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-56 ApjVfiRM@l F& Wise 2000/08/10 : CIA-RDP96-00787R000300230001-5 binary number target (i.e., one or zero). Sequential analysis is particularly sensitive to whether there is a "burst" of AC and can also determine to within statistical limits if no AC is present. 2.3 Results The experienced receiver again produced significant evidence of AC of binary targets. That receiver's hit rate of 51.6% before the application of sequential analysis was improved to 76% as a result of the analysis. The other two receivers scored at chance expectation. 2.4 Conclusions We confirmed earlier results that it is possible to enhance detection of AC with binary targets using sequential analysis. A major difficulty, however, is that the receivers had to register a guess (i.e. by pressing a computer mouse button) approximately 200 times for each sequential analysis trial. Thus the technique, while capable of enhancing the detection of AC of binary targets, is particularly inefficient due to excessive time expenditures. 3. AC in Lucid Dreams 3.1 Abstract Throughout human experience, people have reported various types of AC in dreams, and laboratory experiments in the 1970s confirmed that AC may occur in dreams. A lucid dream is defined as one in which a dreamer becomes aware that she or he is dreaming. Extensive research has confirmed the exis- tence of lucid dreaming, and that it is possible for the dreamer to signal the waking world about his or her knowledge about the dream. The purpose of this pilot study was to determine if AC could occur during lucid dreaming. We found that AC can occur in lucid dreams. Because the dream-trials did not take place in the laboratory, there was some difficulty in interpreting the results; however, it was clear that lucid dreams do not inhibit AC functioning. Because of the success of this experiment, we will be repeating it in an appropriate sleep laboratory. 3.2 Approach This experiment was designed as a pilot effort Seven receivers, three experienced in lucid dreaming and four experienced as AC receivers, participated in the study. The four AC receivers were first trained in lucid dreaming before the AC trials began. During each trial, a target was selected randomly from the established pool of National Geographic magazine photographs and doubly sealed in two opaque envelopes. The dreamer/receiver placed the envelope next to the bed and was instructed, when a dream became lucid, to "open" the dream envelope (i.e., not the real envelope) while still dreaming, study its content, and report the experience upon waking. The target was provided as feedback once the data had been presented to the experimenter. Our standard rank-order analysis was performed to determine if AC occurred in the study. Since the trials were conducted in each receiver's own bedroom rather than under laboratory conditions, it was difficult to "induce" a lucid dream on demand. Thus, the total number of trials was small (i.e., 21). Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-57 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 3.3 Results Our analysis confirmed that robust AC occurred during the study; however, the number trials was insuf- ficientjto ascertain whether the lucid dream state improved AC functioning. 3.4 Conclusions Because the size of the AC effect seen in this study is commensurate with that seen in other AC experi- ments;! we conclude that the lucid dream state, at least, does not hinder AC functioning. In the body of the report we suggest a refined experiment to increase the number of lucid dream trials in order to de- termir}e if lucid dreaming might enhance AC. 4. M,agnetoencephalograph 4.1 Abstract In this: experiment, we attempted to replicate a study in which we found that slow brain-wave patterns appeared to be affected by an isolated flashing light. We were unable to confirm that result; however, important insights resulted from our effort. We did not attempt to show behavioral evidence of AC while measuring the brain waves. That oversight prevented us from determining if the target system was valid for AC. An analogy might be that it would be a mistake to use only a light stimulus in a study of the brain's response to audio information. In addition, we found post hoc that in general the statistical na- ture of brain waves might have fundamentally prevented us from correctly measuring the instantaneous slow rhythms. We suggest that an appropriate follow-on experiment, which remedies these two over- sights, be conducted, because the statistical evidence for AC in other experiments strongly suggests that the central nervous system must be involved at some level. 4.2 Approach Eight individuals were exposed to approximately 1,000 isolated light flashes while their brain activity was being monitored by a magnetoencephalograph, an instrument that measures the magnetic fields produced by active neurons in the brain. The receivers were chosen to participate in the study based on their successful participation in other AC investigations. We searched for subtle changes in their brain activity by measuring various parameters of their alpha rhythms immediately before and immediately after each light flash. A large alpha rhythm usually indicates that the brain is not particularly attentive to external events, moving the body, or thinking. More traditional central nervous system research has shown that any of these activities can cause a change in the alpha rhythm; therefore, it seems reasonable to expect that if AC is a genuine phenomenon, then it too should induce changes in the alpha rhythm. 'Ib assure that any observed effects were not due to an artifact, an equivalent amount of data (i.e., con- trol data) was collected without any receiver under the magneteoencephalograph. 4.3 Results The earlier results could not be verified in this experiment. The mathematical analysis, which had been used originally, did not reveal any unexpected changes in the subtle properties of the alpha rhythm. The control data also showed no unexpected results. We did, however, notice a difficulty in this analysis. Because brain waves are always present, there is substantial activity that is considered to be noise (i.e., activity that is not directly related to reactions to Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-58 OP r ApaR aq FFnar Release 2000/08/10 : CIA-RDP96-007878000300230001-5 stimuli). Our data contained substantial noise and, unfortunately, our analysis technique was so sensi- tive to it that any brain response to the isolated flashing lights would not have been observed. Fortu- nately, we have saved all the raw data from this experiment, so all that is required is to reanalyze the data with improved techniques. We are currently engaged in that task. 4.4 Conclusion Until this new analysis is complete, we are unable to determine whether the brain responds to isolated stimuli. In the body of the report, we suggest that an improved protocol be implemented as part of the continuing research effort. 5. Enhancing the Detection of AC with Binary Coding 5.1 Abstract The literature reports many attempts at using various statistical approaches to enhance the detection of AC. In this experiment, we used a standard technique from information theory (i.e., error correction through redundancy coding). We were unable to demonstrate that this particular procedure was suc- cessful. As a result of this experiment, we identified a number of improvements that might be applied in new studies. For example, in our study, the statistical technique required special targets, which have not been part of our usual collection. A replication will use a pool of targets that have been successfully used in other experiments. We also learned that our statistical procedure was not sensitive to correct AC responses that happened not to be part of the statistical procedure. We have identified a number of new approaches that correct this problem. 5.2 Approach Five receivers, who had previously demonstrated AC ability, contributed eight trials each. For each trial, all receivers worked alone from their homes and, at a convenient time, conducted an AC trial for a target that was located no less than 500 km away. The targets, which were photographs from the Nation- al Geographic magazine, were chosen in accordance with specific design criteria and were available for one week for each trial. lb use error correcting coding, we identified a series of questions that per- tained to the presence or absence of specified target elements. In this way, a target element, for exam- ple water, could correspond to a single binary bit in the error correcting code. That is, if water were present in the target, the value of one would be assigned to it, otherwise it would be assigned a value of zero. We created ten different sets of five target elements and chose photographs that matched the presence/absence criteria. The presence or absence of particular target elements was dictated by the requirements of the 5-bit binary error correcting code that we used in this study. The principle behind error correcting coding in an AC application is that a receiver could "miss" one of the target elements but still arrive at the correct target. Error correction is a common technique found in the computer industry and in deep space communications. We were adapting its use for AC experiments. After a receiver had completed an AC trial, the response was sent by facsimile to an experimenter in our laboratory in Menlo Park, CA. By return facsimile, the receiver was sent five questions that required yes/no answers for the presence or absence of the target elements. Upon the receipt of the completed questionn- aire, the experimenter sent the photograph back as feedback. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-58 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report Three separate analyses were performed on these data. We used blind rank-ordering of the target and three decoys for each trial to determine if AC occurred during the experiment. In addition, an indepen- dent amalyst and the receiver separately answered the appropriate questions for each trial. The ana- lyst's answers were compared with the receiver's answers after applying the error correcting code. 5.3 Results We were unable to confirm the existence of AC in this experiment using the blind rank-order analysis. While the receivers' answers to the questionnaires tended to be much more accurate than those provided by the independent analyst, no answers were good enough to indicate AC using the error correcting coding. 5.4 Conclusions As was the case in the target-dependencies experiment, we speculate that the receivers, who had rarely been exposed to the type of targets that were used, were unable to discriminate the AC from their inter- nal, non-AC experiences. We expand this point in the body of the report. In addition, we noticed that the answer to a single question depended upon the correct perception of a single target element (i.e., water). This element might not be sensed, but others-not included in the five questions-might be sensed. The method could not take into account these other responses. In the body of the report, we suggest a new experiment that improves the target pool and connects the individual coding bits to classed of elements, rather than single elements. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-,p Ar , &J Far, Jse 2000/08/10 : CIA-RDP96-00787R000300230001-5 IV. TARGET DEPENDENCIES This section comprises the final report for SOW items and I. Objective There are two objectives of this pilot study: (1) Explore the effects of target properties on AC quality. (2) Determine the degree to which AC quality depends upon a sender. 2. Introduction The field of parapsychology has been interested in improving the quality of responses to target material since the 1930's, when J. B. Rhine first began systematic laboratory studies of extra sensory perception. Since that time, much of the field's effort has been oriented toward psychological factors that may influ- ence AC. In this section, we review the pertinent literature that categorizes targets that have been used successfully in AC experiments. At a recent conference, Delanoy reported on a survey of the literature for successful AC experiments.1 She categorized the target material according to perceptual, psychological, and physical characteristics. Except for trends related to dynamic, multi-sensory targets, she was unable to observe systematic cor- relations of AC quality with her target categories. Watt examined the AC-target question from a psychological perspective.2 She concluded that the best AC targets should be those that are psychologically meaningful, have emotional impact, and contain human interest; those targets that have physical features that stand out from their backgrounds or con- tain movement, novelty, and incongruity also should be good targets. The difficulty with both the survey of the experimental literature and the psychologically oriented theoretical approach is that understanding the sources of the variation in AC quality is problematical. Using a vision analogy, energy sources of visual material are easily understood (i.e., photons); yet, the percept of vision is not well understood. Psychological and possibly physiological factors influence what we "see." In AC research, the same difficulty arises. Until we understand what factors influence the AC percept, results of systematic studies of AC are difficult to interpret. Yet, in a few cases, some progress has been realized. In 1990, Honorton et al. conducted a careful meta- analysis of the experimental Ganzfeld literature.3 In Ganzfeld experiments, receivers are placed in a state of mild sensory isolation and asked to describe their mental imagery. After each trial, the analysis is performed by the receiver, who is asked to rank order four pre-defined targets, which include the actual target and three decoys; the chance first-place rank hit rate is 0.25. In 355 trials collected from 241 different receivers, Honorton et W. found a hit rate of 031 (z = 3.89, p G 5 X 10'5) for an effect Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001+ Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report size of 0.141. In addition, he found that AC quality was significantly enhanced when the targets were video clips from popular movies (i.e., dynamic) as opposed to static photographs (i.e., effect sizes of 0.32 and 0.05, respectively). All trials were conducted with a sender. In a carefully conducted meta-analysis, Honorton and Ferrari report significant hitting in forced- choice, precognition experiments.4 They analyzed 53 years of experiments conducted by 62 different investigators using a limited set of symbols (Zener cards) as target material. Fifty thousand receivers contributed a total of approximately 2 x 106 individual trials. The overall effect size was 0.020 corre- sponding to a p -value of 6.3 x 10-25. Similarly, in an earlier review article, Honorton analyzed 7.5 x 105 forced-choice Zener card trials that were collected from 1934 to 1939 and found a significant overall effect 'size of 0.016?0.0015 Puthoff and Targ published the results of 39 AC real-time trials where the targets were natural scenes in the San Francisco Bay area .6 The effect size for the 39 trials was 1.15. Table I summarizes these results for each target type: Table 1. Effect Size as a Function of Target Type Target Type Thals Effect Size* Symbols (Real-Time) 7.5 x 105 0.016 ? 0.001 Symbols (Precognitive) 2.0 x 106 0.020 ? 0.001 Static Photographs 165 0.05 ? 0.08 Dynamic Photographs 190 0.32 ? 0.07 Static Natural Scenes 39 1.15 ? 0.16 "Significance maybe computed as z = Effect Size /Error Shown. The effect sizes shown in Table 1 are qualitatively monotonically related to target "complexity;" an ap- propriate quantitative description for target type is currently unknown. Target "complexity," however, was one of the experimentally observed and theoretically conceived attributes examined by Delanoy and Watt, respectively. A number of confounds exist in this database for the effect-size measures. For example, in all but the Puthoi'f and Thrg study (where targets were natural scenes), the receivers were unselected. That is, they did not participate in the various experiments on the basis of their known ability as receivers. So, is the large effect size for the Puthoff and Thrg study because of the accomplished receivers, the natural-scene targets, or some combination of both? While there are other exceptions, the preponderance of the data was from unselected individuals. In many of the trials, a sender was concentrating on the target materi- al, and as in most perception experiments, `psychological factors and boredom contributed to the vari- ance in the effect sizes. . In this', pilot experiment, we applied one physical measure to static and dynamic photographs to quantify the relationship between target type and AC quality. By careful selection of target content, we minimized the psychologiral factors in perception. In addition, we minimized individual differences by conducting many trials with each receiver and by only choosing receivers who had previously demonstrated excellent AC skill. M Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-52 ARPi-PY&YFFn%rAs se 2000/08/10 : CIA-RDP96-00787R000300230001-5 Because the historical database included trials with and without senders, we explored the effects of a sender on AC quality, as well. 3. Approach 3.1 Target-pool Selection The static target material for this pilot study was a set of 50 National Geographic magazine photographs. This set was divided into 10 sets of five photographs that were determined to be visually dissimilar by a fuzzy set analysis.? The dynamic target material was four sets of five 60 to 90 second clips from popular video movies. These clips were selected because they had the following characteristics: ? Were thematically coherent. ? Contained obvious geometric elements (e.g., wings of aircraft). ? Were emotionally neutral in that they did not contain obvious arousing material. The intent of these selection criteria was to control for cognitive surprise, to provide target elements that are easily sketched, and to control for psychological factors such as perceptual defensiveness. 3.2 Target Preparation The target variable that was considered in this experiment was the total change of Shannon entropy per unit area, per unit time. We chose this quantity because it was qualitatively related to the "information" contained in the target types shown in Table 1, and because it represented a potential physical variable that is important in the detection of traditional sensory stimuli. In the case of image data, the entropy is defined as: Nk -1 Sk = - I pjk logz(pjk ), = 0 if pjk = 0, j-o wherepjk is the probability of finding image intensityj of color k. In a standard, digitized, true color image, each pixel (i.e., picture element) contains eight binary bits of red, green, and blue intensity, re- spectively. That is, Nk is 256 (i.e., 28) for each k, k = r, & b. The total change of the entropy in differential form is given by: dSk = IVSk 1 ? d + Has tit. (1) That is, the total change of Shannon entropy is the change because of spatial variations in the static targets added to the change resulting from frame-to-frame variations in the video targets. We must specify the spatial and temporal resolution before we can compute the total change of entropy for a real image. Henceforth, we drop the color index, k, and assume that all quantities are computed for each color and summed. 3.2.1 Static Targets Tb select the 50 static targets, 100 National Geographic magazine photographs were scanned at 100 dots per inch (dpi) for eight bits of information of red, green, and blue intensity. At one centimeter spatial resolu- tion, this scanning density provides 1,550 pixels for each 1-cm2 macro-pixel to compute the j . Approved For Release 2000/08/10 : CIA-RDP96-00787R0003002300014 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report For a specified resolution, the target photograph was divided into an integral number of 1-cm2 macro- pixels excluding a thin border, if necessary. The entropy for the (ii) macro-pixel was computed as: N-1 Si j = - I pj logZ(pj where pf was computed empirically from the pixels in the (i, j) macro-pixel only. For example, consider the target photograph shown in Figure 1. CPYRGHT Figure 2 shows the probability density for the green intensity for macro-pixel (3,3), which is shown as a white ''square in the upper left-hand corner of Figure 1.` The probability density indicate that most of the intensity in this patch is near zero value (i.e., no intensity of green in this case). In a similar fashion, Sy is calculated for the entire scene. ? The priginal photograph was 8.5 X 11 inches, and we have standardized on one centimeter resolution. . - - Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-14 ApM*WFfrQrFP*pase 2000/08/10 : CIA-RDP96-00787R000300230001-5 Figure 2. Green Intensity Distribution for the City Target (Macro-pixel 3,3) We used a standard algorithm to compute the 2-dimensional spatial gradient of the entropy. Figure 3 shows contours of constant change of entropy (calculated from Equation 1) for the city target. The total change per unit area is 1.88 bits/cm." CPYRGHT 3. MY with Mosque ( 133 1 00=9511M The city target was chosen as an example because it was known (qualitatively) to be a "good" static photograph for AC trials in earlier research. Figure 4 shows contours of constant change of entropy for a photograph that was known not to be a "good" AC target. * In this formalism, entropy is in units of bits and the maximum entropy is 24 bits. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001i Aproved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Approved Final Report CPYRGHT Figure 4. Pacific Islands (I AS 1.45 bits) For comparison, we show in Figure 5 the traditional Zener cards, which were used as targets in most of the forced-choice experiments shown in Table 1. Figure 5. Zener Target Cards (Average. SAS],=;0.15 bits) 3.2.2 Dynamic Targets The total change of entropy for the dynamic targets was calculated in much the same way. The video targets were digitized at approximately one frame per second. The spatial term of Equation 1 was com- puted' exactly as it was for the static targets. The second term was computed from differences between adjacent 1-second frames. Or, Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-516 Aim( FF Ase 2000/08/10 : CIA-RDP96-00787R000300230001-5 ) - S11(t) (2) aat1 M d dt(t) - Islio + AtAt 11 where At is one over the digitizing frame rate (i.e., one second). We can see immediately that the dy- namic targets have a larger 45 than do the static ones because Equation 2 is zero for all static targets. 3.2.3 Cluster Analysis Using Equations 1 and 2, we computed AS for all the static and dynamic targets. These targets were grouped, using standard cluster analysis, into relatively orthoginal clusters of relatively constant AS. Fuzzy set analysis and inspection were used to construct packets of five visually dissimilar targets from within each cluster. Our interim report, which is dated 15 February 1992, details the cluster analysis.8 Figures 6 and 7 show the dusters from that report for the dynamic and static targets, respectively. -- ---- , ~ qqN , - - - - - - - - - - - - - I - - - - - - - - - - - - - Figure 6. Cluster Diagram for Dynamic Targets For ease of reading, Figure 7 shows only those 50 static targets that were used to form the constant entropy clusters, rather than the whole set of 100. We show the computed AS at the end of each duster leaf. 3.3 Target Selection For a specified target type (e.g., static photographs), a target pack was selected randomly and one target of the five within that pack was also chosen randomly. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001* Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report , Nrl 1 1 girl ; , earl , , , arl arl , awl ooto ooro , , --I 1 EIJT Figure 7. Cluster Diagram for Static Targets ft Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-518 nice aTFinelReepolease 2000/08/10 : CIA-RDP96-007878000300230001-5 Aap 3.4 Receiver Selection Each of five experienced receivers, who have produced significant AC effect sizes in previous investigations, contributed 40 AC trials (ie., ten trials under each of the conditions shown in Table 2). 'Iivo of the receivers resided in California while the other three resided in Kansas, New York, and Virginia. Experiment Conditions Condition Target Type Sender 1 Static Yes 2 Static No 3 Dynamic Yes 4 Dynamic No 3.5 Sender Selection The sender for all trials was the principal investigator (PI), who was in Lititz, Pennsylvania. 3.6 Session Protocol 3.6.1 Target Preparation Prior to beginning the experiment, an experiment coordinator randomly generated a unique set of 20 static and 20 dynamic targets for each of the five receivers. After a target was selected, it was immedi- ately returned to the pool of possible targets and so could be used again. Within each target type, a counter balanced set of sender/no sender conditions was also generated. A copy of each target was placed in an envelope and a trial number, 1 through 40, was written on the outside. Those envelopes containing targets from the no-sender condition were sealed while those for the sender condition re- mained unsealed. Each set of 40 targets was packaged separately and shipped to the PI in Pennsylvania. 3.6.2 Trial Schedule The experiment was conducted over a five month period. Individual schedules were developed with each receiver so as to cause as little inconvenience to their daily routine as possible. 3.6.3 Session Sequence For each trial and for each receiver, the PI proceeded as follows: ? Selected the appropriately numbered envelope from the box for the appropriate receiver. ? In the sender condition, looked at the selected target for 15 minutes and attempted to "transmit" it to the intended receiver during that time period. ? In the no-sender condition for the static targets, placed the unopened envelope on an uncluttered desk in the PI's office for the 15 minute trial period. In the no-sender condition for the dynamic tar- gets, played the video repeatedly for 15 minutes with the sound turned off and the TV monitor in another room. ? At the conclusion of the 15 minute trial period and after the receipt of the receiver's response by fac- simile, sent a copy of the target material (i.e., either a photograph or video tape) to the receiver by mail. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001.4 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report During each trial, the receiver took these actions: ? At the prearranged time, withdrew to a quiet lighted room in his or her home and sat at a desk. ? Fora period lasting up to 30 minutes, wrote and drew his or her impressions of the intended target material, which was located in Lititz, PA. ? At the end of the AC trial, sent a copy of the response to the PI by facsimile machine. ? By mail, obtained a copy of the target as feedback for the trial. The target copy and original response were subsequently sent to the experiment coordinator in Menlo Park, CA. We did not provide specific instructions beyond logistical information to the receivers, because the re- ceivers were all experienced at this type of task. When; the experiment coordinator received the receiver's response, all identifying information (i.e. name;, date, and time of trial) was removed from the response. Periodically during the course of the experiment the experiment coordinator provided an analyst, who was blind to the target choice, with a set of responses and associated target packs for analysis. Each target pack consisted of the real target and four decoy targets of the same target type and similar OS. 3.7 Analysis 3.7.1 Rank-Order For each trial, there was a single response and its associated target pack (i.e., either static or dynamic). During the first part of the analysis, a judge, who was blind to the condition and target for the trial, was asked to rank-order the targets within the given pack. This was a forced rank, so regardless of the quali- ty of match between the response and targets within the pack, the judge had to assign a first place match to a response, a second place match to a response, and so on for each of the five targets. The output from this part of the analysis is a rank-order number (i.e., one to five, one corresponding to a first place match) for the correct target. As was described above, the targets within each pack were chosen to be visually different from one another, but they all possessed similar AS. Thus, the rank number was not biased because of entropy considerations. For each receiver, target type, and condition there are 10 such rank-order numbers that constitute a block of data. A rank-order effect size was computed for a block as: Eij = (3) where hj is the average rank for target type i and sender condition j, and io is the expected average rank,'which for this study is equal to three for all cases. In Equation 3, Nis the number of possible ranks and is equal to five throughout this study. Reversing the sign in the numerator, Equation 3 reduces to: 3.7.21 Analysis of Variance A two-way analysis of variance (ANOVA) was computed for each receiver. The two primary variables weretarget type and sender condition (i.e., ANOVA main effects). Each of these variables possessed two states as shown in Table 2. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-520 Approved For Release 2000/08/10: CIA-RDl%6 7 6Da ]kB6G6giWe Possible Effect of Geomagnetic Fluctuations on the Timing of Epileptic Seizures JAMES SPOTTISWOODE ERIK TAUBOLL MICHAEL DUCHOWNY VERNON NEPPE ADDRESSES: Science Applications International Corporation, Cognitive Science Laboratory, 1010 El Camino Real Suite 330, Menlo Park, California 94025, USA (S.J.P.Spottiswoode, BSc); Department of Neurology, Rik- shospitalet 0027, Oslo 1, Norway (E. Tauboll, MD); Department of Neurology, Miami Children's Hospital, Salo- man Klein Pavilion, Miami, Florida 33155 USA (M.S.Duchowny, MD); Division of Neuropsychiatry, Universi- ty of Washington, Seattle, Washington 98195 USA (V.M.Neppe, MD). Correspondence to S.J.P.Spottiswoode. Running head: Geomagnetism and epileptic seizure. 1 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Approved For Release 2000/08/10 CIA-RDP9&-40718 ?R4?QQM i 6%&zure Abstract Some reports have suggested that epileptic seizures might occur more frequently at times of enhanced disturbance of the geomagnetic field. This study examines this putative association using 4161 seizures from 22 epileptic patients where the seizure times were known to within a day or be'tter. A measure of the geomagnetic fluctuation level for the seizure day, and the days preceding the seizures, was derived from the geomagnetic index ap. This daily index was significantly higher on the seizure days than on the day prior to the seizures (p = 0.007) and slightly higher than for the preceding 10 days (p = 0.1). The effect size for the increase for the increase of geomagnetic activity on seizure days from the previous days was inhomo- geneous across this group of patients (p = 0.04), suggesting an uncontrolled factor. However, a regression of age, sex, seizure type and frequency onto effect size failed to reveal any signif- icant loadings. Key words: humans; geomagnetism; epilepsy; seizure; magnetic field MR~ a Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 2 Approved For Release 2000/08/10: CIA-RDPG9e?P MA 9Pf 3rAre Introduction The reasons for the precise timing of epileptic seizures in most patients remain largely unknown. Statistical studies of seizure timing have failed to identify clearly non-random patterns such as clustering or periodicity in many patients.1,2 Several explanations for this have been suggested, including the postulation of an inherently random endogenous mechanism3 and the possibility that seizure occurrence might be more or less tightly coupled to an exogenous variable which itself had nearly random statistics. In considering the second of these hypotheses several workers have looked for a suitable environmental stimulus in the very low frequency region of the electromagnetic (EM) spectrum. EM waves with frequencies of 104 Hz or less have several natural sources, including lightning discharges and ionospheric phenomena, and exhibit a complex distribution in time.4 These long wavelength EM emissions are detectable everywhere on the globe and penetrate buildings and conducting structures with little attenuation. Additionally there is some evidence that such low frequency EM fields can interact with the functioning of biological systems, though the question is far from settled.5'6 A connection between the triggering of epileptic seizure and low frequency EM radiation therefore has a certain prima facie plausibility. Some reports have suggested that epileptic seizure frequency may be correlated with disturbances of the geomagnetic field (GMF).7'5'9 Fluctuations in the GMF are primarily driven by changes in the sun's activity and major solar storms give rise to magnetic field changes of up to 1000 nT at the earth's surface and cover a range of frequencies from approximately 20 1Hz to 10 Hz .4 The literature on the effects of magnetic field exposure upon epileptic seizure, while not extensive, contains some suggestive avenues of research. Venkatraman5 originally suggested that there might be an association between magnetic storms and epileptic attacks but did not provide any statistics to support this conclusion. Rajaram & Mitra6 reported that monthly averages of admissions of epileptic cases rose during periods of increased GMF variation. However, no attempt was made to control for other factors which influence hospital admissions. According to Keshavan et all a decrease in convulsive threshold in rats was observed during the GMF variation associated with a solar eclipse. Persinger'? has suggested that increases in the GMF noise level suppress nocturnal melatonin levels, precipitating seizures and consequent cardiovascular instability. Significant correlations have also been reported between epileptic seizure onset and 10 kHz and 28 kHz atmospherics.11 However a laboratory study of audiogenic seizure susceptible rats failed to find an association between EM at these frequencies and seizure timing.12 There is also Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-53 Approved For Release 2000/08/10 : CIA-RDP95eOfi7 QDQ8a i0'8zure evidence, that exposure to relatively intense (105 nT) 60 Hz magnetic fields may actually inhibit electrically kindled seizures in rats.13 Numerous biological effects from exposure to weak VLF and ULF magnetic fields have been reported, as is demonstrated in reviews of this literature such as those by Adey5 and Marino and Becker.6 Interest in the area has recently been stimulated by concern with the possible carcinogenicity of the 50 and 60 Hz magnetic fields associated with power generation and distribution. However, the physical mechanisms which might account for biological sensitivity to weak, low frequency, magnetic fields such as the GMF remain obscure. Adair14 has calculated the electric fields and other effects in cells and cell membranes consequent upon 60 Hz magnetic fields of larger amplitude (and frequency) than GMF fields. He finds that the induced electric fields are considerably smaller than the fields due to thermal noise. However his arguments do not entirely rule out interactions involving larger multicellular receptors. It is also possible that the putative association between enhanced GMF disturbance and epileptic seizure may not be caused by the magnetic field itself, but rather by some other environmental parameter15 which co-varies with the GMF changes. The physics of possible mechanisms for electromagnetic triggering of epileptic seizure is not well enough understood to suggest what frequency or amplitude of EM radiation ',might be responsible for such an effect. While the epidemiological literature suggests that a weak connection between seizure timing and enhanced GMF activity may exist, the; evidence is not statistically assessable. This study examines one of the hypotheses raised by the earlier literature, specifically whether epileptic seizure timing in humans is associated with increased GMF fluctuations at the time of the seizures. Seizure diaries from 22 epileptic patients, containing timings of 4101 seizures, were analyzed to see whether these events occurred at times of enhanced GMF activity. By using seizure diaries, rather than hospital admission records, many potentially confounding factors can be avoided. In the light of the earlier studies it was hypothesized that the days on which epileptic seizures occurred would show higher levels of GMF activity than that of the preceding days. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 4 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 3.7.3 Post-Hoc Assessment Rank-order analysis does not usually indicate the absolute quality of the AC. For example, a response which is a near-perfect description of the target receives a rank of one. Yet a response which barely matches the target, may also receive a rank of one. Table 3 shows the rating scale that we used to per- form a post hoc assessment of the quality of the AC responses regardless of their rank. The quality of an AC response is defined as its visual correspondence with the intended target. Score Description 7 Excellent correspondence, including good analytical detail, with essentially no incorrect information. 6 Good correspondence with good analytical information and relatively little incorrect information. 5 Good correspondence with unambiguous unique matchable elements, but some incorrect information. 4 Good correspondence with several matchable elements intermixed with incorrect information. 3 Mixture of correct and incorrect elements, but enough of the former to indicate receiver has made contact with the target. 2 Some correct elements, but not sufficient to suggest results beyond chance expectation. 1 Very little correspondence. 0 No correspondence. lb apply this subjective scale to a target-response trial, an analyst begins with a score of seven and deter- mines if the description for that score is correct. If not, then the analyst tries a score of six and so on. In this way the scale is traversed from seven toward zero until the score-description is correct for the trial. Figures 8 through 10 illustrate the application of this scale and show that the quality of an AC response is not necessarily indicated by its first-place rank. All three examples were given a rank of one in a blind analysis. These examples were chosen from the experiment which is being described in this section (i.e., Section IV). The response to the waterfall target in Figure 8 included a number of pages of material about a city and other man-made activity. In all of our analyses, we strictly adhere to the concept that any material a receiver deletes from the response prior to feedback is not counted in the analysis. Thus, the response in Figure 8 is considered as complete. The other examples are shown in their entirety. The scale shown in Table 3 can be divided into two sections, 0-3 and 4-7. The upper portion of the scale indicates clear contact, presumably by AC means, with the intended target material, while the remain- der of the scale indicates little or no contact. We used this scale to provide assessment scores to examine the correlation with the target entropy. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5i CPYRGHT arr Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 1) City, buildings seems to be a big leap from what I am feeling about the target. I'll restart 2) Troubled by city feeling. Could be that the uprights are natural rather than man- made. In which case the city interpretation is incorrect and I am feeling MESA. I'll check verticals. 3) DELETE Lights, structure, structures, building, and city. We gots a waterfall, dude. Figure 8. Target and Response with a Post Hoc Rating of 7 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-b A eCroicai FoLRelease 2000/08/10 : CIA-RDP96-00787R000300230001-5 CPYRGHT long white rectangular box like an upside-down sheet cake same box two circular shapes in front, like stepping stones in a garden Iona hollow tube, like crashing surf on a beach - "Hawaii Pipeline" Figure 9. Target and Response with a Post Hoc Rating of 4 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-523 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 cynical Final Report CPYRGHT BEGIN-10:30 AM puffy balls - almost cotton-like. Cottony puffy, splotches. Movement - whizzing through these cottony puffs fast. Damp- ness. A long walkway & metal girders. I keep wanting to say - specifically - air- field landing strip. Flat land. Big airplanes would land here like naval carriers. Has a broken white line down the center of strip & you see it straight on - like you would be coming in for a landing. Figure 10. Target and Response with a Post Hoc Rating of 1 Approved For Release 2000/08/10 : CIA-RDP96-00787R0003002300012 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 4. Hypotheses 4.1 Null Hypothesis The overall null hypothesis was that i = 0. 4.2 Sender and Target Condition Using an F-test we tested the hypothesis that the quality of AC does not depend upon a sender regard- less of target type. Similarly, we used an F-test to test the hypothesis that the quality of AC does not depend upon target type regardless of the sender condition. The ANOVA also tests for potential interactions between the target and sender conditions. For exam- ple, it might be that a sender is required for dynamic targets and not for static ones. 4.3 Target Entropy The AC quality (i.e., scores greater than three from the post hoc scale in Table 3) of each trial was corre- lated with targetd S. A significant correlation would indicate that target entropy and AC quality may be linearly related. 5. Results and Discussion 5.1 Effect Size Analysis Five receivers completed 40 trials each. Table 4 shows the effect sizes (i.e., z /jn) computed for the 10 trials in each cell. The shaded cells indicate 1-tailed significant results. Receiver 009 showed significant evidence for AC in the static target, no-sender condition (p < 0.02); receiver 372 in the static target, sender condition (p< 0.01); and receiver 518 in the static target, no-sender condition (p < 0.05). See the underscored values in Table 4. Receiver Sender Static No Sender Dynamic No Sender Static Sender Dynamic 009 -0.071 0.141 LL -0.141 131 -0.071 0.495 -0.071 0.212 372 0.707 -0.283 0.141 -0.354 389 0.141 0.000 0.212 0.000 518 -0.088 0.283 0.530 -0.495 5.2 Analysis of Variance Table 5 shows the results of an ANOVA on these data. Since there were 10 trials within each cell, the degrees of freedom are the same for all receivers and, therefore, are only shown in the column headings. Two receivers show significant main effects. Receiver 372 showed a tendency to favor static over dy- namic targets (i.e., p !!:t:: 0.03), and receiver 518 showed a tendency to favor no sender (i.e., p < 0.04). Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-525 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 ec1nlcal Final Report See the underscored values in Table 5. That is, for these receivers the ANOVA hypothesis that the data are d rawn from the same distribution is rejected. There were no significant interactions between target type and sender condition. ANOVA Results Receiver Sender Condition Target Type Interaction F(1,36) P -Value F(1,36) P -Value F(1,36) P -Value 009 038 0.5 4 0.68 0.42 2.08 0.16 131 0.18 0.67 1.66 0.21 0.18 0.67 372 1.01 0.32 5.47 Q,Q 0.61 0.44 389 0.01 0.91 0.33 0.57 0.01 0.91 518 4.43 Q.114 0.97 033 0.06 0.81 When we combined the data for static targets regardless of the sender condition (i.e., 100 trials), the sum-o -ranks was 265 (i.e., exact sum-of-rank probability ofp < 0.0073, effect size = 0.248). The total sum-o ranks for the dynamic targets was 300 0.5 effect = (i.e., p < - 00; size - 0.000). 5.3 Post Hoc Assessment Tivo analysts independently rated all 100 trials (i.e., 20 each from five receivers) from the static-target sessions using the post hoc rating scale shown in Table 3. All differences of assignments were resolved in discussion, thus the resulting scores represented a reasonable consensus of the visual quality of the AC for each trial. We has specified in advance that for the correlation with the change of target entropy, we would only use the section of the post hoc rating scale that represented definitive, albeit subjective, AC contact with the target (i.e., scores four through seven). Figure 11 shows a scatter diagram for the past hoc rating and the associated AS for the 28 trials with static targets that met this requirement. Shown also is a linear least- squares 11 fit to the data and the linear correlation coefficient correlation (i.e., r = 0.461, df = 26). This strong correlation suggests that AS is an intrinsic property of a static target and that the quality of an AC response will be enhanced for targets with large AS. This correlation, however, might be a result of AS and the past hoc rating independently correlating with the targets' visual complexity. For exam- ple, an analyst is able to find more matching elements (i.e., a higherpost hoc rating) in a visually complex target than in a visually simple one. Similarly, AS may be larger for more complex targets. If these hy- potheses were true, the correlation shown in Figure 11 would not necessarily support the hypothesis that AS is an important intrinsic target property for successful AC. Tb check the validity of the correlation, we used a definition of visual complexity that was derived from a fuzzy set' representation of the target pool? We had previously coded by consensus 131 different poten- tial target elements for their visual impact on each of the targets in the pool. It is beyond the scope of this report to provide the details of this technique since the details may be found in reference 7. It suf- Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001z--5 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report fices to say, however, that the sigma-count (i.e., the sum of the membership values over all 131 visual elements) for each target is proportional to its visual complexity. A list of these target elements may be found in Appendix A. Figure 11. Correlation of Post Hoc Score with Static Target AS We computed the liner correlation coefficient for target complexity with the assigned post hoc rating. For all 100 s tic targets used in this study we found r = 0.049, df = 98, and for target complexity with the measured AS, we found r = -0.031, df = 98.* On closer inspection neither of these small correlations is surprising. While it is true that an analyst will find more matchable elements in a complex target, so also are there many elements that do not match. Since the rating scale (i.e., Table 3) is sensitive to correct and incorrect elements, the analyst is not biased by visual complexity. The change of Shannon entropy is derived from the intensities of the three primary colors (i.e., Equa- tion 1 on page 13) and is unrelated to large-scale objects or meaning, which is inherent in the definition of visual complexity. Thus, we would not expect a correlation between AS and visual complexity. Visual complexity, therefore, cannot account for the correlation shown in Figure 11; thus, we are able to conclude that the quality of an AC response depends upon the spatial information (i.e., change of Shan- non entropy) in a static target. A single analyst scored the 100 responses from the dynamic tar ets using the post hoc scale in Table 3. Figure 12 shows the scatter diagram for the post hoc scores and the associated AS for the 24 trials with a score greater than three for the dynamic targets. We found a linear correlation of r = 0.043) df = 22. ? Usingjustthe28datapointsinFgurell,wefmdr--0.216,df-26andr-0.003,df=26fortheoorrelationwiththeposthoc score and AS, respectively. Since these correlations are negative or very small, they do not alter the conclusion. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-527 A l m epo Ftn,o elle~ase 2000/08/10 : CIA-RDP96-00787R000300230001-5 0 0 0 o 0 0 8 8 0 2 ~ ir r = 0.043 df =22 5 6 7 $ Post Hoc Score Figure 12. Correlation of Post Hoc Score with Dynamic Target AS. This small correlation is not consistent with the result derived from the static targets; therefore, we will examine this case carefully. The total sum of ranks for the dynamic-target case was exactly mean chance expectation, which may indicate that little AC was observed (see Section 5.2, above). Tiwenty-four trials, however, received a post hoc score of four or more. We see in Figure 12 that only two trials re- ceived a score of seven, and those trials were among the lowest AS. It may be that the small correlation is strongly influenced by these two data points, and a more accurate determination of a putative effect with dynamic targets requires more trials in a future experiment. To determine if there is a trend with- out the' two data points with a score of seven, we computed the correlation for the remainder of the data (i.e., r 0.293, df = 20). Given the past hoc nature of this calculation, all we can conclude is that we should'conduct a similar experiment with dynamic targets to determine if a fundamental correlation between AS and AC quality exists, as it does with the static targets. There is a potentially more important reason that robust AC was not observed in the dynamic target set. Most ptevious research has considered AC from a "systems" perspective in that the target and receiver are thought of as a single AC unit.12 This is not particularly productive if we are searching for intrinsic properties of target systems to guide target selection. An intrinsic target property is one that is in- herently tied to the target (e.g., size, distance from the receiver, activity, entropy) and devoid of any externerpretation. Interpretations, such as emotional impact, can be considered as extrinsic prop- erties of thetaz' et or, moire precisely, intrm cis properties of the receiver. Extrinsic target properties are critical when AC is viewed from a systems point of view; however, if these properties can be con- trolled in experiments, then it is possible to examine intrinsic target properties with little confounding interference from the extrinsic ones. As an aid in understanding extrinsic noise properties of targets, we define target pool bandwidth as a qualitative indicator of the number of disparate target elements in the pool. The dynamic targets, which were clips from video movies, represent a large-bandwidth pool; such disparate scenarios as Superman Approved For Release 2000/08/10 : CIA-RDP96-00787 R00030023000 lip Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report in space, a nature segment on the Grand Canyon, and a James Bond thriller can be included in the same target pool. Conversely, the well-known Zener cards represent a vary narrow target bandwidth. The static targets, which are constructed from a collection of National Geographic magazine photographs, represent an intermediate bandwidth; the size and general content of the material is roughly the same throw out the pool. We hypothesize that the bandwidth of the tar ge t pool is a source of intrinsic noise in the receiver. We assume that the information that is gained by AC is small compared to other sensory mechanisms, and the primary mental task for a receiver is to discriminate the AC data from internally generated, target- unrelated information. For large bandwidth target pools that may contain almost anything, a receiver is unable to censor his/her internal experience. Thus, target-related and target-unrelated material are equally reported; therefore, large bandwidth pools are extrinsically noisy. Small bandwidth pools are also extrinsically noisy but for a different reason. If a receiver is cognizant of all of a limited set of target elements (e.g., Zener cards), then he/she has an internal discrimination problem. All target possibili- ties are experienced with equal intensity because of knowledge about the pool and vivid short-term memory. Assuming there is weak AC information about the specific target, then target-extrinsic noise is generated because of the very low signal-to-noise ratio. Most of our receivers have participated in many earlier experiments which used the static target pool, and were unfamiliar with target pools with large bandwidths such as the dynamic pool. Historically, we have observed AC effect sizes for static targets 50% to 100% larger than we found in this experiment. The current protocol did not include monitoring the AC trials, and the receivers were blind to the target tVe. It is impossible to determine from this experiment which factor was predominant, but if the band- width argument is correct, we would expect a decrease in functioning for even the static targets because receivers would not be able to self-censor their responses.* We recommend that a new target pool be developed that limits the bandwidth of the dynamic targets and that the static targets be specific frames from within the dynamic target pool. In this way, we can control for target bandwidth effects between the target types. We recommend that a new experiment be conducted with these new target pools. 5.4 Overall Conclusions Based upon the results of this pilot experiment, we provide the following tentative conclusions: ? The ANOVA results suggest that a sender is not fundaments required for AC. ? Subject to the caveat suggested in the previous section, the ANOVA results suggest that AC quality does not depend upon target type. ? AC quality for static targets is proportional to a target's spatial information (i.e., AS). Because of the importance of determining if AS is an intrinsic target property for all AC targets, we urge that this study be repeated with the improvements discussed above. * It is important torecognize that limited, or evencomplete, knowledge of the targetpool cannotbias theblind rank-order statis- ticbecause itisa differential measurewithin the pool. It may, however, change the mean of theposthoc scores, butcorrelations are insensitive to means. Thus, correlations based upon the post hoc assessment remain valid. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 29 Ap~pproved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 'Thchnical Final Report Approved For Release 2000/08/10 : CIA-RDP96-00787R00030023000' 5 nicarFinarReport se 2000/08/10 : CIA-RDP96-007878000300230001-5 ABcT V. ENHANCING DETECTION OF AC OF BINARY TARGETS This section constitutes the final report for SOW item 6.23.3. 1. Objective The objective of this investigation was to replicate and extend an earlier study that enhanced the detec- tion of AC of binary targets. 2. Background In 1984, Puthoff used a majority vote procedure to statistically enhance the detection AC of binary tar- gets.9 The chance probability of guessing a binary target correctly is 0.50. In Puthoff's experiment, his best receiver, using AC methods, increased the probability to 60%. Using a majorityvote of five guesses per bit, the probability of guessing the target correctly was increased by 18.3% from 60 to 71 percent. In fact, if the probability of guessing a binary target is given by P =Po+6, (4) where 6 is a non-negative constant much, much less than unity and PO = 0.5, then it can be shown that a majority vote procedure is the most efficient method for obtaining an arbitrarily accurate guess. Let n be the number of bits in a majorityvote procedure (i.e., n is assumed to be odd). Then the majority vote proba- bility is given by a binomial sum as: n "-1 P(n) = p" + (n n l) p"-' (1 - p) + ... + n 2 1 pi' (1 - p) Z , where p is the single bit probability given by Equation 4. By choosing n large, P(n) can approach unity. The problem is that a majorityvote procedure is predicated on the assumption that a is not a function of time, an assumption that is known not to be true in AC experiments. Ryzl attempted to solve this problem by modifying a majority vote scheme to include on-line checks.10 He was able to demonstrate a 100% accu- rate guess of 15 dedmal digits encoded as 50 binary digits (p = 10-15). In 1985, Puthoff, May, and Thomson used a well-known technique called sequential analysis (SA) and, for one receiver, realized a 3.7% enhancement 53.6 to 55.6 percent in a binary AC experiment.11 Dif- fering from the usual statistical analysis, SA does not require that the sample size be specified in ad- vance; however, by adjusting certain SA parameters, it is possible to set the expected number of trials in the processes. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-531 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Techftical Final Report In this',pilot experiment, we set the expected number of trials in an SA algorithm to match the temporal variation of 6(t). Thus, we expected to realize a significant enhancement of binary AC. 3. Approach 3.1 Theoretical Let p be the probability that a binary random variable has the value one. With SA, we test the hypothesis Hp: p = pr, against the hypothesis HI: p = pj. After accepting Hp as true, define a to tie the probability that Hp was false and that HI was true (i.e., Type I error). Likewise, after accepting HI as true, define ig to be the probability that HI was false and that Hp was true (i.e., Type II error). Let n be the current sample number. With parameterspp, P1, a, and/, SA defines two lines as follows: y1 = a n + bl and yo =an - bo, where ln(-?O) 1-P1 ln(~) bo = A In a and b 1 = G , where (5) d = ln(~ - + In/' - Po P1 Let Nbe the accumulated number of ones in a Bernoulli sampling situation. Then the general SA deci- sion algorithm is as follows: ? Collect one binary sample and add its value (i.e., zero or one) to the accumulated number of ones, N. ? Compare the accumulated value to y1 and yo in Equation 5. ? If N y1 (n), then stop the sampling and conclude that hypothesis HI is true wiith a risk of being wrong of/i. ? IfNT yo (n), then stop the sampling and conclude that hypothesisH0 is true with a risk of beingwrong of a. ? If yp(n) < N < yj (n), then continuing sampling. In the general theory of SA, this decision process always converges, and the expected number of samples for a decision in favor of each hypothesis is given by. EHO (n) - (1 - a) ln(1) + a In(1j) and Po In(o) - (1 - Po) ln(r.) (6) ln(1p"-a) + (1 - )In('a?) EH, (n) _ I-PO P1 in( ) - (1 - 01-71 For an arbitrary value of p, we compute the probability that the SA algorithm will decide in favor of hypothesis H1 (i.e., the operating characteristic function-OC) as: Approved For Release 2000/08/10 : CIA-RDP96-00787R0003002300015 Approved for Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 ec nica nal Report k OC(p) = 1 - (41)h 1-() 1-p0 -1 where p is given by () p(h) _ , where -- s h 5 +00. (pp)~ - (ti-Pol 1-pl) h (7) 3.2 A Two-tailed Example of Sequential Analysis In this section we modify the formalism of SA to include a measure of the difference between the accu- mulated number of ones and the expected number of ones. This will allow a two-tailed application of SA. The only modification that is necessary to Equation 5 is that the slope, a, is now given by. ln(1-p1) (8) In this example we assume that a = P, so that the curves (see Figure 13) that define the decision algo- rithm are symmetric. Let 8 be the accumulated excess number of ones (i.e., the number of ones minus the expected number of ones). In the two-tailed case, the two hypotheses that are tested by SA become Ho:p=po, andH1:p=P1 orp =1 -P1. Figure 13. Tivo-tailed SA Decision Graph When 8 enters either Region 1 or 2, stop the sampling and assume HI is true with a Type II error of Likewise, if 8 enters Region 3, stop the sampling and assume Ho is true with a Type I error of a. 3.3 Hypotheses The two hypotheses that were tested in this experiment are: (1) Ho:p=po=0.5,and Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 33 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report (2) H1:P=P1 =0.6or,p=1 -pl =0.4. We chose Pp = 0.5 because that is the expected hit rate for binary guessing, and we chose p! = 0.6 because that was the apprmtimate individual hit rate reported in Ryzl's and in Puthoff's i periment. We adjusted the values of a and A such that the expected number of samples per SA decision was approximately 40, a value that is consistent with anecdotal reports of how many trials can be collectu1 during a single session before A a receiver becomes bored with the task. Figure 14 shows the OC curve for these parameters. 0.4 a 0 a p0 = 0.50 pl = 0.60 a = 0.20 F-1 0 =0.20 Majority Vote of 5 --. 0.0 0.2 0.4 0.6 0.8 1.0 Event Probability -+- Sequential Analysis Figure 14. Operating Characteristic Function -1 Tail For co'1inparison, we also show in Figure 14 the majority-vote-of-five curve that was successfully employed in Puthoff's experiment. Using the SA method, we expect an enhancement of approximately 42% at, 0.6 hitting rate compared to the 18% seen by Puthoff. In addition, our two-tailed formalism allows receiver to use AC to detect either a binary one or a binary zero. Another advantage of SA over majority vote can be see from Figure 14. Forp less than 0.5, the Type I error is sharply reduced. Thus, the false-positive decisions are reduced accordingly. Figure 15 shows the OC for the 2-tailed SA scheme displayed in Figure 13. For this calculation, we as- sume that the accumulated deviations result in an extreme decision (i.e., lines =L yl). That is, under the null hypothesis, 80% of the decisions will be on the inner decision lines in Figure 13. Of the 20% re- maining decisions, 50% will strike the upper decision line. This curve, therefore, demonstrates that under the null hypothesis there will be no enhancement. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001 A Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 3.4 Protocol SA Parameters p0 =0.40 p1 =0.60 a =0.20 I; =0.20 0.0 0.2 0.4 0.6 0.8 1.0 Event Probability Figure 15. Operating Characteristic Function - 2 -Tail 3.4.1 Receiver Selection Three receivers participated in this study. One (receiver 531) was selected because that individual had pro- duced statistically significant results in earlier similar experiments.12,13 TWo receivers (7 and 83) were se- lected because of their interest and because of successes in free-response AC experiments. 3.4.2 Target Selection A Sun Microsystem's SPARC workstation used a feedback shift register algorithm to generate a single binary target for each SA decision trial.14 3.4.3 Trial Definition A trial was defined as an assertive SA decision. That is, eitherp = pi or p =1- pi. Decisions resulting in p = pj were tabulated, but otherwise ignored. Each receiver contributed 100 trials. 3.4.4 Sample Definition An experimental control program oscillated a single binary bit between one and zero as rapidly as pos- sible. When a mouse button was pressed, the state of that oscillating bit represented the value of the single sample. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-535 A ecltlnical Fin or Re lease port : CIA-RDP96-007878000300230001-5 3.4.5 Trial Protocol Each trial proceeded as follows: (1) The receiver started the experimental control program, which asked for the receiver's name. (2) The program determined if the current trial was a continuation of an earlier trial, or the beginning of a new one. (3) If the trial was being continued from an earlier session, the program read the previously saved data and indicated that the receiver may begin. (4) $f the trial was a new one, the computer randomly displayed a new target and indicated that the receiver may begin. (5) The receiver pressed a mouse button to indicate when the oscillating brit matched the predeter- mined trial target bit. We emphasized that the task for the receiver is a precognition one; press the r1ouse button at a time when the oscillating bit matches the displayed target bit. (6) The value of this sample bit was used as input to the SA algorithm. (7) The receiver ended the session at any time (i.e., either within a trial or at the end of one). 3.4.6 Analysis The analysis was defined by the SA algorithm. The control program recorded the number of matches between SA decisions (i.e., either one or zero) and the trial target bit. It also recorded the total number of button presses and the number ofpn decisions that occurred during the 100 SA trials. The binomial distribution was used to calculate a p-value for each receiver, but the normal approxima- tion to the binomial distribution was used to compute the effect sizes. 4. Rgsults and Discussion Tables 6 through 8 show the results for receivers 7, 83, and 531, respectively. The z-scores for the binomial methods of analysis are shown for comparison only, since SA does not specify the number of samples, the results tend to be inflated from their correct value. The binomial (decision) method included only those samples that led to a definite SA decision, whereas the binomial (all) method included all samples. Analysis Method Hits Thals Rate Z-Score e Sequential Analysis 49 101 0.485 -0.299 -0.030 B L inomial (decision) 2,256 4,569 0.494 -0.&43 -0.125 Binomial (all) s 7,856 15,747 0.499 -0.2.-9 -0.002 Receiver 7 inadvertently produced one extra trial; however, it did not affect the: overall score of 'mean chance expectation (i.e., rate = 0.50). As shown in Section V.3.3, SA did not inflate the chance results beyond what was expected.. Mq Approved For Release 2000/08/10 : CIA-RDP96-00787R00030023000'I -5 ApDrov~ed Eor Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 ec n cal lnal sport Analysis Method Hits Trials Rate Z-Score 8 Sequential Analysis 44 100 0.440 -1.20 -0.120 Binomial (decision) 1,916 3,966 0.483 -2.13 -0.034 Binomial (all) 9,422 18,937 0.498 -0.68 -0.005 Receiver 83 produced an overall score of mean chance expectation. Analysis Method Hits Trials Rate Z-Score e Sequential Analysis 76 100 0.760 5.20 0.520 Binomial (decision) 2,842 5,059 0.562 8.79 0.124 Binomial (all) 11,008 21,337 0.516 4.65 0.032 Receiver 531 produced an overall significant score (i.e. Z = 5.2 p G 1 X 10-? e = 052). This receiver is experienced at computer tasks and the result is consistent with his historical performance. A raw hit rate of 0.516 is what is usually seen,12 and the effect size of 0.032 is consistent with other forced choice AC experiments. Although only one receiver of three produced significant evidence of AC, the result is illustrative of the technique, and because of 531's previous performance, we consider that this result is not likely to be spurious. While a 16-fold enhancement of effect size was realized by the SA method, it is particularly inefficient; to obtain 100 decisions, 531 pressed the mouse button 21,333 times for an efficiency of 0.47%. It is possible that the efficiency could be improved if the basic SA method could include some adaptive method. That is, the parameters of the analysis could be modified on the basis of the recent scoring rate. If sufficient improvement could be realized, this method might be incorporated as an aid in decision making in practical applications. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-537 A7pprpved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 ech lcal Final Report Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001 5 Apf&x l ~{>; &A%se 2000/08/10 : CIA-RDP96-00787R000300230001-5 VI. MAGNETOENCEPHALOGRAPH This section comprises the final report for SOW item 6.2.1. 1. Introduction In a series of electroencephalograph (EEG) experiments conducted at SRI International beginning in 1974, the central nervous system (CNS) of individuals was found to respond to remote and isolated visu- al stimuli (i.e., a flashing light).15,16.17 In the first experiment, during randomly interleaved 10-second epochs (i.e., trials), either a flashing light (16 Hz) or no light was present in a sensorially and physically isolated room. Significant decreases of occipital alpha power of isolated receivers were observed by Rebert and Tiirner.15 Tivo replications were conducted in collaboration with Galin and Ornstein at the Langley Porter Neuropsychiatric Institute. As reported by May et al., the results were inconclusive; the first replication confirmed the Rebert and Tbrner finding, a decrease of alpha power concomitant with the flashing light, but the second replication attempt found an increase in alpha power.17 Under another program in FY 1988, SRI International and a biophysics group at a national laboratory conducted an experiment using the magnetoencephalograph (MEG) technique. This experiment was designed as a conceptual extension of the May et al. EEG experiment, although there were significant differences in the protocol. Tivo types of stimuli were randomly presented to an isolated sender while MEG data were collected from a receiver. The experimental stimulus (i.e., remote stimulus) was a 5-cm square, linear, vertical sinusoidal grating lasting 100 milliseconds. The second stimulus, a control stimu- lus (i.e., pseudostimulus), was simply a time marker corresponding to a blank screen in the data stream, and was also presented to the sender. There was no change in the alpha power, as reported by May et al., but a post hoc analysis revealed a root-mean-square average phase shift of the dominant alpha fre- quency-18 A key result of that experiment was that similar "anomalous" phase shifts were obtained for the remote stimuli and the pseudostimuli. Three candidate explanations for these results were sug- gested. The observed phase shifts might have been: ? Spurious (i.e., statistical deviations within chance expectations) ? Electromagnetic artifacts ? Evidence of anomalous cognition In order to determine which of these three candidate explanations was correct, we replicated the study at the national laboratory as part of this current effort. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-539 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report 2. Approach In this section, we provide details of the replication of the 1988 MEG experiment. 2.1 Replication Protocol 2.1.1 Number of Trials We assumed that the observed trial effect sizes that were reported for the previous MEG study resulted from a putative AC effect.18 Under the remote stimulus condition (i.e., approximately 1,100 trials) we found that the trial-level effect size was 0.060 ? 0.030. In statistical terms, this did not exceed mean chance'' expectation. Tb determine the number of trials necessary to provide a confident replication of the previous experi- ment, we conservatively used the observed effect size minus one standard deviaidon (i.e., 0.030). Using traditional statistical power analysis, we found that the probability of observing a significant AC effect in 1,100 trials was 0.258. Conversely, if we require 95% confidence, a significant AC effect could be observed in 12,026 trials, or approximately 120 blocks of 100 trials each. Twelve individuals were initially identified as receivers for the formal replication; however, because of scheduling difficulties only eight participated in the study. Seven receivers contributed ten blocks each, and one receiver contributed five blocks. The statistical power for 7,500 trials was 0.83, which is the probability of a significant :replication over the total of 75 blocks. In this case, a given receiver had a 60% chance of demonstrating an independently significant result if the AC hypothesis is true. 2.1.2 Receiver Selection Eight experienced receivers, who either participated in the earlier MEG study or were known to be "good" receivers from other investigations, participated in the study. 2.1.3 Sender Selection An SAIL experimenter acted as sender throughout the study. While it is assumed that a sender is not necessary for AC, it may have a vital psychological function. 2.1.4 Stimuli The following two types of stimuli were generated by a PC, and consisted of an internal image that could be sent to a standard TV monitor for display: ? Remote stimuli (RS). A low spatial-frequency sinusoidal grating lasting 100 milliseconds was used as a remote stimulus. ? Pseudo Stimuli (PS). All data bytes corresponding to the pseudo stimuli were zero. Thus, the entire video image was a blank screen corresponding to a "time marker" in the data. An HP workstation controlled the collection of data and the presentation of the stimuli. Using a multi- ple congruent pseudo random algorithm (i.e., Rn+1 = ao x Rn + b0, where ao and bo are constants, and 0 s R < 1.0), the nth + 1 stimulus was generated 3.0 + 4.5 x R,a+1 seconds after the nth stimulus. The algorithm was seeded from the system clock. The HP notified the PC of the type and time for a stimulus. The PC waited until the next vertical retrace signal from its hardware-video-output board; switched pointers within the retrace cycle from the blank inter-stimulus (IS) frame buffer to one which contains Approved For Release 2000/08/10 : CIA-RDP96-00787R00030023000140 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report either the RS or PS; and reset the buffer pointers after 100 ms (i.e., the stimulus duration = 100 ms). Figure 16 shows this sequence in graphical form. Stimulus Buffer Pointers Standard 30 Hz Type RS/PS RS Buffer Interleaved Stimulus Initiation IS Buffer Output Buffer PS Buffer V Figure 16. Sequence of Events for Stimuli Generation 2.1.5 Placement of the Seven-Sensor MEG Array The placement of the seven-sensor MEG array was determined by an individual receiver's response to a direct light stimulus. While being stimulated by randomly interleaved low and high spatial-frequency gratings, sufficient stimuli (e.g., 30 to 50 of each type) were collected to produce good signal-to-noise responses. The position of the sensor array, relative to head-based coordinates, was recorded manually on a skull cap, so that the array could be repositioned accurately during subsequent experimental blocks. The array positions that were used during the RS blocks were determined by the maximum re- sponse to these direct stimuli. For this portion of the experiment, the stimuli were generated three to four times faster (i.e., - 1 per second) than in the AC portion of the experiment. 2.1.6 Session Protocol The session protocol was a follows: (1) Using the marking on the skull cap, the MEG array was repositioned as close as possible to the original calibration location. (2) Its position was confirmed with direct stimuli, and adjustments were made, if they were necessary. (3) The designated sender was positioned in front of the remote monitor, which was located approxi- mately 40 in from the receiver. (4) The video monitor, which presented the direct stimuli, was turned off. (5) The receiver was instructed to relax with eyes closed. In addition, the receiver was given a few possible strategies that included focusing attention on the display that the sender was observing, on the sender, or on both. (6) The receiver was notified, by intercom, that the run was about to begin. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-541 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report (7) The run began and seven channels of MEG data and one channel of stimulus data were collected for two minutes. The raw data were saved to disk, and the appropriate parameters for the next run were entered into the log book and the control program. (8) After five runs, an experimenter quietly entered the MEG room, checked the MEG position, and readjusted it, if necessary. No communication about the status of the e3 periment was provided. (9) Five additional runs were collected. (10) At the end of the block, the receiver entered the control room and was shown a computer display of the results of the last run. The experimenter pointed out interesting portions of the display, but cautioned that the final results required careful analysis of all the runs, not just the last one. 2.1.7 Controls Tivo types of controls were used in this experiment to assure the validity of the experimental results: ? Within-block. The data in the inter-stimulus times (IS) were used as a within-block control. ? Between-block. Using a counterbalanced random protocol, either immediately before or immediate- ly after each 20-minute experiment block, an additional block of ten runs was taken under the same conditions as the experiment block, but without the receiver under the MEG. The sender, however, was "sending" as before. 2.1.811 Data Recording Along with the experimental parameters, eight channels of 200 per second data were digitally recorded for later analysis (i.e., seven channels of MEG data and one channel of stimulus data). 2.1.9 Analysis Overview A block of data was ten, 2-minute runs. Each block contained approximately 100 RS and 100 PS stimuli, respectively, from each of the seven sensors. The following was computed for each stimulus type and for each sensor: ? Time averages for 0.5 second prestimulus to 0.5 second poststimulus. ? Separate average power spectra for the prestimulus and poststimulus periods. ? Averages of the phase shift observed at the dominant a-frequency, which was determined from the centroid of the peak with the largest area above "background" for experiment blocks; 10.0 Hz was used. for the between-block controls. The relative phase shift for a single stimulus is defined in Figure 17. The RMS average was computed over the total number of stimuli in the block. The RIMS average phase was the dependent variable for the block. A Monte Carlo calculation was used to determine whether the observed phase shifts deviated from those observed at random times throughout the rest of the block-data record. Each Monte Carlo pass computed the RMS phase omcr random entry points, which were determined by the same timing algorithm described above, into the sune 20-minute data set. The timing algorithm was the same one used during the data collection; however, a new seed started the process on each pass. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-g2 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 Technical Final Report x(t) linear Process Y(t) Let: X(v) - FFT [x(t)l Then: Y(v) - FFT [y(r)l Phase - y'(v) - tan -t Int H(v) e v Poststimulus: y(t) *4Y H(v) = Y Gain - IH(v)I Figure 17. Phase Calculation for a Single Stimulus Statistics (e.g., p-values, z-scores) were computed from the distribution of RMS phases derived from the Monte-Carlo-pass distribution. Conceptually, a 2-tailed z-score was calculated from a Monte Carlo distribution of phase shifts in the following way: Let pp and o y be the mean and standard deviation of the Monte Carlo phase shift dis- tribution, and To be the observed RMS phase shift. Since the distribution of averages is approximately normal, compute: Z = VI0 ,U and P = J e 2dg. z Since we did not specify a direction for a change in phase, the p-value for the block was given by. p=2xP, and the two-tailed z-score was computed from the inverse normal distribution for P In the experiment, the empirical value of P was used. That is, the number of Monte Carlo-derived RMS phases that were greater than or equal to the observed RMS phase was divided by the total number of Monte Carlo passes. There- fore, the 1-a error estimate in P were computed from the binomial distribution for proportions. Or P(1 -P) 1-a error in P = M where M is the number of Monte Carlo passes. For this replication, the analyst was "blind" to the identity of the receiver, the date, the experiment condition (i.e., experimental or control run), and the stimulus type. Details of the Analysis Consider N blocks of experimental data. Let ?j, be the number of remote stimuli r for block j, and njp be the number of pseudo stimuli p in block j. Similarly, define rjr and rjp as the corresponding effect sizes for block j. We define the weighted effect size for each stimulus type, k, as N rk = >WJkE/k, Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 43 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 chflical Final Report nJk Wjk= N E nJk k=r,p. Tests Against the Null Hypothesis: The Average Effect Size e = 0. Since the experimental effect sizes, ejk, are derived from normally distributed data (i.e., Monte Carlo calculations of the RMS phase shift), then we know the standard error for each elk is nJk Thus, the variance of the weighted average effect size is The z-sccore associated with a is N ,/ Var((k) = I w Var( k) = N 1 J-1 n1k -R Var(1k) (1) (2) Equation 2 is used to test the average effect sizes of the RS and PS for the experimental and control conditions against the null hypothesis of i = 0 for the experimental and control conditions. Tests Against the the Null Hypothesis: s(RS) - e(PS) = 0. Within a given condition we cannot assume that the phase shifts from an RS are independent from those associated with a PS. Thus, hypotheses tests th at do not account for potential correlations between the RS and PS are i iappropriate. Because of the simplicity of the individual ejk, we can compute the exact variance for the differences as follows. Let thedifference between the effect sizes for RS and PS be d1 = e1, - e1P. Since there usually are a different number of stimuli for RS and PS, we define a weighting factor for the dj as nj S1J= N Z nJ J-1 n= nJ. X nJP 1 nJ, + nJP Then the weighted mean difference is given by Approved For Release 2000/08/10 : CIA-RDP96-00787R0003002300OW AwchnYreai Final ReR port se 2000/08/10 :CIA-RDP96-007878000300230001-5 The variance ofd is given by: N Var(3) Q; Var(d), -, Var(d,) = Var(e,,) + Var(E, p) - 2 Cov(E,,,e,,), Cov(EJ?E/P) = Q,,, Var(E,,) - Var(e1,). Combining these equations with the definition for the variance of the effect size, gives the Yar(d) as N r Var(f) _ QJ L 1L + j - 2 Q,,, Var(E,,) ? Var(ej )], J-i a 17 Yar(a) (3) rests Against the Null Hypothesis: e(Experiment) -e(Control) = 0. Tb compare each stimulus type in the experimental and control conditions, we assume that the data are independent. Thus, the z-score for the difference is given by 17 a Var(tk(e)) + Var(-r (c)) L "Jk() E ~,k~c) J-1 J-1 (4) where e and c represent the experiment and after-block control conditions, respectively, d is the weighted difference for the stimulus type in the experiment and control conditions. Equation 4 is used to test the difference between experimental blocks and their corresponding control blocks. , , " - " ' Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-545 Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 ct~nical Final Report Table ',9 summarizes these results. Hypothesis Testing for Each Receiver Hypothesis Thst Quantity 1. RS(e) have no effect. n J, (e) 1sX P00` eIROp ft re ' s TV monitors. The four elements in the judging pool are presented in one of four CPYRGHAT Iparoved For Release 2000/08/10: CIA-RDP96-00787RQ94Q D3APMJJ random sequences. The receiver is prompted to identify whatever correspondences they perceive between their ganzfeld mentation and each of the four potential targets. The receiver is given the option to view any or all of the elements in the judging pool as many times as desired, then procedes to perform the blind judging task. The program displays a judging scale (Figure 8) on the receiver's monitor for each of the four possible targets in the judging pool. The judging scale shows a brief descriptive name for each target, a thermometer-style rating scale, and three buttons. Using a mini joystick, the receiver rates the degree of perceived similarity between each potential target and their mentation. The scale ranges from 0% to 100% and the current value of the scale is displayed both numerically and graphically as the receiver clicks either the left or right arrow buttons. Figure 8. Video Ganzfeld Judging Scale Judging Scale -~'ii111Y111 W Yli?~`:IiY W ki:.i~::i11Y V:`~'riYit[-2]YYIY~:iYYYk'L'vi ?::O {.hii:W..{x:?:S'{{#?ti:?.v'Y{i:ti??}.; {,{.. vif?',.. ..?.~4.: i.:?~. ::~'{i:::{v\v;{4:{?::{{.ii.{.i: n{?ii`.{vv:ii'.{v:.v'+T::..tii`8, y:i:?ki^:iif}{{?:'. When the receiver is satisfied with the rating assigned, she or he presses the "OK" button. The judging procedure is repeated for each of the four potential targets in the judging pool. The program checks for tied ratings and prompts the receiver to re-rate in the event of a tie. Once the receiver has rated all of the elements in the judging pool, the program converts the ratings to ranks and stores the ratings and ranks as fields in the session database record. The program calculates a standardized rating (z-score) based on the difference between the rating assigned to the correct target and the mean of the "three decoy ratings divided by the standard deviation of of all four ratings (Stanford & Sargent, 1983). The program times the duration of the judging procedure from initial presentation of the four judging pool elements to completion and adds it to the session database record. ApprovedA~WdFgfiW current session," "Session log," and "Check System." The abort session option is CPYRGAIPproved For Release 2000/08/10 : CIA-RDP96-00787R0840n~e9 ,0c oc P 10 used to terminate an ongoing session prior to completion. Premature termination of a session may only occur in the event of a protocol violation (e.g., sender or receiver leaving their respective rooms after the beginning of the session), equipment failure, or an emergency situation. When "Abort session" is selected, the system displays a dialog box prompting the experimenter to enter his or her security password and indicate the specific reason for terminating the session. This information, along with the the participant's ID, and the date and time are written to a series abort file. Abort session is not available after the blind judging procedure has been completed. Session log enables the experimenter to register comments concerning the current session and "Check System" performs diagnostics on the audiovisual and randomization functions of the system. TargIt Stimuli Target Pool Following Honorton, et al., (1990), target stimuli consist of brief (35-80 sec.) video excerpts from a variety of films and documentaries. Two target pools, each containing 40 targets' (10 judging pools of four targets each), have been prepared. Each target pool is stored on one, 90-min..5 in. VHS videocassette tape. Digital addresses on each videocassette enable frame-accurate access of targets via the video ganzfeld/PC-VCR computer link. A unique digital header is recorded on each videocassette and is read by the computer at the onset of each experimental session. Accidental insertion of a videocassette other than that containing the designated target pool is automatically detected and results in termination of the session. Based on an analysis of target success-rates in the PRL experiments, approximately half of the targets were taken from among the most successful PRL dynamic targets. The remainder of the targets are new. Pool A will be exclusively used for the Novice Screening Series and Pool B'B will be used for the Sender Comparison Series. Since the latter series will include sessions in which the sender will be exposed only to the audio soundtrack portion of the target, the elements in Pool B include a high proportion of targets with descriptive narration. Measurement of Target Attributes The quantification of complex target material has long eluded investigators of anomalous communication. The quantitative characterization of target attributes is important for a number of reasons, for example: ? Development of more statistically powerful methods, for assessing target/description correspondences, ? Detection and elimination of targets associated with strong response bias (i.e., targets that tend to be selected or rejected because of their intrinsic characteristics), ? Detection and elimination of targets that activate perceptual defense, Approves Forieleasgetl '`b e$j less y P99 t8 X00300230001-5 CPYRG' roved For Release 2000/08/10: CIA-RDP96-00787R D MQ&F15 11 ? Identification of elements of target environments that may be especially amenable to retrieval via anomalous communication. Recently, major advances have been made with regard to certain aspects of this problem as it specifically applies to remote viewing studies (May, et al., 1985; 1990). While aspects of May's conceptual schema can also be applied to ganzfeld research, there are two aspects of the latter that call for a somewhat different approach: (1) The standard ganzfeld mentation protocol focuses upon the elicitation of unconstrained spontaneous imagery rather than an explicit focus upon describing the target. (2) The video targets are themselves quite different from those typically used in remote viewing research: They include auditory components (e.g., music, dialogue, narration, sound effects), occasionally major transitions in perspective, highly evocative dramatic and comedic scenes, etc. For these reasons, we have adopted a somewhat different approach, consisting of two distinct aspects: (1) Specific descriptors tailored to the content of the target pools, and (2) generic characteristics derived from environmental psychology. Content-based Descriptors Each target has been coded with respect to Theme, Tone, and Content. Each item is coded Table 1. Content-based Descriptors Nature/wildlife Fantasy/religion/mythology Aggression/battles/warfare/conflict Social interactions Sports/athletics/acrobatics Art/dance/music Places/travel/exploration Cartoons/animation People Animals Fantasy/mythical characters Water Rocks/hills/mountains Trees/flowers/foliage Land vehicles/scenes Terrestrial flight scenes Underwater vehicles/scenes Architecture/urban scenes CPYRffoved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 on a four-point scale, where 0 = absent, 1= present, 2= prominent, and 3=dominant. (See Table 2). The content-based descriptors are used (a) in construction of orthoginal judging pools and (b) in exploratory analysis of target attribute correlates of anomalous communication. Generic Characterization of Targets based on Environmental Psychology Approaches The above approach represents what Mehrablan and Russell (1974) describe as "the most common, but least parsimonious, approach... the use of the everyday language of specific events and entities" (p. 6). They point out that this approach does not permit comparison across environments, ".... and it is impossible to analyze behavioral changes as functions of changes in environments so described" (p. 6). Mehrabian and Russell survey a wide array of evidence pointing to the advantage of generic characterization of environments in terms of the primary emotional responses they elicit and a (psychologically-based) measure of information rate. Their general framework is illustrated below in Figure 9. Figure 9. Mehrabian & Russell Framework for Characterizing Environments THE ENVIRONMENT Sense modality variables (e.g., color and temperature) Information rate (characterizing the spatial and temporal relationships among the stimulus components of an environment) Ch4racteristic emotions associated with PERSONALITY After Figure 1.1 of Mehrabian & Russell (1974). PRIMARY EMOTIONAL RESPONSES Pleasure Arousal Dominance BEHAVIORAL RESPONSES Approach-avoidance (which includes physical approach, exploration, affiliation, performance, or other verbal and non- verbal communications of preference) Within this framework, environments are coded using semantic differential scales measuring the three primary emotional responses (pleasure or evaluation, arousal or activity, dominance or potency) and information rate. The scales are reproduced in the appendix. Each of the targets has been coded on these four scales. We believe that this approach may provide a basis for broader comparison across laboratories and target sets than more traditional methods. It of course remains to be seen how useful it will be as a predictor of success in anomalous communication. Approved For Release 2000/08/10 : CIA-RDP96-00787R000300230001-5 MW CPYRG~Tpproved For Release 2000/08/10 : CIA-RDP96-00787R MW AW175 13 Predictor Measures Extraversion and Openness to Experience Performance in anomalous communication tasks has been found to correlate with the psychological trait of extraversion in a recent meta-analysis of 15 studies by five independent investigators (Honorton, Ferrari, & Bern, 1990). The mean correlation is small (r = .20) but consistent across investigators, studies, and personality measures. While the meta-analysis provides strong evidence that a relationship exists between anomalous communication and extraversion, it is silent as to the nature of the relationship. Extraversion is commonly associated with sociability (gregariousness), but it is now known that there are at least five other components of extraversion. For this reason, we have chosen the NEO Personality Inventory (Costa & McRae, 1985), an instrument that measues six facets of extraversion. Recent research implicates sensation seeking as an instrumental factor in the ganzfeld experience (Glicksohn, 1991) and we are especially interested in the possibility that it also correlates with performance in anomalous communication tasks. We also will use the NEO PI Openness scale, and its six facets, because a number of studies have indicated a relationship between anomalous communication and various measures of openness to experience. Table 24ists the six facets of extraversion and openness. Table 2. Facets of Extraversion and Openness 1. Warmth 2. Gregariousness 3. Assertiveness 4. Activity 5. Excitement Seeking 6. Positive Emotions 1. Fantasy 2. Aesthetics 3. Feelings 4. Actions 5. Ideas 6. Values A computer program scores the questionnaire and presents graphic profiles for each of the six facets of extraversion and openness. Statistical power analysis (Cohen, 1977) indicates that a sample size of 200 subjects will achieve a 90% likelihood of detecting a correlation of .2 at p