# INTUITIVE DATA SORTING: AN INFORMATIONAL MODEL OF PSYCHOENERGETIC FUNCTIONING. SRI PROJECT 1291.

Document Type:

Collection:

Document Number (FOIA) /ESDN (CREST):

CIA-RDP96-00787R000500140001-3

Release Decision:

RIFPUB

Original Classification:

K

Document Page Count:

37

Document Creation Date:

November 4, 2016

Document Release Date:

October 21, 1998

Sequence Number:

1

Case Number:

Publication Date:

December 1, 1986

Content Type:

RP

File:

Attachment | Size |
---|---|

CIA-RDP96-00787R000500140001-3.pdf | 1.25 MB |

Body:

Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Final Report- -Objective E, Tasks 3 and 4
December 1986
19601-
INTUITIVE DATA SORTING: AN INFORMATIONAL
MODEL OF PSYCHOENERGETIC FUNCTIONING
PETER J. McNELIS, DSW
CONTRACTING OFFICER'S TECHNICAL REPRESENTATIVE
333 Ravenswood Avenue
Menlo Park, California 94025 U.S.A.
(415) 326-6200
Cable: SRI INTL MPK
TW X : 910-373--2046
oved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
ved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Final Report- -Objective E, Tasks 3 and 4 December 1986
Covering the Period 1 October 1985 to 30 September 1986
INTUITIVE DATA SORTING: AN INFORMATIONAL
MODEL OF PSYCHOENERGETIC FUNCTIONING
PETER J. McNELIS, DSW
CONTRACTING OFFICER'S TECHNICAL REPRESENTATIVE
ROBERT S. LEONARD, Executive Director
Geoscience and Engineering Center
333 Ravenswood Avenue Menlo Park, California 94025 ? U.S.A.
(415) 326-6200 ? Cable: SRI INTL MPK ? TWX: 910-373-2046
oved For Release 2000/08/07 CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
We develop a comprehensive model of psychoenergetic functioning called Intuitive Data
Sorting (IDS). Extending purely philosophical arguments, we derive specific mathematical
predictions for the interpretation of random number generator experiments. Two experiments
are analyzed: (1) a pseudorandom number generator (PRNG) experiment conducted at SRI
International, and (2) a random number generator experiment conducted at the Princeton
Engineering Anomalies Research laboratory at Princeton University. Preliminary results from
the PRNG experiment are in statistical agreement with the IDS model. We show, however,
that the Princeton University RNG data were collected under unfavorable conditions to serve
as a test of the IDS model; we recommend an RNG experiment protocol that will allow a
more favorable test.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
TABLE OF CONTENTS
LIST OF ILLUSTRATIONS ................................................... iv
LIST OF TABLES ........................................................... v
I INTRODUCTION ................................................... 1
A. A Conceptual Thought Experiment ................................ 2
B. A Practical Thought Experiment ................................... . 3
II BACKGROUND .................................................... 6
III METHOD OF APPROACH ........................................... 8
A. Theoretical Considerations ....................................... 8
1. RA Data Reduction .......................................... 8
2. Mean Chance Expectation .................................... 9
a. Theoretical Considerations for Overall Mean Chance Expectation .. 9
b. Theoretical Considerations for an RA Interaction ............... 11
c. Theoretical Considerations for the IDS Model ................. 14
d. IDS vs. RA in Perspective .................................. 17
3. Analysis and Hypothesis Testing ................................ 18
B. The SRI Pseudorandom Data ..................................... 20
1. Justification for a Pseudorandom Number Generator Experiment ..... 20
2. PRNG Experiment Description ................................. 20
C. The Princeton Engineering Anomalies Research Data ................. 21
IV RESULTS AND DISCUSSION ........................................ 24
A. The PRNG Experiment Results .................................... 24
B. The PEAR Results .............................................. 25
1. Data Reduction and Analysis .................................. 25
2. Discussion of the PEAR Results ................................ 27
V CONCLUSIONS AND RECOMMENDATIONS ........................... 29
REFERENCES ............................................................... 30
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
LIST OF ILLUSTRATIONS
1. Generalized Decision Process .............................................. 1
2. A Conceptual Thought Experiment ......................................... 3
3. A Practical Thought Experiment ............................................ 4
4. One RA Model Compared with MCE ....................................... 14
5. One IDS Model Compared to MCE and One RA Hypothesis .................... 16
6. MCE and PK- Data ...................................................... 28
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
LIST OF TABLES
1. Exact and Approximate Values of ln(JApI) ................................... 11
2. Raw Data for Operator 10--Princeton University Data .......................... 23
3. MCE for Each Data Set .................................................. 25
4. Results of the Analysis of the PEAR RNG Data ............................... 26
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
I INTRODUCTION
Since 1979, SRI International has been constructing a model of psychoenergetic
functioning that may provide an explanation for a broad range of experimental data. * The
idea first occurred to us as part of an interpretation of an experiment we conducted that
year.I t In that experiment, it appeared that individuals were able to make decisions
(psychoenergetically) based upon information propagating backward in time. This unorthodox
concept, in its generalized form, is shown schematically in Figure 1 as one of the inputs to a
decision process.
FIGURE 1 GENERALIZED DECISION PROCESS
For example, suppose that you had to decide if it were safe to cross a highway. Clearly,
one input to that decision is real-time analysis: There is a continuous stream of traffic; don't
cross the road. Suppose that there was only one car on the road, but it was at some distance
away. Experience might suggest that you are unable to run fast enough to avoid being hit;
don't cross. Without defining intuition, it might tell you not to cross the highway even though
* This report constitutes deliverable "a"--final report on RA perturbation of device-driven
random sequences--for Objective E, Task 4, Conduct a retrospective test and analysis
of the Intuitive Data Sorting (IDS) model, and the deliverable--final report on RA
activity upon pseudorandom number generators--Objective E, Task 3, Investigate RA
activity by examining statistical changes of state of pseudorandom number generators.
t References are listed at the end of this report.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
all other indicators suggest that it is safe. We recognize that all these processes are
intermixed, and that there may be others that we have not yet considered.
Our model proposes that there is one other source of input to the decision process that
has not yet been considered--information propagating backward in time from the future. You
have an impression (maybe a visual experience) that indicates that you are dead on the
highway; don't cross the road.
While there are specific examples of information propagating backward in time in physics
(e.g., the Dirac equation--a positron traveling forward in time is mathematically identical to
an electron traveling backward in time), the idea that it is possible at the macroscopic level is
not generally accepted. It is beyond the scope of this report to discuss the profound
implications for physics and philosophy if it were true.
Since the late 1930s, however, the parapsychological research journals have been
reporting evidence that information from the future is available in the present--at least
statistically. We are not able, at this time, to provide a complete analysis of this literature,
which claims evidence for precognition--the parapsychological term for accessing information
from the future. Rather, we provide a brief discussion of two of the pertinent reports.
In 1983, C. T. Tart reviewed the precognition "forced choice" literature as part of an
investigation to determine information rates for both real-time and precognition experiments2
In this context, "forced choice" implies that the subject was aware of the limited number of
possible targets, but was blind to the actual target chosen for any single trial. Tart found 32
studies, of this type, that claimed statistically significant evidence for precognition.
In 1986, Nelson et al. reported the results of over 400 "free-response" precognition
trials demonstrating strong statistical evidence for information flow from the future.3, 4
"Free-response" experiments differ from "forced-choice" in that the target material is
relatively unbounded (i.e., while the general nature of the target material might be known by
the subject, the range of target possibilities is very large and is unknown to the subject).
Because these reports provide strongly suggestive evidence for macroscopic information
flow from the future, we proceed with the development of the model.
A. A Conceptual Thought Experiment
Consider an arbitrarily complex experiment as shown in Figure 2. Further,
suppose that this experiment has one result and that there is a single decision required before
the experiment is conducted. (We recognize that complex experiments produce more than
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
one result and require many decisions for their design; therefore, we are considering only the
primary result and decision, and are including the lesser results and decisions as part of the
complexity of the experiment.) There are no limitations placed upon the complex
experiment. In principle, it might be conducted by many researchers at different laboratories
for many years.
To be specific, let us assume that the single decision involves turning a knob to
one of five settings, 1 through 5. Also consider that the single result is a meter reading in
which we assume a high reading is "good." Our model proposes that the following internal
dialogue by the decision maker is possible:
If I were to conduct this experiment with the knob set at 2, let me
"peek" into the future and examine the result. I perceive the meter
showing a low reading and thus I reject that option. Suppose I were to
conduct this experiment with the knob set at 4. I perceive the meter
showing a very high reading. Because I like that result, I will set the
knob at 4, and conduct the experiment!
For the above dialogue to be sensible, we must be able to "sample" the
future--reject the unwanted ones, and "select" a preferred one. When the model is
formulated mathematically and applied to a specific set of experiments, it will provide
compelling evidence (see below) that this unorthodox idea is possible.
By describing a practical thought experiment, we will illustrate an important
consequence of the model. If we are allowed to "sample" the future, then what was
previously thought of as a cause-and-effect relationship might be confused with an
informational relationship involving no causality at all.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Suppose that we wanted to demonstrate that cyclamates cause cancer in laboratory
rats. Further, assume that normal double blind protocols are in effect. A laboratory assistant
is asked to administer injections to two separate groups of rats--one a control group, the
other a target group. By double blind, we mean that the assistant does not know which is the
control group, nor which injections will be cyclamates rather than biologically neutral
cyclamate placebos.
Let us assume that the experiment produces a statistically significant (e.g., p <
0.01) separation between the control and target groups (i.e., the probability of a target rat
contracting cancer is significantly larger than for a control rat). This result is shown
schematically in the lower half of Figure 3. The conclusion that is drawn from this
experiment is that whatever was in the syringe caused cancer in the rats.
To illustrate an informational explanation of this same result, we describe a totally
fraudulent way in which this experiment might be conducted. Suppose that the researcher is
motivated to cheat, and in doing so is able to select from a normal population of rats those
that are predisposed toward cancer. (Because this is a thought experiment, we can invent a
mechanism by which he/she could do this.) We assume that this researcher is sophisticated in
statistical matters and sorts the selected rats into target and control groups in such a way (by
cross mixing them) to produce the identical results described above. Furthermore, to
preclude any possible confounding of the fraudulent outcome, the researcher replaces the
contents of the syringes with distilled water. Then he/she lets the experiment proceed as
above.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
The statistically significant result (p < 0.01) would still be interpreted as before.
Whatever was in the syringe caused cancer in the rats--in this case distilled water! Thus, a
purely informational process (knowledge of the individual rats' predisposition toward cancer) is
mistaken as a causal one (distilled water causes cancer in rats).
The researcher, in the above example, need not be fraudulent to produce the
identical result. We propose that by statistically "peeking" into the future he/she is able select
the predisposed rats. For example, having picked a rat, he could sample the future to
determine if that rat is likely to contract cancer. If so, he would add it to the target group; if
not, it would be included in the control group. Thus, this researcher has "simply"
psychoenergetically sorted rats (see Figure 1). We call this putative ability Intuitive Data
Sorting (IDS).
Because large bodies of research have concluded some form of cause-and-effect
relationship based upon statistical hypothesis testing, it is important to determine if our model
has any scientific basis. What follows is the application of our model to a large body of
psycho energetic experimental data.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
II BACKGROUND
In 1969, Schmidt5 introduced a type of psychoenergetic experiment in which individuals
were asked to "modify" the statistics of "true" random number generators (RNG), i.e.,
devices that generate sequences of numbers based on some fundamental random process such
as radioactive decay or thermal noise. Since publication of that important initial paper, we
have been able to locate (through 1984) 56 pertinent references in English language journals
and reports describing a total of 332 individual binary RNG experiments. We simulated an
additional 95 nonsignificant experiments to account for a "filedrawer" problem (i.e.,
experiments that were conducted and not published because they were not significant).6 By
including these simulated studies in the data base, we refrain from using an artificially inflated
data base because of selected reporting of "successful" studies. In the preliminary analysis of
these data, we calculated that the probability that the observed deviations occurred by normal
statistical fluctuations alone was p < 3.9 X 10-18 during experimental conditions, and p G
0.78 under control conditions. Clearly, there is a statistical anomaly within these data.
Since 1969, there has been considerable discussion about mechanisms that can explain
these RNG results.5, 7, 8 (For the purpose of this report, we assume that artifact and incorrect
statistics have been accounted for. Radin has shown that this assumption is true to first
order;6 a detailed meta-analysis is now in progress to determine the overall validity of the
RNG data base.) The most frequently proposed explanations are remote action (RA) and
precognition. Under an RA hypothesis, by definition, a participant "forces" a physical
modification in a source of random signals so as to produce a change in the output statistics.
Alternatively, under the IDS hypothesis, we propose that humans can make decisions (by
psychoenergetic means) to take advantage of the natural and unperturbed fluctuations of a
system. In the context of an RNG experiment, it appears that individuals can anticipate
locally deviant subsequences from within larger and unperturbed sequences and make
decisions based upon that knowledge. Suppose that an individual is asked to "make" the
RNG produce more binary ones (1s) than zeros (0s). Rather than "causing" the device to
produce binary ones, we suggest that the participant has simply initiated the run by
anticipating when the RNG was going to produce a series of ones as part of its natural
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
binomial fluctuation. Thus, the participant has capitalized upon natural events, rather than
"causing" anything to occur.
In our final report to a client in FY 1984, we applied the IDS formalism to the data
base of RNG experiments described above.9 We found that the data were described by the
IDS model rather than the RA hypothesis. There were problems, however, with the
comparison of the IDS model to the previously published RNG data base. The IDS formalism
is derived from the assumption that the sequence length, [n], results from a single press of a
button. None of the experiments in the data base were reported in that way. All of the data
were aggregates over many button presses. While we were able to draw conclusions based
upon averaged data, (e.g., on the average, IDS appears to account for the results in the
historical data base), the ideal test of IDS must be conducted using data resulting from single
button presses.
There are two questions that should be addressed when conducting single button press
experiments:
? Is IDS possible under conditions that preclude any RA? If so, are there
any sequence length dependencies in the ability?
? Can IDS provide an interpretation of RNG data?
We have addressed both issues as separate tasks for FY 1986. The first task was to
examine the IDS model, at SRI under a condition that precludes RA--using pseudorandom
number generators. To know if IDS is possible, in principle, we must conduct experiments
that have as few confounding factors as possible. The second was to examine the IDS model
using "true" RNGs. The only data of this type that was available was collected by the
Princeton Engineering Anomalies Research (PEAR) Laboratory, in the school of Engineering
at Princeton University.10 We therefore let a subcontract to this laboratory to provide us with
their single button press RNG data.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
III METHOD OF APPROACH
Our method of approaching the data generated at SRI and Princeton proceeds in three
phases:
? Theoretical considerations--a derivation of mean chance expectation for
the various models under study.
e The SRI data--details of the pseudorandom data collection and display.
? The PEAR data--details of the Princeton "true" RNG data collection.
The data analysis proceeds in three steps, data reduction, definition of mean chance
expectation, and analysis and hypothesis testing.
1. RA Data Reduction
For each button press, the raw data consist of a sequence length [n] and the
number of "hits" [h] (e.g., binary is). We transform these data into logarithmic form for
analysis by computing the following quantities for each data point:
h
p= n
lnlopl = lnl\n - 0.5)I
p - 0.5
z = 1 V
2 n
If the linear correlation coefficient for all pairs of data points, In(n) and InjApl, is
significant, then a straight line may be fit to the data, and thus, the raw data are reduced to
two coefficients--the slope and the intercept.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
2. Mean Chance Expectation
In the absence of all forms of psychoenergetic functioning, we must determine,
theoretically, the mean chance expectation (MCE) for the type of data described above.
a. Theoretical Considerations for Overall Mean Chance Expectation
We define the sequence length, [n], as the number of samples collected from
an RNG as a result of a single button press. If we consider all possible values of sequence
lengths, it is convenient to use the binomial statistic in its exact form for n < 200, and use the
normal approximation for larger sequence lengths. We use the observed fractional hit rate, p
(hits/trials), minus the expected fractional hit rate, po (equal to 0.5 for binary RNGs), as the
dependent variable, ip = p - po, and [n] as the independent variable.
For the "continuous" region where n > 200, Op is normally distributed about
a mean of zero with a standard deviation of o'o given by
where q0 = (1 - po). For convenience, we shall examine the statistical properties of [n] and
Op in logarithmic form--In(n) and ln(Op), respectively, and, without loss of generality,
consider only the absolute value of Op, IOpl. The expected value of InjApj is given by
00 -0.5 02
f In Iopl e Q? d (Op)
00 0.5 (A)2
f e- a?/ d(Op)
Following the usual definition of a z-score, let
(1)
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Substituting into Equation (1), we find that
In IApI = In aao+InIzI .
In I Ap I = 0.5 In po qo + In
- 0.5 Inn (2)
Equation (2) is a simple way to express the null hypothesis of no
psychoenergetic functioning for the RNG data. If many RNG runs of varying sequence lengths
are conducted in the absence of all psychoenergetic phenomena, the natural logarithm of the
sequence lengths and their associated logs of IOpl's are linearly related--having a slope of
-0.5, and an intercept determined by po, qo and the average value of lnjzj--a known
constant.
Given the "unboundedness" of jzj (i.e., 0 < jzi < oo), it may surprise some
readers that there is a linear relationship between the expected value of InjApj and In(n). To
demonstrate this linear relationship from a different perspective, we calculate the expected
linear correlation coefficient, r, under the null hypothesis:9*
-0.5774 X (Sx- SM)
19.7392 + (Sx- SM) 2
(3)
Sx and SM are the logarithms of the maximum and minimum sequence lengths, respectively.
Equations (2) and (3) form the basic set of relationships that describe, in
detail, the expected results under the null hypothesis.
A significant note must be added at this point. Equation (2) represents the
MCE under a normal approximation. As we will show, for any actual experimental case, the
The authors wish to acknowledge and thank Dr. J. Utts for deriving this relationship.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
difference between normality and the exact binomial calculation becomes important in
determining the expected slope of the MCE line. Table 1 shows a comparison between the
normal approximation and the exact binomial calculation for the expected value of InIOpl as
a function of selected sequence lengths.
EXACT AND APPROXIMATE VALUES OF InIOpI
Sequence
Length
Exact
Binomial
Normal
Approx.
Percent
Error
200
-3.7880
-3.9774
4.76
2000
-5.0504
-5.1288
1.53
10000
-5.8922
-5.9335
0.70
100000
-7.0689
-7.0848
0.24
For any given experiment, the MCE line should be computed in the following
? Compute the exact value for 1nIOpj for each sequence used in the
experiment.
? Fit the above result with a weighted straight line. The weighting
factor for each sequence length is the number of trials, for that
sequence length, that were conducted in the experiment under
study.
? Use the slope and intercept from the above for the MCE values.
b. Theoretical Considerations for an RA Interaction
Because in the most general case of RA a subject could "perturb" the RNG
device in any way, the Gaussian in Equation (1) must be replaced by an arbitrary function,
f(Op,n) . Or,
00
r
J In I op I f (Ap,n) d(ip)
In JOpJ =
00
ff(p,n) d(Op)
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
To evaluate this general relationship, we must assume some specific form for f(Ap,n). From
this point on in the development of the model, we will assume that RA induces a minor
perturbation in the physical system. Thus, we assume that f(Ap,n) remains Gaussian, but RA
shifts its mean--slightly.
We consider a class of RA models in which RA perturbs (to a small degree)
a binomial distribution by shifting its mean. The theoretical task is to evaluate the expected
value for the lnjApj using Equation (1). Define
AP = p - ps
A Ps = pa - PO
L Pa = P - pa = AP - AP,
As before, po is the mean of an unperturbed distribution, and now pa is the mean of a shifted
distribution. Apa is the difference between the observed fractional hitting rate, p, and the
shifted mean, pa, and Aps is the shift in the means of the two distributions. We have derived
the expected value of lnIApj elsewhere,9 but we present a simplified version of it here.
For this calculation, we assume that the RNG is binary
(i.e., po = 0.5). Under these conditions, the expected value of lnjApI is given by
In I Op I = 0.5 In(0.25) + In [zf(Aps) + zs ] - 0.5 In(n) , (4)
2
f(Aps) = -1- 4Aps
This latter term arises because the variance of a binomial distribution depends
upon the mean of that distribution. If the shift of the distribution is small (i.e., Aps 0),
then f (Aps) s ? 1. For any RA model under consideration, we must define how the mean of
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
the shifted distribution behaves as a function of the sequence length, [n], and knowing that
[z] is distributed normally, the evaluation of Equation (4) reduces to the evaluation of the
expectation of In (zf (Ops) + zs). Or,
F2 00 -0.5 Z 2
In [ z f(A P) + zs] _ fln [ z f(A P) + zs] e d z 'rr 0
(5)
Let us now consider a general class of binary RNG models where the mean
of a binomial distribution is shifted by an amount:
= 0.5 (1+a)
A ps = pa - 0.5 = 0.5 a ,
where [a] is an RA strength parameter. That is to say, the properties of the RNG device are
modified by RA such that the mean probability of producing a hit has been shifted from 0.5
to pa. There are many other models that we might consider, but this particular one has been
proposed by the PEAR group. They report that if n = 200, [a] is approximately 0.001.10, 11
In other words, this model implies that individuals cause the mean probability
of producing a hit to be constant, regardless of the sequence length. (The logarithm term in
Equation 4 does retain an n-dependency through the zs term even though Apa does not.) RA
perturbs the device on a bit-by-bit basis that is independent of the number of bits in the
sequence. Figure 4 shows the result of evaluating Equation (4) for various values of [a]
compared to the MCE line.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
3 x 10 2~
p- 0.5 (1 + a)
2.0 x 10 3
Sequence Length (n)
8.0 x 10 5
FIGURE 4 ONE RA MODEL COMPARED WITH MCE
2.5 x 10 2
c. Theoretical Considerations for the IDS Model
Equation (2) represents the MCE condition in which no psychoenergetic
interaction is present. Because one premise of the IDS model is that no causal interactions
occur, there is only one term in Equation (2) that can possibly account for a non-causal, but
psychoenergetic effect. The 0.5 ln(pogo) and the -0.5 ln(n) terms are consequences of the
standard deviation for the unperturbed distribution of Op; therefore, the only remaining term,
the expected value for lnjzi, must contain the IDS considerations.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
In general terms, the expected value for lnizJ is given by
00
f ln(z) g(z,n) dz
in Iz1 _
(6)
CO
f g(z,n) dz
The function g(z,n) is the distribution that reflects a subject's ability to select subsequences,
leading to z-scores. For example, if a subject were always able to select subsequences
(regardless of sequence length) leading to z = 2.15, then g(z,n) contains no n-dependence
and is a Dirac delta function at z = 2.15. If g(z,n) is simply a Gaussian with a mean of zero
and a standard deviation of one, then Equation (6) reduces back to the null hypothesis. To
determine predictions under the IDS model, Equation (6) must be evaluated for different
assumptions for the function g(z,n).
Consider the case in which g(z,n) is not a function of the sequence length.
Then, g(z,n) = y(z). Equation (6) becomes a sequence length independent constant, which
can only affect the intercept of a straight line, regardless of any details of y(z). Figure 5
shows the RA hypothesis from Figure 4 and an example of the IDS case described above.
Suppose, however, g(z,n) does contain some n-dependencies. Even under
the IDS hypothesis, we might expect this to be the case. Recalling that IDS is a
psychoenergetic decision algorithm, we can imagine that better "decisions" result when a lot of
"information" is available, or the reverse: "decisions" are inaccurate if there is not enough
"information." It is possible that inaccurate "decisions" are made in the presence of too
much "information" as well.
Elementary information theory tells us that the amount of information in a
sequence of length [n] is proportional to [n] if the bits in the sequence are independent.
Therefore, we can equate "information" in the above paragraph with the sequence length.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
RA:
pa= 0.5 (1 + a)
IDS (n Independent)
2.5 x 10 2
2.0 x 10 3
Sequence Length (n)
a
0.01
0.0001
MCE
8.0 x 10 5
FIGURE 5 ONE IDS MODEL COMPARED TO MCE AND ONE RA HYPOTHESIS
We now examine the case in which g(z,n) retains a sequence length
dependency. To define g(z,n), we transform the coordinates similar to what was done in the
RA derivation. Let
Oz = z - zo
OZs = Za - Zo
OZa = Z - Za = AZ - Az,
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
where Za is the mean of a "shifted" Gaussian z-score distribution having a standard deviation
of o. Zo is the mean of the "unshifted" z-score distribution (i.e., zo = 0.0). Using this
transformation, g(z,n) becomes
-o.S Aza n) l1 `2
g(z,n) = e o (n)
Let zeta be given by
Then Equation (6) becomes
J
Aza
Cr
00
f -O.S~2
J 1n(a + ~s) e 4
(7)
. We notice that Equation (7) is of the same form as Equation (5)--including
their relative n-dependencies. Thus, depending upon the parameters involved in both
equations, we would be generally unable to differentiate between RA and n-dependent IDS.
The IDS model appears to have the sensitivity to differentiate between some
causal processes and some informational ones. In particular, the most obvious causal model
(i.e., the "interaction" is, on the average, the same for each bit in the sequence) contains
vastly different predictions than the most obvious IDS model (i.e., sequence length
independent of the z-scores).
There is another class of RA models that have attracted attention:
n
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
This particular form of [pa] leads directly to an n-independent zs = a. F(Ap5) in Equation
(5) becomes
f(Ap$) nz 11/2 1
This function is approximately equal to one for most actual cases (i.e., a = 0.1, n > 200).
We note here, that it may be possible to separate even this RA model from IDS at small
sequence lengths. Thus, Equation (5) reduces to a sequence length independent constant and
is equivalent to the IDS case of g(z,n) = y(z) in Equation (6). Therefore, RA (of this form)
and IDS are indistinguishable. All other forms of n-dependent RA contain a strong
n-dependence in the zs term.
To separate RA from IDS in this case, requires a different approach
altogether. Suppose that RNG experiments continue to produce significant results that appear
to be independent of all known physical parameters (e.g., source type--(3 decay, noise,
distance, shielding, etc.). Further, suppose that PRNG experiments continue to produce
significant results. We would then argue that IDS is the preferred choice for a mechanism
because:
? Precognition can be shown to be true by separate and fundamentally
different experiments.
? IDS can be shown to be possible (PRNG results).
? Results of one of our RNG experiments' suggest that human-mediated
RA must be able to switch on and off within 1 ms.
? There is no known force in nature that can interact equally with the
weak nuclear and electromagnetic forces.
To us, it would seem more parsimonious to assume that humans are more able to anticipate
the unperturbed natural fluctuations of an RNG, rather than generate "forces" that must
conform to such a set of attributes.
3. Analysis and Hypothesis Testing
The first step, after data reduction, is to determine if we are justified in continuing
with a linear analysis. To accomplish this, we calculate the linear correlation coefficient for
all data points [ln(n), lnjApl], and determine if there is a significant correlation (i.e., [r]
significantly different than 0.0) and, as a side interest, determine if the correlation is
significantly different than expected [i.e., Equation (3)].
If the correlation is significantly different than 0.0, we fit the data with a straight
line, where the intercept is determined at the average value of ln(n) for the data set. By
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
transforming the data about the average value of In(n) slope and intercept hypotheses testing
may be done separately. To determine if the observed line is significantly different from the
MCE line, we use an ANOVA technique.12 The F-ratios (from the ANOVA) for the two
tests are given below.
Let nk = ln(sk ), where sk is the sequence length for the kth data point. Let n be the average
value for the nk over [m] data points, or
m
n = m Z nk
k=1
The F-ratios are given by
F (slope) _
2 m 2 _ 2)
l
m (b - b') X (F nk - n
k=1 m
F (intercept) =
df1 = 1; df2 = (m - 2)
2
m (a-a')
df1 = 1; df2 = (n - 2)
where [a] and [b] are the intercept and slope for [m] data points, (ln(n), lnj 0 p1).
[a'] and [b'] are the intercept (at nk = n) and slope for the MCE line. A is given by
E M2 + ma 2 + b2 ~nk + 2 a b Fnk - 2 a YMk - 2 b YnkMk
k=1 k=1 k=1 k=1 k=1
Mk is the observed value of lnjOpl for the kth data point.
If the F-ratio for the slope is not significant, then we can conclude that the
z-scores do not have a sequence length dependency. Furthermore, if the F-ratio for the
intercept is significant, then there is evidence in support of the IDS model. Deviations from
this scenario will be discussed below.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
In 1985, Radin and May described a protocol that could be used as the pseudorandom
portion of a comprehensive test of the IDS model, and presented pilot data that appeared to
support the IDS model.13 We have modified and extended that original proposal to form the
basis of our pseudorandom investigation.
1. Justification for a Pseudorandom Number Generator Experiment
We have proposed an elaborate model (IDS) that predicts significantly different
results from those expected from an RA interaction. As part of a systematic investigation of
the validity of the model, we must determine if an IDS "interaction" can be demonstrated, in
principle, under conditions that preclude RA.
Because there has been no evidence to date to support the idea that computer
hardware is susceptible to a putative RA interaction, we assume that a purely pseudorandom
number generator (PRNG), which is seeded by a computer clock, constitutes an environment
that precludes RA.
2. PRNG Experiment Description
The primary concept for this experiment is to study IOpI as a function of sequence
length. To accomplish this, we design a PRNG experiment in which a single trial contains the
following steps:
? Select a sequence length, [n], from a limited menu of lengths.
? Collect [n] bits from a PRNG that has been seeded from a
computer system clock at the moment when the participant presses
a button.
? Calculate jApj and a z-score, and display the z-score to the
participant.
? Store raw data for later analysis.
Since the data will be analyzed in logarithmic form, we chose the following 10
sequence lengths because they are relatively evenly spaced when they are expressed as
logarithms, and they did not allow Ap = 0.0 (i.e., the lnjApI will remain finite): 101, 201,
501, 1001, 3501, 7001, 10001, 35001, 70001, and 100001. While we may have wanted to
explore larger sequence lengths than 100001, we were limited by the speed of the Sun
Microsystems Model 3/160-C computer, and by human factor considerations. The delay
between a button press and the display of the result was approximately 1.5 seconds. This
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
delay was constant in spite of the actual sequence length chosen, in order for a given trial to
maintain a double blind condition with respect to sequence length.
At the first trial of a series, the above sequences were placed in a random order
and stored as part of a data file. The sequence length, for a given trial, was taken in order
from this randomized list. Thus, the sequence length ordering was repeated every 10 trials.
The PRNG that was used (a Kendel shift register feedback algorithm) has been
studied extensively theoretically by Lewis14 and experimentally by May.' This algorithm meets
the accepted tests for "randomness" 15 and was checked further in this particular experiment
using control trials. The low order 15 bits from the system clock were used as seeds for the
PRNG.
For each trial, the seed, sequence length, number of ones in the sequence,
z-score, time (to the nearest second), and date of the trial were stored as part of a data file
for later analysis.
C. The Princeton Engineering Anomalies Research Data*
In order not to "pre-select" data to support (or not) a given hypothesis, we obtained all
of the data to date (September 1986) from one of PEAR'S best RNG participants. A subset
of these data has been reported previously, but we have a complete set as of this writing.1o
A single trial was defined as a continuous collection of binary bits from a "true" RNG
(i.e., the sequence was derived from the noise associated with a back-biased PN junction).
For the data under study there were two trial lengths, 200 bits and 2000 bits. To avoid
problems of a possible single dimensional "bias" in the hardware, a target bit was toggled at a
rate such that each new bit from the generator was compared to the one's (is) complement
of the previous target bit. Data were collected in two fundamental modes, manual and
automatic. In the manual mode, a single button press resulted in a single trial. During the
automatic mode, a single button press resulted in 50 consecutive trials. In the IDS formalism,
is the total number of bits (sequence length) resulting in a single button press is the
independent variable. Thus, there are only 4 allowed values for this independent variable:
200, 2000,, 10000, and 100000.
* We wish to express our appreciation to R. Jahn, R. Nelson and B. Dunne for providing
access. to their data and for their assistance in transferring it to our computer system.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
For all sequence lengths, data were collected as a function of "aim"
? PK+ -- The participant attempts to force matches between the target bit
and "response" bit from the generator.
? PK- -- The participant attempts to force mismatches between target bit
and "response" bit from the generator.
? BL -- The participant makes no attempt to modify the bits from the
generator. These data are referred to by PEAR as the baseline data.
The order of "aim" was determined in two modes: the volitional modes, in which the
participant chooses the order of the triad, and the random mode, in which an RNG
determines the order of the triad. Table 2 shows the data files that were used for the IDS
analysis. A few of the button presses produced results in which the number of matches
between target and response bits was exactly equal to one-half of the sequence length. Those
special cases are ignored in our analysis to avoid computing logarithms of zero. The analysis
will focus upon the specific aim regardless of volitional/random control of the aim.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
RAW DATA FOR OPERATOR 10--PRINCETON UNIVERSITY DATA
Sequence
Button
Number
Data Set
Length
Presses
z=0
200
5918
332
All PK ?
2000
15014
286
PK)
(A
10000
2065
13
_
1000000
597
3
200
3088
162
All PK +
2000
7219
131
(PK+)
10000
1028
9
1000000
299
1
200
2830
170
All PK -
2000
7795
155
(PK-)
10000
1037
4
1000000
298
2
200
1471
79
Volitional PK +
2000
4913
87
PK+)
(V
10000
330
4
_
1000000
105
0
200
985
65
Volitional PK
2000
5203
97
PK-)
(V
10000
328
1
_
1000000
100
0
200
1617
83
Random PK+
2000
2306
44
PK+)
(R
10000
698
5
_
1000000
194
1
200
1845
105
Random PK-
2000
2592
58
PK-)
(R
10000
709
3
_
1000000
198
2
200
0
0
All Baseline
2000
2451
49
(BL)
10000
1170
12
1000000
350
0
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
The results of the SRI pseudorandom number generator experiment and the PEAR
experiment will be discussed separately here and then compared in Chapter V, Conclusions
and Recommendations.
A. The PRNG Experiment Results
In 1985, Radin and May reported pilot results for two participants (I.D. 105 and I.D.
531) who were selected on the basis of past successful performances in similar tasks.' For
example, Participant 531 was the most significant contributor in our 1979 RNG experiment.
In the 1985 pilot experiment, Participants 531 and 105 contributed 500 and 298 trials,
respectively. The analysis showed that neither of the participants produced sequence length
dependencies different from MCE (i.e., a slope of -0.5). However, the analysis revealed that
both individuals showed independently significant evidence for IDS (i.e., the intercepts were
significantly above MCE at the p < 0.005 level for each participant). Thus, our tentative
conclusion from these data is that IDS appears possible, at least with these two participants.
During the FY 1986 program, we conducted the experiment in two phases: a screening
and an experiment phase. For the pilot phases, we asked 20 individuals to contribute 100
trials each under the protocol described above. All but 4 of them completed this task. For
availability reasons, the 4 participants contributed varying numbers of trials (less than 100).
We had decided to select 7 individuals from within the pilot group to participate in a formal
PRNG IDS experiment. The criterion for being included in the formal group was that the
participant had to produce a significant increase above MCE of the variance of the z-score
distribution over 100 trials (the MCE variance = 1.0).
Of the 16 participants who finished the 100 trial series, only one, 531, met the above
requirement (variance = 1.37, p < 0.008). The second best performer, however, produced a
variance = 1.21 (p < 0.07). Judging from the 1984 study, we would not expect to see a
significant intercept with only 100 trials, and none were observed.
While it is particularly interesting that Participant 531 maintains his/her consistent
performance, we felt that we should continue the pilot screening until we are able to select 7
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
significant participants. Thus, at this point, we do not have any results to report for the
formal experiment.
Using the exact binomial procedure described in Chapter III, we have computed
the MCE lines for the various data sets (see Table 3). These MCE lines were used as the
basis for the IDS analysis that follows.
Table 3
MCE FOR EACH DATA SET
Data Set
Variable
A PK
PK+
PK-
BL
V PK+
V PK-
R PK+
R PK-
Intercept (X=0)
-0.9514
-0.9502
-0.9525
-1.0923
-0.9379
-0.9498
-0.9575
-0.9545
Intercept (X-bar)
-4.8845
-4.8688
-4.8999
-5.4789
-4.8738
-4.9533
-4.8616
-4.8331
Slope
-0.5380
-0.5381
-0.5378
-0.5208
-0.5402
-0.5387
-0.5365
-0.5370
X-bar
7.263
7.233
7.293
8.420
7.242
7.397
7.219
7.164
We have analyzed the PEAR data from a "top down" perspective (i.e., beginning
with the most combined data and ending with the most condition specific data). The first
requirement from Chapter III is that we must determine if a linear analysis is appropriate for
these data. For all data sets, the linear coefficient was strongly significant (i.e., r s~ -0.6 for
all data sets) when compared to r = 0. Therefore, we are justified in continuing with the IDS
analysis. Table 4 shows the results of this analysis (the data set abbreviations are taken from
Table 2).
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
RESULTS OF THE ANALYSIS OF THE PEAR RNG DATA
Data Set
Variable
A PK
PK+
PK-
BL
V PK+
V PK-
R PK+
R PK-
Trials
23594
11634
11960
3971
6819
6616
4815
5344
Mean-Z
-0.0067
0.0180
-0.0308
-0.0097
0.0185
-0.010
0.018
-0.056
Z-Variance
1.018
1.016
1.019
1.035
1.002
1.025
1.036
1.012
p-of-Variance
0.024
0.110
0.070
0.062
0.454
0.083
0.038
0.274
Stouffer's Z
-1.031
1.946
-3.367
-0.613
1.516
-0.821
1.219
-4.124
p-of-Z
0.849
0.026
3.80(-4)
0.730
0.065
0.794
0.038
1.86(-5)
MCE
X-bar
7.263
7.233
7.293
8.420
7.242
7.397
7.219
7.164
Intercept
-4.8606
-4.8420
-4.8754
-5.4778
-4.8502
-4.9349
-4.8305
-4.8017
Slope
-0.5380
-0.5381
-0.5379
-0.5208
-0.5402
-0.5387
-0.5365
-0.5370
DATA
Intercept
-4.8606
-4.8469
-4.8737
-5.4580
-4.8574
-4.9338
-4.8319
-4.7989
Slope
-0.5317
-0.5471
-0.5157
-0.5276
-0.5449
-0.5167
-0.5486
-0.5144
F-Intercept
0.06
0.32
0.03
0.21
0.42
0.01
0.01
0.024
p-value
0.784
0.570
0.859
0.649
0.626
0.920
0.913
0.878
F-Slope
2.01
2.09
12.20
0.26
0.24
4.16
2.20
8.34
p-value
0.155
0.148
4.79(-4)
0.609
0.626
0.041
0.137
0.004
df-2
23592
11632
11958
3969
6817
6614
4813
5342
The meaning of MCE and data variables and their associated F-ratios have been discussed in
Chapter III. The z-score variables are included here in order to make comparisons with the
results published previously for part of these data.10 The mean z-score is the average value of
the z-scores for all the trials shown in each column; the MCE is zero. The variance for the
z-score is calculated for the same data; the MCE is one. Note that both of these quantities
were not calculated as absolute values. The Stouffer's z-score is the proper way of combining
all the z-scores to test against the MCE hypothesis: no psychoenergetic functioning. The
Stouffer's z-score is given by16
n
Z Zk
where [n] is the number of trials.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
We note that our analysis is consistent with PEAR's in that the difference between
the PK+ data (the participant was tasked to force more matches than MCE regardless of the
origin of the "aim" request--volitional or random) and PK- data as measured by the
Stouffer's z-score (zaiff = 3.75, p < 8.6 x 10-5) is highly significant. While the magnitude of
the effect is small, it is, nonetheless, persistent and statistically robust.
We are able to discuss the IDS analysis for all of the PEAR data by examining the
most deviant (from MCE) data set. Figure 6 shows the best fit line for the PK- data
compared to its MCE. We plot this particular data set because it produced the most
significant deviation (p < 4.79 x 10-4) from the MCE slope.
While the slope of the data line is significantly more positive than the MCE slope,
the intercept is not significantly different. In effect, the data line has rotated about its
intercept point (i.e., sequence length = 1470). Yet, the Stouffer's z-score indicates strong
evidence for some form of psychoenergetic functioning (p < 3.8 x 10-4) for this data set.
The resolution of this apparent inconsistency involves understanding a fundamental, and
unfortunate, problem with these PEAR data in general.
Their data were not collected to provide a specific test of our IDS model. Thus,
the sequence lengths that were chosen and, more importantly, the number of trials collected
at each sequence length, were not optimized for our test. In the extreme, if all the data were
collected at a single sequence length, our IDS analysis is completely inappropriate (i.e., the
IDS formalism requires testing as a function of sequence length). To first order, these PEAR
data suffer from the same problem. Sixty-five percent of the total data shown in Figure 6
were collected at a single sequence length (i.e., 2000). When we examine these data at each
sequence length, we find that the Stouffer's z-scores are -2.98, -0.80, -2.87, and -2.69 for
sequence lengths of 200, 2000, 10000, and 100000, respectively. Thus, most of the data for
this set are not significant, even though when they are combined across sequence lengths, they
are highly significant.
The situation described above is similar for all data sets; none of the data sets
produced significant intercepts. Because the data were not collected uniformly as a function
of sequence length, it is difficult to interpret the results of analysis. We feel that it is
premature to speculate upon forms of either RA or IDS models that can fit these data.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
2.5 x 10 2
2.0 x 10 3
Sequence Length (n)
8.0 x 10 5
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
V CONCLUSIONS AND RECOMMENDATIONS
In this report, we have developed a detailed model (IDS) for psychoenergetic
functioning. In particular, we have provided a mathematical formalism by which random
number generator experiments may be used to test the concept.
Preliminary results using pseudorandom number generators (PRNG), indicate that an IDS
ability appears possible. Many more data are required using PRNGs to confirm these
preliminary results. It is anticipated that by 3rd quarter, FY 1987, we will be able to
complete the PRNG experiment.
The PEAR data represent the largest amount of RNG data currently available. The
PEAR group have reported strong statistical evidence of psychoenergetic functioning within this
massive data base (i.e., 1.12 x 108 binary bits). It is unfortunate that the data were collected
in such a way that an IDS analysis is inconclusive.
We strongly recommend that RNG data be collected with an equal number of trials as a
function of sequence length. The protocol should be similar to the one in use in our PRNG
experiment, in which a double-blind condition is maintained with respect to sequence length.
We conclude with some speculation. Suppose, after many experiments of different
varieties, we could demonstrate that the philosophical concepts behind IDS were true. We
would call into question any experimental results from any discipline that claim
cause-and-effect relations based upon statistical inference.
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
1. May, E. C., Humphrey, B. S., and Hubbard, G. S., "Electronic System Perturbation
Techniques," Final Report, SRI International, Menlo Park, California (September 1980).
2. Tart, C. T., "Information Acquisition Rates in Forced-Choice ESP Experiments:
Precognition Does Not Work as Well as Present-Time ESP," JASPR, Vol. 77, No. 4,
pp. 293-310 (October 1983).
3. Nelson, R. D., Jahn, R. G., and Dunne, B. J., "Operator-Related Anomalies in Physical
Systems and Information Processes," JASPR, Vol. 53, No. 803, pp. 261-285 (April
1986).
4. Dunne, B. J., Jahn, R. G, and Nelson, R. D, "Precognitive Remote Perception,"
Technical Note PEAR 83003, Princeton Engineering Anomalies Research Laboratory,
Princeton University School of Engineering/Applied Science, Princeton, New Jersey
983)
5. Schmidt, H., "Precognition of a Quantum Process," Journal of Parapsychology, Vol. 33,
pp. 99-108 (1969).
6. Radin, D. I., May, E. C., and Thomson, M. J., "Psi Experiments with Random Number
Generators: Meta-Analysis Part 1," Proceedings of the Presented Papers of the 28th
Annual Parapsychological Association Convention, pp. 199-234, Tufts University,
Medford, Massachusetts (August 1985).
7. Schmidt, H., "A PK Test with Electronic Equipment," Journal of Parapsychology, Vol.
34, pp. 175-181 (1970).
8. Tart, C. T., "Laboratory PK: Frequency of Manifestation and Resemblance to
Precognition," Research in Parapsychology 1982, W. G. Roll, J. Beloff, and R. A.
White (Eds.) Scarecrow Press, Metuchen, New Jersey, pp. 101-102 (1982).
9. May, E. C., Radin, D. I., Hubbard, G. S., and Humphrey, B. S., "Psi Experiments with
Random Number Generators: An Informational Model," Final Report, SRI Project
8067, SRI International, Menlo Park, California (October 1985).
10. Nelson, R. D., Dunne, B. J., and Jahn, R. G., "An REG Experiment with Large Data
Base Capability, III: Operator Related Anomalies," Technical Note PEAR 84003,
Princeton Engineering Anomalies Research Laboratory, Princeton University School of
Engineering/Applied Science, Princeton, New Jersey (1984).
11. Jahn, R. G., "The Persistent Paradox of Psychic Phenomena: An Engineering
Perspective," Proceedings of the IEEE, Vol. 70, No. 2, pp. 136-170 (1982).
12. Cooper, B. E., Statistics for Experimentalists, pp. 219-221, Pergamon Press Inc.,
Maxwell House, Fairview Park, Elmsford, New York (1969).
13. Raclin, D. I. and May, E. C., "Testing the Intuitive Data Sorting Model with
Pseudorandom Number Generators: A proposed Method," Proceedings of the
Presented Papers of the 29th Annual Parapsychological Association Convention, pp.
539-554, Sonoma State University, Rohnert Park, California (August, 1986).
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3
14. Lewis, T. G. and Payne, W. H., "Generalized Feedback Shift Register Pseudorandom
Number Algorithm," JAC, Vol. 20, No. 3 (July 1973).
15. Lewis, T. G., Distribution Sampling for Computer Simulation, Lexington Books,
Lexington, Massachusetts (1975).
16. Rosenthal, R., Meta-Analytic Procedures for Social Research, Sage Publications, Beverly
Hills, California (1984).
Approved For Release 2000/08/07 : CIA-RDP96-00787R000500140001-3