Electrodermal Presentiments of Future Emotions
DEAN I. RADIN
Institute of Noetic Sciences
101 San Antonion Rd.
Petaluma, CA 94952
Abstract—In previously reported double-blind experiments, electrodermal
activity (EDA) monitored during display of randomly selected photographs
showed that EDA was higher before emotional photos than before calm photos
(p¼0.002). This differential effect, suggestive of precognition, was dubbed
‘‘presentiment.’’ Three new double-blind experiments were conducted in an
attempt to replicate the original studies using the same basic design, but with
new physiological monitoring hardware, software, stimulus photos, subject
populations, and testing environments.
The three replications involved 109 participants who together contributed
3,709 trials. The new studies again showed higher EDA before emotional
photos than before calm photos ( p¼0.001). All four experiments combined
involved 133 participants and 4,569 trials; the associated weighted mean effect
size (per trial) was e¼0.064 60.015, over 4 standard errors from a null effect.
As a more general test, presentiment predicts a positive correlation between
pre-stimulus EDA and independently assessed emotionality ratings of the photo
targets. The observed correlation across all four experiments was significantly
positive ( p¼0.008).
Consideration of alternative explanations, including expectation, sensory
cues, hardware or software artifacts, inappropriate analyses, and anticipatory
strategies, revealed no suitable candidates that could systematically generate
the observed results. This series of four experiments, supported by successful
replications conducted by other investigators, appears to demonstrate a small
magnitude but statistically robust form of precognition in the human autonomic
Keywords: electrodermal activity—precognition—autonomic nervous system
Many people have experienced intuitive hunches or forebodings about future
events that later turned out to be correct. Most such hunches can be attributed to
unconscious inferences, others are undoubtedly coincidences, instances of
selective memory, or due to forgotten expertise. However, sometimes a hunch
seems so intrinsically unlikely and yet turns out to be valid, that one wonders
whether such experiences, often on the edge of conscious awareness, might
involve perception of future information. In a series of experiments designed to
test this idea under double-blind conditions, I explored whether the human
Journal of Scientiﬁc Exploration, Vol. 18, No. 2, pp. 253–273, 2004 0892-3310/04
autonomic nervous system would be able to correctly anticipate exposure to
randomly selected calm or emotional photographs (Radin, 1997).
Those initial studies provided evidence for what I called presentiment. I used
this term, in contrast to precognition, as the latter implies conscious awareness
(i.e., pre-cognition) of future events. Publication of the initial results in this
journal prompted a number of other researchers to attempt to replicate the effect.
Some of the replications focused primarily on electrodermal activity (EDA), as
in the original studies (Bierman, 2000; Bierman & Radin, 1997, 1998; Norfolk,
1999; Parkhomtchouk et al. 2002; Spottiswoode & May, 2003; Wildey, 2001).
Others explored different physiological measurements, including functional
magnetic resonance imaging (MRI; Bierman & Scholte, 2002) and heart rate
variability (McCraty, 2002). All of the replications reported results consistent
with the original findings.
This paper reviews the results of presentiment experiments that I conducted
from 1996 through 2000, only the first of which was previously published
(Radin, 1997). All these experiments were primarily proof-oriented rep-lications
using different hardware and software implementations, subject populations,
environmental conditions, and photo stimuli. Experiment 1 was the initial series
of tests, Experiment 2 was a straightforward replication, Experiment 3 was
conducted as a proof-of-principle demonstration in an industrial research
laboratory, and Experiment 4 explored the use of a custom-designed
psychophysiological monitoring device. To avoid repeating descriptions of
common design elements, a brief overview of the general procedure will be
described first and then appropriate details added as each experiment is
Basic Experimental Procedure
A participant (P) is seated in front of a computer monitor displaying a black
screen. The experimenter attaches EDA electrodes (Ag-AgCl, 8 mm diameter)
to the volar surfaces of the distal phalanges of the index and middle fingers of
P’s non-dominant hand. The electrodes are secured with Velcro straps, and
electrode gel is used to enhance contact with the skin. After the experimenter
ensures that the physiological hardware is recording the EDA data properly, P
is instructed to press a button at will. When this occurs (see Figure 1), the
computer waits 5 seconds, then it selects a photo at random from a large pool of
possibilities, displays it for 3 seconds, and then the screen goes blank again for
10 seconds. After the 10-second ‘‘cool-down’’ period, the computer instructs P to
press the button again when ready to begin the next trial. A typical session may
last 30 minutes, during which time 40 trials are run. The design is double-blind
in the sense that neither P nor the experimenter know in advance which photos
will be displayed in a given session, or in what sequential order.
254 D. I. Radin
The concept of presentiment postulates that present physiological states are
correlated with near-term future experiences. When those futures involve
emotional experiences, as evoked through the use of photographs of varying
emotionality, the correlation is postulated to be detectable as a present-time
arousal of the autonomic nervous system. This idea leads to three progressively
more general hypotheses.
Hypothesis 1. This hypothesis states that EDA before display of emotional
photos will be larger than EDA before display of calm photos. This differential
prediction is tested by partitioning all trials in an experiment into emotional and
calm subsets of equal size. These subsets are formed based upon prior
assessments of the emotionality of the target photos. The probability of the
resulting differences in EDA is formally tested using a nonparametric statistical
method known as randomized permutation analysis (Blair & Karniski, 1993).
Hypothesis 2. This hypothesis states that as the contrast between emotional
and calm trials increases, the magnitude of the presentiment effect will also
increase, and vice versa. This is tested by sorting all trials according to their pre-
assessed emotionality ratings, then comparing the top 1% most emotional vs. the
1% most calm trials, then the top 2%, and so on up to 50% (which is then the
same as Hypothesis 1). No specific prediction is made for the emotional contrast
level that would show the largest effect, but it is expected that the peak might
fall somewhere between 5% and 25%. This is because a few photos with very
high or very low emotionality assessments would provide strong emotional
contrasts, but at the cost of low statistical power. And an assessment made with
Fig. 1. Illustration of basic experimental protocol.
Electrodermal Presentiments 255
many photos of varying emotionality would provide greater statistical power,
but at the cost of weaker emotional contrast.
Hypothesis 3. This hypothesis generalizes the first two hypotheses and
predicts a positive correlation between the pre-assessed emotionality ratings and
changes in EDA prior to the stimuli.
Method of Analysis
Electrodermal activity (EDA) refers to variations in electrical resistance of the
skin. These fluctuations are due to activity of the eccrine sweat glands, which are
activated by the sympathetic nervous system (Bouscein, 1992). The specific
form of EDA used in these studies was skin conductance level (SCL). SCL has
been used as the principal physiological measurement in these and in many of
the replication attempts primarily because SCL is a conveniently measured and
widely used indicator of overall autonomic activity. All SCL data in these
experiments were uniformly analyzed using a simplification of the technique
employed in the initial studies (Radin, 1997).
An SCL ‘‘sample’’ refers to an instantaneous measurement of absolute skin
conductance. For the sake of exposition, let us assume in this discussion that one
trial is 18 seconds in length (as shown in Figure 1) and that the sampling rate is 5
Hz; thus each trial consists of 90 samples. Let us further assume that an experiment
consists of 20 participants, each of whom runs 30 trials, for a total of 600 trials.
To analyze the results of an experiment, each SCL sample in each trial is
normalized as z
, where in our example irefers to samples 1–90, j¼
1 to 600 trials, x
refers to the raw SCL value for sample iin trial j,m
average of the 25 samples in the 5-second pre-stimulus period in trial j(i.e., from
the starting button press to just before the photo appears), and s
is the standard
deviation of those same 25 samples. Then, all normalized trials z
are clamped to
zero after the button press as p
), where z
refers to the first
normalized SCL sample after the button press in trial j, and iranges across all
values are thus changes in normalized SCL (i.e., SCL). Normalized
SCL is of interest, rather than absolute SCL, because otherwise a few Ps with
highly labile SCL signals would overwhelm the data from other, less labile Ps.
And change in SCL is of interest because we are interested in event-related
responses, i.e., how autonomic arousal fluctuates from the moment P decides to
begin each trial, rather than in P’s general level of sympathetic arousal or in
spontaneous fluctuations in SCL.
To determine the statistical likelihood of the differences in SCL observed
before emotional and calm trials, randomized permutation analysis was
employed as follows (Blair & Karniski, 1993):
1. The value S
was determined for each trial j, where the sum was
taken over pre-stimulus samples i¼1 to 25 per trial, starting just after the
initiating button press and ending just before stimulus onset.
256 D. I. Radin
2. All trials jwere then sorted by each trial photo’s pre-assessed emotionality
ratings, in ascending order.
3. The sorted list in step 2 was divided into two equal subsets such that trials
1toj/2 were defined as calm and trials ( j/2) þ1tojwere defined as
4. The difference D¼P
was determined, where c¼1toj/2 and
5. The order of the emotionality ratings were randomly scrambled and steps
2 through 4 repeated 1,000 times, each time keeping track of the
difference, say D9. Then the mean l
and standard deviation r
of all the
D9values were determined.
6. Now z¼(Dl
and its associated one-tailed probability was
calculated. This pvalue represents the probability of observing a difference
as large or larger than the observed D.
This process was then repeated for the 1% most emotional vs. 1% most calm
trials, then the 2% most emotional and calm trials, etc., up to the 50% split as
described in the above six steps. In this way, a total of 50 zscores were
generated, one for each ‘‘emotionality contrast’’ percentage from 1% to 50%.
Then effect size per experiment was calculated as e¼z/ﬃﬃﬃﬃ
p, where Nwas the
number of trials in each experiment.
Experiment 1a used a 66 Mhz desktop personal computer (PC) to control the
test. Experiment 1b was similar to 1a except that it presented the stimulus photo
for 1 second rather than 3 seconds. Experiment 1c was identical to 1a except that
the experiment was run on a 75 Mhz notebook PC. Experiment 1d investigated
combinations of three simultaneous physiological measures, including SCL.
Because tests 1a and 1c used identical designs, data from those two tests were
pooled for the present analysis (N¼900), and because the other two tests were
primarily designed to explore variations on a theme, those data (N¼280 trials)
were excluded. It is worth mentioning that both of the excluded datasets
individually provided positive evidence for the presentiment hypothesis (Radin,
Participants. Experiments 1a and 1c involved 24 Ps recruited from friends,
staff, faculty, and students visiting the Consciousness Research Laboratory,
University of Nevada, Las Vegas (UNLV). Ps were restricted to adult
volunteers, and all were required to sign an informed consent explaining that
photos portraying a wide range of emotions would be displayed. Immediately
before starting the experiment, Ps were asked to provide a verbal affirmation to
Electrodermal Presentiments 257
Procedure. P was asked to sit in an office chair approximately 2 feet in front
of a color computer monitor. The experimenter attached electrodes to record
SCL (as previously described), heart rate, and peripheral blood flow, the latter
two using a photoplethysmograph on the fingertip of the fourth digit of the non-
dominant hand. These signals were monitored by a J&J Engineering (Poulsbo,
WA) Model I-330 physiological data acquisition system (SCL digitized with 12-
bit resolution using the constant voltage method at 0.3 volts; conductance
measures range from 1 to 100 lS with 60.05% accuracy). The physiological
monitor and experiment protocol were controlled by a DOS-based program
written in Microsoft QuickBasic 4.5 by the author. SCL samples were recorded
at 5 Hz.
P was instructed to rest the hand with the electrodes in her lap (the female
gender will be used hereafter to avoid awkward phrasing). In her dominant hand,
she held a computer mouse with her index finger resting on the left mouse button.
When ready to begin each trial, she pressed the mouse button and waited to see
a picture on the computer monitor. After the button press, the computer selected
a target photo at random, there was a 5-second delay during which the screen
remained black, the selected picture was displayed for 3 seconds (as illustrated in
Figure 1), and then the screen went black again. After a 10-second cool-down
period, a message appeared on the screen alerting P to begin the next trial by
pressing the mouse button at will. SCL was continuously monitored during each
trial but not between trials in this experiment. P viewed 41 pictures in a single
session, one picture at a time. The experimenter talked P through the first trial to
ensure that the procedure was understood, and then the remaining 40 trials were
conducted by P alone. Only the last 40 trials were used for subsequent analysis.
If P needed to stretch or move between trials, she was asked to do so as
needed and then settle down before continuing. After P indicated that she
understood the procedure, the experimenter retired behind an opaque wall-
screen and P conducted the remaining 40 trials unobserved and at will. To
enhance the display contrast of the stimulus pictures during the experiment, and
to reduce possible electrical interference with the monitoring equipment, the
overhead fluorescent lights were turned off and a dim red incandescent lamp (10
watts) was turned on. The laboratory was air conditioned to approximately 728F
and humidity levels were generally dry.
Targets. At the beginning of each trial, the QuickBasic 4.5 pseudorandom
number generator (PRNG) was re-seeded with the computer’s clock time at the
moment of the button press, and the target was selected out of a pool of 120
digitized color photographs. Calm targets included photos of landscapes, nature
scenes, and people; emotional targets included erotic, violent, and accident
scenes. Most of the calm pictures were selected from a Corel Professional Photo
CD-ROM. Most of the emotional pictures were selected from photo archives
accessible over the Internet.
Pictures were displayed in color, at 600 3800 screen resolution, in a screen
area about 6 inches wide by 4 inches high. If during a session a given target was
258 D. I. Radin
randomly selected again, another picture with the closest pre-assessed subjective
emotionality rating—but not previously shown in that session—was selected in
its place. Participants were unaware of the size of the target pool, or the ratio of
calm to emotional photos, thus selecting targets without replacement would not
have biased their expectations about the upcoming targets. This target-selection
strategy was used to help ensure that each successive trial was as novel as
possible. Novelty was important because the experiment was designed to evoke
an orienting response (Kimmel et al., 1979), and novelty is one factor known to
stimulate such responses.
To provide the independent subjective assessments of the target pictures, three
men and three women were independently asked to examine each of the 120
pictures presented in randomized orders, and to rate each picture from 1 (calm)
to 5 (emotional). Their average ratings were used to assign an emotionality value
to each photo.
Results. Figure 2 graphically summarizes the results of Experiment 1, split
into two equally sized datasets, one calm and the other emotional according to
the target emotionality ratings. The vertical line in the graph shows the moment
of stimulus onset. Skin conductance is expected to react 2 to 3 seconds after an
emotional stimulus, and this response is evident in Figure 2. The difference in
SCL before emotional and calm stimuli ( p¼0.002) is the hypothesized
Fig. 2. Average normalized change in skin conductance level (SCL) for calm and emotional trials
in Experiment 1. A total of 860 trials were contributed by 24 participants. The pre-stimulus
period is indicated by negative seconds; stimulus onset was at time 0; stimulus offset at 3
seconds. Randomized permutation analysis indicated that the observed difference in pre-
stimulus SCL is associated with z¼2.92, p¼0.002 (one-tailed).
Electrodermal Presentiments 259
Participants. Participants in this study included 50 volunteers who ran the
test at the Consciousness Research Laboratory, UNLV, and six who ran the same
test at Interval Research Corporation, Palo Alto, CA, using the same
physiological equipment and (nearly) the same stimulus pool. Most participants
at UNLV ran 40 trials per session; those at Interval ran 30 trials per session. The
testing environment at UNLV was the same as described in Experiment 1; trials
at Interval were conducted on a 300 MHz desktop PC. The testing environment
at Interval was a small office where participants sat in front of a computer
monitor at a desk, in an ordinary office chair. During the experiment, office
lights were turned off and windows were blocked. The office was air
conditioned to about 728F and humidity levels were comfortably moderate.
Procedure. The same electrodes, physiological hardware, and software from
Experiment 1 were used in this study. However, unlike in Experiment 1, where
the target photo was determined by the time of each button press used to initiate
a trial, in this study the PRNG selected the target immediately prior to its
display, thus ensuring that when P started each trial, the future emotional
experience really was in the future, and not determined at the beginning of
To accomplish this, the computer program created a new seed-number
immediately prior to stimulus onset by adding the value of the PC’s internal
clock time to the instantaneous values of three continuously monitored
physiological signals: SCL, heart rate, and peripheral blood flow. This sum
formed a seed-number used to initialize the PRNG, which was used in turn to
select the target photo. The physiological signals were used to help form the
PRNG seed-number because the clock time component of the seed-number was
more or less determined even 5 seconds after the button press, whereas the
instantaneous values of the physiological variables were not. It should be noted
that the correlation between seed-numbers and the target photos selected by this
PRNG was effectively zero (r¼0.0007); thus P could not infer the identity of
the target even if the PC’s clock time and the instantaneous physiological
measurements were explicitly known (these values were not available to P).
After the target photo was selected, it was retrieved from the PC’s hard disk
and displayed for 3 seconds. Through this process, from the moment a button
press initiated a trial to just before the stimulus was displayed, the target was not
yet determined. Nor were there sounds due to movements of the computer’s hard
disk, or electromagnetic changes in the computer monitor display, or any other
sensory hints that might have provided clues about the identity of the upcoming
In addition, even though participants did not know the size or composition of
the target pool, target photos were randomly selected in such a way as to prevent
statistical hints about the future targets from accumulating over the course of
a session. As in Experiment 1, to avoid repeating a stimulus photo that had
260 D. I. Radin
already been selected, the computer noted the selected target’s emotionality
rating and randomly selected another picture with the nearest rating, and used
that instead. In this way, the probability of observing a calm or emotional target
on any given trial was held constant throughout the experiment, and no targets
Targets. Thirty new pictures were added to the photo pool from Experiment
1, bringing the total to 150. Five men and five women were asked to
independently examine the pictures, one at a time, in random order. The rating
dimension consisted of a 100 point scale, and the rating method asked each
person to view a picture on a computer screen and move a pointer across a sliding
scale to indicate his or her assessment. Each person’s assessments were
individually normalized into standard normal deviates (z scores), and z scores
per target were combined across all 10 people to provide a distribution of target
Results. Figure 3 shows the results of Experiment 2. As in Experiment 1,
SCL was slightly higher for emotional vs. calm photos ( p¼0.11); this rise
started immediately after P pressed the button used to initiate each trial.
This experiment used new hardware and software, a new photo pool, and three
new participant populations and testing environments. Also, rather than running
40 trials per participant, 30 trials were used to alleviate the physiological
accommodation observed in some participants in the first two experiments.
Fig. 3. Average normalized change in skin conductance level (SCL) for calm and emotional trials
in Experiment 2. This experiment consisted of a total of 2,059 trials, contributed by 56
participants. Randomized permutation analysis indicates that the difference in pre-stimulus
SCL curves is associated with z¼1.23, p¼0.110 (one-tailed).
Electrodermal Presentiments 261
Participants. Forty-seven volunteers were recruited from staff and visitors to
Interval Research Corporation, Palo Alto, CA, and from participants at seminars
held in Port Antonio, Jamaica, and Esalen Institute in Big Sur, CA.
The test environment at Interval Research was described in Experiment 2. The
test environment for trials collected in Jamaica was in a closet in a small cottage
next to the ocean. The lights were turned off and there were no windows nearby,
so the room was dimly lit. Participants sat in a straight-back chair in front of
a laptop PC screen. The test environment at Esalen Institute was a bedroom in
a house overlooking the ocean. Window shades in the room were drawn,
dimming ambient illumination. Participants sat in a straight-back chair in front
of a laptop screen. At Esalen, the atmosphere was moderately humid and the
temperature was about 758F. In Jamaica, the atmosphere was very humid and the
temperature was about 858F.
Procedure. Trials contributed at Interval were conducted on a 300 MHz
desktop PC; the other trials were run on a 233 MHz laptop PC. In both cases the
Windows NT4.0 operating system was used. The physiology equipment used was
a J&J Engineering Model I-330C2 (six channel, battery-powered, psychophys-
iological monitor with 12-bit resolution, measuring SCL in the range 1 to 100 lS
with an accuracy of 60.5%, using the constant current method with 2.5 lA for
excitation). Electrodermal measurements were continuously collected at 10 Hz for
the duration of the session, including between trials. The program used a 6-second
pre-stimulus period, in contrast to the 5-second period used in Experiments 1 and 2.
A new controlling program was written for this experiment in Microsoft
Visual Cþþ 5.0. The program allowed use of either the Cþþ PRNG or a noise-
based truly random number generator (RNG) to select the targets. All trials run
at the Interval office used an Orion RNG (ICATT interactive media,
Amsterdam, The Netherlands). The Orion is an electronic noise-based,
truly random generator that passes Marsaglia’s Diehard test, a gold-standard
randomness testing suite (Marsaglia, no date). All trials run in Jamaica and
Esalen used the Cþþ PRNG. Targets were selected by re-seeding the PRNG with
the computer system’s clock (with 1 millisecond resolution) or by sampling from
the true RNG, immediately before the stimulus picture was displayed.
Targets. The target pool consisted of the 80 most calm and the 40 most
emotional pictures from the International Affective Picture System (IAPS,
Bradley et al., 1993; Ito et al., 1998), where ‘‘calm’’ and ‘‘emotional’’ were
defined by the emotionality ratings (averaged across gender) that accompany the
IAPS picture set. In an attempt to further enhance the contrast between the
emotional and calm targets, participants wore headphones that played one of 20
randomly selected noxious sounds for 3 seconds during presentation of
emotional pictures (i.e., screams, sirens, explosions, etc.). Calm pictures were
presented in silence.
The use of a 2:1 ratio of calm to emotional photos would seem to add noise to
the analysis of Hypothesis 1 (which is therefore a conservative test), since that
analysis sorts targets by their pre-assessed emotionality ratings and splits the
262 D. I. Radin
data into two equal subsets. But in practice, people show strong idiosyncratic
responses to photos, and this significantly blurs a nominal dichotomy into
a continuous scale of emotionality. Thus, splitting the trials into two equal
subsets does not introduce as much noise as one might expect.
Results. Figure 4 shows the results of Experiment 3. In this study because
data were recorded continuously, it was possible to evaluate and graph SCL
values before the trial-initiating button press. This graph shows that mean SCL
levels were indistinguishable before the button press, but as in the two previous
studies the signals began to differentiate according to the future stimulus starting
immediately after the trial began ( p¼0.0004).
Participants. Participants in this study were recruited from visitors to the
Boundary Institute, Los Altos, CA. The test environment was similar to that at
Interval Research except that the controlling computer was a 300 MHz laptop
PC. All tests were conducted in an office where participants sat in an office chair
in front of the laptop at a desk. The office lights were turned off and a window
was shaded by miniblinds. The office was air conditioned to about 728F and
humidity levels were comfortably moderate.
Procedure. The physiological equipment was an experimental eight-channel
SCL monitor custom-designed by a team at Interval Research Corporation (the
device used the constant current excitation method at 2.5 lA and had 16-bit
Fig. 4. Average normalized change in skin conductance level (SCL) for calm and emotional trials
in Experiment 3. Each trial was initiated at 6 seconds; stimulus onset was at 0 seconds. A
total of 1,410 trials were contributed by 47 participants. Randomized permutation analysis
indicates that the difference in pre-stimulus curves is associated with z¼3.34, p¼0.0004
Electrodermal Presentiments 263
resolution). Software drivers for the device were written at Interval Research
Corporation in Microsoft Visual Basic 5.0; the experiment itself was written in
Visual Basic 5.0 by the author. SCL measurements were continuously collected
at 10 Hz for the duration of the session. The experimental design employed a
5-second pre-stimulus period.
The SCL electrodes were also custom-made. Each electrode consisted of
a gold dot, 10 mm in diameter, deposited on a flexible printed circuit board
material. Two such electrodes were attached to the first two fingers without
electrode paste, using a Velcro fastener. The experimental electrodes and the
physiological monitor were designed with the goal of eventually creating an
inexpensive research package that would allow independent investigators to
easily and quickly conduct replications of the presentiment experiment. This
goal was eventually abandoned because it was felt that if positive evidence for
presentiment were obtained using these devices, it would be criticized on the
basis that the data were collected using non-standard equipment and electrodes.
Targets. As in Experiment 3, the IAPS photos were used, but in this case the
targets were selected uniformly at random, with replacement, from the entire set
of some 700 IAPS photos (the size of the IAPS pool changes with each new
update). Targets were selected by the Visual Basic 5.0 PRNG based on the
system clock time immediately after the last SCL sample of the pre-stimulus
period had been collected.
Results. Figure 5 shows the results of Experiment 4. Note that unlike the
previous three experiments, this study showed a positive post-stimulus response
Fig. 5. Average normalized change in skin conductance level (SCL) for calm and emotional trials
in Experiment 4. A total of 240 trials were contributed by six participants. Randomized
permutation analysis indicates that the difference in pre-stimulus curves is associated with
z¼0.59, p¼0.28 (one-tailed).
264 D. I. Radin
not only for emotional trials, but also for calm trials. As postulated by
Hypothesis 2, the weak contrast between post-stimulus calm and emotional trials
may have accounted for the weak differential results observed in the pre-
stimulus period ( p¼0.28).
Table 1 summarizes key elements of the four experiments. In Experiments 1a
and 1c a total of 900 trials were run, but the file containing data from one session
of 40 trials was corrupted, so a total of 860 trials were available for analysis. In
Experiment 2, fifty people contributed 40-trial sessions at UNLV and six people
contributed 30-trial sessions at Interval Research Corporation. A small number
of trials did not record properly, leading to a total of 2,059 usable trials. Most of
the failures were due to one of the SCL electrodes spontaneously breaking
contact with the participant’s skin; others were due to equipment failures
possibly caused by power spikes or by the PC’s operating system freezing for
unknown reasons. With use of new physiological equipment, uninterruptible
power supplies, and newly designed software, all sessions run in Experiments 3
and 4 were recorded properly and all data were analyzable.
Table 2 shows the results of testing Hypothesis 1 for each experiment, for the
three replication experiments combined, and for all four experiments combined.
The original experiment produced a significant effect size of e¼0.100, p¼0.002.
The three replication experiments resulted in a combined effect size of about half
that magnitude, but with four times as many trials it was statistically more
significant, e¼0.049, p¼0.001. Over all four experiments, the combined effect
size was e¼0.060, p¼3310
. The overall mean weighted effect size (weighted
by number of trials per experiment) was e¼0.064 60.015, p¼1.3 310
Figure 6 shows the weighted mean effect size (weighted per number of trials)
for both pre-stimulus and post-stimulus SCL data combined across the four
Informational Summary of the Four Experiments
Experiment Location Trials Participants
1a,c UNLV 860 24 20 or 40 120 custom Conductance J&J 1330
2 UNLV & IRC 2,059 56 30 or 40 150 custom Conductance J&J 1330
3a IRC 570 19 30 IAPS Resistance J&J 1330C2
3b Esalen 180 6 30 IAPS Resistance J&J 1330C2
3c Jamaica 660 22 30 IAPS Resistance J&J 1330C2
4 Boundary 240 6 40 IAPS Resistance Custom
Total 4,569 133
Electrodermal Presentiments 265
experiments and for emotionality contrasts ranging from 1% to 50%. As
a measure of post-stimulus activity, the sum of SCL values from 2.5 to 3.5
seconds after stimulus onset was used.
At the 50% emotional vs. calm contrast level (i.e., Hypothesis 1) the post-
stimulus effect size was more than 20 standard deviations from chance. For
Hypothesis 2, the post-stimulus effect size was observed to peak at the 5%
emotionality contrast and then decline with increasing percentage. And for
Hypothesis 3, the expectation that more emotional targets would result in larger
post-stimulus EDA was confirmed with the correlation r¼0.28, t¼19.7,
N¼4,569, p’0. The strong positive effect size for post-stimulus SCL
indicates that an orienting response was observed as expected for emotional
What is unexpected is that the weighted mean effect size for pre-stimulus
SCL was significantly above zero for all levels of emotional contrast beyond
5%. To explore whether these results were possibly driven by a few outliers, an
analysis was performed based upon the median values of pre-stimulus SCL,
rather than the sum. This resulted in an overall mean weighted effect size e¼
0.049, about 2.9 standard errors above chance, and an unweighted Stouffer Z
score ¼3.31 ( p¼0.0005). This indicates that the observed results are not due to
outliers, but reflect a generalized phenomenon.
The peak weighted effect size of e¼0.23 at a 1% contrast was 2.2 standard
errors above chance, but given the 50 overlapping and therefore dependent
statistical tests, this is not a particularly persuasive result. The next highest peak
of e¼0.15 at an 8% emotional contrast level was 4.05 standard errors above
chance, which is more interesting. The weighted mean effect size then declines
to e¼0.06 at the 50% contrast. Thus, in accordance with Hypothesis 2, higher
emotional contrasts did seem to result in higher differential effects.
As the most general test, Hypothesis 3 predicted a positive correlation
between the pre-assessed emotionality ratings vs. pre-stimulus SCL levels.
The resulting correlation was small in magnitude, but as predicted it was
Results of Experiments
E1 E2 E3 E4 E2–4 E1–4
z2.92 1.23 3.34 0.59 2.98 4.04
N860 2059 1410 240 3709 4569
p0.002 0.11 0.0004 0.28 0.001 0.00003
e0.100 0.027 0.089 0.038 0.049 0.060
Note: Results of each experiment (E), experiments 2–4 combined, and all four experiments
combined, in terms of zscore, number of trials, pvalue, and effect size, e.
266 D. I. Radin
significantly positive: r¼0.04, t¼2.42, N¼4,569, p¼0.008. Support for
Hypothesis 3 indicates that prior to viewing randomly selected emotional
photos, participants’autonomic nervous systems not only responded more than
before calm photos, but responded more in proportion to the independently
assessed emotionality levels of the future stimuli.
In individual experiments, for Experiment 1 r¼0.04, t¼1.28, N¼860, p¼
0.10. For Experiment 2, r¼0.03, t¼1.29, N¼2059, p¼0.10. For Experiment 3,
r¼0.05, t¼1.93, N¼1,410, p¼0.03. And for Experiment 4, r¼0.02, t¼
0.34, N¼240, p¼0.64.
Commonly proposed alternative explanations for the observed results include
(1) sensory and statistical cues about the upcoming targets; (2) data collection,
measurement, and/or analytical artifacts; (3) selective reporting biases; (4)
participant or experimenter fraud; and (5) conscious or unconscious anticipatory
strategies. All of these factors were considered in the process of designing and
running these experiments. Each explanatory category is addressed in turn.
Sensory or statistical cues. If the computer’s hard disk retrieved the target
photo immediately after the button was pressed to begin a trial, and if the calm
targets differed from the emotional targets either in terms of where they were
Fig. 6. Weighted mean effect size and one standard error bars for post-stimulus and pre-stimulus
SCL, in terms of emotional contrast percentage. An emotional contrast percentage of 1%
represents those trials with the most emotional 1% and most calm 1% of the trials, or 2% of
all available trials. The 50% contrast level includes all trials.
Electrodermal Presentiments 267
located on the disk or their size, then the participant might have learned to
associate the computer’s hard disk sounds with different upcoming targets. To
avoid such possibilities, in all experiments the targets were not retrieved off the
hard disk until immediately before they were displayed. Also, recall that in
Experiments 2 through 4 the targets were not even selected until immediately
before they were displayed. In short, the software in all of these experiments was
designed to ensure that there were no differences in sounds, displays, or other
physical cues until just before the target was displayed. Given that evidence for
presentiment effects in these studies began 5 to 6 seconds prior to the stimulus,
sensory cues are not a viable explanation.
Statistical cueing might occur if the sequence of targets was non-random. To
circumvent this possibility, first the PRNGs and true RNG used in these
experiments were checked for sequential randomness before they were used, and
all proved to be adequately random under long-term calibration conditions.
Second, the majority of participants in these tests ran a single session of 20 to 40
trials, an insufficient number of trials for most people to learn sequential biases,
unless the biases are extreme. And third, examination of the actual sequence of
targets used in these experiments showed that the autocorrelations to lag 15 were
all in alignment with chance expectation.
Hardware, software, and analytical artifacts. To avoid the possibility that
any given implementation of the experiment might have introduced hardware or
software-specific artifacts, three different physiological monitoring systems,
three software programs and PC operating systems, many different computers,
and four types of random number generators (three PRNGs and a true RNG)
were employed to provide replications using different experimental setups.
In all experiments, the software was designed to ensure that the physiological
data representing the pre-stimulus period were already recorded in the
computer’s memory before the target was selected (Experiments 2–4) or
displayed (all experiments). The software in all experiments marked the SCL
data in real-time with the current condition of the test (i.e., pre-stimulus,
stimulus, or post-stimulus) to ensure correct synchronization with external
events. As a result, the hardware, software, and data collection mechanisms used
in these experiments are unlikely to be sources of systematic bias that could
explain the observed results.
To avoid possible violations of distributional assumptions associated with
parametric statistical tests, nonparametric randomized permutation analysis was
used to evaluate the results.
Selective reporting. Because results similar to those presented here can
undoubtedly be mimicked by carefully selecting data, special care was taken to
analyze all available trials in all of the author’s presentiment experiments
conducted so far. The only data excluded were 280 trials from Experiments 1b
and 1d, since they were collected with different experimental designs or
intentions, and in any case both of those experiments had produced positive
effects (Radin, 1997).
268 D. I. Radin
Participant or experimenter fraud. Participant fraud would have been
difficult to perpetrate in this experiment. If P had tried to tamper with the
equipment after the experiment began, the physical movements would have been
detected by the physiological monitor as large, sustained artifacts. No such
artifacts were evident in the data. Further, P had no access to the equipment in
advance of or after experimental sessions, and P typically ran a single session.
So it would have been difficult to arrange for more sophisticated forms of
tampering. If P had described her experience to another P who was going to run
the test later, it is conceivable that the second P might exhibit higher overall
sympathetic arousal due to advance knowledge about the photo stimuli, but that
could not explain the differential effect observed in these experiments. The
question of experimenter fraud can only be resolved through successful
independent replications, which are now available.
Anticipatory strategies. This is the most common and prima facie the most
plausible conventional explanation for the observed effects. The idea assumes
the use of an experimental design with dichotomous stimuli: emotional vs. calm,
or stimulus vs. no stimulus. With such a design, it is conceivable that on
sequential trials the participant’s EDA would increase with each successive calm
trial, since anticipation would keep increasing until the emotional trial appeared.
Thus, EDA would peak on the emotional trial and then reset back to zero on the
next trial, as illustrated in Figure 7.
If this strategy were followed either consciously or unconsciously, then EDA
averaged across all emotional trials would be higher than EDA averaged across
Fig. 7. Simulation of an idealized anticipatory strategy. The y-axis is the degree of autonomic
arousal, the x-axis is successive trials. The top of each ‘‘arousal’’ ramp represents an
emotional trial; all other trials are calm.
Electrodermal Presentiments 269
calm trials. Simulation and analytical studies have confirmed the existence of
this bias (e.g., Dalkvist et al., 2002; Wackermann, 2002). The same simulations
also show that with longer sessions, or after pooling trials across many
participants, that these biases can become vanishingly small.
Anticipatory simulation studies are valuable in highlighting worst-case
scenarios, but they also oversimplify what actually occurs in these experiments.
For example, as previously mentioned, people’s idiosyncratic responses to
photos inevitably blur the dichotomous distinction between calm and emotional
targets. A more realistic anticipatory simulation might use targets with
a continuous range of emotionality, and it would adjust the arousal value for
trial Nþ1 up or down according to the emotionality rating of trial N.
But idealized simulations and strategies aside, we can investigate whether
anticipatory strategies can explain the observed results by examining the actual
data. To do this, data from Experiments 2 and 3 were pooled; this provided a set
of 3,469 trials, contributed by a total of 103 participants. The pre-stimulus sums
of SCL (i.e., PSCL) prior to stimulus onset in each of the two experiments
were normalized separately and then combined.
Then the targets were separated
into two classes: emotional, defined as those trials with the top 26% emotionality
ratings, and calm, defined as those 74% with lower emotionality ratings. These
percentages were selected to create about a 1:3 ratio of emotional to calm targets
to ensure that there would be an adequate number of calm targets in a row to test
the expectations of an anticipatory strategy. Based on this definition of
emotional and calm targets, 13 of the 103 participants were identified who
independently obtained significant ( p,0.05) emotional vs. calm differences in
their pre-stimulus responses. Together these people contributed a total of 450
trials, and as a group they represented (by selection) extremely strong evidence
An anticipatory strategy supposes a positive trend between the number of
calm trials before an emotional trial vs. PSCL for each of those trials, as
illustrated in Figure 7. Note that this trend, which can be evaluated with a simple
linear correlation, cannot include the emotional trial itself, since that would
confound testing an anticipatory strategy with a genuine presentiment effect.
Figure 8 shows the observed means of PSCL for calm trials 1 to 13 steps
before an emotional trial, and for the emotional trial itself (the ‘‘0’’ point on the
x-axis), with one standard error bars. The error bars become progressively
smaller because the number of sequential calm trials before an emotional trial
decreases with the number of total trials. For example, there are many more
cases of say, the sequence C–E (one step away) than there are of the sequence,
C–C–C–C–C–E (five steps away).
The weighted linear correlation between mean PSCL and trial number for
steps 13 !1 was positive, but not significantly so (r¼0.29, p¼0.17). Notice
that with one exception, all of the mean PSCL values prior to the emotional
trial were negative, and three were significantly negative, including the trial
immediately preceding the emotional trial. Thus, contrary to the expectations of
270 D. I. Radin
an anticipatory strategy, a subset of participants specifically selected for
exhibiting strong differential results suggestive of a genuine presentiment effect
showed relaxation responses before the emotional target rather than progressive
arousal. In sum, while idealized anticipatory strategies might provide an
explanation of the observed results in principle, the actual data did not indicate
that such strategies were employed.
While working at Interval Research Corporation, I was asked to arrange for an
independent critical evaluation of the methodology and results of Experiment 3.
Because assessments of controversial evidence tend to differ depending upon
whether opinions are presented in public or in private, to encourage an unbiased
review it was agreed in advance that the identities of the reviewers and the
specifics of their written reports would remain confidential.
Six experienced scientists were identified to conduct the review. Three were
sympathetic to the possibility of precognition, and three were skeptical. The
group’s expertise included statistics, experimental psychology, personality and
cognitive psychology, psychophysiology, computer science, and physics. The
reviewers unanimously agreed that they could not identify any methodological
flaws that could explain the observed outcomes. However, their personal
opinions about whether the data persuasively demonstrated precognition fell into
close alignment with their a priori beliefs about the possibility of precognition.
This outcome demonstrated one of the key difficulties encountered when trying
to achieve scientific consensus about controversial ideas, especially ideas that
are not yet supported by well-accepted theories. Members of the Society for
Scientific Exploration are undoubtedly familiar with this syndrome.
Fig. 8. Average PSCL, and one standard error bars, for up to 13 calm trials prior to an emotional
trial (at 0), for 13 participants in Experiments 2 and 3, each of whom showed an
independently significant presentiment effect.
Electrodermal Presentiments 271
Four double-blind experiments, using different hardware and software
implementations, subject populations, environmental conditions, and photo-
graphic stimuli, explored the possibility that some intuitive hunches, as reflected
by fluctuations in the autonomic nervous system, may involve unconscious
perception of future experiences. Overall the experiments supported this idea.
Consideration of alternative explanations suggests that the observed effects were
not due to known artifacts. It is tempting to speculate about possible theoretical
explanations for these effects, especially the possibility that presentiment may be
one way that the time-symmetries pervading fundamental physics manifest in
human experience. But further speculations will be reserved for future
Temperature and humidity levels are reported because those factors are known
to affect electrodermal responses, and this information may prove to be useful
in future replications.
To demonstrate that the QuickBasic 4.5 PRNG used in this experiment did not
introduce an artifactual relationship between the physiological state and the
resulting target photos, the PRNG was seeded with a number (N¼1), and then
used to generate one random number from 1 to 150, using the same
programming code as employed in the experiment. This created a seed-
number, target-number pair. This process was repeated for seed-numbers N¼2
to 5000, and then a correlation was determined between the resulting pairs. If
the seed-number was associated with the resulting target selection, then
a positive correlation would be predicted. But no such relationship was found,
r¼0.0007, p¼0.96, two-tailed.
The reason for the normalization was that the pre-stimulus SCL values were
formed by summing the change in normalized SCL over all samples in the
presentiment period, and given that the sampling rates in Experiments 2 and 3
were different (5 Hz and 10 Hz, respectively), those sums were different.
Experiment 1 was supported by the Bigelow Foundation of Las Vegas,
Nevada. Part of Experiment 2 was supported by the Institut fr Grenzgebiete der
Psychologie und Psychohygiene of Freiburg, Germany. Part of Experiment 2
and all of Experiment 3 were supported by Interval Research Corporation.
Experiment 4 was supported by Interval Research Corporation and Boundary
Institute. I thank Steven Rubin, Paul Korff, and Lee Felsenstein for their efforts
in developing the hardware and software used in Experiment 4, and Richard
Shoup, Thomas Etter, Edwin May, Russell Targ, and Dick Bierman for many
valuable discussions about the design and analyses of these experiments.
272 D. I. Radin
Bierman, D. J. (2000). Anomalous baseline effects in mainstream emotion research using
psychophysiological variables. Paper presented at the Proceedings of the 43rd Annual Convention
of the Parapsychological Association, Cary, NC (pp. 34–47).
Bierman, D. J., & Radin, D. I. (1997). Anomalous anticipatory response on randomized future
conditions. Perceptual and Motor Skills, 84, 689–690.
Bierman, D. J., & Radin, D. I. (1998). Conscious and anomalous non-conscious emotional processes:
A reversal of the arrow of time? In Toward a Science of Consciousness, Tucson III (pp. 367–386).
MIT University Press.
Bierman, D. J., & Scholte, H. S. (2002). Anomalous anticipatory brain activation preceding exposure
of emotional and neutral pictures. Paper presented at Toward a Science of Consciousness, Tucson
IV, Tucson, AZ.
Blair, R. C., & Karniski, W. (1993). An alternative method for signiﬁcance testing of waveform
difference potentials. Psychophysiology, 30, 518–524.
Bouscein, W. (1992). Electrodermal Activity. Plenum.
Bradley, M. M., Greenwald, M. K., & Hamm, A. O. (1993). Affective picture processing. In
Birbaumer, N., & Ohman, A. (Eds.), The Structure of Emotion. Hogrefe & Huber Publishers.
Dalkvist, J., Westerlund, J., & Bierman, D. J. (2002). A computational expectation bias as revealed by
simulations of presentiment experiments. Paper presented at the Proceedings of the 45th Annual
Convention of the Parapsychological Association, Cary, NC (pp. 62–79).
Ito, T. A., Cacioppo, J. T., & Lang, P. J. (1998). Eliciting affect using the International Affective
Picture System: Bivariate evaluation and ambivalence. Personality and Social Psychology
Bulletin, 24, 856–879.
Kimmel, D. H., van Olst, E. H., & Orlebeke, J. F. (1979). The Orienting Reﬂex in Humans. Erlbaum.
McCraty, R., Atkinson, M., & Bradley, R. T. (2004). Electrophysiological evidence of intuition: Part
1. The surprising role of the heart. Journal of Alternative and Complementary Medicine, 10, 133–
Norfolk, C. (1999). Can future emotions be perceived unconsciously? An investigation into the
presentiment effect with reference to extraversion. Unpublished manuscript, Department of
Psychology, University of Edinburgh.
Parkhomtchouk, D. V., Kotake, J., Zhang, T., Chen, W., Kokubo, H., & Yamamoto, M. (2002). An
attempt to reproduce the presentiment EDA response. Journal of International Society of Life
Information Science, 20, 190–194.
Radin, D. I. (1997). Unconscious perception of future emotions: An experiment in presentiment.
Journal of Scientiﬁc Exploration, 11, 163–180.
Spottiswoode, S. J. P., & May, E. C. (2003). Skin conductance prestimulus response. Journal of
Scientiﬁc Exploration, 17, 617–642.
Wackermann, J. (2002). On cumulative effects and averaging artefacts in randomised S-R
experimental designs. Paper presented at the Proceedings of the 45th Annual Convention of the
Parapsychological Association, Cary, NC (pp. 293–305).
Wildey, C. (2001). Impulse Response of Biological Systems. Master’s thesis, Department of Electrical
Engineering, University of Texas at Arlington.
Electrodermal Presentiments 273