Explore: The Journal of Science and Healing
(2009) 5, 272-276
Sensing the sending of SMS messages: an
automated test
Rupert
Sheldrake,
Perrott-Warrick Project,
Leonidas Avraamides, and Matous Novák
Mobifi Ltd,
Abstract.
Objective:
To carry out automated experiments to test for telepathy in connection with
text messages.
Method:
Subjects, aged from 11 to 72, registered online with the names and mobile
telephone numbers of 3 senders. A computer selected a sender at random, and
asked him to send an SMS message to the subject via the computer. The computer
then asked the subject to guess the sender’s name, and delivered the message
after receiving the guess. A test consisted of 9 trials.
Interactions evaluated: The effects of subjects’
sex and age and the effects of delay on guesses.
Main outcome measure: The proportion of correct
guesses of the sender’s name, compared with the 33.3% mean chance expectation.
Results: In 886 trials there were 336 hits (37.9%),
significantly above the 33.3% chance level (p
= .001). The hit rate in incomplete tests was 38.4% (p = .03) showing that optional stopping could not explain the
positive results. Most tests were unsupervised, which left open the possibility
of cheating, but high-scoring subjects were retested under filmed conditions,
where no cheating was detected, with 19 hits in 43 trials (44.2%; p = 0.09).
Key
words: SMS
messages, telepathy, ESP, automated test, internet experiment.
INTRODUCTION
Most people believe they have experienced telepathy or
other forms of “extrasensory perception” (ESP) or psi (e.g.
A small
minority of scientists have been investigating telepathy experimentally for
more than 125 years (Radin, 2007). From the 1880s to the 1940s, the most
popular experimental method involved card-guessing tests. During this period,
142 published articles described 3.6 million such trials, with positive hit
rates that were statistically significant, even though the average effect was
small, less than 2% above the level expected by chance (Pratt, et al., 1966).
In the
1960s and 1970s, controlled studies of dreams provided a new approach. Could people pick up images telepathically
when dreaming in a laboratory, while a “sender” in another room concentrated on
a randomly chosen image? In a meta-analysis of the 25 published studies on
dream telepathy, covering a total of 450 trials, the overall hit rate was
significantly above chance expectation (Radin, 1997).
Since the
1970s, the principal method used by parapsychologists for investigating
telepathy was called the Ganzfeld, and involved a mild form of sensory
deprivation. Participants sat in a relaxed state in dim red light with halved
ping-pong balls over their eyes. In another room, a “sender” concentrated on a
picture or video clip, selected at random from a pool of possible targets.
After the session, the participant was shown four pictures or video clips, and
asked to pick which one most closely corresponded to impressions he or she may
have received during the test session. By chance, participants would select the
target picture roughly one time in four, with a hit rate of 25%. In 1985, a
meta-analysis covering 28 studies showed a hit rate of 37% (Honorton, 1985). A
leading member of CSICOP published a meta-analysis of the same data (Hyman,
1985), and also found that the odds against chance were astronomical.
Hyman and
Honorton jointly drew up a set of guidelines for further research (Hyman and
Honorton, 1986) in which they recommended rigorous precautions against sensory
leakage, full documentation of all experimental procedures, extensive security
procedures to prevent fraud, and complete specifications about what statistical
tests were to be used.
Investigators
in Honorton’s laboratory carried out a new series of Ganzfeld experiments
following these recommendations; in a meta-analysis of the results from 354
such sessions, the average hit rate was 32% (effect size .28; p < .01; Bem and Honorton,
1994). In 1999, a new meta-analysis that
included studies from other laboratories claimed there was no significant
effect (Milton & Wiseman, 1999). However, Milton and Wiseman’s analysis
involved questionable statistical methods (Storm & Ertel, 2001) and
excluded a recent series of studies from
However,
the Ganzfeld procedure bears little resemblance to apparent telepathy in
everyday life. Most people do not lie in dim red light with halved ping-pong
balls over their eyes trying to pick up images from a sender in another room
watching a video. Also, in some Ganzfeld tests, the “senders” and “receivers”
were strangers, whereas in real life apparent telepathy generally takes place
between people who know each other well (Sheldrake, 2003).
In the
modern world, the most commonly reported kind of telepathy occurs in connection
with telephone calls (Brown & Sheldrake, 2001; Sheldrake, 2000, 2003). Most
people claim to have thought of someone for no apparent reason, then that
person called; or they knew who was calling when the phone rang before
answering it or looking at a caller identification display. Many people claim to have had similar experiences
with e-mails (Sheldrake, 2003).
Such
impressions of telepathy could be illusory. Perhaps people frequently think
about others, and then a person they are thinking about calls. They may imagine
that telepathy is involved, but forget all the times they thought about people
who did not call. Thus apparent telepathy could arise from a combination of
chance coincidence and selective memory.
An illusion of telepathy could also arise if a person had an unconscious
expectation of a call, based on an implicit knowledge of the caller's behaviour.
Telephone
telepathy has recently been investigated experimentally in a series of
randomized trials (Sheldrake & Smart, 2003a, 2003b, Sheldrake et al.,
2004). Participants received a call from one of four potential callers. The
potential callers were nominated in advance by the participants themselves, and
were usually people they knew well. Callers and participants were in some cases
thousands of kilometres away from each other. In a given trial, the caller was
picked at random by the experimenter. When the telephone rang, the participant
guessed who was calling before the other person spoke. The guess was either
right or wrong. By chance, participants
would have been right about one time in four.
In a total of 271 videotaped trials, 45% of the guesses were hits
(effect size .45; p < 1 x 10-6;
Sheldrake & Smart, 2003b). In a replication at the
A related
form of apparent telepathy occurs in connection with emails: someone thinks of
a person for no apparent reason and soon afterwards receives an email from that
person. In an experimental investigation
of email telepathy, following a similar procedure to the telephone telepathy
tests, participants had four potential emailers. In each trial the sender was
selected at random. Before receiving the email from that sender, the
participant guessed whom it would be from. In 137 videotaped trials, this hit
rate of 47% was significantly above the chance level of 25% (d = .5, p < 1 x 10-6; Sheldrake & Smart, 2005).
A similar
kind of apparent telepathy happens quite frequently among people who
communicate by text (SMS) messages on mobile telephones. In this paper we
describe an automated SMS experiment for which participants chose three family
members or friends as potential senders. For each trial the sender was selected
at random by the computer. All trials were coordinated automatically, and the
data were stored in an online database.
In the
light of the results from telephone and email telepathy experiments, our
hypothesis was that hit rates in the automated text message telepathy test
would be above chance, but that the effect size would be smaller than in the
telephone and email telepathy tests for two reasons.
First, in
experiments on telephone and email telepathy, participants were recruited on
the basis of their apparent telepathic sensitivity, whereas in the experiment
described in this paper, there was no such selection.
Second, in
the telephone and email experiments, participants were asked to guess who was
calling or emailing while the other person was actually on the telephone or
writing the email. In the text message tests, participants were asked to guess
who had sent a text message at least several minutes, and in some cases more
than half an hour before they made their guess. Thus the telepathic contact was
not simultaneous. We hypothesized that this would reduce the effect.
We did not
expect this preliminary experiment to provide conclusive evidence for or
against the existence of telepathy. Most of the tests were unsupervised, and
carried out in real-life conditions, not in laboratories. Rather, we wanted to
explore a radically new procedure that enables tests to occur under
“ecological” rather than artificial conditions. This automated method minimized
experimenter effects; it enabled people to take part in the experiment while
going about their normal lives; and it permitted widespread participation.
However,
it left open the possibility that some participants could have cheated. We
therefore retested some of the highest-scoring subjects under videotaped
conditions to control for this possibility.
METHODS
Procedure
Participants registered online through Rupert
Sheldrake’s web site ( www.sheldrake.org
) The subjects entered their first and
second names, sex, age, email address and mobile telephone number, and also
entered the names of three contacts
(first names only) together with their mobile telephone numbers. In
cases there two of the contacts had the same first name, in registering they
gave them different names, e.g. John and Johnt, when registering and when
guessing.
All
participants received a welcome SMS message, and the subject was told that she
could stop the test at any time by texting STOP to the system. Then the test
proceeded as follows:
1. After a random time delay of between 1 and 4
minutes, the system selected one of the contacts at random and sent him a
message reading, “This is the telepathy test.
Please send an SMS reply which will be forwarded to [subject’s name] do
not attempt to contact [subject’s name] directly. Thank you.”
2. The contact then wrote a message to the
subject and sent it back to the system, which then immediately sent a text
message to the subject reading “[subject’s name] one of your contacts has sent
you a message. Please reply and guess
who has sent it.”
3. The subject then sent a text message back to
the system with the name she guessed.
When this had been received by the system and the data recorded, the
message from the contact was sent on to the subject, who therefore received
immediate feedback as to whether the guess was correct or not.
4. After a randomized delay this process was
repeated until the subject had completed 9 trials, at which stage the test was
complete. She then received a message
saying “Thank you for taking part in this test.
You scored [number of hits] correct out of 9 trials. Please text the word START
if you wish to do the test again.”
The
data were stored on an online database, accessible by the use of a
password. When group leaders were
recruiting subjects, they asked them all to register with the same group name,
and by using this group name as a password, the leader could access the data
from all members of his group. The database displayed a chronological list of
results, with one line per test, giving the subjects name, the subject’s sex
and age, the date and time at which the test started, the number of trials and
the number of hits. Also, separate columns showed the number of trials and hits
and that occurred less than 3 minutes after the contact sent a message to the
subject, those that occurred 3 – 10 minutes afterwards, and those that occurred
more than 10 minutes afterwards. In
addition, for each test it was possible to display all the details trial by
trial, and a full list of all messages sent and received by the system during
the test, together with the time at which they were sent or received, recorded
to the nearest second.
System
and Programming
The system was developed and operated by Mobifi
Ltd (
The
programming environment was Microsoft Visual Studio (Microsoft Corporation,
Parts
of the application were programmed in VB Script (Visual Basic Script). This
script was used to present the results on the web and for communicating with
external services such as 2Ergo Plc, the company that provided the direct
sending of SMS text messages to Vodafone
The
application sent SMS text messages using HTTP Post directly from the
application to an external gateway provided by 2Ergo, with a direct connection
to Vodafone
When
the participants replied to the application, they did so by sending a message
to a Virtual Mobile Number (VMN), which was provided by the mobile network
operator, T-Mobile. These replies were then sent to the application via HTTP.
The
application was hosted on a secure server, a Dell PowerEdge 1750, located in a
data centre in Byfleet,
The SMS system worked reliably only within the
When
all the work scholars’ subjects had completed the tests, they retested the two
highest-scoring subjects while filming them in their homes, to ensure that the
subjects were not receiving any telephone messages, emails, instant messages or
other forms of communication during the test.
The work scholar set up the camera, and then left the room so that the
participants were alone during the test. On completion of the project, work
scholars provided RS with a short report and videotapes or DVDs of the filmed
experiments. One of the work scholars
was able to film only one of the highest scoring participants, so there were 7
filmed tests in all.
The videos were assessed blind by Pamela Smart,
RS’s Research Assistant. Any trials in
which the subjects went off camera, received other telephone calls or messages,
or in which another person was present in the room, were disqualified.
The results were analysed using the binomial
test; the chance probability of a hit was 0.33. Single-sided p values were used. The null hypothesis that
the hit rate in the tests would not be significantly different from the chance
level. For comparisons of
different sets of data, e.g. from male and female subjects, the 2 x 2
chi-squared test was used. Cohen’s effect size d was calculated according to the formula d = (p (hits observed) - .33) / sqrt (.33 x .66), where p (hits observed) is the proportion of
hits.
RESULTS
Overall
Results
In a total of 886 trials, there were 336 hits
(37.9%), significantly more than the 33.3% expected by chance (p = .001). Not all subjects completed
the 9 trials in a test. Their reasons for breaking off were varied. In some
cases one of their contacts did not respond: in others they had to stop the
test because they ran out of time; in some there were technical faults with the
system that terminated the test too soon.
Altogether,
there were 66 complete tests with a total of 594 trials, in which there were
224 hits (37.7%; p = .009). Of these
tests, 16 had hit rates below the chance level of 3; 22 were at the chance
level and 28 were above it.
In
the incomplete tests there were 292 trials, with 112 hits (38.4%; p = .03). There was no significant
difference (p = .85) between the hit
rates in complete and incomplete tests.
Effects
of Delayed Guesses
The beginning of each trial took place with a
random time delay after the ending of the previous trial; there were also
variable delays in the responses of the contacts to the request to send a
message, some of which were due to unpredictable delays in the deliveries of
text messages by the telephone system, and also variable delays in the
responses of the contacts.
After
the contact sent his message to the system, the system sent a text message
within a few seconds to the subject asking her to guess whom the message was
from. There were variable delays before the subject made her guess and sent it
back to the system, variable delays in the transmission of the text message to
the subject by the telephone system, and variable delays in the subject
responding to the text message.
The
exact times at which all messages were sent and received were recorded on the
database to the nearest second. The delays in responses by the subjects after
the contacts sent their messages were grouped into three categories: delays of
less than three minutes, three to10 minutes, and more than 10 minutes. In 695
out of 886 trials (78.4%) the time between the contact
sending his message and the subject making her guess was less than 3 minutes
(Table 1). This hit rate in these trials was 38.7%.
In
only 80 trials the delay was between three and 10 minutes, and here the hit
rate was 31.3%, slightly below the chance level, although not significantly so.
In 107 trials the delays were more than 10 minutes; the hit rate was 39.2%,
slightly higher than in trials with less than three minutes delay, but this
difference was not significant.
Table 1.
Hit rates in trials with different delays between to contact sending a message
and the subject responding.
|
Delay,
min |
Trials,
No. |
Hits,
No. |
Hits,
% |
|
<3 |
695 |
269 |
38.7 |
|
3-10 |
80 |
25 |
31.3 |
|
>10 |
107 |
42 |
39.2 |
Effects
of Subjects’ Sex and Age
With male subjects there were 323 trials with
128 hits (39.7%; p = .002); and with
female subjects 463 trials with 168 hits (36.2%; p = .07). The higher hit rate with males than females was not
statistically significant (p = .29).
The
hit rates for subjects of different ages are shown in Table 2. The highest were in the 15-19 and 40-49 age
groups; the lowest in the group aged over 60 years.
Table 2. Hit rates with subjects of different ages
|
Age,
y |
Trials,
No. |
Hits,
No. |
Hits,
% |
|
10-14 |
60 |
21 |
35.0 |
|
15-19 |
380 |
151 |
39.7 |
|
20-29 |
66 |
25 |
37.9 |
|
30-39 |
115 |
41 |
35.7 |
|
40-49 |
143 |
55 |
38.5 |
|
50-59 |
87 |
32 |
36.8 |
|
>60 |
35 |
11 |
31.4 |
Filmed
Tests
Seven high-scoring subjects were retested under
videotaped conditions to ensure that they were not receiving emails, instant
messages, text messages or direct communications from other people during the
experimental period. The videotapes were evaluated blind.
Two
tests were disqualified because other people entered the room and talked to the
subjects. In the other five tests, the videotapes showed no signs of any
conventional kinds of communication with other people. In their initial,
unfilmed tests, these five subjects took part in a total of 44 trials, with 28
hits (63.6%). In the filmed tests the hit rate was 19 out of 43 trials (44.2%)
(p = .08). (In one of the unfilmed
trials and two of the filmed trials, the tests terminated after only eight
rather than nine trials).
DISCUSSION
Our hypothesis was
that hit rates in an automated SMS telepathy test would be above chance levels,
but the effect size would be smaller than in previous telephone and email
telepathy experiments. We found that the overall hit rate of 37.9% was indeed
higher than the chance level of 33.3% (p = .001), and the effect size (d = .1) was indeed smaller than in
telephone telepathy tests (d = .45;
Sheldrake and Smart, 2003b) and email telepathy tests (d = .5; Sheldrake and Smart, 2005).
We expected a lower effect size for two reasons: the subjects in this
SMS experiment were not selected on the basis of apparent telepathic ability,
and also the subjects guessed who had sent them a message after the message had
been sent, whereas the telephone and email experiments were simultaneous: the
senders were thinking about the subject while she was making her guess.
We
expected that the longer the delay in the response of the subject, the lower
the hit rate would be. For delays between three and 10 minutes, hit rates were
in fact at the chance level. But for
delays of over 10 minutes, the hit rates were slightly higher than in trials
with less than three minutes delay.
However, this difference was not significant, and it is therefore
impossible to interpret. In future studies, it would be possible to introduce
random delays into the system so that a wide range of delayed responses could
be studied systematically in order to find out whether seemingly telepathic
responses fall off with time after the message is sent.
‘Optional
stopping’ could cause a possible artefact if participants who were not scoring
well quit the test. If this were the case, incomplete tests should have a lower
hit rate than complete tests. In the present experiment, the hit rate in
incomplete tests was slightly higher than in the complete tests, but the
difference was not statistically significant.
Hence ‘optional stopping’ cannot explain the positive results reported
here. Nor can any kind of ‘file-drawer
effect’ whereby only favourable data are reported. All the data from this
experiment are included.
Most
of these tests were unsupervised, and therefore the possibility arises that
some of the subjects were cheating. We
cannot rule out this possibility, but we think it unlikely. First the effect size was small, so if people
were cheating, they were not doing so very enthusiastically, or very few were
doing so, otherwise the hit rates would have been higher. Second, people had no motive to cheat and
most subjects were curious to find out how they would perform in this
test. Nevertheless, for a test of this
kind to provide strong evidence for the reality of telepathy, the tests would
have to be performed under more rigorous conditions. That is why some of the higher-scoring
subjects were selected for further tests while being filmed. In these
circumstances, the hit rates were still above chance (44.2%) but the sample
size was too small for this result to achieve statistical significance. However, the hit rate was lower than in these
subjects’ preliminary unsupervised tests.
Ertel (2005) observed a similar drop in hit rates when he retested
previously unsupervised participants under controlled conditions. This
exploratory experiment suggests that the best use of unsupervised tests with
widespread participation would be to identify talented subjects for more
rigorous testing under controlled conditions.
However,
unsupervised tests are potentially valuable in their own right for a comparison
of variables that effect hit rates. In
this experiment, we studied only sex and age, and found no significant
differences. In future studies it would be possible to investigate other
variables, such as the relationship between the contacts and the subjects and
the distance between them at the time of the trial. In previous research on telephone and email
telepathy (Sheldrake and Smart, 2003b, 2005) and with internet-based telepathy
tests (Sheldrake and Lambert 2007) hit rates were highest when contact and
subjects knew each other well, and especially when they were members of the
same family. The effects did not fall
off with distance, even when the participants were as far away as possible,
near the antipodes.
This
experiment shows that modern technologies such as mobile telephones and text
messages provide good opportunities for ‘real world’ as opposed to laboratory
research.
Acknowledgements
We thank all those who participated in these tests,
Pam Smart for her blind evaluation of the videotapes, and to the work scholars
who helped to recruit and test participants: Sophie Adler, Timothy Harrison,
Matthew Jensen and Florence Maclean. This work was supported by grants from the
Bial Foundation,
References
Bem, D. J., & Honorton, C. (1994). Does psi exist? Replicable evidence for an anomalous process of information
transfer. Psychological Bulletin,
115, 4-18.
Bem,
D. J., Palmer, J., & Broughton, R. (2001). Updating the ganzfeld
database: a victim of its own success? Journal
of Parapsychology, 65, 207-218.
Blackmore,
S. (1997). Probability misjudgement and belief in the
paranormal. British
Journal of Psychology, 88.
683-689.
Brown, D., & Sheldrake,
R. (2001) The anticipation of telephone calls:
a survey in
Carter,
C. (2007). Parapsychology
and the Skeptics.
Ertel, S. (2005). "Psi test feats achieved alone at home: Do they
disappear under lab control?" Australian
Journal of Parapsychology, 10,
149-164.
Honorton,
C. (1985). Meta-analysis of psi ganzfeld research: a response to Hyman. Journal of Parapsychology, 49, 51-91.
Humphrey, N. (1995). Soul Searching: Human Nature and Supernatural Belief.
Hyman,
R. (1985) The ganzfeld psi experiment: a critical
appraisal. Journal of Parapsychology,
49, 3-49.
Hyman,
R., & Honorton, C. (1986) A joint communiqué: the
psi ganzfeld controversy. Journal of
Parapsychology, 50, 351-362.
Lobach, E.., & Bierman, D. J. (2004).
Who’s calling at this hour? Local sidereal time and telephone
telepathy. Proceedings of the
Parapsychological Association Annual Convention, 2004, 91-97.
Milton,
J. (1999). Should ganzfeld research continue to be crucial in the search for a
replicable psi effect? Journal of
Parapsychology, 60, 309-333.
Milton, J., & Wisemani,
R. (1999). Does psi exist? Lack of replication
of an anomalous process of information transfer. Psychological Bulletin, 125,
387-391.
Pratt,
J. G., Rhine, J. B., Smith, B. M., Stuart, G. E., & Greenwood, J. A. (1966)
Extra-sensory perception after sixty
years.
Radin,
D. (1997) The conscious universe: the scientific truth of
psychic phenomena.
Radin,
D (2007) Entangled Minds: Extrasensory
Experiences in a Quantum Reality.
Sheldrake,
R. (2000). Telepathic
telephone calls: two surveys. Journal of
the Society for Psychical Research, 64, 224–232.
Sheldrake,
R. (2003). The sense of being stared at,
and other aspects of the extended mind.
Sheldrake,
R., Godwin, H., & Rockell,
S. (2004) A filmed experiment on telephone telepathy
with the Nolan sisters. Journal of the
Society for Psychical Research, 68, 168-172.
Sheldrake, R. and LAMBERT, M. (2007) An automated online telepathy test. Journal of Scientific Exploration, 21, 511-522.
Sheldrake,
R., & Smart, P. (2003a) Experimental tests
for telephone telepathy. Journal of the
Society for Psychical Research, 67, 174-199.
Sheldrake,
R., & Smart, P. (2003b) Videotaped
experiments on telephone telepathy. Journal of Parapsychology, 67,
147-166.
Sheldrake,
R., & Smart, P. (2005) Testing for telepathy
in connection with e-mails. Perceptual
and Motor Skills, 101, 771-786
Storm, L. & Ertel, S. (2001) Comments
on Milton and Wiseman’s meta-analysis of ganzfeld research. Psychological Bulletin, 127. 424-533