Scientific Papers on The Sense of Being Stared At

The Sense of Being Stared At: An Automated Test on the Internet

Journal of the Society for Psychical Research, (2008) 72, 86-97
by Rupert Sheldrake, Charles Overby and Ashwin Beeharee
Full Text PDF

Abstract
In previous research on the sense of being stared at participants worked in pairs, with the starer behind the staree. In a series of 20 randomized trials, the starer looked or did not look at the staree, who had to guess "looking" or "not looking". We here describe an automated, internet-based version of this standard staring experiment. In 498 tests, each with 20 trials, the computer gave an automatic sound signal to indicate when each trial began. The average hit rate was 53.0% (p <1x10-6); 268 participants scored above the chance level of 10 out of 20, 150 below, and 80 at the chance level. There was no significant difference between male and female starees, and little effect of starees' age. The highest hit rates were with parent-child participants. Hit rates were significantly higher when starees received trial-by-trial feedback, but there was no increase in the second half of the test compared with the first. Although these tests were unsupervised, the results replicated many of the features of previous tests and illustrate the potential for carrying out research through the internet, enabling widespread participation.

The Sense of Being Stared At: Do Hit Rates Improve as Tests Go On?

Journal of the Society for Psychical Research, (2008) 72, 98-106
by Rupert Sheldrake
Full Text PDF

Abstract
Simple experiments on the sense of being stared at have given repeatable, positive results that are highly significant statistically. In these experiments, people work in pairs. The staree sits with his or her back to the starer, who either looks at the back of the staree's neck, or looks away, in a random sequence. In each trial, the staree has to guess whether or not the starer is looking. However, when Marks & Colwell (2001) and Lobach & Bierman (2004) conducted tests of this kind, some of their experiments gave results not significantly different from chance, and they attempted to explain the positive results in staring tests as artifacts. Their hypotheses predict that positive scores should arise only in trials with feedback, only in trials with one particular kind of randomization, and that scores should increase towards the end of the experimental session. I have examined the data from the first and second halves of more than 19,000 trials to test these predictions. Both with and without feedback, and also with different randomization methods, the scores were positive and statistically significant in both the first and the second halves of tests. With feedback there was a small increase in scores in the second halves, but this was not statistically significant. Without feedback, there was a tendency for the scores to decline. In a trial-by-trial analysis of one large-scale experiment, the highest hit rate occurred in the very first trial for starees who were about to receive feedback, before any feedback had actually been given! Thus the beneficial effect of feedback may not depend so much on the feedback itself as the state of mind of the participants.

Investigating scopaesthesia: Attentional Transitions, Controls : and Error Rates in Repeated Tests

Journal of Scientific Exploration 22, 517-527 (2008)
by Rupert Sheldrake
Full Text PDF

Abstract
The sense of being stared at, or scopesthesia, was investigated experimentally with participants working in pairs. Two participants were tested repeatedly and the effect of attentional transition was investigated. In some tests, in the pre-trial period the starer stared at the staree, who was blindfolded, and in others the starer did not stare during the pre-trial period. Their overall hit rate in these attentional transition tests was 52.8% (2,800 trials; p=0.002), but there was no significant difference in hit rates between the two kinds of test. Participants were given trial-by-trial feedback, so if there was any learning, there should have been a progressive increase in hit rates. This did not happen. The participants also took part in a control tests in which there was no staring at all. In these tests hit rates were at chance levels, indicating that other forms of ESP, such as telepathy and clairvoyance, could not account for the results in scopesthesia tests. There were only 3 recording errors in 2,800 trials (0.1%), and two of these cancelled out, leaving a net error rate of 0.04%.

Special Edition of the Journal of Consciousness Studies

A special edition of the Journal of Consciousness Studies (2005) Vol 12 No. 6
Editorial Introduction by Anthony Freeman: The Sense of Being Glared At
Rupert's papers from the Journal:
Part 1: Is it Real or Illusory? | PDF
Part 2: Its Implications for Theories of Vision | PDF
The Non-Visual Detection of Staring - Response to Commentators | PDF

The complete edition, entitled Sheldrake and His Critics: The Sense of Being Glared At is available in paperback.

Introduction
In 1981 Rupert Sheldrake outraged the scientific establishment with his hypothesis of morphic resonance. Subsequently he devoted his research to pioneering science, winning popular acclaim and continued establishment opprobium with a series of ground-breaking works. In this special edition of JCS, Rupert summarises his case for the 'non-visual detection of staring'. His claims are scrutinised by fourteen critics, to whom Rupert then responds. Anthony Freeman, in his editorial introduction, explores the concept of "heresy" in science and in religion and asks why it provokes such hostility.

Experiments on the Sense of Being Stared At: The Elimination of Possible Artefacts

Journal of the Society for Psychical Research, Vol. 65, pp.122-137 (2001)
by Rupert Sheldrake
Full Text PDF

Abstract
The sense of being stared at from behind can be investigated by means of simple experiments in which subjects and lookers work in pairs, with the looker sitting behind the subject. In a random sequence of trials the looker either looks at the back of the subject, or looks away and thinks of something else. In each trial the subject guesses whether or not he or she is being looked at. There is a 50% probability of getting it right by chance. More than 15,000 trials have already been conducted, involving more than 700 subjects, with extremely significant excess of correct over incorrect guesses (Sheldrake, 1999), indicating that people really can tell when they are being looked at from behind. In this paper I discuss possible artifacts that could have affected these results and describe the results of experiments carried out in a school in London in which I investigated the effects of blindfolding subjects and giving them feedback about whether their guesses were correct or not. Blindfolding and feedback had no significant effects. Under all conditions the scores in looking trials were positive and statistically significant, and in not-looking trials at chance levels. I also describe the results of a series of experiments carried out in schools in Ireland with blindfolded subjects who were not given feedback. The significant positive scores in these experiments confirmed that the feeling of being stared at from behind does not depend on visual clues, nor does it depend on the subjects knowing if their guesses are right or wrong.

Follow-up Research on the Feeling of Being Stared At

Skeptical Inquirer (2000), March/April, 58-61
by Rupert Sheldrake
Full Text PDF

Abstract
Two recent articles in the Skeptical Inquirer have claimed that the feeling of being stared at is an illusion. Both have attempted to refute my own experimental research on the subject, which indicates that many people do indeed have an unexplained ability to detect stares.

A variety of surveys have shown that most people believe they can feel unseen stares (Sheldrake 1994). In his article "Can we tell when someone is staring at us?" (March/April 2000 SI) Robert A. Baker, a CSICOP Fellow, dismissed this belief as false. "Skeptics.... believe that it is nothing more than a superstition and/or a response to subtle signals from the environment." (Baker 2000, p. 40). He claimed to provide empirical evidence to support his presuppositions.

The Sense of Being Stared At: Experiments in Schools

Journal of the Society for Psychical Research 62: 311-323 (1998)
by Rupert Sheldrake
Full Text PDF

Abstract
Simple experiments to test whether or not people can tell when they are being stared at from behind were carried out in schools in Germany and the United States. Lookers and subjects worked in pairs, with the lookers sitting behind the subjects. In a series of trials the lookers either looked or did not look at the subjects in a random sequence determined by tossing a coin. In each trial, the subjects guessed whether or not they were being looked at. The results show an overall positive effect, with 56.9% correct guesses as opposed to 50% expected by chance. 97 of the subjects were right more often than they were wrong, and 42 were wrong more often than they were right. This positive effect was highly significant statistically (p=3x10-6). The data showed a consistent pattern. There was a positive effect when the subjects were being looked at, while the guesses were not significantly different from chance when they were not being looked at. In one school in Germany where sensitive subjects were tested repeatedly, 71.2% of the guesses were correct, and two students were right about 90% of the time. Possible sources of artefacts in these experiments are examined, and the implications of the results are discussed.

The Sense of Being Stared At Does Not Depend On Known Sensory Clues

Biology Forum 93 209-224
by Rupert Sheldrake
Full Text PDF

Abstract
The "sense of being stared at" can be investigated by means of simple experiments in which subjects and lookers work in pairs, with the looker sitting behind the subject. In a random sequence of trials, the looker either looks at the back of the subject, or looks away and thought of something else. More than 15,000 trials have already been conducted, involving more than 700 subjects, with an extremely significant excess of correct over incorrect guesses (Sheldrake [1999]). This effect was still apparent in experiments in which subjects were blindfolded and given no feedback, showing it did not depend on visual clues, nor on the subjects knowing if their guesses were right or wrong (Sheldrake [2000]). In this paper I describe experiments I conducted in schools in England in which the subjects were not only blindfolded and given no feedback, but looked at through closed windows. There was again a very significant excess of correct over incorrect guesses (p<0.004). At my request, teachers in Canada, Germany and the United States carried out similar experiments and found an even more significant positive effect than in my own experiments (p< 0.0002). The fact that positive results were still obtained when visual clues had been effectively eliminated by blindfolds, and auditory and olfactory clues by closed windows, implies that the sense of being stared at does not depend on the known senses. I conclude that peoples' ability to know when they are being looked at depends on an influence at present unknown to science.

The Sense of Being Stared At Confirmed by Simple Experiments

Biology Forum 92: 53-76 (1999)
by Rupert Sheldrake
Full Text PDF

Abstract
The feeling of being stared from behind is well known all over the world, and most people claim to have experienced it themselves. There have been surprisingly few empirical investigations of this phenomenon. I describe a simple experimental procedure with subjects and lookers working in pairs. In a random sequence of trials, the looker either looked at the back of the subject, or looked away and thought of something else. Such experiments showed a very significant excess of correct over incorrect guesses. When subjects were being looked at, they guessed correctly about 60% of the time, whereas in control trials, when they were not being looked at, their guesses were close to the chance level of 50%. The same pattern of results was found in my own experiments with adult subjects, with two different procedures: in experiments conducted in schools in Connecticut, USA: in experiments conducted by volunteers all around the world; and in a previous series of experiments in schools in Germany and the USA. All these sets of data showed a highly significant effect. Taken together they showed that in looking trials, 427 people were more often right than wrong, as opposed to 157 who were more often wrong than right. This difference is extremely significant (p<1x10-25). In the control trials, there was no significant difference between the number of people who were more often right than wrong (294) and more often wrong than right (287). These results suggest that the feeling of being looked at from behind is a real phenomenon that depends on factors as yet unknown to science. Non-human animals may also share this kind of sensitivity, which may be of evolutionary sugnificance in the relationships between predators and prey.

Related Research by Others
Distant intentionality and the feeling of being stared at: Two meta-analyses

British Journal of Psychology, 1 May 2004, vol. 95, no. 2, pp. 235-247(13)
by Schmidt S.; Schneider R.; Utts J.; Walach H.
PDF

Abstract
Findings in parapsychology suggest an effect of distant intentionality. Two laboratory set-ups explored this topic by measuring the effect of a distant intention on psychophysiological variables. The 'Direct Mental Interaction in Living Systems' experiment investigates the effect of various intentions on the electrodermal activity of a remote subject. The 'Remote Staring' experiment examines whether gazing by an observer covaries with the electrodermal activity of the person being observed. Two meta-analyses were conducted. A small significant effect size (d =.11, p = .001) was found in 36 studies on 'direct mental interaction', while a best-evidence-synthesis of 7 studies yielded d = .05 (p = .50). In 15 remote staring studies a mean effect size of d = 0.13 (p = .01) was obtained. It is concluded that there are hints of an effect, but also a shortage of independent replications and theoretical concepts.

Biology Forum 92: 53-76 (1999)
by Rupert Sheldrake


1. The Feeling of Being Stared At

In folklore all over the world, it is supposed that an influence can be transmitted through the eyes, capable of affecting that which is looked upon. In India people will travel hundreds of kilometres for the darshan, the look, of a holy man or woman because this look is believed to confer blessings. Conversely, looks of anger or envy are widely feared. There are world-wide beliefs in the Evil Eye, as shown by the protective measures taken against it, in charms, talismans and amulets (Elsworthy [1895]; Heaton [1978]).

Probably the prevalence of such beliefs and their classification as superstitions have contributed to a taboo against investigating the possible reality of the power of looks, including the well-known "sense of being stared at". Even parapsychologists have until recently neglected this phenomenon, with fewer than a dozen papers published on the subject in the last 100 years (Sheldrake [1997]).

In spite of this neglect by scientifically-minded researchers, most people have had direct experience of this phenomenon. According to recent surveys, between 70 and 97% of the population in Europe and North America claim to have experienced reactions to being looked at from behind (Braud, Shafer and Andrews [1990]; Sheldrake [1994]; Cottrell, Winer and Smith [1996]).

Scientific research on this subject was probably set back for decades by two American sceptics, Tichener [1898] and Coover [1913], who claimed to have shown the phenomenon to be illusory. Tichener carried out experiments with himself as the looker. His subjects were students of his who claimed they could sometimes tell when they were being looked at. He published no quantitative data, nor gave details of his experimental methods. He merely announced that his results confirmed his negative expectations.

By contrast, Coover described an elegantly simple experiment, and published quantitative data. Like Tichener, he was himself the looker, and his subjects were his own students. He claimed that there was no significant effect of his looking, and concluded that popular belief in the sense of being stared at was "groundless". (Coover [1911]).

Apparently, 48 years elapsed before the next report in the scientific literature. Poortman [1959] described some trials he carried out with himself as the subject and a woman friend as the looker, following a similar procedure to Coover. She was a City Councillor in The Hague, Holland and was accustomed to attracting the attention of other council members by the power of her gaze. Poortman was right significantly more often than wrong in guessing when she looked at him (Sheldrake [1994]).

The experimental design with subject-looker pairs used by Coover and Poortman has remained the basis of most subsequent research on the subject, including the experiments described in this paper. In the basic Coover-Poortman procedure, the subject is looked at from behind in a series of trials, randomly interspersed with an equal number of control periods when the subject is not looked at. At the end of each test period, the subject guesses whether or not he or she was being looked at. The response is scored right or wrong and written down, and the next trial begins.

In the 30 years following Poortman's paper, until around 1990, apparently the only advances in research on this subject were achieved in two unpublished student projects. The first, by Peterson [1978], at the University of Edinburgh, involved experiments in which the looker and the subject were separated by a one-way mirror. The looker was invisible to the subject, and sat in a closed booth. The results were positive and statistically significant.

Williams [1983], a student at the University of Adelaide, Australia, found a statistically significant effect when the subject was looked at through closed-circuit TV, and the looker was in a different room.

In the 1990s, there has been an increasing interest in this field of research among parapsychologists. Practically all recent laboratory research has followed Williams in using closed circuit TV with the subjects and lookers in separate rooms. But in these new studies, the subjects were not asked to guess whether or not they were being looked at. Instead they could relax, even read a magazine, while their galvanic skin response was recorded automatically, as in lie-detector tests. Does looking at such subjects at random intervals through closed-circuit TV significantly affect the subjects' electrodermal responses?

In most investigations these experiments have given significant positive results (Braud, Shafer and Andrews [1990, 1993a, 1993b]; Schlitz and LaBerge [1994, 1997]). Wiseman and Smith [1994] also found a significant positive effect, but they speculated that it might be an artifact. Only Wiseman et al. [1995] obtained overall non-significant results.

In a recent closed-circuit TV experiment conducted in the laboratory of Wiseman, at the University of Hertfordshire, England, Schlitz obtained her usual positive results with herself as the experimenter and looker. Meanwhile, Wiseman, a sceptic, obtained non-significant results with himself as the experimenter and looker, under otherwise identical conditions (Wiseman and Schlitz [1997]). Thus there was a striking experimenter effect, with the sceptic seemingly inhibiting the manifestation of this effect when he himself was the looker.

This kind of laboratory research has an important role to play. But I believe it is worth taking further the simple procedures pioneered by Coover and Poortman, since these open up research on this topic to much wider participation, and are far easier and cheaper to do (Sheldrake [1994]).

In this paper I describe the results of experiments carried out by myself, by participants in seminars and lectures I have given, by teachers and pupils in schools in Connecticut, USA, and by volunteer investigators all around the world, recruited through the New Scientist magazine, Discovery Channel television and the world wide web. Elsewhere, I have summarized experiments carried out by teachers and their pupils in other schools in the United States and Germany (Sheldrake [1997]). The overall conclusion is that there is a highly significant tendency for people to know when they are being looked at. back to the top

2. The Looker Sits Behind the Subject

People worked in pairs, one (the subject) sitting with his or her back towards the other (the looker). The distance between them was 1 metre or more. They sat in places where there were no reflective surfaces (such as mirrors or windows) that could have enabled the subject to see the looker. In a series of trials, in a random sequence, the looker either looked at the back of the subject or looked away, and was instructed to think of something else.

The looker indicated to the subject when a trial was beginning by a tap or click, and the subject then guessed whether he or she was being looked at or not.

The looker recorded the result on a score sheet with two columns, the first headed "Looking (heads)", the second "Not looking (tails)", entering a tick or a cross in the appropriate column, depending on whether the guess was right or wrong. (For specimen score sheets, see Sheldrake [1994] ).

My own experiments:

In the experiments I conducted myself, the random sequence was decided by tossing a coin before each trial: heads meant "Look" and tails "Don't look". The looker then told the subject whether the answer was correct or not. In the first set of experiments, the looker signalled to the subject that the test period was beginning by a tap, hitting a table, chair or book with a pen or pencil. This method left open the possibility that information might be communicated consciously or unconsciously by the intensity of the tap, so in the second set of experiments mechanical clickers were used instead, to give a standard clicking sound. These clickers were plastic clips, used for holding garments on to coat hangers by Marks and Spencers, a British clothing and food store.

Usually subjects indicated their guess within 10 seconds, but if they had not done so already were asked to do so after 20 seconds. The procedure was therefore quite fast, and most pairs could easily complete 10-20 trials within 10 minutes. They then exchanged roles and carried out a new series of trials.

The first set of data in Table 1 came from experiments I conducted myself with members of my family and friends, with myself as a looker or subject. The other experiments were carried out with sets of looker-subject pairs at workshops I gave in Stockholm, Sweden; in Bremen, Hamburg, Stuttgart and Todtmoos Germany; during courses in Assisi, Italy for St Thomas University, Florida, and at Schumacher College, Dartington, Devon; at lectures to the Association for the Scientific Study of Anomalous Phenomena (ASSAP) in London; at the 1994 Annual Conference of the British Scientific and Medical Network held at Stow, Buckinghamshire; at the Fifth Annual Conference of the Institute of Noetic Sciences in Boca Raton, Florida, USA; and at Eton College, Berkshire.

Experiments in schools in Connecticut:

As part of their course work for a Master's degree in Science Education at South Connecticut State University, in a programme called LEARNScience, teachers were required to carry out this experiment in their schools. (An earlier version of this experiment was carried out by teachers in a previous LEARNScience course in 1996, and the results are described in Sheldrake [1998]). This experiment was coordinated by James Trifone and Dr Harris Stone, and all the teachers were sent instructions written by myself together with a set of 24 differently randomized score sheets, each of which was for 20 trials. Their students worked in pairs following the procedures described above, but the subjects were not told after each trial whether their guess was right or wrong. In some schools, instead of the looker giving the signal for the beginning of each trial by a click or bleep, the teacher gave a signal for the whole class by ringing a bell. All pairs in the class then did the trials at the same time. After each pair had completed the 20 trials in the random sequence set out on their score sheet, they changed places and did another set of 20 trials using a differently-randomized score sheet. Most of the children who took part in these experiments were aged between 7 and 15. The experiments were carried out in May-June 1997.

All the teachers were required to send in their results, and all the score sheets were sent to me for analysis.

Experiments carried out in other schools and by amateur researchers:

In an article about my research on the feeling of being stared at in the popular science magazine New Scientist (Webb [1997]) readers were invited to try the experiment for themselves and were sent instruction sheets on request, or else they were able to download them from the New Scientist world wide web site. I subsequently established a world wide web site of my, own giving these instructions together with a full set of randomized score sheets (www.sheldrake.org). Those who carried out the experiment were asked to send in their results to me, together with the full set of score sheets. In this paper I summarize the results that have been submitted in this way, up to December 1998. Most of these experiments were carried out in schools by the following teachers: W. Gribben (St Helen's College, St Helens, UK); W. Cotton (John Port School, Derby, UK); Susan Kemp (St James School, Twickenham, UK); J. Goddard (Ewell Castle School, Ewell, UK); T. Wilson (International School of Helsinki); D.Mitchell (Ueda Nishi High School, Nagano, Japan). Some experiments were done by groups of adults: the Warman family of Clitheroe, UK; the City Adult Learning Centre (CALC), Toronto, Canada (coordinator: W. Chappell); and the Psychic Study Group (PSG) in Chesterfield, UK (coordinator: W. Eyre). In addition, experiments was carried out by viewers of the science news programme on the Discovery TV channel broadcast from Toronto, Canada (coordinator: Jagg Griffiths) following a programme on October 16 1997 in which I described the experiment. Detailed instructions were made available to interested viewers via the world wide web, and results were sent in by email.

Analysis of results:

The numbers of right and wrong guesses from each series carried out by each looker-subject pair were tabulated in three columns; "Looking", "Not looking" and "Total", enabling the total number of right and wrong guesses in each column to be obtained. For each series of trials by each looker-subject pair, in each column, the data were also scored as follows:

      + if there were more right than wrong guesses
      - if there were more wrong than right guesses
      = if the number of right and wrong guesses was the same.

The number of + and - scores in the "total" columns in the tables are not the sum or the average of those in the "looking" and "not looking" columns, because for each subject on each column these scores were evaluated separately. Thus, for example, if a subject was right in 7 out of 10 trials when being looked at (+), and in 4 out of 10 trials when not looked at (-), the total score would be 11 correct guesses out of 20 (+).

Statistical analysis was carried out in two ways. First, the total number of right and wrong guesses for each set of subjects was compared using the the paired-sample t-test, with the numbers of right and wrong guesses for each set of subjects as the paired sample.

Second, the chi-squared test was used to compare the total number of + and - scores. The = scores were disregarded. The null hypothesis was that by chance alone the number of + and - scores would be equal.

For the comparison of two sets of scores 2 x 2 contingency tables were used (Campbell [1989]). The significance was assessed by the chi-squared test, with the null hypothesis that the proportions of right and wrong guesses in both sets were equal.

3. Subjects Guess Better When Stared At Than In Control Trials

The most striking and consistent feature of the data was the tendency for subjects to make more correct guesses when they were being looked at than when they were not (Tables 1 and 2). Overall, 58.5% of the guesses were correct when they were being looked at, compared with 48.2% in the control periods.

However, the total numbers of right and wrong guesses do not provide the most reliable way of evaluating the data. Some looker-subject pairs carried out more trials than others, and therefore had a disproportionate influence on the totals. Also some subjects had far more or less success than others in guessing correctly, and again had an exaggerated effect on the totals.

Table 1

Staring Experiments in which the looker signalled the beginning of each trial by means of a tapping noise. Numbers of right and wrong guesses (above) and numbers of looker-subject pairs (below) with more right than wrong guesses (+), more wrong than right guesses (=). (For the totals, the percentage of right guesses is shown in parentheses).

A more reliable way of evaluating the data was to give equal weight to each subject. This was done by scoring each subject positive, negative or equal, depending on whether he or she made more correct guesses than incorrect (+), more incorrect than correct (-) or equal numbers of both (=). By chance alone, there should be approximately equal numbers of positive and negative scores.

Under looking conditions, the predominance of positive scores (159+ and 63-) was extremely significant statistically (p = 1x10-10). By contrast, under not-looking conditions, the difference between the positive and negative scores (101+ and 117-) was not significant statistically.

The pattern of results under looking and not-looking conditions was also compared using 2 x 2 contingency-table chi-squared tests (Campbell [1989]). The significance of the difference between looking and not-looking conditions was very high (p = 5x10-5 for experiments with taps; p = 4x10-5 for experiments with clicks; and p = 2x10-8 for the overall results).

Table 2

Overall results: When the results from the looking and not-looking trials were added together, there were more correct than incorrect guesses (Tables 1 and 2). Overall, the percentage of correct guesses was 53.4 %, compared with 50% expected by chance. This is not a large effect, but it showed up quite consistently. In only one out of the 10 sets of data (from Bremen, Table1) was the number of wrong guesses greater than the number of correct guesses.

Looking at the overall results in terms of +, - and = scores, it is clear that more subjects scored positively than negatively, or in other words more people were right more often than they were wrong (Tables 1 and 2). This pattern was apparent is all sets of data, including those from Bremen. Both in the experiments where the beginnings of trials were signalled by taps and in those signalled by clicks, the greater number of positive than negative scores was highly significant statistically (p = 0.0002 and p = 0.001 respectively). Taken together (Table 5) the total number of positive scores was 141, compared with 70 negative scores. In other words, 141 subjects were more often right than wrong and 70 more often wrong than right. This difference was very significant (p = 1x10-6).

No significant difference between tapping and clicking signals:

If tapping had permitted the conscious or unconscious transfer of information from looker to subject, then the proportion of positive scores (Table 1) should have been higher than with standard mechanical clicks (Table 2). But this was not the case. There was no significant statistical difference between the two sets of data.

4. Confirmation from Experiments in Schools in Connecticut

Eight teachers submitted results from properly-conducted experiments, summarized in Table 3. The overall percentage of correct guesses was 55.3, somewhat higher than in my own experiments. In all eight sets of data, there were more correct than incorrect guesses. In total, 149 subjects were more often right than wrong, compared with 74 who were more often wrong than right. This difference was very significant (p=5x10-7).

Table 3

As in my own experiments, in the looking trials there was a highly significant tendency for subjects to guess correctly, whereas in the non-looking trials the scores were much closer to chance levels, although they were significantly positive.

Five other teachers misunderstood my instructions and used score sheets with the same randomization for all looker-subject pairs. Since this could have enabled subjects to obtain clues from other pairs in the classroom, or memorize part of the random sequence, the data from these teachers were excluded from the summary in Table 3. However, their results showed a very similar pattern, with 56.8 per cent correct guesses overall, and with 30 subjects right more often than they were wrong and 10 wrong more often than right (p = 0.002). There was the usual difference between the looking and not-looking trials, but it was even more pronounced than in the data in Table 3, with 61.6% correct guesses in the looking trials and 51.9% in the not-looking trials.

5. Confirmation From Experiments Carried Out By Volunteers

Altogether I received 10 complete sets of data from volunteer experimenters, mostly recruited through the New Scientist magazine and Discovery Channel television. (I also received several other reports that were incomplete, containing only the total scores rather than the detailed results, and hence these could not be included in this analysis.) The results are summarized in Table 4. In all cases there were more correct than incorrect guesses. Overall, 53.2% of the guesses were correct, and 121 subjects were right more often than they were wrong, compared with 60 who were wrong more often than right. This difference was highly significant statistically (p=6x10-6).

Table 4

Once again, the positive scores in the looking trials were highly significant, while in the not-looking trials the differences were not significant.

When these results are inspected more closely, it turns out that in three locations (Port School, Chesterfield PSG and Ueda Nishi High School) there were more more looking than not-looking trials, whereas there should have been more or less the same number of both. This reveals the existence of an error in the randomization procedure or a bias in the reporting of results. The results were therefore recalculated omitting the data from these three locations. The general pattern remained the same, with a very significant excess of positive over negative scores in the looking trials (58+ 26-; p=0.0004); no significant difference in the not-looking trials, and a very significant overall result (52+ 21-; p = 0.0003).

6. The Repeatability of the Results

Essentially the same pattern of results has appeared in four different series of experiments: in an initial series of experiments in schools in Germany and the United States (Sheldrake [1988]); in my own experiments (Tables 1 and 2); the Connecticut school experiments (Table 3); and the experiments by volunteers (Table 4). In all these sets of data, subjects scored very significantly above chance levels in the looking trials, whereas in the control trials their results were close to the 50% expected by chance.

In some of the experiments (Sheldrake [1988] and Tables 1 and 2), the randomization was done by the lookers tossing a coin trial by trial, the number of trials each looker-subject pair carried out varied, and the subjects were given feedback after each trial as to whether their guess was correct or not. In the other experiments (Tables 3 and 4), the randomization was done in advance, each looker-subject pair was asked to carry out a fixed number of trials, according to the score sheet they were supplied with, and there was no feedback. These differences in procedure had little or no effect on the overall pattern.

A criticism often levelled by sceptics against data that seem to show surprising or controversial effects is that researchers tend to publish only positive results while leaving negative data unpublished in their file drawers. This "file-drawer effect" can lead to a bias in the published data in the direction of the researchers' beliefs and expectations. This criticism is no doubt of general validity, and probably applies to much of the scientific literature, even in uncontroversial areas. But does it apply to the data reported in this paper?

I have included all the experiments I performed myself, and all the experiments that did not involve procedural errors in the Connecticut schools. (As described above, the excluded data from Connecticut showed the same pattern too, so even if these had been included they would not have changed the overall results.) So the file-drawer effect cannot explain the data in Tables 1, 2 and 3. In Table 4, I have included all the data I received from volunteer experimenters except for sets of data that were unusable because they were incomplete. But it is possible that some volunteers who carried out the experiment and failed to find any evidence for the feeling of being looked at did not bother to send in their results. Hence there could have been a kind of file-drawer effect owing to the selective reporting of results. I doubt if this could have had much effect because the overall pattern agrees with that in the other series of experiments. Nevertheless, because of this possible effect, I omit this series of experiments from the summary in Table 5.

Table 5

The cumulative pattern of results in Table 5 is very clear. There was a highly significant overall effect, whereby 55% of the subjects' guesses were correct as opposed to 50% expected by chance. A total of 387 subjects were more often right than wrong, as opposed to 186 who were more often wrong than right, with the probability of this being a chance result less than 1x10-15.

The most remarkable feature of the results is the way that the positive scores in looking trials were staggeringly significant, while there was no significant difference in the not-looking trials (Table 5). In the looking trials, 427 people were more right than wrong, as opposed to 157 who were more wrong than right. In the not-looking trials there was practically no difference (296+ 284-). This pattern can be seen at a glance in Figure 1.

Figure 1



The superior performance of subjects in looking trials did not depend on a minority of particularly sensitive subjects, but rather represented a general tendency for subjects to score better when they were being looked at than when they were not. This is clearly shown by the distribution curves in Figure 2, based on data from the experiments in Connecticut schools. The other sets of data gave very similar distribution curves. In the control trials, the distribution curve centred on a score of 5 out of 10, the level expected by chance. In the looking trials, the entire curve was shifted to the right, with its peak at a score of 6 out of 10.

Figure 2

Fig. 2 - Distribution of scores in looking and not-looking trials at schools in Connecticut. The ordinate represents the number of subjects who had a given score, and the abscissa the number of correct guesses out of ten.

Why should there be such a striking difference between the looking and the not-looking trials?

If there really is a tendency for people to know when they are being looked at, they would indeed tend to be right when they are being looked at. But in the control trials when they are not being looked at, they are being asked to detect the absence of an effect, which has no parallel in real-life conditions. And indeed under these conditions, the results were no better than chance; the subjects were just guessing.

Even the supposedly negative results of Coover [1913] show a similar pattern. I have scored his subjects using the same procedure I used here. Not only was his overall result positive (5+ 3- 2=), but that there was also a marked tendency for people to be right when they were actually being looked at (7+ 2- 2=) while they were at the chance level in the not-looking trials (5+ 5- 0=).

If people really can tell when they are being looked at, why is the effect detected in these experiments relatively small? In the trials reported in this paper, only about 55% of the guesses were correct, as opposed to the 50% expected by chance. There could be several reasons.

First, under the artificial conditions of experiments, people are being asked to do consciously what they may usually do unconsciously. Self-consciousness may interfere with their sensitivity.

Second, some of the subjects and lookers may have become bored or distracted during the experimental sessions, reducing the rate of success.

Third, some people are better as lookers or as subjects than others, and the inclusion of ineffective lookers and insensitive subjects in the trials may have diluted the effect. In a series of trials with selected lookers and subjects, up to 90% of the guesses were correct (Sheldrake [1998]).

Interestingly, the highly positive results with selected and experienced subjects show not only that some people are better than others, but that people can improve their scores with practice, both in looking and in not-looking trials. Although naive subjects tend to score best in looking trials, and at chance levels in not-looking trials (Figure 1), those who are tested repeatedly tend to improve in both kinds of trials. It seems possible to learn to detect the difference between being looked at and not being looked at.

7. Can the Results Be Explained As Artifacts?

Although these simple experiments gives such repeatable results, do they really show a mysterious effect of the gaze at a distance? Or can they be explained in terms of artifacts or subtle sensory cues?

First, consider the possibility that an artifact could have arisen if subjects tended to say "looking" in most of the trials because of some inherent bias in favor of saying "looking". Such a tendency to say "looking" more often than "not looking" would make subjects seem successful in looking trials. But it would at the same time make them equally unsuccessful in not-looking trials. The results do not bear out this idea. The positive scores in the looking trials were not offset by negative scores in the not-looking trials. Rather, in the not-looking trials, the scores were at chance levels (Tables 1-5; Figures 1 and 2).

Second, there is the possibility that some subjects were picking up unconscious auditory or olfactory cues from the looker, or that they were cheating by peeping to see what the looker was doing, or that some lookers were secretly signalling to the subject whether they were looking or not. Such hypotheses could explain the highly significant success in the looking trials, but they cannot explain why cheating or subtle cues should have failed to let the subjects know when they were not being looked at.

The pattern of results argues against these sceptical hypotheses. Nevertheless, these possibilities need to be tested as rigorously as possible, and in a new series of experiments I have modified the procedure to take account of all sceptical hypotheses so far proposed.

To prevent any possible peeping or cues from peripheral vision, the subjects are blindfolded.
To prevent subjects learning how to pick up subtle cues, no feedback is given, so the subjects have no way of knowing during the trial whether their guesses are right or wrong.
To block possible auditory or olfactory cues, the lookers are indoors and look out of closed windows. The blindfolded subjects are outdoors, with their backs to the windows and sitting up to100 metres away.
To prevent any possible signalling by the way in which the looker signals the beginning of a trial, the signalling is done by myself or by others supervising the experiment, and the trials are done simultaneously by several looker-subject pairs, with each looker following a different random sequence of looking and not-looking trials.

In such experiments, the general pattern of results is similar to that described in this paper (Sheldrake, in preparation).

8. Does Vision Involve "Extramission"?

If further studies confirm the reality of the ability to detect an unseen gaze in a way that cannot be explained in terms of artifacts or sensory information, the implications for our understanding of the nature of the human mind will be very far-reaching (Abraham, McKenna and Sheldrake [1992]; Sheldrake {1994]). Any hypothesis capable of explaining this effect would need to postulate an influence of the mind of the looker on the person being stared at. This would in turn raise the question of the nature of vision.

As every psychology student learns, so-called primitive people and children often believe that vision involves "extramission", a sending out of influences from the eye of the looker. Psychology students are taught that the correct theory is the intromission theory, according to which light comes into the eye but nothing goes out. Nevertheless, recent surveys carried out by Winer and his colleagues at Ohio State University have shown that most children and many adults believe that vision involves both intromission and extramission (Cottrell and Winer [1994]; Winer and Cottrell [1996]). Winer and Cottrell [1996, p.139] confessed that they were shocked by these findings, and dismayed to find that these beliefs were surprisingly resistant to eradication by education: "One of the most interesting findings was that the belief in the ability to feel stares, which occurs at a high level among children as well as adults, seems, if anything, to increase with age, as if irrationality were increasing rather than declining between childhood and adulthood!"

But what if the beliefs of children, so-called primitive people and the majority of our fellow citizens turn out to be correct rather than incorrect, rational rather than irrational?

Profound theoretical questions are at stake. Experiments on the sense of being stared at are simple and inexpensive to carry out. They work well in schools. Many people could participate in further research on these questions, including students in search of interesting projects.

Anecdotal evidence suggests that the feeling of being looked at is not confined to human beings, but that a variety of other species seem to react to the gaze of unseen humans. Cottrell, Winer and Smith [1996] found that over a third of the American children and adults they surveyed said that they could feel the stare of an animal, and over half said that animals could feel their stare.

Animals may also be able to feel when other non-human animals are looking at them. If an animal could feel the look of a hidden predator and escape as a consequence, this would be of selective advantage, and evolution may have favoured the development of this sensitivity in many species. The biological and evolutionary implications of this phenomenon are at present unexplored.

Acknowledgements

I thank all the people who conducted and took part in these experiments, and am grateful to Nicholas Humphrey, Tom Merriam and Michael Morgan for helpful discussions and advice on statistical analysis. I thank the Institute of Noetic Sciences, Sausalito, CA and the Lifebridge Foundation, New York for financial support

References

Abraham, R., McKenna, T. and Sheldrake, R. [1992], Trialogues at the Edge of the West. Bear and Co., Santa Fe.

Braud, W., Shafer, D. and Andrews, S. [1990], Electrodermal correlates of remote attention: Autonomic reactions to an unseen gaze. Proceedings of Presented Papers, Parapsychology Association 33rd Annual Convention, Chevy Chase, MD, pp14-28.

Braud, W., Shafer, D. and Andrews, S. [1993a], Reactions to an unseen gaze (remote attention): A review, with new data on autonomic staring detection. Journal of Parapsychology 57: 373-390.

Braud, W., Shafer, D. and Andrews, S. [1993b], Further studies of autonomic detection of remote staring: replications, new control procedures, and personality correlates. Journal of Parapsychology 57: 391-409.

Campbell, R.C. [1989], Statistics for Biologists. Cambridge University Press, Cambridge.

Coover, J.E. [1913], The feeling of being stared at. American Journal of Psychology 24: 570-575.

Cottrell, J.E. and Winer, G.A. [1994], Development in the understanding of perception: the decline of extramission perception beliefs. Developmental Psychology 30: 218-228.

Cottrell, J.E. and Winer, G.A. and Smith, M.C. [1996], Beliefs of children and adults about feeling stares of unseen others. Developmental Psychology 32: 50-61.

Elsworthy, F. [1895], The Evil Eye. Murray, London.

Heaton, J.M. [1978], The Eye: Phenomenology and Psychology of Function and Disorder. Tavistock Press, London.

Howat, S.J., Delanoy, D.L. and Morris, R. [1994], Remote staring detection and personality correlates. Journal of Scientific Exploration 8: 582.

Poortman, J.J. [1959], The feeling of being stared at. Journal of the Society of Psychical Research 40: 4-12.

Schlitz, M. & LaBerge, S. [1994], Autonomic detection of remote observation: two conceptual replications. Proceedings of Presented Papers, Parapsychology Association 37th Annual Convention, Amsterdam, pp. 352-60.

Schlitz, M. & LaBerge, S. [1997], Covert observation increases skin conductance in subjects unaware of when they are being observed: a replication. Journal of Parapsychology 61: 185-194.

Sheldrake, R. [1994], Seven Experiments that Could Change theWorld, Chapter 4. Fourth Estate, London.

Sheldrake, R. [1998], The sense of being stared at: Experiments in schools. Journal of the Society of Psychical Research 62: 311-323.

Titchener, E.B. [1898], The feeling of being stared at. Science New Series 8: 895-897.

Webb, J. [1997] Are you looking at me? New Scientist 26 July, p. 39.

Williams, L. [1983], Minimal cue perception of the regard of others: the feeling of being stared at. Paper presented at the 10th Annual Conference of the Southeastern Regional Parapsychological Association, Carrolltown, GA, Feb 11-12.

Winer, G.A. and Cottrell, J.E. [1996a], Effects of drawing on directional representations of the process of vision. Journal of Educational Psychology 88: 387-396

Winer, G.A. and Cottrell, J.E. [1996b], Does anything leave the eye when we see? Extramission beliefs of children and adults. Current Directions in Psychological Science 5: 137-142.

Wiseman, R. and Schlitz, M. [1997], Experimenter effects and the remote detection of staring. Journal of Parapsychology 61: 197-207.

Wiseman, R. and Smith, M.D. [1994], A further look at the detection of unseen gaze. Proceedings of Presented Papers, Parapsychology Association 37th Annual Convention, Amsterdam, pp. 465-78.

Wiseman, R., Smith, M.D., Freedman, D., Wasserman, T. and Hurst, C. [1995], Examining the remote staring effect: two further experiments. Proceedings of Presented Papers, Parapsychology Association 38th Annual Convention, pp. 480-490.

Biology Forum 93 209-224
by Rupert Sheldrake


1. Simple Experiments Show That People Can Tell When they are Being Stared At from Behind

Many people have had the experience of turning round with the feeling that someone is looking at them from behind, to find that this is in fact the case.Surveys show that between 70 and 97% of the population in Europe and North America have had personal experience of this phenomenon (Braud, Shafer and Andrews [1990]; Sheldrake [1994], Cottrell, Winer and Smith [1996]).

I have developed a simple experimental procedure to test whether people really can tell when they are being looked at from behind (Sheldrake [1994], [1998], [1999]).Participants work in pairs, with the looker sitting behind the subject.

In a randomized series of trials, the looker either looks at the back of the subject's neck, or looks away and thinks of something else. The results are repeatable, consistent and positive. More than 15,000 trials have already been conducted, involving more than 700 subjects (Sheldrake [1999]). Overall, there was an extremely significant positive effect (p< 1x10-15), indicating that people really can tell when they are being looked at from behind.

The data revealed a characteristic pattern whereby the scores in the 'looking' trials were very significantly above the chance level, whereas in the control 'not-looking' trials the scores were not significantly different from chance (Sheldrake [1999]). This pattern of results makes sense if the sense of being stared at is a real phenomenon. It would be expected to work when people were actually being stared at, as they were in the looking trials.

By contrast, in the control trials when they were not being looked at, subjects were being asked to try and detect the absence of an effect, which has no parallel in real-life situations; and under these conditions the results were close to chance levels.They were just guessing.

If subjects were cheating or receiving subtle sensory cues then they would have been expected to obtain positive scores in both the looking and the not-looking trials. But this is not what happened. The pattern of results does not support the idea that they depended on cheating or subtle sensory clues (Sheldrake [1998, 1999]).

However, in the first sets of experiments, the subjects were not wearing blindfolds and were given feedback after each trial as to whether their guess was right or wrong. So the possibility remained open that some of them could have been peeping or learning from the feedback how to associate subtle cues with the looking trials, even though it is difficult to see how this possibility could be compatible with the pattern of results. In subsequent experiments I have found this same characteristic pattern even when subjects were blindfolded and deprived of feedback (Sheldrake [2000]), showing that the effect does not depend on visual clues, nor on feedback.

Nevertheless, when the lookers and subjects are in the same room, it is difficult to eliminate the possibility that the effect depends on subtle auditory or even olfactory clues.In order to test these possibilities, I have carried out experiments, described in this paper, in which the lookers and subjects were separated by closed windows, effectively eliminating any possible role of smells or sounds.In these experiments the subjects were also blindfolded and deprived of feedback.I also describe the results of independent tests carried out at my request in Canada, Germany and the United States.

2. Subjects are Looked At Through Windows

Experiments in London : My experiments were carried out between March and June, 1997 at University College School (UCS) Junior Branch, a boys' school in Hampstead, London and at New End School, a primary school for both boys and girls, also in Hampstead. Each experiment took place with a different class: at UCS in the second form (age 8-9) or in the third form (age 9-10), and at New End School with class 4 (age 8-9). The experiments were supervised by myself and the class teachers: Mark Albibi, Mark Lall Chopra, Heidi Gregory, Yvonne Gregory at UCS, and Lynn Gavin at New End. Before the experiment began, I gave a brief introductory talk explaining and demonstrating the procedure. I also carried out an experiment at UCS during the lunch break with volunteers, who were boys from forms 2 and 3 who had already been tested with the rest of their class.

As in my previous experiments on the sense of being stared at (Sheldrake [1999]), the children worked in pairs, one (the subject) sitting with his back towards the other (the looker). But rather than being in the same room, the subjects and lookers were separated by closed windows. The lookers were inside the school laboratory or in a classroom, while the subjects sat in a row on chairs outdoors, in the playground, with their backs to the windows. The lookers were arranged in a row in the same order as their partners, so they could look straight out of the window at them.

In most experiments the distance between lookers and subjects was 3-5 metres, and the lookers were looking out of ground-floor windows, but in one case (class 2C at UCS) the lookers were in a first-floor classroom and the subjects were 100 metres away, at the other side of the playground.

In all cases, the subjects wore blindfolds, kindly supplied by Virgin Atlantic Airways, of the type widely used by air passengers in order to sleep on planes.

In a set of 20 trials, in a random sequence, the looker either looked at the back of the subject or looked away, and was instructed to think of something else. The random sequence was set out on previously prepared instruction sheets, with 24 different random sequences of looking and not-looking trials, compiled on the basis of standard random number tables. These sheets were given to the lookers only after the subject was in place and unable to see the sheet. Subjects were given score sheets, with a list of numbers from 1 to 20, one for each trial, and a space in which they entered their guess for each trial: if they guessed that they were being looked at they put a tick, and if they guessed they were not being looked at, they put a cross.

Before the test began, both lookers and subjects were asked to write their names and the names of their partners on their sheets. I myself was indoors with the lookers, and told them which trial was about to begin, so they could consult their instruction sheets and see whether or not they were to look in this trial. Each looker had a different instruction sheet with a different randomization. When they were ready, either looking at their partners or looking away, I said "Start", and pressed a button which rang an electric chime in the playground, indicating to the subjects that the trial was beginning.

At the end of the10-second trial period, the subjects were told by the teacher supervising them in the playground to write down their guess. To do this they had to raise their blindfold, replacing it when they had written down their tick or cross. When all the subjects were ready, with their blindfolds in place, for the next trial, the teacher gave me a signal, and I instructed the lookers to get ready for the next trial.

(A slightly different procedure was used with classes 3A and 3Q at UCS, in that beside each subject sat a scorer, who also had his back to the window. The subjects told their scorers their guesses, and the scorers recorded these guesses on the score sheets.)

In these experiments the subjects received no feedback about whether their guesses were right or wrong during the course of the session. Nor did the lookers know what guesses the subjects had made.

At the end of the session, when all 20 trials were completed, the lookers and subjects gave their sheets to me, and I stapled them together in pairs, so that each subject's guesses could be compared with the looker's instructions, to see how many guesses were right or wrong. The subject and looker then changed roles, and the procedure was repeated.

Experiments in other countries :

After completing this series of experiments in London, in order to find out if the results could be replicated independently, and I asked three schoolteachers in other countries to carry out similar tests in which the lookers and subjects were separated by windows. All three teachers had already carried out staring experiments with the lookers and subjects in the same room, and were familiar with this experimental method. The details of these experiments were as follows:

  1. Laura Beatty, a teacher in Connecticut, USA, carried out this experiment with her fourth grade class (aged 8-9) in School, in November 1997. The lookers and subjects were separated by a glass partition wall. Beatty gave the signal for the beginning of each trial. The subjects did not wear blindfolds, but their eyes were closed during the trial and they had their backs to the lookers so they could not see them.
  2. Helmut Lasarcyk, a teacher in the Stormarnschule, Ahrensburg, Germany, did the experiment in June 1997 with 9 pupils from grade 12 (aged 18).The subjects were outdoors, separated by a window from the lookers in a room on the ground floor.They did not wear blindfolds. The teacher stood in a doorway overlooking both sides and gave verbal signals for the beginning of each trial.
  3. Greg Wisnicki, a teacher at Sinclair Secondary School, Port Hope, Ontario, Canada, carried out his tests between September 1997 and April 1998 with pupils aged 14-18, who volunteered to participate in the context of a directed studies programme. The trials took place in the school chemistry laboratory, which was adjacent to a preparation room with a two-sided chemical fume hood in between. The subjects were looked at through the fume hood, with the plexiglass shields raised. Thus in this experimental set-up they were not looked at through windows, but possible olfactory and auditory clues were reduced or eliminated in other ways, namely by switching on the fume hood extraction fan, thus removing possible odours that could have travelled from lookers to subjects, and also providing background noise that reduced the possibility of auditory clues. In addition the subjects wore industrial sound eliminators, and they were also blindfolded. They sat with their backs to the starers, 2.5 metres away. One subject at a time was tested, and was looked at by two starers, who were given a freshly generated set of instructions that indicated whether they should look or not look at the subject in a series of 10 trials. These random sequences were determined by a random number generator on a calculator. On a given day, odd numbers were taken to mean staring and even numbers not staring, and on the subsequent day the meaning of odd and even numbers was reversed. The lookers and subjects were signalled at the beginning of each trial by a signaller, who was out of sight. The signal to the lookers was seeing a card with the number of the trial on it; the signal to the subject was the activation of a small electric motor held in the left hand. In looking trials the lookers looked at the subject through the fume hood; in the not looking trials they looked elsewhere. At the end of each 10-second trial period, the subject indicated his guess by moving a toggle switch forward or backward For the first 50 subjects, moving the toggle switch forward meant they guessed they were being looked at, and backward not looked at. For the next 50 subjects these meanings were reversed; and for the last 23 subjects they were reversed again. When the toggle switch was moved forwards a red light flashed and when backwards a green light flashed, and these guesses were recorded by a recorder who was out of sight of the lookers. Thus the lookers did not know what the subject's guesses were, nor whether they were right or wrong, and the recorder did not know whether the subjects had been looking or not. The results from each session were later collated with the instructions to the starers, and the data tabulated.

Analysis of the data:
The numbers of right and wrong guesses from each set of trials carried out by each looker-subject pair were tabulated in three columns;
"Looking",    "Not looking" and    "Total",
enabling the total number of right and wrong guesses in each column to be obtained.

For each set of trials, in each column, the data were also scored as follows:

+ if the subject made more right than wrong guesses
- if the subject made more wrong than right guesses
= if the number of right and wrong guesses was the same.

Statistical analysis was carried out in three ways.
    First, the chi-squared test was used to compare the total numbers of right and wrong guesses in each column. The null hypothesis was that the numbers of right and wrong guesses would be the same.
    Second, the chi-squared test was used to compare the total numbers of + and - scores. The = scores were disregarded. The null hypothesis was that by chance the number of + and - scores would be equal.
I am indebted to Professor Nicholas Humphrey for suggesting this method of analysis.
    Third, the proportion of right guesses by each subject was compared with the chance level of 50% using the the paired-sample t-test.
The null hypothesis was that the proportion of right guesses would be 50%.
I am grateful to Professor Patrick Bateson, F.R.S. for suggesting this method.
For the comparison of two sets of scores, (for example the scores with and without blindfolds) 2 x 2 contingency tables were used (Campbell [1989]), with the null hypothesis that the proportions of right and wrong guesses in both sets were equal.

3. Subjects Guess Better When Looked At Than in Control Trials

The subjects made more correct guesses when they were being looked at than when they were not both in my own experiments in London schools (Table 1) and in the tests in other countries (Table 2). Overall, the percentage of correct guesses was 55.2% in the looking trials and 50.8% in the control trials.

Both in London and in other parts of the world the scores in the looking trials were very significantly above chance levels, while in the not looking trials they were not significantly different from the chance level of 50% (Tables 1 and 2).

Overall, combining the results of looking and not-looking trials, there were 2544 correct guesses as opposed to 2254 incorrect guesses (53.0% correct), and this difference was very significant statistically (p<0.00003).


A very similar pattern was apparent when the data were analysed in an alternative way whereby each subject was scored positive (+) if more guesses were right than wrong and negative (-) if more guesses were wrong than right (Tables 1 and 2). Combining the data from Tables 1 and 2, in the looking trials the scores were 160+ 82- , an extremely significant difference (p<1x10-6), while in the not-looking trials the scores showed only a small excess of positive over negative scores (134+ 110-) that was not significant statistically (Figure 1). For the looking and not-looking trials combined the excess of positive over negative scores (150+ 94-) was significant at the p<0.0004 level.

The data were also analyzed by means of a paired-sample t test subject by subject. By this method too the excess of positive over negative guesses was very significant in the looking trials, not significant in the not-looking trials, and significant for the looking and not-looking trials combined (Tables 1 and 2).

An inspection of the detailed data from the different classes in London (Table 1) and from the schools in other parts of the world (Table 2) shows that this general pattern was consistent and repeatable. In all cases there was an overall excess of positive over negative scores. back to the top

4. Factors Affecting Subjects' Scores

These experiments show that people really can tell when they are being stared at from behind, even when they are looked at through windows. They confirm the results of previous experiments carried out in the same room, with very significantly more positive than negative scores in the looking trials, and scores close to chance levels in the not-looking trials (Sheldrake [1998, 1999]).

The experiments reported in this paper (Table 1) impose even more rigorous safeguards against possible artefacts in that the subjects were blindfolded and given no feedback and separated from the lookers by closed windows which effectively eliminated the possibility of auditory and olfactory clues. In the Canadian experiments these possibilities were reduced yet further by the extraction of any possible smells through the fume cupboard, by the subjects wearing industrial sound eliminators, by the background noise of the extractor fan, but still the results showed the usual pattern with positive scores in the looking trials and scores close to chance levels in the not-looking trials (Table 2).

The main difference between the results of the experiments in the same room from those that involved looking through windows is that the positive scores were generally higher in the looking trials in the same room: 59% correct (Sheldrake [1999]) as opposed to 55% through windows. The difference in scores was statistically significant (p<0.05). At first sight this results suggests that looking through windows weakened the ability of subjects to know when they were being looked at. However, they may be another reason for this difference, due to the fact that in same-room experiments, the scores are generally higher when the subjects are given feedback (Sheldrake [2000]), which could have encouraged them and helped them to take more interest in the experiment. The overall figure of 59% correct guesses in the same-room experiments included two kinds of procedure, one with and the other without feedback. The percentage of correct guesses in the no-feedback same-room experiments was 54.2% (from the combined results in Tables 3 and 4 in Sheldrake [1999]). Thus in the experiments described in this paper, in which the subject were not in the same room and received no feedback, the percentage of correct guesses was very similar to the same-room experiments without feedback. The apparent lowering of the subjects' scores seems to be explicable in terms of the lack of feedback rather than in terms of looking through windows. I conclude that the elimination of possible auditory and olfactory clues made very little difference to the subjects' performance.

There are two other differences between these experiments carried out through windows and experiments in the same room which could have affected the results, perhaps reducing the success rate. First, in the experiments reported in this paper the lookers received no feedback about the subjects' guesses, whereas in previous experiments the lookers did receive such feedback (Sheldrake [1998], [1999], [2000]) and this could have affected the motivation or the concentration of the lookers. Second, in my experiments reported here the subjects and lookers sat in rows, and although lookers were told to look only at their own partner, in several cases I noticed that lookers were glancing at other peoples' partners, which could have weakened the staring effect for a looker's own partner (thus reducing positive scores in looking trials), and confused other lookers' partners in trials in which they were not supposed to be looked at (thus increasing negative scores in not-looking trials). Nevertheless any such effects seem to have been small. back to the top.

5.Confirmation from Experiments Using Closed Circuit Television

The experiments described in this paper confirm that people can indeed tell when they are being looked at from behind in a way that cannot be explained in terms of the normal senses.

This conclusion is confirmed by an independent series of experiments in which subjects were looked at through closed circuit television (CCTV), while their galvanic skin response was recorded continuously, as in lie-detector tests. In these experiments, the lookers and subjects were in different rooms, and the subjects could not have received any clues about when they were being looked at through normal sensory channels. The subjects in these CCTV experiments were not asked to make conscious guesses about when they were being looked at or not; their physiological responses were unconscious. Yet there were significant differences differences in their skin resistance when they were being looked at on a TV monitor in another room (Braud, Shafer and Andrews [1990], [1993a], [1993b]; Schlitz and LaBerge [1994], [1997]; Wiseman and Schlitz [1997]).

So far, practically all the experimental investigations of the effects of being stared at have involved unselected groups of subjects and lookers. The overall positive effects are an average, and of course include data from subjects whose scores were at or below chance levels. There may well be consistent differences between subjects in their sensitivity to being looked at, and some subjects' sensitivity may also improve with practice. In experiments in a German school in which sensitive subjects were selected and then tested repeatedly, the overall percentage of correct guesses was more than 70%, and two of the subjects were right more than 85% of the time (Sheldrake [1998]). Interestingly, the scores of these repeatedly-tested subjects were high not only in the looking trials, but also in the not-looking trials, suggesting that through practice people can learn to tell when they are not being looked at by detecting the difference from when they are being looked at.

There may also be differences between people in their effectiveness as lookers, and there is already evidence from the CCTV experiments of Wiseman and Schlitz [1997] that this is the case. When Schlitz was the looker, there were significant differences in the skin resistance of subjects when she looked at them through the TV monitor, while there was no significant effect when Wiseman (a sceptic) was the looker.

However, even with unselected lookers and subjects the evidence for the reality of the effect of being looked at is already very strong, and the results summarized in this paper indicate that this effect does not depend on normal channels of sensory communication. It could be described as a kind of extrasensory perception or sixth sense, but these terms merely restate the fact that it cannot be explained in terms of the known senses.

I conclude that peoples' ability to know when they are being looked at depends on an influence at present unknown to science. Such influences may play an important part in predator-prey relationships ( [1999]), and have far-reaching implications for our understanding of the nature of the mind (Abraham, McKenna and Sheldrake [1992]; Sheldrake [1994]). back to the top

Acknowledgements

I am grateful to all the people who took part in these experiments, to John Hubbard, the Headmaster of University College School Junior Branch for enbling me to do these experiments in his school, to the class teachers who helped conduct the tests, and to Laura Beatty, Helmut Lasarcyk and Greg Wisnicki for sending me their results and agreeing to their publication in this paper. I thank the Institute of Noetic Sciences, Sausalito, CA, the Lifebridge Foundation, New York and the Bial Foundation, Portugal, for financial support.

References

Abraham, R., McKenna, T. and Sheldrake, R. [1992], Trialogues at the Edge of the West. Bear and Co., Santa Fe.

Braud, W., Shafer, D. and Andrews, S. [1990], Electrodermal correlates of remote attention: Autonomic reactions to an unseen gaze. Proceedings of Presented Papers, Parapsychology Association 33rd Annual Convention, Chevy Chase, MD, pp14-28.

Braud, W., Shafer, D. and Andrews, S. [1993a], Reactions to an unseen gaze (remote attention): A review, with new data on autonomic staring detection. Journal of Parapsychology 57: 373-390.

Braud, W., Shafer, D. and Andrews, S. [1993b], Further studies of autonomic detection of remote staring: replications, new control procedures, and personality correlates. Journal of Parapsychology 57: 391-409.

Campbell, R.C. [1989], Statistics for Biologists Cambridge University Press, Cambridge.

Cottrell, J.E. and Winer, G.A. and Smith, M.C. [1996], Beliefs of children and adults about feeling stares of unseen others. Developmental Psychology 32: 50-61.

Schlitz, M. & LaBerge, S. [1994], Autonomic detection of remote observation: two conceptual replications. Proceedings of Presented Papers, Parapsychology Association 37th Annual Convention, Amsterdam, pp. 352-60.

Schlitz, M. & LaBerge, S. [1997], Covert observation increases skin conductance in subjects unaware of when they are being observed: a replication. Journal of Parapsychology 61: 185-194.

Sheldrake, R. [1994], Seven Experiments that Could Change theWorld, Chapter 4. Fourth Estate, London.

Sheldrake, R. [1998], The sense of being stared at: Experiments in schools. Journal of the Society of Psychical Research 62: 311-323.

Sheldrake, R. [2000], The sense of being stared at: Effects of blindfolding subjects and giving them feedback. Journal of the Society of Psychical Research (in the press).

Wiseman, R. and Schlitz, M. [1997], Experimenter effects and the remote detection of staring. Journal of Parapsychology 61: 197-207.

Journal of Scientific Exploration 22, 517-527 (2008)
by Rupert Sheldrake and Pamela Smart

Introduction

The sense of being stared at is well known. Most people claim to have turned around to find that someone was looking at them; most people also claim to have caused other people to turn round by looking at them. (Sheldrake, 2003). Scopesthesia is a newly coined scientific term for this phenomenon (Carpenter, 2005), which is also referred to in the research literature as "unseen gaze detection" (Wiseman & Smith, 1994), "staring detection" (Braud, 2005), "non-visual staring detection" (Sheldrake, 2005b) or "remote staring detection" (Baker, 2005).

The simplest tests for this phenomenon involve people working in pairs. One person, the staree, sits with her back to the other, and usually wears a blindfold. In a randomized sequence, the starer either stares at the back of the staree's neck or looks away and thinks of something else. The beginning of each trial is signalled by a click, beep or bell. In over 30,000 trials of this kind the overall hit rate was 55%, very significantly above the mean chance expectation of 50%. The hit rates were also significantly above chance in several studies in which starers and starees were separated by windows or one-way mirrors (for a review, see Sheldrake, 2005a).

It is a general principle of sensory physiology that the senses detect changes or differences. Perhaps the same principles apply to scopesthesia. In all the tests conducted so far, before the trial the starer was not looking at the staree. Then in looking trials, he started looking at her, while in not-looking trials he continued not looking. Thus at the beginning of looking trials there was a change in his attention, and in not-looking trials there was not. In the tests described in this paper we explored whether starees were more sensitive when the starer changed from not looking to looking, or vice versa. In half the tests, all 20 of the trials in the test began with the starer looking. Then, as a signal was given to the staree, the starer either looked or did not look according to a randomized schedule. Thus in the looking trials there was no change, and in the not-looking trials there was a transition of attention. In the other tests, as in standard staring experiments, the situation was opposite: in looking trials there was a change of attention from not-looking to looking, while in not-looking trials there was no transition.

Most experiments on scopesthesia have tested participants only once or twice. In the present series of investigations, the same participants were tested repeatedly. This enabled us to find out whether or not there were any changes over time in the staree's hit rates. In these tests, starees received immediate feedback after each trial as to whether their guess was correct or not. Their hit rates would increase if they were improving with practice, or decline if they were becoming bored with the testing process or losing their ability to make accurate guesses for any other reason.

We also carried out control tests to find out what patterns of guessing occurred in the absence of scopesthesia. Control experiments enable several fundamental questions to be explored. Are above-chance results in scopesthesia tests due to some ability other than scopesthesia? If so, what? First, significant positive results may depend on a detection of subtle sensory cues. Second, the staree might pick up the starer's intentions telepathically, rather than through staring itself. Third, the staree might be picking up the starer's written instructions by clairvoyance. Fourth, in tests in which trial-by-trial feedback is given, starees might be picking up the answers precognitively.

If positive results in scopesthesia tests depend on any of these other kinds of information transfer, then when scopesthesia is eliminated in control experiments, the hit rates should still be above chance.

One way to carry out control experiments would be to tell the starees that these are standard staring tests and that they will be stared at, or not stared at, in a random sequence. In fact, they are not looked at in any of the trials.

We did not want to base an experiment on deception. Instead, in our control tests, the starer and staree sat with their backs to each other. The starer did not look at the staree at all, and the staree knew that this was the case. The staree was asked to guess what instruction the starer was receiving in each trial. Using standard randomized instruction sheets, the starer looked at the instruction, signalled the beginning of the trial to the staree by means of a standard mechanical click, and the staree then guessed whether the instruction was "look" or "no". The staree received immediate feedback as to whether the guess was correct or not. We analysed the results of tape-recorded experiments to find out how frequently errors occurred in the recording of data.

Method

Participants

There were two participants in these tests who took turns at being starer and staree: Pam Smart (PS) and her 14-year old niece, JM, who was paid for her participation. These participants had already taken part in standard staring tests, and were familiar with the general procedure.

Tests

The tests took place in JM's family's house with both participants in the same room, sitting about 2 metres apart. The staree was blindfolded. Each test consisted of 20 trials, and was conducted in accordance with a randomized instruction sheet. There were 20 different randomized sheets altogether, and the sequence of looking and not-looking trials was determined by a random number generator. These 20 sheets were used repeatedly, but in different orders each time. Some sheets had equal numbers of looking and not-looking trials, while others had unequal numbers. Thus, by chance in some sets of data there were unequal numbers of looking and not-looking trials.

The participants carried out a series of between 14 and 16 tests in a session, and in each test there were 20 trials. After one 20-trial test was completed, the participants changed roles. The sessions occurred at roughly weekly or two-weekly intervals. The dates of these sessions are given in Table 1. For the tests on attentional transitions there were 10 sessions, followed by 3 sessions of control tests. All sessions began at 4 pm apart from the session of February 15, which began at 1 pm.

Just before the beginning of each trial, the starer looked at the instruction sheet and read the instruction "look" or "no", then signalled the beginning of the trial by means of a mechanical clicker, which gave a sound of standard intensity. For the tests on stimulus transitions, each trial was divided into two parts, the first lasting 3 seconds. The beginning of the second part of the trial was signalled by another click. A device used for training dogs emitted these clicks: the first was produced by pushing in a metal flange, and the second by releasing it three seconds later.

There were two kinds of test: in "looking tests", during the first part of each trial, the starer looked at the staree, and then, 3 seconds later, as the second click was sounded, followed the randomized instruction "look" or "no". These were designated L tests. In the second kind of test, designated N ("not-looking"), during the first part of each trial the starer did not look at the staree, and then after 3 seconds either looked or continued not looking in accordance with the randomized instruction. Thus in L tests, at the second click the starer either continued to look, or changed to not looking; in the N tests at the second click the starer either continued not looking, or changed to looking. The staree knew whether the test was L or N.

PS determined at random, by the toss of a coin, whether the first test in a session was L or NL, and then each staree alternated between L and NL tests throughout the session. The starees knew whether they were taking part in an L or N test.

The staree guessed out loud "looking" or "not looking" within 10 seconds of hearing the second click, and received immediate feedback as to whether this guess was correct or not. The starer recorded the result on the instruction sheet, and proceeded to the next trial.

In the control tests, the starer and starees sat with their backs to each other. As usual the staree was blindfolded and just before each trial the starer looked at the instruction sheet to see if the trial was "look" or "no", and signalled the beginning of each trial by means of a mechanical clicker. Unlike the dog clicker used in the stimulus transition tests, this clicking device emitted a standard single click. The starer did not look at the staree at any stage during the test. The staree was asked to guess what instruction the starer had received, and made this guess within 10 seconds of the trial beginning.

Error detection

All attentional transition tests were tape-recorded so that the pattern of clicks and responses could be evaluated independently at a later date "in a "blind" fashion". This evaluation was carried out by Kayleigh Allenby (KA), who did not know either of the participants and lived 200 miles away. KA listened to the tapes of the trials, noting down the trial number and date, and then recording what guess the staree made in each trial. While Rupert Sheldrake (RS) had the original score sheets, PS then entered the guesses recorded by KA on duplicates of the original score sheets for each test. RS then compared these score sheets with the originals so that discrepancies could be detected. When such discrepancies were found, RS listened to the tape recording of that test to determine whether there was any error in the evaluation of the tape by the evaluator. If there had not, then the discrepancy was due to a recording error by the starer.

Scoring and statistics

As usual in staring tests, the number of correct and incorrect guesses in looking and not-looking trials were tabulated separately, along with the total for each test (Sheldrake, 2000). As in previous research the totals were also evaluated by means of the sign method, with scores of 11 or more out of 20 given a "+" sign, scores of 9 or less a "-" sign and scores of 10 an "=" sign. The advantage of the sign method is that it gives an equal weighting to each test. The chance expectation was that 50% of the guesses would be correct, and also that the number of + signs would be equal to the number of signs, ignoring the number of = signs. The null hypothesis was tested using the binomial test. For comparisons of data from tests under different conditions the 2x2 chi squared test was used.

Results

Attentional transition tests

The overall hit rate for all the attentional transition tests was 1477/2800, or 52.8% (p=0.002). By the sign method the results were 69+34-23= (p=0.0005).
For the L tests, in which the staree was looked at during the first 3 seconds of the trial, the hit rate was 52.9% (p=0.01) or 39+23-13= (p=0.02) (Table 2).
For the N tests, in which the staree was not looked at during the first 3 seconds of the trial, the hit rate was 52.5% (p=0.03) or 36+16-13= (p=0.004). The hit rate for the L tests was significantly greater than for the N tests on the basis of scores (p=0.03) but not on the basis of signs.
The two starees had slightly different hit rates (Table 1): overall, PS scored 53.2% (p=0.01) or 37+21-12= (p=0.02) and JM 52.3% (p=0.04) or 38+18-14= (p=0.005). These differences between the two starees's scores were not statistically significant.
In the L tests, the hit rates in looking trials were slightly higher than in not-looking trials, 53.5% and 52.3% respectively. In the N tests, the reverse was the case, with 52.4% in looking and 52.7% in not-looking trials. Overall, the score was slightly higher in looking trials (53.0%) than in not-looking trials (52.5%) but these differences were not statistically significant.

In most previous staring experiments, the total number of "looking" guesses was greater than the total number of "not looking" guesses; in other words there was a response bias in favour of "looking" (Sheldrake, 2005). In these tests, the total number of "looking" guesses was 757 (looking/hits) + 651 (not looking/misses) = 1408/2800 or 50.3%, not significantly different from the chance level of 50%. However, this average figure conceals a striking difference between the two starees. PS guessed "looking" in 53.2% of the trials, while JM did so in only 47.4% of the trials, a significant difference (p=0.002).

Control tests

The results of these control tests are shown in Table 3. The overall hit rate of 49.3% was not significantly different from the chance level of 50%, nor were the scores of the individual participants: JM's hit rate was 48.5% and PS's 50.0%.

Both participants scored below the chance level in looking trials, and above the chance level in not-looking trials. This effect was due to a response bias whereby both of them guessed "not looking" more often than "looking": only 42.2% of JM's guesses were "looking" and 46.2% of PS's. Overall, the percentage of "looking" guesses was 44.3%.

Changes with time

The hit rates in the 10 sessions of attentional transition tests and for the 3 sessions of control tests that followed them are shown in Figure 1.

There was no systematic trend in the data, neither a regular improvement from session to session, nor a regular decline. In the first three attentional transition tests and in the control tests the fluctuations for the two starees moved in similar directions, but in most other sessions they moved in opposite directions.

Within each session, each staree took part in 3 or 4 L and N tests. The average data testwise are shown in Figure 2. Again there was no clear trend. In the L tests the hit rates were higher in tests 3 and 4 than in the first 2 tests, but this was not the case in the N tests.

Error rates

The number of errors in recording the data was determined from tape recordings of all the attentional transition tests. Out of a total of 2,800 trials, there were 5 discrepancies. Of these, two were owing to errors made by the evaluator in writing down the guesses on the tape recording. Three were errors made by the starer in noting down the staree's guesses, an overall error rate of 0.1%. All three errors were in not-looking trials; two were false positives and one a false negative, giving an overall error of just one false positive, giving a net error rate of 0.04%.

Discussion

This is one of the first studies on scopesthesia in which the same participants have been tested repeatedly. A surprising result was that there was no obvious tendency for hit rates to increase or decline (Fig.1). Since the starees were receiving trial-by-trial feedback, their hit rates might have been expected to improve with practice, but this was not the case. In the first 3 sessions and in the control sessions, the fluctuations were similar with both starees, suggesting that external factors may have influenced both similarly, but there was no similar pattern in the other sessions. Within sessions, there was a tendency for hit rates to increase in L tests but not in N tests (Fig. 2). The lack of systematic trends suggests either that there was no tendency to improve with practice, or that any such tendency was offset by a countervailing tendency, such as boredom.

A second surprise was that the number of errors in recording the data was so low. The net error rate of 0.04% was negligible.

The fact that hit rates were at chance levels in the control tests shows that the above-chance hit rates in the scopesthesia tests cannot be ascribed to telepathy, clairvoyance or precognition, or to any subtle sensory cues that were common to both kinds of test.

There was surprisingly little effect of attentional transition, with very similar overall results from L and N tests. The attentional transitions within these two kinds of tests did not result in higher hit rates than the trials in which there was no transition; indeed there was a tendency for the reverse to be the case. In the L tests, all of which started with the starer looking at the staree, there was no transition in the looking trials, because the starer simply continued to look. The transition occurred in the not-looking trials when the starer stopped looking. Yet the looking trials gave a slightly higher hit rate than not-looking trials. Conversely in the N trials, all of which started with the starer not looking at the staree, not-looking trials involved no transition and gave a slightly higher hit rate than looking trials.

This lack of effect of the transitions is not what we expected. It implies either that scopesthesia differs from other senses in not responding to changes or differences, or that the tests we carried out were too insensitive to detect them, or inappropriately designed. Probably the best way to detect such transitions would not be to signal when they occur, as we did, but to create a situation in which the transitions occurred unpredictably, and to monitor people's response to them physiologically, for example by the galvanic skin response.

Overall, the pattern of results differed from the typical pattern in staring experiments, where there are usually above chance hit rates in looking trials, around 60%, and chance-level hit rates, around 50%, in not-looking trials (Sheldrake, 2005). Here, the score was only slightly higher in looking trials (53.0%) than in not-looking trials (52.5%) and the difference was not statistically significant.

The usual pattern in standard staring experiments could arise because of a response bias in favour of saying "looking" rather than "not looking". In the absence of scopesthesia, a 5% bias would give a 55% hit rate in looking trials and 45% in not-looking trials, with an overall average of 50% (Schmidt, 2001). If staring detection occurred in both kinds of trial at 5% above chance, then the hit rate would be 60% in looking trials and 50% in not-looking trials, as observed.

In the attentional transition experiment described in the present paper, there was no significant overall response bias: 50.3% of the guesses were "looking". Thus in looking trials, taking into account the response bias, the hit rate was 2.7% above chance, and for not-looking trials 2.8%, not significantly different. However, the two starees had significantly different response biases. PS's response bias followed the more common pattern in that it was in favour of looking. With her response bias of 3.2% in favour of "looking", the hit rate in looking trials of 56.3% was 3.1% above this chance level. In the not-looking trials her response bias would give a chance hit rate of 46.8%; the actual hit rate of 50.0% was 3.2% above this. JM's response bias went in the opposite direction: only 47.4% of her guesses were "looking", or in other words her response bias was 2.6%. Her hit rate in looking trials was 49.6% and in not-looking trials was 55.0%, which were 2.2% and 2.4% respectively above the chance level expected on the basis of her response bias.

These results closely fit a simple model for the probability P of a hit in looking and not-looking trials:

P (hit/looking) = 1/2 + b + s
P (hit/not-looking) = 1/2 - b + s

Where b is the response bias, positive when the percentage of looking guesses is greater than 50%, and s is the effect of scopesthesia, with equal contributions in looking and not-looking trials.

In the control tests, both starees showed a response bias in favour of saying "not looking", with "not looking" guesses making up 55.7% of the total. This bias may well have reflected the fact that they both knew that in these tests they were never being looked at.

The fact that both participants served as starees in over 70 tests each makes them unusually experienced, and the results in these experiments may not be representative of naïve participants with little or no previous experience. This is something that only further research can reveal.

A possible problem with the attentional transition test described here was that the starers gave two signals to the starees in each trial using a mechanical clicking device: the first signal indicated that the pre-trial period had begun. In the L tests this meant that the starer was looking at the staree, in the N tests she was not. The second click indicated the beginning of the randomized trial in which the starer would either be looking or not looking. The starer estimated the 3-second interval between the two clicks. This raises the possibility that she might have given subtle cues by unconsciously varying the length of the interval. However, the starees themselves were not aware of any differences of this kind. But perhaps they picked up subtle cues unconsciously. Unfortunately, we were unable to resolve this question definitively by a precise timing of the click interval trial by trial because the tape recordings were inadvertently discarded.

Starees might also have been influenced by other subtle cues such as slight sounds from the starer as she turned her head. Another possible flaw was that we reused the same 20 randomized sheets and hence it is conceivable that starees might have unconsciously remembered the randomized sequences after they were exposed to them repeatedly. If so, the feedback they received should have enabled them to improve their scores very considerably with practice. But this did not happen. In future experiments, possible auditory cueing should be minimized either by the use of sound-proof windows separating the participants, or by the use of ear plugs or headphones. Also a fresh randomization procedure should be used for each test.

Because the double-click procedure is potentially capable of introducing artefacts, it should be avoided in any further research on attentional transitions. A better method would be to use an electronic beeper that emits two beeps with a 3-second interval between them. An even simpler procedure would be for the starer to give just one signal at the beginning of the trial, having looked for three seconds previously in the L tests, or not looked in the N tests.

Acknowledgements

We thank David Acunzo for helpful comments on drafts of this paper. We are grateful for financial support from the Institute of Noetic Sciences, Petaluma, California, the Perrott-Warrick Fund, administered by Trinity College, Cambridge, and Mr Addison Fischer, of Naples, Florida.

References

Baker, I.S. (2005) Nomenclature and methodology. Journal of Consciousness Studies, 12, 56-63.

Braud, W. ( 2005) The sense of being stared at: fictional, physical, perceptual or attentional/intentional? Journal of Consciousness Studies, 12, 66-70.

Carpenter, R.H.S. (2005) Does scopesthesia imply extramission? Journal of Consciousness Studies, 12, 76-77

Schmidt, S. (2001) Empirische Testung der Theorie der morphischen Resonanz - Können wir entdecken wenn wir angeblikt werden? Forschende Komplentärmedizin, 8, 48-50 [In German].

Sheldrake, R. (2000) The 'sense of being stared at' does not depend on known sensory clues. Biology Forum, 93, 209-224.

Sheldrake, R. (2003) The Sense of Being Stared At, And Other Aspects of the Extended Mind. London: Hutchinson.

Sheldrake, R. (2005) The Sense of Being Stared At, Part 1: Is it real or illusory? Journal of Consciousness Studies, 12, 10-31.

Wiseman, R. & Smith, M. (1994) A further look at the detection of unseen gaze. Proceedings of the Parapsychological Association 37th Annual Convention. Parapsychological Association. pp. 465-478.

TABLE 1

Dates of sessions of attentional transition tests (sessions 1-9) and control tests (sessions 10-12). All took place in 2005.

Session Date Session Date
1 Jan 31 8 April 12
2 Feb 7 9 April 18
3 Feb 15 10 April 25
4 Feb 21 Controls
5 Feb 28 11 May 16
6 March 7 12 May 23
7 March 14 23 June 6

TABLE 2

Scores in staring tests where trials began with 3 seconds looking (L) or not looking (N) with 2 starees (PS and JM). The numbers of hits and misses are shown for looking trials, not-looking trials and totals. The signs indicate the number of tests in which the hit rate was 11/20 or more (+), 9/20 or less (-) or 10/20 (=).

Staree Test Looking Not Looking Totals
hit miss hit miss hit miss hit% signs
PS L 215 156 178 171 393 327 54.6 21+11-4=
PS N 189 158 163 170 352 328 51.8 16+10-8=
JM L 189 195 212 184 401 379 51.4 18+12-9=
JM N 164 163 167 126 331 289 53.4 20+6-5=
Totals
PS   404 314 341 341 745 655 53.2 37+21-12=
JM   353 358 379 310 732 668 52.3 38+18-14=
  L 404 351 390 355 794 706 52.9 39+23-13=
  N 353 321 330 296 683 617 52.5 36+16-13
Grand Totals 757 672 720 651 1477 1323 52.8 75+39-26=

TABLE 3

Control tests. The numbers of hits and misses are shown for looking trials, not-looking trials and totals. The signs indicated the number of tests in which the hit rate was 11/20 or more (+), 9/20 or less (-) or 10/20 (=).

Staree Looking Not Looking Totals
hits misses hits misses hits misses signs
JM 95 138 128 99 223 237 8+13-2=
% 40.8   54.0   48.5    
PS 111 129 129 111 240 240 11+12-1=
% 46.3   53.8   50.0    
Total 206 267 257 210 463 477 19+25-3=
% 43.5   55.0   49.3    

Journal of the Society for Psychical Research 62: 311-323 (1998)
by Rupert Sheldrake

Introduction

Many people have had the experience of turning round with the feeling that someone is looking at them from behind, to find that this is in fact the case. Conversely, many people have found that they can sometimes make people turn around just by looking at them. Surveys show that between 70 and 97% of the population in Europe and North America have had personal experience of this phenomenon (Braud, Shafer and Andrews, 1990; Sheldrake, 1994; Cottrell, Winer and Smith, 1996). Indeed it seems to be well known all over the world.

In spite of its familiarity, there has been very little research on this subject, with fewer than a dozen reported experimental investigations over the past 100 years, including some that are buried in unpublished student theses.

The first two reports, by Titchener (1898) and Coover (1913) were negative. Both investigators were sceptics. Titchener, one of the founders of experimental psychology in the United States, carried out experiments with himself as the looker. His subjects were students who claimed they could sometimes tell when they were being looked at. He gave no details of these experiments save to say that they were negative, fully confirming his expectations. He continued:

If the scientific reader object that this result might have been forseen, and that these experiments were, therefore, a waste of time, I can only reply that they seem to me to have their justification in the breaking down of a superstition which has deep and widespread roots in the popular consciousness. No scientifically-minded psychologist believes in telepathy. At the same time the disproof of it in a given case may start a student upon the straight scientific path, and the time spent may thus be repayed to science a hundredfold. (Titchener, 1898; p. 897).

Coover (1913), following Tichener's lead, carried out a series of experiments with his students at Stanford, and obtained statistically non-significant results. He concluded that the widespread belief in the feeling of being stared at was "groundless".

As far as I know, no further experiments were reported in the scientific literature until Poortman (1959) described some trials he carried out with a woman friend. He was right more often than wrong in guessing when she looked at him.

The experimental design used by Coover and Poortman, involves subjects and lookers working in pairs. The subject is looked at from behind in a series of trials, randomly interspersed with an equal number of control trials when the subject is not looked at. At the end of each trial period, the subject says whether or not he or she is being looked at. The response is scored right or wrong and written down, and the next trial begins.

Nearly two decades passed after Poortman's report before Peterson (1978), in an unpublished Master's Thesis at the University of Edinburgh, described experiments in which the looker and the subject were separated by a one-way mirror. The looker was invisible to the subject, and sat in a closed booth that reduced, if not elimated, any possible cues from the looker to the subject through sounds and smells. The subjects were right significantly more often than not.

The next student project, by Williams (1983) at the University of Adelaide, Australia, found a statistically significant effect when subjects and lookers were in entirely different rooms. The subject was looked at through closed-circuit TV.

Almost all subsequent published research has involved closed circuit TV, with the additional sophistication that the subjects are not even asked to guess whether they are being looked at or not. Electrodes are attached to the subjects' fingers and their galvanic skin response is recorded automatically, both when the subject is being looked at through closed circuit TV and in the randomly interspersed control periods. As in lie-detector tests, this is a simple way of monitoring responses of the autonomic nervous system. Does looking at the subjects, or not looking, significantly affects the subjects' unconscious responses?

In most investigations these experiments have given significant positive results (Braud, Shafer and Andrews 1990, 1993a, 1993b; Schlitz and LaBerge, 1994). In another study there were tendencies in a positive direction, but these were not statistically significant (Howat, Delanoy and Morris, 1994). Wiseman and Smith (1994) found a significant positive effect, but they speculated that this might be an artefact. Wiseman et al (1995) obtained overall nonsignificant results, but found a puzzling, statistically significant correlation between subjects' ability to detect an unseen stare and their reported level of luckiness.

In a recent closed-circuit TV experiment conducted in Wiseman's laboratory at the University of Hertfordshire, Schlitz obtained her usual positive results while Wiseman himself, a sceptic, obtained non-significant results (Wiseman and Schlitz, 1997). Subjects from a common pool were randomly assigned to the two experimenters, who were themselves the lookers, and the experiments were conducted under identical conditions. This evidence shows a striking experimenter effect, perhaps because of the transmission of positive or sceptical attitudes by the respective experimenters as they explained the experiment to the subjects before the trials began, and/or because of the differing effectiveness of the two experimenters as lookers.

I believe that this kind of laboratory research has an important role to play. But I also think that it is worth looking again at simple experiments of the kind described by Coover and Poortman, since these allow far more people to be tested, and open up research on this topic to much wider participation.

I have developed a version of this basic Coover-Poortman experiment which can be carried out by people working in pairs at home, at school, or indeed almost anywhere (Sheldrake, 1994). The procedure involves people working in pairs, with the looker sitting behind the subject. In a randomized series of trials, the looker either looks at the back of the subject's neck, or looks away and thinks of something else. These experiments are so simple that students can do them either as projects or classroom experiments in schools.

At least two student projects have already been carried out following this procedure. The first was by Michael Mastrandrea, an eighth-grade student, in Nueva Middle School, Hillsborough, California, and gave statistically significant positive results, with 54.1% correct guesses, 4.1% above the level expected by chance (Mastandrea, 1991). In Port Hope, Ontario, Canada, Jasmine James and Elaine Yau, senior students at high school, won the first prize in their local Science Fair with a staring experiment that again gave statistically significant positive results, with 55.0% correct guesses (James and Yau, 1996). Both these projects were, of course, carried out under the guidance of science teachers.

Several teachers in Germany and the United States have organized such experiments with their students, and have kindly sent me their results, which I describe in this paper. The overall conclusion from these experiments is that there is indeed a highly significant tendency for people to know when they are being looked at.

A re-examination of the supposedly negative findings of Coover (1913) shows that his data in fact agree well with the results reported here.

Methods

The basic experimental procedure is as follows. People work in pairs, one (the subject) sitting with his or her back towards the other (the looker). The distance between them is 2 metres or more. It is important to choose a place where there are no reflective surfaces (such as mirrors or windows) that would enable the subject to see the looker. The looker is equipped with a score sheet, a pen or pencil, a coin, and in some cases a means of making a mechanical sound, such as a clicker or bleeper.

In a series of trials, in a random sequence, the looker either looks at the back of the subject or looks away and thinks of something else. The random sequence is decided by tossing a coin before each trial: heads means "Look"; tails means "Don't look". The looker indicates to the subject when a trial is beginning by a click or bleep, and the subject then guesses whether he or she is being looked at or not. Alternatively, the signal for the beginning of each trial is given to whole class by the teacher.

The looker records the result on a score sheet with two columns, the first headed "Looking (heads)", the second "Not looking (tails)", entering a tick or a cross in the appropriate column, depending on whether the guess is right or wrong. (For a specimen score sheet, see Sheldrake, 1994). The looker then tells the subject whether the answer was correct or not. This feedback helps to maintain the subject's interest, and may also enable subjects to learn how to respond more accurately.

In one school, in Freiburg, Germany, the experiments were not carried out with looker-subject pairs as described above. Instead, the students were tested as subjects in groups of three, with their teacher as the looker.

Usually subjects indicate their guess within 10 seconds, but if they have not done so already are asked to do so after 20 seconds. The procedure is therefore quite fast, and most pairs can easily complete 10-20 trials within 10 minutes. They can then exchange roles and carry out a new series of trials. The numbers of right and wrong guesses from each series of trials carried out by each looker-subject pair are tabulated in three columns; "Looking", "Not looking" and "Total", enabling the total number of right and wrong guesses in each column to be obtained. For each series of trials by each looker-subject pair, in each column, the data are also scored as follows:

      + if there are more right guesses than wrong guesses,
      - if there are more wrong guesses than right guesses, or
      = if the number of right and wrong guesses is the same.

Statistical analysis was carried out using the chi-squared test to compare the number of + and - scores. The = scores were disregarded. The null hypothesis was that by chance alone the number of + and - scores would be equal.

For the comparison of two sets of scores (for example the scores in the "Looking" and "Not looking" columns) 2 x 2 contingency tables were used (Campbell, 1989). The significance was assessed by the chi-squared test, with the null hypothesis that the proportions of right and wrong guesses in both sets were equal.

The experimenters, schools and procedural details were as follows:

America

The teachers were participants in an interdisciplinary graduate programme of Southern Connecticut State University, called LEARNscience, and were working towards a Master's degree in Science Education. This research was organized by James Trifone, the Academic Coordinator.

  1. MaryEllen McKee, Old Greenwich School, Greenwich, CT. The subjects were Grade 3 students (8-9 years old), and Ms McKee was the looker. The signal for the beginning of each test was given by a click from a mechanical clicker.
  2. Bonnie Maur, Chalk Hill Middle School, Monroe, CT. The looker-subject pairs were grade 6 students (11-12 years old). The looking trials for all pairs took place simultaneously, and Ms Maur signalled the beginning of each test for the whole class by means of a buzzer.
  3. Kathleen Robinson, Stepney Elementary School, Monroe, CT. The subjects were Grade 3 students (8-9 years old). The signal for the beginning of each test for the whole class was given by the ringing of a bell.
  4. Tracy Tishion, Whisconier Middle School, Brookfield, CT. Some of the subjects were Grade 5 students (all 10 years old), tested in the classroom, and some were friends and members of her family. She was the looker. The signal for the beiginning of each test was given by ringing a bell.
  5. Elaine Bamford, Eric G. Norfeldt Elementary School, West Hartford, CT. The looker-subject pairs were Grade 4 students (9-10 years old). The looking trials for all pairs took place simultaneously, and Ms Bamford signalled the beginning of each trial for the whole class by means of a clicker.

Germany

  1. Helmut Lasarcyk, at Stormarnschule, Ahrensburg, Schleswig-Hollstein. One experiment was carried out with Grade 8 students (13-14 years old), and three with Grade 12 students (17-18 years old). Both classes did the experiment on January 19, 1996, and the Grade 12 students were tested again on March 5 and May 5 1996, but with different combinations of students in pairs. The signal for the beginning of each test was given by standard clicking noises from Biros.
  2. Dolfi Wilke, at Geshwister-Scholl-Schule, Konstanz, Baden-Württemberg. The experiment was carried out with 11-16 year old students on June 16 1995. The signal for the beginning of each test was given by a mechanical clicker.
  3. Rolf Robischon, at Johannes-Grundschule, Freiburg im Breisgau, Baden-Württemberg. The experiments were carried out with Grade 3 students (8-9 years old), whom Herr Robischon had been teaching since they were in the Kindergarten. Three students at a time served as subjects, with Herr Robischon as the looker. In each experiment there were 24 trials, and two experiments (with different subjects) were conducted on March 8, March 22, April 26 and May 3, 1995. The 3 students sat with their backs to Herr Robischon and with their eyes closed. In each trial, one of the children was looked at while the others were not. The random sequence in which they were to be looked at was noted in advance, before the subjects had been selected.
    The beginning of each trial was signalled by Herr Robischon saying "Anfang" (start), and the end by his saying "Danke" (thank you). Thus all 3 subjects heard the same words in the same intonation at the same time. The children put their hands up at the end the trial period if they thought they had been looked at, and kept them down if they thought they had not.
    Four children were tested in 4 experiments each; 2 children were tested in 2 experiments each; and 4 children were tested only once each. In the data shown in Table 4, the results for the 2 children tested twice were combined, as were the data for the 4 children tested once.

Results

Basic experiments

The pattern of results was very similar in the United States and Germany (Tables 1 and 2). In both there was an extremely significant excess of positive over negative scores in the looking trials, no significant difference between the positive and negative scores in the not-looking trials, and an overall positive result, which was highly significant statistically.

Table 1
Staring Experiments in Schools in Germany

Above: Numbers of right and wrong guesses (percentage of right guesses shown in parentheses). Below: total number of subjects with more right than wrong guesses (+), more wrong than right guesses (-) or equal numbers of right and wrong guesses (=).

Table 2

Staring Experiments in Schools in Connecticut, USA

Above: Numbers of right and wrong guesses (percentage of right guesses shown in parentheses). Below: total number of subjects with more right than wrong guesses (+), more wrong than right guesses (-) or equal numbers of right and wrong guesses (=).

A comparison of the pattern of results from America and Germany showed that they were not significantly different from each other (Table 3). The large excess of positive over negative scores in the combined results (97+ 42-) was very significant statistically (p = 3x10-6).

Table 3

Comparison of the Results from Germany and the United States

Total number of subjects with more right than wrong guesses (+), more wrong than right guesses (-) or equal numbers of right and wrong guesses (=).

Repeated experiments with selected children

In the experiments carried out by Rolf Robischon, in Freiburg im Breisgau, Germany the subjects were 8-9 year old students who knew him very well: he had been teaching them since they were in the Kindergarten. He tested them three at a time with himself as the looker, in each trial looking at one of the children and not at the other two. In a preliminary experiment (where the tests were done with groups of 5), there were 59.2% correct guesses. Those with the highest scores were selected as subjects for the subsequent experiments.

The results (Table 4) show that some of these selected children were remarkably good at knowing whether or not they were being looked at. The best two, Benjamin and Dirk, were right 93.8% and 87.5% of the time respectively. In the overall results for all the children, 71.2% of the guesses were correct, and the total scores (21+ and 1-) were highly significant statistically (p = 2x10-5).

Table 4

The results of a series of experiments carried out by Rolf Robischon in Freiburg im Breisgau, Germany with 8-9 year-old children selected for their sensitivity to being looked at.

In these experiments Herr Robischon was the looker, and three children at a time were the subjects. The children kept their eyes shut throughout. In each trial, in a random sequence, one of the children was looked at and the other two were not looked at. For each child there were therefore approximately twice as many non-looking as looking trials.

Above: Numbers of right and wrong guesses (percentage of right guesses shown in parentheses). Below: total number of subjects with more right than wrong guesses (+), more wrong than right guesses (-) or equal numbers of right and wrong guesses (=).

These results are not directly comparable with those carried out in other schools because they involved a different procedure, with the children being tested in groups of 3, and looked at by their teacher, whom they knew very well. Moreover the repetition of the tests may have enabled the children to improve by practice. But they indicate that with a combination of sensitive subjects and effective lookers, far higher scores can be achieved than with unselected lookers and subjects.

Discussion

The overall pattern of results

The results of these experiments summarized in Tables 1 and 2 show a remarkably consistent pattern, even though they were conducted in different schools, in different countries, with children of different ages, and with different methods of signalling the beginning of a trial.

First, there were more correct (56.9%) than incorrect guesses (43.1%). And more subjects scored positively than negatively. Overall, 97 subjects made more right than wrong guesses, 42 made more wrong than right guesses and 21 had an equal number of right and wrong guesses (Table 3). For the purpose of statistical analysis, the null hypothesis is that by chance alone the number of subjects with positive scores should equal the number with negative scores. In fact the number with positive scores is very significantly greater than the number with negative scores, with only a 3 in a million probability of being due to chance (Table 3).

Second, there was a striking difference between people's responses in the trials when they were being looked at and in the control trials, when they were not being looked at. When they were being looked at, they were right more often than not. Overall, this effect was highly significant, with odds against chance of 10 billion to one (Table 3). In the control periods their guesses were not significantly different from chance.

I have myself conducted experiments with a total of 242 looker-subject pairs, and the pattern of results was essentially the same as that in these school experiments. In the looking trials there was very significantly more subjects with positive then negative scores, 159 as opposed to 63. In the not-looking trials there was no significant difference. Overall, there were 141 subjects with positive scores and 70 with negative scores, with odds of a million to 1 against this result being due to chance (Sheldrake, in preparation).

Comparison with previous experiments

In the light of these results, it is interesting to look again at the results of the two previous reports of experiments of this type, those by Coover (1913) and Poortman (1959). I have scored their data by the same system I used for the school results (Table 5). It turns out that Coover's results are similar to those described here, both in the overall score (5+ 3- 2=) and in the marked tendency for people to be right when they were being looked at (7+ 2- 2=) whereas they were around chance levels in the not-looking trials (5+ 5- 0=). Unfortunately, Coover, following Tichener's strongly sceptical lead, pronounced his results negative, and probably helped set back research on this subject for decades.

Poortman's findings were positive, but the pattern of results differs from Coover's and those summarized in Tables 1 and 2. The guesses were correct slightly more often when the subject was not being looked at. They were, however, conducted with the same looker-subject pair on a series of occasions, and therefore differ from Coover's and the classroom experiments, in which each looker-subject pair was tested only once. This repeated testing could have enabled the subject to learn the difference between the feeling of being stared at and the feeling of not being stared at. The schoolchildren in Freiburg were also tested repeatedly, and they too performed about as well when they were not being looked at as when they were (Table 4).

Could these findings be artefacts?

What do these results mean? Can people really tell, by some unexplained power, when they are being looked at from behind? Or could the data have arisen as an artefact or as a result of subtle cues?

One possible artefact could arise if the subjects were biassed towards guessing that they were being looked at, whether they were or not. In fact subjects did guess they were being looked at more often than not. From the data in Tables 1 and 2, the overall proportion of "looking" guesses was 55.9%. But if this bias were expressed equally in both looking and not-looking trials, it would be reflected in an excess of correct guesses in looking trials, offset by an excess of incorrect guesses in the control trials. The total scores should show no significant deviation from chance. But in fact the positive scores in the looking trials were not cancelled out by equal and opposite negative scores in the not-looking trials, and the overall results were positive. The positive scores in the looking trials were highly significant statistically, whereas there was no significant difference in the not-looking trials (Tables 1,2 and 3).

Can these results be explained by subtle sensory cues? One possibility is that the some subjects see what the looker is doing by means of peripheral vision. For anyone who has actually been a subject in such experiments (as I have myself on many occasions) this is implausible, because it is not in fact possible to see what the looker is doing, seated directly behind one's back. Nevertheless, I am currently testing this possibility in further experiments in which the subjects are blindfolded.

The possibility that sensory cues are transmitted by sounds or smells from the looker, or by infra-red radiation from the looker's face, are harder to rule out conclusively in tests such as these where the subject and looker are in the same room. To test this possibility, I am carrying out further experiments in which the subjects are viewed through closed windows. And there is already evidence from closed-circuit TV experiments that the effect still persists even when all possible sources of subtle sensory cues have been eliminated. (This research was summarized in the Introduction).

The most plausible way in which lookers could unconsciously have given subtle clues to subjects could have been by means of the clicker with which they signalled the beginning of each trial. Although this mechanical signal could not in itself have given any clues because of its standard sound, clues could have been conveyed by the position in which it was clicked, or by longer or shorter delays before it was clicked. Fortunately this argument can be tested empirically. In three of the American schools the individual lookers did not themselves signal the beginning of each trial. Rather, the teacher signalled to the whole class when each trial was beginning. Thus no subtle cues could be conveyed by lookers to subjects through the signalling process. In spite of the elimination of this possible source of artefact, in these schools (Chalk Hill, Stepney and West Hartford) there was a high proportion of positive scores: taken together, there were 29 subjects who were more often right than wrong, and 7 more often wrong than right, in other words over 4 times more positive than negative scores. By contrast, in the other American schools, there were only 1.4 times more positive than negative scores and if the German schools (Table 1) are included as well, this figure is 2.3. Thus the evidence is strongly against this artefact hypothesis.

Finally, there is the possibility that some of the children were cheating. For example, the subjects might have peeped the see if the looker was looking at them or not, or the looker could have whispered or given some other signal. But if cheating was going on, the scores should have been boosted in both looking and not-looking trials. The same should be true if subtle cues were involved. Yet there were highly significant positive scores only in the looking trials, while in the not-looking trials the scores were no better than chance (Tables 1, 2 and 3). Cheating or subtle cues should not boost scores only in looking trials. A cheat would also know when not-looking trials were going on. And any cues that indicated when the looker was looking would - by their absence - indicate when he or she was not looking. Thus cheating and sensory cues are not plausible explanations for the pattern of response actually observed with these subjects.

Whatever the explanation of these results, they show a strikingly repeatable effect. That such simple experiments give consistent results is encouraging. Through further research it should be possible to discover either how this pattern of results arises as an artefact, or to establish that the sense of being stared at is a real effect that may have as yet no explanation in terms of established science.

Why is the effect so small?

If some people really can tell when they are being looked at from behind, why was the positive effect in most of these experiments so small, with only 56.9% correct guesses overall? There are several possibilities:

  1. As in all other human abilities, people probably differ in their effectiveness as lookers and in their sensitivity as subjects. In most of the experiments conducted here, no attempt was made to select effective lookers or sensitive subjects. These results therefore represent an average over a wide range of abilities. By selecting effective lookers and sensitive subjects, a much larger effect could probably. There is already evidence that this is the case from the pioneering experiments of Rolf Robischon (Table 4).
  2. The feeling of being looked at occurs in real-life conditions spontaneously, rather than when a person is consciously trying to detect it. Under the artificial conditions of experiments, the conscious mind may inhibit or interfere with a sensitivity that is normally unconscious.
  3. If there is indeed a sense of being stared at, looking at someone may give a detectable stimulus. But when a person is not being looked at, there is no such stimulus. Thus there is an asymmetry between the looking and not-looking trials. In the not-looking trials, subjects are being asked to detect an absence of the feeling of being looked at, an artificial situation with no parallel in real-life experience. And indeed, under these conditions, naive subjects' guesses were no better than chance. However, with practice, subjects may learn to detect the difference between presence and absence of the gaze, and hence experienced subjects could become as successful under non-looking as under looking conditions. And this was the pattern of response shown by experienced subjects (Tables 4 and 5).

Table 5

Staring Experiments by Coover (1913) and Poortman (1959)

Numbers of right and wrong guesses (above) and numbers of experiments (below) with more right than wrong guesses (+), more wrong than right guesses (-) or equal numbers of right and wrong guesses (=). (The percentage of right guesses is shown in parentheses).

Implications

If further studies confirm the reality of the ability to detect an unseen gaze in a way that cannot be explained in terms of artefacts, cheating or sensory information, the implications for our understanding of the nature of the human mind will be very far-reaching (Abraham, McKenna and Sheldrake, 1992; Sheldrake, 1994). Any hypothesis capable of explaining this effect would need to postulate an influence of the mind of the looker that acted at a distance on the person being stared at.

This effect also raises the question of its evolutionary origins. Is it confined to human beings, or can animals also tell when they are being looked at by people or by other animals? So far, there seem to have been no experimental investigations of animals' abilities in this respect. But clearly this ability could be of evolutionary advantage, for example if it enabled animals to detect the gaze of an unseen predator. The feeling of being looked at could be deeply rooted in our biological heritage.

The potential for further experiments in schools

The results described in this paper show that a very simple experiment can give remarkably consistent results in schools, even in primary schools. The experiment concerns a phenomenon with which most children are familiar and in which many are interested. It also demonstrates how scientific methods can be applied to the the investigation of an unexplained effect. In further research it should be possible to eliminate any possible sensory cues by staring at the backs of blindfolded subjects through closed windows.

Acknowledgements

I am grateful to the students who took part in these experiments and to the teachers who carried them out, sent me their results and kindly agreed to my publishing their findings. I thank Professors Nicholas Humphrey and Michael Morgan for helpful discussions and advice on statistical analysis. This work was supported by the Institute of Noetic Sciences, Sausalito, CA and the Lifebridge Foundation, New York.

References

Abraham, R., McKenna, T. and Sheldrake, R. (1992) Trialogues at the Edge of the West. Santa Fe: Bear and Co.

Braud, W, Shafer, D. and Andrews, S. (1990) Electrodermal correlates of remote attention: Autonomic reactions to an unseen gaze. Proceedings of Presented Papers, Parapsychology Association 33rd Annual Convention, Chevy Chase, MD, pp. 14-28.

Braud, W, Shafer, D. and Andrews (1993a) Reactions to an unseen gaze (remote attention): A review, with new data on autonomic staring detection. Journal of Parapsychology 57, 373-90.

Braud, W, Shafer, D. and Andrews (1993b) Further studies of autonomic detection of remote staring: replications, new control procedures, and personality correlates. Journal of Parapsychology 57, 391-409.

Campbell, R.C. (1989) Statistics for Biologists. Cambridge University Press, Cambridge.

Coover, J.E. (1913) The feeling of being stared at. American Journal of Psychology 24, 570-5.

Cottrell, J.E., Winer, G.A. and Smith, M.C. (1996) Beliefs of children and adults about feeling stares of unseen others. Developmental Psychology 32, 50-61.

Howat, S.J., Delanoy, D.L. and Moris, R. (1994) Remote staring detection and personality correlates. Journal of Scientific Exploration 8 , 582.

James, J. and Yau, E. (1996) An experiment on the sense of being stared at. Report submitted to the 1996 Science Fair, Durham, Ontario.

Mastrandrea, M. (1991) The feeling of being stared at. Project Report, Nueva Middle School, Hillsborough, CA.

Poortman, J.J. (1959) The feeling of being stared at. Journal of the Society for Psychical Research 40, 4-12.

Schlitz, M and LaBerge, S. (1994) Autonomic detection of remote observation: Two conceptual replications. Proceedings of Presented Papers, Parapsychology Association 37th Annual Convention, Amsterdam, pp. 352-60.

Sheldrake, R. (1994) Seven Experiments that Could Change the World. London: Fourth Estate, Chapter 4.

Titchener, E.B. (1898) The feeling of being stared at. Science New Series 8, 895-7.

Williams, L. (1983) Minimal cue perception of the regard of others: The feeling of being stared at. Paper presented at the 10th Annual Conference of the Southeastern Regional Parapsychological Association, Carrolltown, GA, Feb 11-12.

Wiseman, R. and Schlitz, M (1997) Experimenter effects and the remote detection of staring. Journal of Parapsychology 61, 197-208.

Wiseman, R. and Smith, M.D. (1994) A further look at the detection of unseen gaze. Proceedings of Presented Papers, Parapsychology Association 37th Annual Convention, Amsterdam, pp. 465-78.

Wiseman, R., Smith M.D., Freedman, D., Wasserman, T and Hurst, C. (1995) Examining the remote staring effect: Two further experiments. Proceedings of Presented Papers, Parapsychology Association 38th Annual Convention, pp. 480-490.