The ability to detect unseen staring: A literature review and empirical tests.
There is evidence to suggest that individuals not only believe in their ability to detect an unseen gaze, but may genuinely be able to do so. The present study reviewed past research and sought to determine whether such a phenomenon was empirically demonstrable. In Expt 1, 12 participants responded to 12 sequences (with feedback in the last nine) of 20 trials each, with staring or non-staring episodes based on Sheldrake's random number sequences. No effects were obtained when no feedback was given. With feedback, more accurate than mean chance expectation (MCE) results were obtained on staring, but no difference on non-staring trials. However other `normal' explanations of ESP phenomena discuss the possibility of matching in bias between experimental sequences and participants' representations of randomness. Tests of the sequences found more alternations than expected, a feature typical of subjective randomness, but the increase in accuracy found on staring trials only was nor consistent with this explanation . It was concluded that the improvement in accuracy with feedback is likely to be due to implicit learning, given the structure in non-random sequences. This hypothesis was supported in Expt 2 where 12 participants responded to 12 `genuinely random' sequences, and no differences in accuracy from MCE were obtained.
The fact that some people believe that they can tell when they are being stared at by another person, not directly in their field of vision, was first reported many years ago (Titchener, 1898), and this has been supported in subsequent research with varying incidence rates (e.g. 86% and 68%, Coover, 1913; 74%, Williams, 1983; 92%, Braud, Shafer, & Andrews, 1993a). Titchener (1898) argued that 'no scientifically-minded psychologist believes in telepathy' (p. 897), and he provided a 'normal' explanation of the phenomenon. Based on the assumption that the eye is attracted to movement in a field of vision, he suggested that the starer's gaze is simply drawn to the movement of the staree's head turning in his direction. Mill's method of difference (Mill, 1843) suggests that the cause for an event is likely to be found in conditions that alter as the event alters rather than in conditions which remain the same. Titchener's hypothesis is that the source of causality lies in the staree, not the starer, and he refers to attribution of causality to the starer as a misinterpretation of fact.
Research into the phenomenon, although sparse, has a long history. Titchener (1898) carried Out laboratory trials with participants who said that they knew when they were being stared at, and by others who said that they could cause people to turn around by staring at them. As he expected, no effect was obtained. Coover (1913) conducted several weekly individual sessions lasting 3 or 4 hours with 10 participants sitting with eyes closed and shaded with one hand. They responded to 100 staring or non-staring trials lasting for 15-20 seconds, with each trial beginning with a single and ending with a double pencil tap. Randomness of trial type was achieved by removing a shaken die from a box. Coover reported 50.2% correct responses, which he suggested did not differ significantly from chance levels. However a reanalysis of Coover's data was conducted by the present authors. Accuracy in non-staring trials did not differ significantly from chance levels, but, using a single mean t test, a significant effect (p [less than] .05) was found for staring trials. Another study by Poortman (1959), using a female acquaintance and a total of 89 trials, reported detection rates of 61.7% for staring and 57.1 % for non-staring, regarded by him as non-significant. However these data were reanalysed by Braud et al. (1993 a) and a significant detection effect was obtained.
Positive results have been reported in several recent studies undertaken under more rigorous experimental conditions. Peterson (1978) used a one-way mirror system, with nine participant pairs (who swapped roles) and 36 experimental sessions of 6 minutes each, containing three 30-second staring trials. The task of the staree was to indicate, by pressing a button, whenever he 'sensed being stared at'. Peterson's analysis produced a significant staring detection effect, but no data is given on differences in accuracy between staring and non-staring episodes.
Williams (1983) increased separation by putting starer and staree in rooms 60 feet apart, linked only by a closed-circuit video arrangement, so that the starer could see the staree on a monitor during staring trials. Twenty-eight starees experienced 52 staring and 52 non-staring random trials lasting for 12 seconds with tones signalling the beginning of each trial. A significant accuracy effect was obtained, although it is not possible to ascertain from the data if the result held for staring and non-staring trials separately. The experiment included both sheep (those who believe in ESP) and goats (those who do not), and the significant effect was found for sheep only.
Braud et al. (1993 a) pointed out that, despite support for the ability to detect staring, the effect sizes (ranging from .004 to .42) obtained were not very impressive. This, they argue, is because these laboratory studies 'were designed to encourage deliberate conscious guessing in order to identify staring periods' (p. 376). Such activity is seen as interfering with attention to internal signals, which would be present in staring detection in real life. Thus a radical change in methodology was employed where spontaneous phasic skin resistance response (SSR) was used to measure sympathetic autonomic nervous system arousal. Some starees had undergone special training, to 'help them to increase their sensitivity to internal physiological reactions, to increase their understanding of what it might feel like to be interconnected with other persons, and to help them deal with their possible psychological resistance to such interconnectedness' (p. 377). Training consisted of activities such as experiential exercises (staring into another's eyes in order to increase comfort and closeness with others) and group discussions (e.g. exploring feelings associated with 'merging' with another). Starers and starees were linked via a video set-up similar to that used by Williams (1983). Ten staring and 10 non-staring trials of 30 seconds duration were presented randomly during a 20-minute period individually to 32 participants. Untrained participants displayed significantly greater SSR activity during staring compared to non-staring control periods (59.4% vs. 50% chance level), whereas trained participants displayed significantly lower levels (45.5% vs. 50% chance level) during experimental periods. In addition the effect size was .59 for untrained participants and -- .50 for trained participants. It is argued that these large effect sizes support the more powerful detection ability of unconscious measures of staring.
Braud et al. (1993a) suggest that untrained participants will react in a typical manner by feeling threatened when being stared at (Argyle & Dean, 1965) resulting in sympathetic system arousal. This account may be feasible given the possible role of internal signals. For example, 78% of participants in a study by Thalbourne and Evans (1992) said that they often had a strong emotional reaction (e.g. anxiety, sexual arousal, fear) to being stared at, and 56% said that they had a strong physical reaction (e.g. heart pounding, face flushing). However, trained participants will have greater 'connectedness' with others, and they will have learnt to become more comfortable when being stared at, thus producing lower arousal. This explanation may be less convincing. There is likely to be a 'focus of attention' triggered by the realization that someone is staring, and this suggests an increase in some kind of activity, if internal signals are involved, rather than a decrease. These results were replicated in later work, which also included sham controls in which no staring trials took place (Braud, Shafer, & Andrews, 1993b). On the basis of their meta-analysis the authors argue that these data on staring 'are not merely laboratory curiosities, but may be reliable and robust enough to have important influence in everyday life' (Braud et al., 1993a, p. 388).
Sheldrake (1994) discusses the detection of staring and he gives details of his own research with starees sitting with their backs to starers. He reports accuracy rates no better than mean chance expectation (MCE) with non-feedback, but significantly better than MCE (at 53.1%) when immediate feedback is provided. These data encouraged him to initiate a large-scale study, results of which have been widely reported in the media (e.g. Matthews, The Sunday Telegraph, 1997; Mullins, New Scientist, 1997). He also suggests that people conduct their own experiments, and his 'experimental kit' can be downloaded from the New Scientist web site (http://www.nexvscientist.com/nsplus/rnr/look.html). A list of 24 sequences each consisting of 24 trials is provided, but sequences 13-24 are reproduced from sequences 1-12 with looking and non-looking trials switched. It is suggested that each child in a group is tested on a different sequence, and readers are invited to download the sequences. Another suggestion is to base trials on the toss of a coin. Sheldrake asks for results of such experiments to be sent to him for inclusion in his pooled data set.
Sheldrake suggests two basic methods, both involving no feedback. In the first, pairs of schoolchildren divide into starer and staree, after which the starer sits at least 1 m behind the staree. Using random number sequences to guide staring or nonstaring trials, and a mechanical signalling device such as a clicker, the starer indicates the beginning of each of 20 trials. During the sessions starees wear airline blindfolds, and responses are recorded on data sheets by the starer. The children then exchange roles. In the second method, an attempt is made to isolate participants in that starers sit in rows inside a building looking through a window, and starees sit in rows outside. It is reported (Matthews, 1997) that Sheldrake finds no difference in chance performance for non-staring, but accuracy for staring is said to be statistically significant at around 60%. These data are based on 18000 trials on schoolchildren from a number of countries.
This review of past research reveals mixed results, but there is some support for the commonly held view that there is an ability to detect staring beyond chance levels, and that detection can take place without sensory cues present in normal forms of communication. The present study was undertaken in order to provide an empirical test of the ability to detect unseen staring, and specifically of Sheldrake's observations in controlled laboratory conditions. On the basis of his data Sheldrake expects detection of staring, but not non-staring, without feedback, to be above MCE a pattern obtained by other researchers (Braud et al., 1993a, 1993b; Coover, 1913).
Sheldrake's (1994) finding of increased accuracy with immediate feedback was also tested. Previous research has reported feelings associated with being stared at, for example, 'a state of unpleasant tension or stiffness at the nape of the neck, sometimes accompanied by tingling, which gathers in volume and intensity until a movement which shall relieve it becomes inevitable' (Titchener, 1898, p. 895), or 'a feeling of restlessness' (Coover, 1913, p. 574). Feedback may help in the identification of internal signals associated with being stared at, and so lead to improved performance.
Twelve volunteers, seven men and five women, aged 19--49 years (M = 24.5 years), participated as starees in the study. They were recruited on the basis of their belief in ESP, and their interest and belief in the remote staring detection effect. One of the authors (SS) acted as the starer throughout the study.
Starer and staree were located in adjoining closed rooms linked by a one-way mirror. The intention was to try to mimic the real world, but at the same time preclude the possibility of auditory and visual signals reaching the staree from the starer. In line with this 'real world' approach no attempt was made to isolate the staree in a soundproof environment. Occasional everyday random noise from a nearby corridor could be heard in the staree's room. The observer took great care not to provide auditory cues, but slight noises could not be heard in the experimental room, and it was not possible to see the observer through the one way mirror.
Apparatus and materials
The experimental room (4 X 4 m in size) contained an IBM-compatible PC with a monitor and a response box with two keys (staring and non-staring). The observation room (4 X 3 m in size) also contained an IBM-compatible PC and monitor. The PCs in both rooms were linked so that the display on the monitor in the experimental room could he seen by the starer on the PC in his room. This allowed him to know when each trial began and finished. Sheldrake's sequences were downloaded from the New Scientist web site and they were presented using a Microsoft Quickbasic program.  The participant sat in the middle of the experimental room facing the monitor with his back to the one-way mirror.
Participants were met on each occasion by the experimenter. The purpose of the study, to test for the ability to detect unseen staring, was carefully explained, and participants were asked to try to `listen' to internal signals, and to refrain from conscious `statistical' guessing.
Each trial began when the word 'TRIAL' appeared on the monitor, and lasted for 20 seconds, after which the words 'YES OR NO' appeared, and the participant had to respond by pressing the appropriate response key. There was no time limit for the response, but it was a trigger for a 10-second 'rest period', during which the monitor screen remained blank, before the next trial began. Each participant completed the same 12 sessions (Sheldrake's sequences 1-12) in the same order. There were 20 trials in each session (which lasted for around 15 minutes), and no participant received the same sequence of trials more than once. The master list of trial sequences was kept in the observational room so that the experimenter would know when to stare, during which time he would stare intently at the participant and concentrate on making him aware of his staring. During the non-staring trials he would look directly away from the one-way mirror and think of other things.
During the first three sessions no feedback was given on the accuracy of each trial. Sheldrake's advice is to test each participant on one sequence of 20 trials, and perhaps one more if time permits, with the effect being sufficiently robust to he detected on this small number of trials. This suggests a large effect, and a power analysis in relation to chi-square, with a large effect size (w = .5), at .05 significance level, and N = 60, results in power of .97 (Cohen, 1977). Therefore a large effect should be evident on 60 trials. On the remaining nine sequences (180 trials) the word 'correct' or 'false' appeared on the monitor immediately after each response. However, no information was given on accuracy for each session, and the results, which were retained by the computer program, were not looked at, or analysed, until the study had been completed.
Individual accuracy results for non-staring and staring trials on both non-feedback and feedback conditions are shown in Table 1, along with the results of chi-square statistics on this data (2 x 2 contingency tables with non-staring/staring vs. no/yes).  With non-feedback no significant results were obtained. With feedback, there was one significant result (Case 9), scoring above MCE on both non-staring and staring trials. A similar performance was produced by Case 11 but significance was not attained (p [less than] .055). The same chi-square contingency table cannot be used to examine all the individual data at once, and so total accuracy scores were subjected to single mean t tests. There was no effect on the non-feedback trials, but a significant difference (from MCE), t(11) = 6.19, p [less than] .001, was obtained in the feedback trials. Participants performed at higher than chance level when feedback was given.
An examination of the means in Table 1 shows that accuracy was higher in staring than non-staring trials, both in the non-feedback condition, t(11) = -2.48, p [less than] .031, and in the feedback condition, t(11) = -3.30, p [less than] .007. Such a difference can be produced by a strategy of responding with 'yes' more often than 'no', although overall accuracy will not be altered. Response bias must be taken into account, and therefore single mean t tests were computed on `yes' responses. During the non-feedback trials the response `yes' was given significantly more often compared to MCE, t(11) = 2.48, p [less than] .03. Therefore there is a response bias. That bias is apparent in both staring and non-staring trials, though the mean number of `yes' responses for each participant is above, but not significantly above, MCE (15) for both. During the feedback trials `yes' was also said more often than MCE, t(11) = 3.27, p [less than] .007. However, the departure from MCE is evident on staring trials only, t(11) = 7.10, p [less than] .0001, and the mean number of 'yes' responses on the non-staring trials (45.25) is close to MCE (45).
The first three sessions (60 trials) contained no feedback, but feedback was given on the following nine sessions (180 trials). In order to test for improvement four 'blocks' were created, each consisting of three sessions (60 trials), with Block I (the three non-feedback sessions) used as a baseline measure. The block means for non-staring and staring trials are shown in Table 2 along with the results of single mean t tests. A 2 (non-staring vs. staring) x 4 (blocks 1, 2, 3 and 4) repeated measures ANOVA resulted in a significant main effect for blocks, F(3,33) 5.43, p [less than] .004, and a significant linear trend, F(1,11) = 13.92, p [less than] .003. An examination of the results of single mean t tests confirms the trend for increased accuracy over the blocks.
There is little support for the staring detection effect on the non-feedback trials, in that performance did not differ from MCE. Even assuming a medium effect size (w .3), where 60 trials produces lower power of .64 (Cohen, 1977) at the .05 significance level, the likelihood of non-significant results being obtained for all participants is very low (p [less than] .0001). The occurrence of this highly unlikely event underlines the lack of support for the presence of an effect. Significantly greater accuracy was obtained in staring compared to non-staring trials, but this is due to a bias to respond with 'yes' more often than 'no'. The effect of this would be to improve scores on staring trials at the cost of poorer performance on non-staring trials.
With feedback, the response bias is present on staring, but not on non-staring, trials, and performance on staring trials is, with the exception of one case, always above MCE. In contrast, performance on the non-staring trials is at MCE. Responding with 'yes' more often than 'no' during a staring trial leads of course to higher accuracy. However, there is no corresponding decrease in accuracy during non-staring trials, and therefore it might be concluded that the increased accuracy in staring trials is not due to response bias, but to an ability to detect when a staring trial is taking place. This evidence supports Sheldrake's (1994) findings (an effect with feedback but none without), but not the findings of his later large-scale study, which is reported to have obtained an effect with no feedback.
However, there is another possible 'normal' explanation of these data. Typically in staring research the respondent is asked to choose between two alternatives, and accuracy is derived on the basis of these choices. The underlying premise is that the possible events will be distributed randomly, and so the participant is essentially being asked to produce a sequence similar to the experimental sequence (Brugger, Landis, & Regard, 1990). Two problems have been raised in this context. First, sequences generated by participants who are asked to produce random number sequences are not truly random. There is a tendency to avoid repetitions of the same element (Falk & Konold, 1997; Reichenbach, 1949; Wagenaar, 1972), called the alternation bias (Budescu, 1987; Rapoport & Budescu, 1997). Indeed, Blackmore (Blackmore, 1997; Blackmore & Troscianko, 1985) points to individual differences in this tendency as the basis for an explanation of belief in the paranormal.
Secondly, as Knuth (1997) points out, random number sequences, however generated, are never truly random, because they usually contain bias. There may be (in ESP research) a match between experimental sequence bias and participant subjective bias which produces an erroneously high accuracy rate (Gadin, 1977). Some ESP research has used experimental 'random' sequences, which contain nonrandom features similar to those produced by participants (Brugger et al., 1990).
The possibility of matching between experimental sequences and subjective representations of randomness must be taken into account. It has been suggested (Brugger et al., 1990; Knuth, 1997) that rather than prove the randomicity of sequences, it is essential to disprove non-randomness by testing for bias. Global bias can be measured by matching responses to experimental sequences to unused sequences drawn from the same source (Brugger et al., 1990). Global bias is present if the experimental effect is obtained.
In addition Brugger et al. (1990) argue that it is necessary to check for 'local' bias, or short 'local' patterns, and this entails checking that a sequence is actually random. Knuth's (1997) advice, in a comprehensive discussion of random numbers, is to accept a sequence as random if it passes at least 6 of 11 possible tests. However, these tests involve partitioning sequences into non-overlapping runs. In the feedback sequences, where there may be pattern learning by a participant, it is possible for a pattern to begin at any point, and so overlapping sequences or serial dependency must be considered.
Wagenaar (1970) discusses the probability, in a sequence of black and white dots, of white following white or black following black. The probability of a repetition is 0.5, and in a random sequence of 20 binary digits there should be 9.5 such repetitions. A similar test of randomness of overlapping runs is to look for higher order dependencies, or 'g-tuples' (Rapoport & Budescu, 1992). Rapoport and Budescu (1997) partition binary sequences into, for example, three-tuples. The first three-tuple comprises the first three numbers in a sequence, the second, numbers 2, 3 and 4, and so on.
Global bias was measured by matching the experimental responses to test sequences (Sheldrake's sequences 13-24) which had not been used in the study. Given the relationship mentioned earlier between sequences 1-12 and sequences 13-24, the following method was used. Participant 1: responses to sequence 1 matched to sequence 13, sequence 2 to sequence 14, and so on. Participant 2: responses to sequence 1 matched to sequence 14, sequence 2 to sequence 15, and so on. The experimental responses of the other participants were matched according to this pattern. In addition analyses of repetitions and three-tuples in each experimental sequence (sequence 1-12) were conducted to test for local bias.
Chi-square analyses produced no effects in non-feedback or feedback trials. Single mean t tests on test accuracy scores within feedback and non-feedback conditions also produced no significant results. The means and single mean t tests for each test block are shown in Table 3. Six of the single mean t tests are significant, but tests of trend proved insignificant.
The number of repetitions in the 12 sequences in order are: 6, 6, 6, 8, 8, 8, 8, 9, 9, 7, 7, 7 (M = 7.42). The average probability of a repetition (p = .39) is below what would be expected in a random sequence (p = .5). Also the obtained mean is significantly different from the theoretical mean, t(11) = --6.66,p [less than] .001. In relation to the stare (S)/no-stare(N) dichotomy there are eight possible three-tuples, each of which is equally likely (p = .125), and so they should appear equally often. Each sequence was analysed separately for three-tuples, as this is how they would normally be presented in experimental trials. The three-tuple combinations are shown in Table 4, with the totals from each sequence added together. There is a marked deviation from the expected pattern ([X.sup.2] 26.96, d.f. = 7, p [less than] .O01) , with increased alternation. The most frequent (SNS and NSN) are those which have most alternations, and the least frequent those which have least alternations (SSS, NNN).
Also there are only four same-element four-tuples and no same-element g-tuples longer than four.
In the test for global bias chi-square results were not significant when no feedback was given, and this is similar to the results found in Expt 1. There were also no significant chi-square results in the feedback condition, and a non-significant single mean t test, and both of these results differ from those found in the feedback condition in Expt 1. Also, unlike the results for Expt 1, there was no evidence of a trend for increased accuracy in staring trials over time. The significant results found on single mean t tests within each block are an effect of response bias, with increased accuracy on staring trials balanced by a decreased accuracy in non-staring trials. Therefore global bias cannot account for the improvement obtained on staring trials in the experimental sequences. However, the analysis of the structure of the sequences reveals that there are many more alternations that would be expected in acceptably random sequences. There are fewer repetitions, fewer same element three-tuples, and in the 12 sequences there are only four same-element four-tuples. Increased accuracy due to a matching of subjective and experimental bias seems plausible, but the absence of an effect in the non-feedback condition does not support that explanation.
Another possibility is for the detection of staring trials to have been caused by a kind of bias in structure of the sequences, which favoured both alternating frequently and saying 'yes' more often than 'no'. The trend for increased accuracy across the blocks may have been due to a gradual build-up of response bias. However, an examination of the incidence of three-tuples (Table 4) reveals a high degree of symmetry (e.g. NNN = 12, and SSS = 10), and therefore no bias in the pattern of staring and non-staring trials. Also a repeated measures ANOVA on the number of 'yes' responses in each block proved insignificant, with no evidence of a linear trend for an increase in response bias.
The most likely explanation is that the effect found in the feedback condition was due to pattern learning. For example, given the bias in the sequences, the probability of an alternation after 'two of a kind' is very high. The possibility of participants becoming sensitive to such structure in the sequences is supported by work in the area of implicit learning, said by Reber (1989) to be the inadvertent learning of complex rules. One paradigm, sequence learning tasks, is often used in such research, and learning criteria consists of prediction of the next event, or choice reaction time, where a target stimulus will appear in a complex matrix. Learning is said to be implicit because it takes place unintentionally or incidentally. Some (e.g. Reber, 1989) argue that there is no conscious awareness of learning having taken place, but others disagree (e.g. Dulany, Carlson, & Dewey, 1984). Cleeremans (1993) suggests that 'claims that acquisition is entirely implicit must be taken with caution' (p. 110), and he proposes an information processing mechanism, the simple recurrent network (SRN), described as a connectionist architecture, to account for the processing of sequences. In short, knowledge about sequential structure is acquired in the form of internal representations, which may be 'abstract' or 'exemplar based'. These provide guidance on the prediction of future events, or even micro-rules which govern the probability of local events, sometimes referred to as 'chunking' (e.g. Servan Schreiber & Anderson, 1990). Cleeremans (1993) provides evidence to support the notion that participants become increasingly sensitive to sequential structure, and that this sensitivity extends back as far as three elements in a sequence.
Sequence learning has been investigated using different paradigms, with the most common being reaction time to presentation of a number of possible stimuli (e.g. Lewicki, Czyzewska, & Hoffman, 1987). Typically reaction times decrease over time when there is structure in the sequences. The task in another paradigm is to predict the next event in the sequence, with number of correct predictions as the criterion (e.g. Kushner, Cleeremans, & Reber, 1991), and again an improvement is observed with structured sequences. The experimental situation in the present research would seem to bear more resemblance to the latter paradigm, because participants were not encouraged to respond as quickly as possible. Indeed it could be argued that such an Instruction would lead to interference with attempts to pick up the signals which operate in the detection of unseen staring. Cleeremans (1993) discusses factors which encourage explicit (a conscious search for rules) as opposed to implicit strategies, or a combination of both. These are not relevant to the present research when no feedback is given, since participants have no evidence on which to base any learning. Feedback provides that opportunity, though the experimental situation does not fit easily into the conceptual structure provided by Cleeremans. However he argues that immediate external feedback in a simple situation would promote explicit strategies. In spite of this, post experimental interviews suggested that participants thought that they could detect staring, and there was no hint of conscious awareness of rules governing the sequences.
There is still the question of why there has been improvement in staring and not on the non-staring trials, given that pattern learning should account for improvement in both. A possible explanation, in a study of the ability to detect staring, is that the participants have focused more on the detection of staring than non-staring episodes. Research on implicit learning lends support to the view that the improvement in detection could be due to the learning of patterns in sequences containing structure. In order to test this hypothesis a second experiment was carried out in which participants had to respond with feedback to sequences containing no structure.
Twelve volunteers, two men and 10 women, aged 20-40 years (M 30.8 years), participated as starees, and, as in Expt 1, they were recruited on the basis of their interest and belief in the staring detection effect. One of the authors (DS) acted as the starer throughout the study. The experimental setting and apparatus were as described for experiment one.
Tests of randomness were carried out on 10 new sequences of 20 binary digits derived from a table of random numbers, with the following results.
The number of repetitions in the 10 sequences in order are: 9, 9, 10, 13, 7, 11, 7, 10, 9, 9 (M = 9.4), with the average probability of a repetition (p = .47) close to the theoretical level (p .5). A single mean t test between the obtained mean and the theoretical mean of 9.5 was insignificant.
The total three-tuple combinations are shown in Table 5. The obtained frequencies did not differ significantly from expected frequencies ([X.sup.2] = 4.62, d.f. = 7). The results of these tests confirm that the sequences can be accepted as random.
The procedure was similar to that in Expt 1 except that feedback was given in all 10 sequences. The first sequence was regarded as a practice session, and the remaining nine sequences provided the experimental data.
Individual accuracy scores are shown in Table 6 along with the results of chi-square and single mean t test statistics. No significant chi-square results were obtained. Total accuracy scores did not differ from MCE, but significantly higher accuracy was found on staring trials, t(11) 3.80, p [less than] .003, and significantly lower accuracy on non-staring trials, t(11) = -3.08, p [less than] .01. This result suggests a response bias to say 'yes' more often than MCE, and this was confirmed by the results of a single mean test, t(11) 3.70, p [less than] .004. Also that bias was present in both staring, t(11) = 4.78, p [less than] .001, and non-staring trials, 1(11) = 2.25, p [less than] .046 (single mean t tests).
In order to test for improvement the nine sessions were partitioned into three blocks each comprising three sequences (60 trials), and the block means for staring and non-staring trials are shown in Table 7, along with results of single mean t tests. The increased accuracy on staring trials, and decreased accuracy on non-staring trials, simply reflects response bias. A 2 (non-staring vs. staring) x 3 (blocks 1,2,3) repeated measures ANOVA produced no effect for blocks, and no evidence of a linear trend for improvement.
The results of Expt 2 support the hypothesis that the improvement in accuracy during staring episodes found in Expt 1 is due to the detection and response to structure present in non-random sequences, rather than to an increased ability to detect staring. These data add to the literature on implicit learning, with the introduction of an interesting experimental situation, where learning appears to be implicit, despite circumstances favouring explicit strategies.
The non-feedback trials in Expt 1 provided the best test of the staring detection effect, because pattern recognition should not be possible. No effect was obtained, and so there is no support for the staring detection effect. However, neither is there any support for the theory of bias matching. There is a bias in Sheldrake's sequences, containing features typically found in participants' subjective bias. A participant who alternates may be expected to score above MCE because of the alternation bias in the sequences, although this explanation would not predict differences in performance between staring and non-staring episodes. The effect found in Expt 1 was for staring trials only.
The present results are in line with Sheldrake's (1994) data, where accuracy was no different from MCE when no feedback was given, but significantly higher than MCE with immediate feedback. Sheldrake's positive result may have been due to implicit learning, if non-random sequences were used, but this mechanism would not account for the effect obtained in his later large-scale study when no feedback was provided. Neither could that effect be due to response bias since the statistic used, the t test, looks at the difference between overall correct and incorrect responses for each individual. One possibility is that there may have been threats to the internal validity of his research. For example, cues may inadvertently be given by the child who fills the role of the starer, and who will also have had sight of at least one list before he becomes the staree.
However Sheldrake (1994) does suggest that the effect is difficult to obtain in artificial conditions, presumably as found in the present research. This creates the problem of providing sufficient controls necessary for an empirical test whilst at the same time keeping the situation 'natural', which may render the effect empirically untestable. He also makes the point that participants may need time to become accustomed to an experimental situation, or that they may need time to practise, though that did not appear to be necessary in his or other research which has produced positive results (e.g. Peterson, 1978). Sheldrake (1994) also claims that some people are better than others, and it is possible that all the participants in the present research were low in staring detection ability. In addition the effect, if it exists, may be very small, and a large number of trials and subjects may need to be used in future research for detection to be a realistic possibility.
Bias effects would not appear to be able to account for the data obtained by Braud et al. (1993a, b), whose research provides the clearest support for the staring detection effect. However in real life, the claim is one of an ability to detect and to become aware of an unseen gaze, and the best evidence would need to take this form. There is no evidence in the present research of a general ability to detect unseen staring, and it must be concluded that the positive results obtained in Expt 1 can be explained by the detection and learning of structure present in Sheldrake's non-random sequences.
The authors would like to thank Vicki Bruce, Chris French, Chris McManus, David Marks and one anonymous reviewer for their constructive comments on earlier drafts of this paper.
(*.) Requests for reprints should be addressed to Dr John Colwell, Psychology Academic Group, School of Social Science, Middlesex University, Queensway, Enfield EN3 4SF, UK (e-mail: firstname.lastname@example.org).
(1.) We are grateful to Guthrie Walker for writing the program and for his help in running the studies.
(2.) The chi-square statistic was used to analyse individual data--it controls for response bias though some observers may question the assumption of independence of responses. It provides an indication of individual performance, but she main interest is in overall scores.
(3.) Chi-square has been used, though the assumptions of independence are not fully met. However, chi-square appears to be the only viable option. Accordingly, a more conservative significance level was set. We are grateful to David Budescu for his advice in this matter.
Argyle, M., & Dean, J. (1965). Eye-contact, distance, and affiliation. Sociometry, 28, 289-304.
Blackmore, S. (1997). Probability misjudgment and belief in the paranormal: A newspaper survey. British Journal of Psychology, 88, 683-689.
Blackmore, S., & Troscianko, T. (1985). Belief in the paranormal: Probability judgments, illusory control, and the 'chance baseline shift'. British Journal of Psychology, 76, 459-468.
Braud, W., Shafer, D., & Andrews, S. (1993a). Reactions to an unseen gaze (remote attention): A review, with new data on autonomic staring detection. Journal of Parapsychology, 57, 372-390.
Braud, W., Shafer, D., & Andrews, S. (1993b). Further studies of autonomic detection of remote staring: Replication, new control procedures, and personality correlates. Journal of Parapsychology, 57, 391-409.
Brugger, P., Landis, T., & Regard, M. (1990). A 'sheep-goat effect' in repetition and avoidance: Extrasensory perception as an effect of subjective probability? British Journal of Psychology, 81, 455-468.
Budescu, D. V. (1987). A Markov model for generation of random binary sequences. Journal of Experimental Psychology: Learning, Memory and Cognition, 13, 25-39.
Cleeremans, A. (1993). Mechanisms of implicit learning: Connectionist models of sequence processing. Cambridge, MA: MIT.
Cohen, J. (1977). Statistical power analysis for the behavioral sciences (Rev. ed.) New York: Academic Press.
Coover, J. E. (1913). The feeling of being stared at. American Journal of Psychology, 24, 570-575.
Dulany, D. E., Carlson, R. A., & Dewey, G. I. (1984). A case of syntactical learning and judgment: How conscious and how abstract? Journal of Experimental Psychology: General, 113, 541-545.
Falk, R., & Konold, C. (1997). Making sense of randomness: Implicit encoding as a basis for judgment. Psychological Review, 104, 301-318.
Gatlin, L. L. (1977). Meaningful information creation: An alternative interpretation of the psi phenomenon. Journal of the American Society for Psychical Research, 73, 1-18.
Knuth, D. E. (1997). The art of computer programming: Vol. II. Seminumerical algorithms (3rd ed.). Reading, MA: Addison-Wesley.
Kushner, M., Cleeremans, A., & Reber, A. (1991). Implicit detection of event interdependencies and a PDP model of the process. In Proceedings of the Thirteenth Annual Conference of the Cognitive Science Society. Hillsdale, NJ: Erlbaum.
Lewicki, P., Czyzewska, M., & Hoffman, H. (1987). Unconscious acquisition of complex procedural knowledge. Journal of Experimental Psychology: Learning, Memory and Cognition, 13, 523-530.
Matthews, R. (1997). The Sunday Telegraph, 2 November, p. 6.
Mill, J. S. (1843). A system of logic: Ratiocinative and inductive. London: Longmans, Green & Co.
Mullins, J. (1997). Are you looking at me? New Scientist, 26 July, p. 39.
Peterson, D. M. (1978). Through the looking-glass: An investigation of the faculty of extra-sensory detection of being stared at. Unpublished thesis, University of Edinburgh, Scotland.
Poortman, F. F. (1959). The feeling of being stared at. Journal of the Society for Psychical Research, 40, 4-12.
Rapoport, A., & Budescu, D. F. (1992). Generation of random series in two-person strictly competitive games. Journal of Experimental Psychology: General, 121, 352-363.
Rapoport, A., & Budescu, D. V. (1997). Randomization in individual choice behaviour. Psychological Review, 104, 3, 603-617.
Reber, A. S. (1989). Implicit learning and tacit knowledge. Journal of Experimental Psychology: General, 118, 219-235.
Reichenbach, H. (1949). The theory of probability. Berkeley: University of California Press.
Servan-Schreiber, E., & Anderson, J. R. (1990). Learning artificial grammars with competitive chunking. Journal of Experimental Psychology: Learning, Memory and Cognition, 16, 592-608.
Sheldrake, R. (1994). Seven experiments that could change the world: A do-it-yourself guide to revolutionary science. London: Fourth Estate.
Thalbourne, M. A., & Evans, L. (1992). Attitudes and beliefs about, and reactions to, staring and being stared at. Journal of the Society of Psychical Research, 58, 129, 380-385.
Titchener, E. B. (1898): The feeling of being stared at. Science (New Series), Vol. VIII, 308, 23 December, 895-897.
Wagenaar, W. A. (1970). Subjective randomness and the capacity to generate information. Acta Psychologica, 34, 233-242.
Wagenaar, W. A. (1972). Generation of random sequences by human subjects: A critical survey of literature. Psychological Bulletin, 77 (1), 65-72.
Williams, L. (1983). Minimal cue perception of the regard of others: The feeling of being stared at. Journal of Parapsychology, 47, 59-60.
Individual accuracy scores and the results of chi-square tests and single mean / tests in Expt 1 Non-feedback (60 trials) Non- Total Stare stare [X.sup.2] accuracy Case 1 14 12 1.07 26 2 17 13 0.00 30 3 16 11 0.23 27 4 14 13 0.60 27 5 16 17 0.60 33 6 16 16 0.27 32 7 13 10 3.30 23 8 23 10 0.74 33 9 21 16 3.36 37 10 15 14 0.07 29 11 19 14 0.62 33 12 13 16 0.07 29 M 16.4 13.5 29.9 MCE 15 15 30 t = --0.07 Feedback (180 trials) Non- Total Stare stare [X.sup.2] accuracy Case 1 49 50 1.80 99 2 55 41 0.82 96 3 54 46 2.24 100 4 63 38 2.91 101 5 54 42 0.87 96 6 58 35 0.30 93 7 48 41 0.02 89 8 55 44 1.83 99 9 54 54 7.20 [*] 108 10 53 46 1.81 99 11 53 50 3.76 103 12 48 53 2.70 101 M 53.7 45.0 98.7 MCE 45 45 90 t = 6.19 [**] (*.)p [less than] .01; (**.)p [less than] .001 (two-tailed). Means and single mean t test results for staring and non-staring scores within each block in Expt 1 Mean Mean Mean Block stare t(11) p non-stare t(11) p total t(11) p 1 16.42 1.56 13.50 --2.14 29.92 --.07 2 17.25 3.18 .009 [*] 13.67 --1.54 30.92 .86 3 17.75 4.65 .001 [*] 15.33 .36 33.08 3.94 .002 [*] 4 18.67 6.30 .0001 [*] 16.00 1.06 34.67 5.40 .001 [*] (*.)two-tailed. Note. MCE = 15. Means and single mean t test results for staring and non-staring scores within each block to test for global bias in Expt 1 Mean Mean Mean Block stare t(11) p non-state t(11) p total t(11) p 1 16.83 2.56 .026 [*] 13.25 --2.59 .025 [*] 30.08 .15 2 17.50 3.11 .01 [*] 12.08 --2.57 .026 [*] 29.58 --.29 3 17.67 3.37 .006 [*] 14.58 --0.51 32.25 3.00 .01 [*] 4 15.83 0.81 12.33 --3.05 .011 [*] 28.17 --1.63 [*]two-tailed. Note. MCE = 15 Three-tuples and their incidence in the 12 experimental sequence in Expt 1 SSS SSN SNN NNN NSS NNS SNS NSN Observed 10 31 30 12 31 30 37 35 Expected 27 27 27 27 27 27 27 27 Note. S = staring; N = non-staring. Three-tuples and their incidence in the 10 experimental sequences in Expt 2 SSS SSN SNN NNN NSS NNS SNS NSN Observed 20 21 23 18 19 20 23 27 Expected 22.5 22.5 22.5 22.5 22.5 22.5 22.5 22.5 Note. S = staring: N = non-staring. Individual accuracy scores and the results of chi-square and single mean t tests in Expt 2 No Total Stare (94) stare (86) [X.sup.2] accuracy Case 1 61 30 .00 91 2 49 41 .00 90 3 64 28 .01 92 4 57 27 .61 84 5 46 49 .63 95 6 65 24 .19 89 7 54 44 1.34 98 8 43 41 .80 84 9 49 30 3.12 79 10 59 27 .68 86 11 59 41 2.01 100 12 52 43 .51 95 M 54.8 35.4 90.25 MCE 47 43 90 t = .14 Means and single mean t test results for staring and non-staring scores within each block in Expt 2 Mean Mean staring non-staring Total Block (MCE) t(11) p (MCE) t(11) p (MCE) t(11) p 1 18.42 (17) 1.59 10.92 (13) -2.25 .046 [*] 29.33 (30) -.61 2 21.92 (18) 3.38 .006 [*] 9.33 (12) -2.50 .03 [*] 31.67 (30) 1.67 3 14.50 (12) 4.02 .002 [*] 15.33 (18) -1.77 29.83 (30) .17 (*.)two-tailed.
|Printer friendly Cite/link Email Feedback|
|Author:||Colwell, John; Schroder, Sadi; Sladen, David|
|Publication:||British Journal of Psychology|
|Date:||Feb 1, 2000|
|Previous Article:||Dogs as catalysts for social interactions: Robustness of the effect.|
|Next Article:||Orienting to exogenous cues and attentional bias to affective pictures reflect separate processes.|