Psi and associational processes

Stuart Wilson

If psi (1) is a real phenomenon, then logically it should follow some kind of psychological laws. One kind of model would posit that psi information, once it has entered the cognitive system, is subject to similar laws to other comparable incoming information (e.g., weak sensory stimuli--see Beloff, 1974; Nash, 1986; Schmeidler, 1986 for reviews). Furthermore, if the psi process somehow involves existing cognitive processes, then this could serve to reduce the amount of processing exclusive to psi that would otherwise be required. This is not a new suggestion; in 1946 Tyrell postulated that the psi percipient constructed a "mediating vehicle" which itself is the product of psychological processes. Tyrell was essentially suggesting that psi may "piggy-back" onto normal psychological processes and may influence the organism through its influence on these processes. This is an intuitively appealing notion, for if psi involves such ongoing processes, it might obviate the need for a unique process (or group of processes) dedicated to psi. This would reduce demands on processing capacity, and it may also be the reason why many spontaneous reports of ostensible psi phenomena take the form of normal psychological processes, e.g., associations, memories, feelings, etc. (see Stanford, 1974a). If psi processing could interact with these standard processes appropriately, then they would serve to be useful vehicles as a means for psi to influence the percipient in some way. Eliciting of these normal mental events will not be considered strange; therefore they will not be consciously recognised as being influenced by psi. This might be why spontaneous psi appears to be unconscious (see Stanford 1974a; 1990). A number of researchers have recognised this possibility, and have suggested ways in which psi may interact with cognitive processes. Roll (1966) proposed a model of ESP in which ESP is more like memory than perception. He claimed that certain information represented in some of the individual's memory traces is relevant to the psi source or target, and that extrasensory activation of those traces may result in mediation of the relevant information into consciousness.

Irwin (1979) followed a similar line of reasoning, proposing an information-processing model of psi, in which ESP occurs through paranormal activation of appropriate memory traces, bringing them into consciousness.

Stanford's PMIR (psi-mediated instrumental response) model (Stanford, 1974a, 1974b, 1977, 1982, 1990) also deals with the importance of ongoing processes in the mediation of psi information. Under the PMIR model, a psi-mediated response is accomplished through triggering of various cognitive processes that are already in the repertory of the organism.

One process that may be useful as a psi vehicle is associations. Roll (1966) claimed that mediation of psi information may be governed by the principles of normal cognitive processing, including what he called the "laws" of association. Similarly, in 1973, Stanford suggested that extrasensory information would have to become imposed on or mediated by, or would have to interact with, ongoing associative processes, and his PMIR model also emphasised the role of association in psi processing.

Stanford (1973) used word association as a directed free-response task. The rationale was that word association is a free-response task which is scorable in the same manner as a forced-choice task, thus having the advantages of a free-response test without the drawbacks associated with scoring. It is similar to a free-association task in that participants are given words and asked to respond with the first associated word that comes to mind. It can be scored as a forced-choice task in the sense that all of the responses generated can be categorised. In his 1973 study, Stanford used homophones as stimuli. Participants were instructed to respond with the first associated word that came to mind after hearing the homophone. The categorisation in this case was in terms of two alternative meanings assigned to a particular homophone. Thus responses could be scored as being either one or the other, similar to a forced-choice task.

This approach has several further advantages. Firstly, participants need not know they are participating in a parapsychology study. Consequently, participants do not have to be asked to "turn on" their ESP, which may cause considerable self-consciousness or may even lead to dissonance, depending on their worldview. This may have a considerable detrimental effect on psi performance. A word association task only asks participants to report the first associated word that comes to mind after presentation of a stimulus. In many of the word association experiments used in parapsychology, the assumption is made that the ESP will work automatically on an unconscious level. Stanford (1973) describes work in which participants were made aware of the psi component of the study, but were told that they need not struggle to elicit it as psi would "take care of itself." Thus, Stanford aimed to induce participant motivation to do well in the psi study without invoking potential psi-inhibitive egocentric efforts.

A related advantage of this type of approach is that it is more ecologically valid, reflecting how psi may work in the "real" world (although it may be argued that many "real-world" psi events occur without any of the inherent motivation that Stanford was keen to promote). It is unlikely that the psi experiences typically reported in the spontaneous case literature are a result of a conscious intention on the part of the percipient. Indeed, as stated above, many reported experiences take the form of associations, feelings, etc., suggesting an interaction with ongoing cognitive processes. Finally, this paradigm circumvents the tendency of participants to pattern responses rationally. For example, on a typical Zener card test, a participant may call a circle only because he or she hasn't called it for a while. Again, this may have an effect on psi performance. In a word association task, each trial is independent, and patterning of responses is not a possibility, giving psi a better chance to manifest itself without the burden of conscious rationalisation (though see discussion below on a potential "stacking-effect" relating to non-independence of targets in Experiment 1).

Stanford's (1973) study seemed to suggest that a word association test was a useful way of looking at the way in which psi-related processes might exert a subtle influence on ongoing cognitive processes. An attempted replication was less successful, however, although there were trends suggestive of an effect (Stanford & Schroeter, 1978). Given the conceptual and methodological advantages and the potential usefulness of using word association, we decided to use word association as a non-intentional psi task. (2) In line with Stanford (1973, 1977) we constructed a study using homophones (see above for definition). The primary aim of Experiment 1 was to establish whether or not an effect would appear given this protocol, thus providing evidence that the potential presence of psi-mediated information could have an influence on a particular psychological process (i.e., word association). Also, we took the opportunity to investigate whether our method of selecting stimuli was justified (see below). It was hypothesised that the presence of an "agent" (3) would influence the participant's responses to homophones, in accordance with a randomly chosen "target" interpretation.

The current study differs from Stanford in that participants were not aware they were participating in an ESP test. Although Stanford told his participants that ESP would "take care of itself" without any conscious effort, the very fact that they knew the study was parapsychological may have influenced their approach to the experiment, either through demand characteristics or by other means. The present research attempts to obtain a more valid measure of non-intentional psi by having the participants unaware of the ESP nature of the task.

There were two exploratory aspects to this study; the validity of matching each homophone pair for familiarity was to be determined (see below), as were any potential experimenter effects. In the present study, two researchers took on the roles of experimenter and agent. In accordance with previous research suggesting that different experimenters may influence the outcome of a session in different ways (see White, 1977; Wiseman & Schlitz, 1997), it was decided to investigate in the present study whether any effect of experimenter was present. As the study progressed, a further hypothesis emerged: that responses which appeared to have a bias against them within the sample would result in a greater number of "hits," in accordance with Stanford (1967).

EXPERIMENT 1

Method

Overview. Participants were presented with homophones and asked to report the first associated word that came to mind. Each homophone had two predominant interpretations, and these interpretations were assigned to two lists. Before each session, one list was randomly designated the "target" list. An agent viewed images relating to the target list, and attempted to influence the associations of the participant accordingly.

Participants. Twenty unpaid participants were used, consisting of 18 females and 2 males aged between 19 and 43. All were students or employees of the University of Edinburgh and were known to Stuart Wilson (SW). They were recruited through word of mouth, through e-mail appeals, and through SW's tutorial group. Participants were told that they would be taking part in a study of word association, and at no point were they told of the parapsychological nature of the study. There were two experimenters, SW and Eric Pronto (EP), who both assumed the role of either agent or experimenter during the course of the study.

All sessions were conducted in the Koestler Parapsychology Unit, University of Edinburgh. The rooms used in the present study have been previously used in various ganzfeld experiments, and are considered to be secure (see diagram on p. 133; for a full discussion of the security issues involved in this setup see Dalton et al., 1996).

Materials. Twenty homophone pairs (e.g., mail/male) were selected. These were chosen from a database on the basis that each pair was roughly matched for familiarity. The familiarity ratings were derived from merging three sets of familiarity norms: Paivio (unpublished (4)), Toglia and Battig (1978), and Gilhooly and Logie (1980). The method by which these three sets of norms were merged is described in Coltheart (1981). It was hoped that each interpretation thus had an equi-probable chance of being elicited on any one trial. Familiarity was chosen because this was the measure that showed the closest matches between pairs. (5)

Each homophone's interpretation was randomly assigned by computer to one of two lists (List A or List B) such that one list had a complete set of 20 interpretations, while the other list contained the alternative interpretations.

Homophones were recorded using the voice of Caroline Watt, a research fellow at the Koestler Parapsychology Unit. Words were recorded onto DAT (digital audio tape; Sony DTC A8 DAT recorder). Each word was recorded approximately three times, and the best one chosen for inclusion as a stimulus. Further editing resulted in each word being preceded by 15 s of white noise, followed by a 5-s quiet "answer" period.

Visual cues relating to each meaning of the homophones were obtained from the Internet and were transformed into "jpeg" files consisting of the image and the word it related to. The images could be either illustrations or actual photographs. An example of the two images obtained for the homophone "steal/steel" is given on p. 135.

A computer program was designed in a way such that during the 15 s of white noise on the tape, the visual cue related to the forthcoming homophone was displayed on a 17-inch monitor viewed by the agent. The program also randomly selected List A or List B to be the "target" list for the session in progress.

Procedure. Before commencement of the session, a coin was tossed between SW and EP in order to determine who was to be "agent" and "experimenter" during the session. SW usually tossed the coin, and there was no suggestion that this was anything other than sufficiently random.

Each participant was greeted by SW, who then introduced EP as a co-experimenter. Whoever was designated "agent" then guided the participants into a sound-attenuated experimental room and asked them to sit in the chair and make themselves comfortable. He explained what was going to happen during the session and what the participant's task was. Participants were also told at this point that they were under no obligation to complete the experiment, and could leave if they wished. Having agreed that they wanted to continue (as all participants did), they were instructed to respond to each word that they heard with the first associated word that came to mind. They were not told that the words they would hear were homophones. The agent had the most interaction with the participant, so as to build up some kind of rapport. At no point was the parapsychological nature of the experiment made explicit. The participant clipped on a small microphone, and the recording levels were checked. The agent asked the participant if he/she had any further questions. He then asked how the participant would like the lights (dimmer/ brighter). The participant was instructed to put the headphones on, and was told that the experiment would start shortly.

The agent retired to the "agent's" room, which was approximately 25 m away from where the participant was situated (see Figure 1). above). When the agent was inside (indicated by the ceasing of a flashing light in the experimenter's room showing when the agent's door had been closed), the experimenter (situated in a room adjacent to the participant) simultaneously started the computer program and the DAT recording. The agent then viewed images related to one interpretation of each homophone, and attempted to psychically influence participants' subsequent response.

[FIGURE 1 OMITTED]

The experimenter listened to what the participant was hearing in one ear of his headphones while monitoring responses in the other ear. Each response was noted on a response sheet by the experimenter (who, at this point, was blind to the target list).

After completion of the trials, the experimenter entered the participant's room with his/her responses. He then went through all the homophones and the associated responses with the participant. The participant was asked to clarify which (if any) of the alternative interpretations was meant by the response. This was conducted by presenting two words, each associated with one of the meanings, and asking which of the words was most like the response. For example, if the response to "steel/steal" was "robbery," participants were asked if their response was more like "metal" or like "thief." This served to clarify any ambiguities, and also to get an indication of which words were mis-heard. At this point, the experimenter was still blind to what the target list was and so could not influence the participant towards the targets. Ambiguities were, in fact, relatively rare, and on the majority of the trials when the word was correctly heard it was clear what the association was.

After this, the participant was informed that the session was over. At this point the agent had returned to the experimental suite. The participant was asked if he/she had any questions, which the experimenter would attempt to answer. The majority of the participants asked what the purpose of the study was. The experimenter then went on to describe the psi nature of the experiment, and briefly outlined some past psi research. Any participant who was given this information was requested not to reveal the psi nature of the study to anyone else. The remaining participants who did not specifically ask what the purpose was tended to be close colleagues/students of SW who were aware of his research interests. Although these participants were not informed of the nature of the study directly after the session, they were all subsequently made aware of the psi nature of the study by the first author, either in private discussions or through departmental seminars in which the study was presented. Thus, all participants were made aware of the psi nature of the study, either immediately after their participation or shortly after completion of the study.

Once the participant had left, the experimenter checked the computer to see which list was the target list, and scored the responses accordingly (each response having previously been classified as being related to one interpretation by the experimenter while he was still blind to the target list, thus ruling out any bias in the scoring).

At no point during the session was the participant made aware of the psi nature of the task. It is possible, however, that some worked it out independently, given their knowledge of the parapsychology research unit and SW's involvement with parapsychology. It is worth noting, however, that in those situations in which the participant enquired about the nature of the experiment, nobody mentioned having any idea what the purpose was.

Results

Any responses that were not one of the two meanings assigned to each homophone were discarded, as were occasions when no response was offered. Of the 400 trials (20 subjects with 20 trials per subject), 109 could not be used in the analyses. All results are, therefore, in terms of "appropriate responses." Because the computer randomly chose the "target" list, the probability of any one response being a hit was 0.5.

Planned comparisons. Overall, trial-based analyses were planned to serve as the criterion analyses, although subject-based analyses were also carried out.

Out of 291 appropriate responses, there were 161 hits (55.3%) and 130 misses (44.7%). These figures give a z-score of 1.82 (p = .07, two-tailed). (6) Effect size was calculated, suggesting a small effect (0.1) (see Table 1).

In addition to this, a subject-based analysis was also conducted, which revealed no significant deviation from chance: t(19) = 1.70, p = .11.

Results were then split in terms of whether SW or EP was acting as agent or experimenter. When SW was acting as the agent (and EP was the experimenter), there were 74 hits out of 125 (59.2%, z = 2.05, p = .04).

Additionally, a subject-based analysis on data for trials in which SW acted as agent fell just short of significance: t (8) = 2.24, p = .06.

When EP was acting as the agent (and SW was the experimenter), there were 87 hits out of 166 (52.4%, z = 0.62, p = .54) (see Table 1). The difference between the agent/receiver arrangements was not significant: [chi square] (1, N= 291) = 1.33, p = .25.

Subject-based analysis on trials in which EP acted as agent revealed no significant effect: t (10) = 0.41, p = .69.

Again, a subject-based analysis was conducted to test the difference between the "agents." Due to the small N, a Mann-Whitney test was used, which did not reveal any significant difference: U= 34.00, p = .24.

Post-hoc response bias comparisons. At the conclusion of the study, it became clear that the strategy of matching the interpretations for familiarity had not been successful. Several homophones were eliciting one type of response considerably more than the other. This led us to consider testing for a "response bias" effect.

In 1967, Stanford suggested that psi-mediated impressions may come to the attention of the percipient because they somehow feel out of place in their context. Stanford proposes that the feeling "something terrible has happened to John" is more likely to be noticed or acted upon if it appears inappropriate in its context. On the other hand, if a person often thinks that "something terrible has happened to John" (say, for example, if John has a dangerous job), then this feeling will not be unusual and will most likely be dismissed. In relation to laboratory ESP research, there are certain responses which are more or less favoured (e.g., in an experiment using Zener cards, the star tends to be the most popular choice, with the wavy lines the least popular). Stanford has suggested that responses with a low probability of occurrence may be more accurate than more popular responses, due to the extra "push" of psi triggering a response against which a bias normally exists. Stanford has some experimental evidence for this effect (Stanford, 1967; Stanford, 1973), and, as mentioned previously, his 1973 study made use of homophones as stimuli.

We decided to conduct post-hoc tests on our data in order to examine whether a response-bias effect was present. Although these were post-hoc tests, it is important to note that we did not conduct them with any knowledge concerning the outcomes. We noticed that certain responses were being elicited more than others, and this raised the possibility of a potential psi effect in these data. Ordinarily, a response bias is defined at the individual level as Stanford (1967, 1973) did. In the present situation, we did not have the data to do this. What had caught our attention, however, were instances indicating the existence of sample-wide biases that would still permit at least an indirect assessment of the response-bias effect in our data. We formulated the hypothesis prior to examining any of the data, so in this respect, many of the shortcomings of performing post-hoc tests were avoided. So, in light of Stanford's reasoning, we hypothesised that responses which were found to be the less favoured of the two alternatives would display a greater psi score than their more favoured counterparts. We took all the minority responses (responses which were the less favoured of the two interpretations for each homophone), and found that, out of 70 minority responses, there were 43 hits (61.4%, z = 1.91 p = .06) (see Table 2). This is compared to a z-score of 1.01, (p = .31) when hits on pro-bias responses are taken into account (118 hits out of 221 majority responses, 53.3%).

Again, we divided up the data in terms of agent/experimenter and found that when SW was acting as agent there were 21 hits out of 30 minority responses (70%, z = 2.19, p = .03). When EP was the agent there were 22 hits out of 40 (55%, z = 0.63, p = .53) (see Table 2). Again, the difference between the agents was not found to be significant: [chi square] (1, N= 70) = 1.63, p = .20.

Stacking effect--a possible artefact? Technical limitations in the current study meant that the order in which the homophones were presented could not be separately randomised for each participant. This meant that each homophone interpretation was randomly assigned to a target list (A or B) and each participant was randomly assigned a target list at the outset of the trials. As a result of this technique, the targets were not independent of each other, with half of the participants receiving the same target order while the other half received the alternative target order. Situations in which participants in an ESP test call from the same target order have been criticised as being vulnerable to what is known as the stacking effect (see Burdick & Kelly, 1977). The stacking effect occurs when participants share a particular response bias. This commonality in responding between participants can lead to spuriously high (or low) correlations between target and responses. This is due to the subjects sharing the same calling bias, which may coincide with (or deviate from) the target order (Thalbourne, 1982). In other words, the stacking effect assumes that respondents may share common "guessing habits." Grimmer and White (1986) demonstrated the presence of such habits in an Australian sample. They reported that when participants were asked to guess which two geometrical shapes the experimenter was thinking of, the circle-square guess sequence appeared more frequently than the circle-triangle one.

Naturally, when the target order is random and unique for each respondent, this guessing bias can safely be ignored, since its effect is counterbalanced by the design (especially when the target items have equal probabilities of occurrence). However, if respondents are guessing from a single target sequence, and if they share a common response bias, then the results cannot be thought to be independent from each other (for example, if the circle-square sequence is consistently included in the target sequence, then the tendency to call "circle" followed by "square" will result in hits that are not due to psi, but are an artefact of that particular response bias). In the latter case, the potential increase in the degree of agreement between the results can produce a reduction of the variance within each target, and consequently it may have a substantial yet spurious effect (a stacking effect) on the statistical significance of the tests employed.

Several procedures have been proposed to account for the presence of the stacking effect, such as the Greville formulas (Greville, 1944; Pratt, 1954), the majority-vote technique, or the index of preference (for an overview see Thouless & Brier, 1970; Burdick, 1983). However, these techniques make the assumption that the stacking effect is inevitably present in all cases in which the respondents share a target order. As such, the methods advocated by previous authors have consisted of manipulating the variance estimates and thus making the tests more conservative. However, it is not necessarily the case that the stacking effect becomes manifest whenever single target orders are used, and it may be possible to detect situations in which there is actually a shared response pattern for the participants in the study.

Given that the effect may or may not be present in designs such as the current one (which consisted of two complementary target lists), the main question becomes one of determining whether or not systematic response patterns were present in the data.

Analysis

The issue to address was whether or not the target lists differed from each other. If one particular list was related to a particular calling bias, then it would lead to this list displaying spuriously high (or low) "hits" in relation to the other list.

To analyse this, the two target lists were considered to be "treatments." Participants' guesses were scored against one of these treatments, and the two treatments were compared. This comparison was non-significant, suggesting that guesses relating to each of the target lists were not significantly different from each other: t (18) = 0.08, p = .93.

EXPERIMENT 2

Experiment 2 was an attempt to replicate some of the effects and address some of the issues from Experiment 1. This experiment was conducted using a considerably larger sample size than the previous study, addressing some of the issues related to the small N used in Experiment 1. More specifically, we wished to determine whether the positive (though nonsignificant) trend indicating that homophone association might be influenced by psi would persist. We also were interested in replicating the apparent response-bias effect from Experiment 1.

Some methodological and procedural aspects were modified for convenience or due to necessity (see below). The experiment described below was part of an overall larger experiment conducted as part of the first author's Ph.D. research. Only the results relevant to the psi/word association paradigm will be reported here.

The first methodological issue to be addressed in this follow-up study concerns intelligibility of the stimuli. Many participants in Experiment 1 had difficulty comprehending certain stimuli. Due to the diversity of subjects used in this experiment, it is possible that regional accents may have caused some subjects to misunderstand what word was being presented. Using a computer to generate the verbal stimuli was considered, but none of the programs considered sounded natural, and this would likely have caused more confusion than using a human voice. Instead, a different speaker was used in Experiment 2, and words were electronically treated in an attempt to enhance intelligibility.

An additional aspect of Experiment 2 was an issue related to potential gender differences in psi scoring. This particular aspect was included as a result of some unpublished work by the first author (conducted in the interim period between Experiments 1 and 2) suggesting that females may have obtained higher psi-scores than males when SW was "sending" (Wilson, 2002), although it should be noted that the effect was weak and nonsignificant. This gender effect is something that previously has been found in parapsychology, using different kinds of ESP tasks (see Dalton & Utts, 1995; Freeman, 1967).

Freeman conducted a series of psi experiments with children in the 1960s (1963, 1965, 1966, 1967, 1969) and found that, under certain conditions, males score in the opposite direction from females. Freeman reasoned that the different patterns of scoring for males and females may be indicative of underlying psychological differences between the genders. He then proceeded to conduct ESP studies using tests of verbal and spatial abilities as a possible mediating variable for the observed gender differences. Freeman's studies (1968, 1969) offered a degree of support for this hypothesis, suggesting that gender differences in psi performance may be a function of underlying cognitive factors. (See Kimura [1999] and Halpern [2000] for a discussion on gender differences in cognitive abilities.)

Palmer (1978) reviewed the literature on gender differences in ESP scoring, and concluded that, apart from two notable exceptions, gender did not play a part in ESP studies "although it occasionally has been shown to interact with other predictor variables" (p. 146). However, Rao, Kanthamani, and Norwood (1983) identified a number of studies conducted since Palmer's review that appear to contradict this assessment.

Dalton (1994) and Dalton and Utts (1995) have investigated the effect of same- or mixed-sex agent/receiver pairs in various ganzfeld databases, and have suggested that mixed-gender pairings are more successful than same-gender pairings.

A hypothesis related to differences in psi scores between males and females was added as an exploratory measure that could potentially inform us further about the nature of the agent/receiver pairing in psi experiments. This was in addition to the two main hypotheses relating to the results of Experiment 1, i.e., that agent involvement would influence word associations, and that this effect would be more pronounced for "less-favoured" responses (response-bias hypothesis).

Finally, due to the installation of new digital equipment in the KPU, Experiment 2 incorporated improved target randomisation techniques. (7) Although the stacking effect was not found to influence the outcome of Experiment 1, it is acknowledged that the design was not ideal, and as such, Experiment 2 addressed this issue.

Method

Participants. Fifty participants took part in this study. All were students or employees of the University of Edinburgh and were aged between 18 and 38. There were 14 males and 36 females. They were recruited through word of mouth, through e-mail appeals and through tutorial groups. In the current experiment, there was only one experimenter, who also acted as an agent (SW), because EP was unavailable for this follow-up study.

Materials. The homophones used were identical to those used in Experiment 1 with the following exceptions. Due to a recording error, the homophone "watch" could not be used, and was replaced by "flour/flower." An additional homophone ("serial/cereal") was also used. Additionally, all stimuli were re-recorded by a different speaker. (8) Words were recorded onto DAT (digital audiotape; Sony DTC A8 DAT recorder). The words were then transferred onto computer in the form of ".wav" files. Words were then "de-essed" and compressed in order to make them more easily intelligible. Further editing on the computer resulted in each word being preceded by 15 s of white noise.

The visual cues relating to each meaning of the homophones were identical to those used in Experiment 1, except in the case of the new homophone pairs (flower/flour; cereal/serial). The visual cues relating to these pairs were obtained from the Internet (via the image search at www.altavista.com), and were transformed into "jpeg" files consisting of the image, and the word it related to.

A computer program was written in Visual Basic to present stimuli and record responses. Auditory stimuli were presented to participants via Koss SB30 headphones, which had an attached microphone to record responses.

Procedure. Participants were greeted by SW and led into the experimental suite at the KPU. A brief description of the procedure was then given to each participant, and they were asked whether they wanted to continue (all did). Participants were also told they could leave the experiment at any time. No participants opted to leave the experiment. They were then given a series of questionnaires (results of which will not be reported here) before taking part in two experiments. The psi/word association was randomly chosen to be either the first or second of these experiments. Participants were taken into the receiver's room and told to make themselves comfortable. It was then explained to them that they would hear 15 s of white noise followed by a word. They were instructed to relax during the white noise, and then to say aloud, and as clearly as possible, the first associated word that came to mind after hearing the word. The experimenter then answered any questions before starting the experiment in the experimental room. There was a slight delay in the commencement of the trials in order to give the experimenter time to lock the door to the experimental room and take up his position in the agent's room.

For each trial, the computer randomly selected a homophone (which had two primary interpretations), and randomly selected one of these interpretations as a target. This target interpretation and a visual image related to it were displayed to the agent (always SW) during the period in which the participant was listening to the white noise. As the agent was viewing this image, he attempted to influence the participant's response to the homophone in the direction of the target interpretation. After the white noise was over, the computer began recording the participant's response. (9) Once the session was over, the experimenter went back to the experimental suite, where he informed the participant that this part of the experiment was over.

Afterwards, participants were asked if they had any questions, which SW did his best to answer. No participant explicitly asked whether the study had a parapsychology component. All participants were sent, via e-mail, a full debriefing on what the research was about.

Classifying responses. It was important that the classification of responses into "hits" and "misses" was carried out blind. This meant that the experimenter could not take part in this process. Two judges were used to carry this out. (10) The first judge listened to all the responses and noted them down (with no prior knowledge of either the homophone or the target interpretation). Any ambiguous responses were given to the second judge, who gave her opinion on what the word was. This was usually enough to come to a consensus, but in two cases the response was too unintelligible to make out, and these were discarded.

The judges were then given information about the homophones relating to each response. With this information, they recorded which particular interpretation of the homophone the response was related to (e.g., the response was "envelope" the homophone was "male/ mail," the response relates more to "mail" than "male"). It is important to note that judges were still blind as to the target at this point.

Finally, each judge had to class each response depending upon whether it was a "hit" (response matches target interpretation), a "miss" (response matches the non-target interpretation) or "neither" (response does not match either interpretation). Although initial agreement was high, it was necessary to get a consensus between the judges. The experimenter then collated the trials on which the judges disagreed, asked each judge for an explanation as to why they had classified as they did, and re-presented these to the judges. It became clear at this point that many of the disagreements between the judges were simply due to mistakes made in the judging procedure. There were some instances in which a judge's knowledge caused him or her to rate an item as a "hit" while the other judge rated the same item as a "neither." For example, the homophone "been/bean" often elicited the response "pie." One judge had never heard of a "bean pie," so classed these responses as "neither." However, when the other judge explained what a bean pie was, an agreement was reached. Such instances were clarified during this procedure, and a general consensus was reached. After this, there were only eight instances in which the judges still disagreed. These were subsequently given to a third party (also blind) who made a final decision.

Results

Overall psi scores. As the target interpretation in each trial was selected randomly by the computer, the probability of obtaining a hit by chance is 0.5. It should be noted that this is true regardless of any biases that might exist towards or away from particular responses.

As in Experiment 1, the analysis was conducted on all the "appropriate" responses (i.e., responses that related to one of the two prescribed meanings of the homophone). Overall, there were 905 appropriate responses, of which there were 454 hits and 451 misses. This results in a z-score of 0.1, which is not significant: p = .92). Again, in addition to this trial-based analysis, a subject-based analysis was also conducted, which was nonsignificant: t (49) = 0.11, p = .92.

Response bias. As in Experiment 1 the responses that were favoured less within our sample were isolated and analysed. It is interesting to note that of the 18 homophones that were common to both experiments, the same response bias was observed in 16 of them. The homophone pair meet/meat elicited a different response bias in Experiment 2 than it did in Experiment 1, and the homophone pair some/sum elicited equal responses for the two interpretations in Experiment 2 (whereas in Experiment 1, "some" was less frequently elicited). The correspondence between response patterns between the two experiments supports the notion that the biases under study do, in fact, reflect population biases.

Overall there were 185 minority responses, of which 93 were hits and 92 were misses. This is almost exactly as chance would predict, and is not significant (z = .07, p = .94).

Gender effects. To test this hypothesis a "psi-proportion" was calculated for each participant by dividing the number of hits they achieved by the total number of appropriate responses they gave (i.e., hits + misses).

Table 3 suggests that females score more hits on the psi task than males. An independent samples t-test, with "psi-proportion" as the dependent variable, was carried out to determine whether the difference between males and females was significant. This hypothesis was directional, due to the trends observed in the previous studies. The t-test revealed that the difference was marginally significant: t (48) = 1.64, p = .05, one-tailed.

Discussion

The two experiments described were attempts to investigate the potential role of associative processes in mediating psi inputs. Results from Experiment 1 were encouraging, if inconclusive. Experiment 2 attempted to replicate and extend these findings. Results from this follow-up experiment did not reproduce the response patterns from Experiment 1, although there was evidence of a gender effect.

It is worth briefly speculating on this failure to replicate. Firstly, the most parsimonious explanation is, simply, that there was no effect in the first experiment, and any suggestion of one was the result of a Type 1 error. This error may subsequently have been eliminated in Experiment 2 due to the increased number of trials. This would appear to be the most likely explanation.

Secondly, even if there were above-chance patterns in the data from Experiment 1, the lack of replication of these patterns may be due to the improvement in randomisation procedures in Experiment 2. This may have eliminated any potential biases caused by patterned responses (i.e., the stacking-effect). However, given the analysis conducted on the data from Experiment 1, it is difficult to see where any potential artefact in the initial study would manifest itself, as the lack of true randomisation procedure in Experiment 1 did not seem to lead to any consistent patterning of responses. It is, therefore, unclear whether the randomisation in Experiment 2 actually did "iron-out" an artefact from Experiment 1.

It is also worthy of note that the conditions in which Experiment 1 and Experiment 2 were conducted were slightly different, and this may have been a factor. In Experiment 2, the word association test was one of a number of tasks given to the participants, (11) and as such, the dynamics of the system into which participants were entered were slightly different from those of Experiment 1, which solely consisted of the word association experiment. In addition to this, only one person acted as both experimenter and agent in Experiment 2, whereas Experiment 1 had two individuals taking on these roles respectively. Again, this slightly altered participants' experience of the experimental situation.

It is also worth taking into consideration the pool of participants used. In Experiment 1, all the participants were known to the experimenter (SW). Participants in Experiment 2 were recruited through e-mail appeals, and most were not known to the experimenter prior to testing. It is possible that motivation issues played a part, and that the participants in Experiment 1 were more motivated to do well. A meta-analysis on early ganzfeld research (Honorton et al., 1990) seemed to suggest that agent-receiver pairs who were friends performed better than strangers (although this trend was not observed in a meta-analysis of later ganzfeld work, Bern and Honorton, 1994). With regards to agent/receiver interpersonal dynamics in the experiment under discussion, it is difficult to predict how the relationship might influence non-intentional psi, when the receiver is not aware that he/she is taking part in a psi task, or that an agent even exists. One might expect that, in this situation, the relationship between the two would be less critical than in intentional psi tasks, but, again, this is something that might benefit from closer scrutiny.

A further point of discussion concerns the "experimenter effect." Results of Experiment 1 suggested that different experimenter/agents may obtain different "hit" rates in an ESP test. It was found that when SW was agent, there were more hits than would be expected by chance (59.2%), while when EP was agent, hits were close to chance (52.4%). This looks like an experimenter effect (see Wiseman & Schlitz 1997), although it is important to note that the difference between the scores for each agent was nonsignificant. Possible explanations as to why patterns differed when different agents were used may be found in the fact that most participants were better known to SW than to EP. Indeed, many participants were members of classes taught by SW (see Van Bussbach, 1953, 1955, 1956, 1959, 1961, for a suggestion that teacher-pupil pairings may be successful in psi tasks).

It should be noted that there was no attempt to replicate this effect in Experiment 2 due to resource and time constraints. Thus, the validity of this observed pattern must be in serious doubt, given the replication failure for the other effects found initially in the first experiment.

Gender effect. Related to the issue of interpersonal factors is the gender effect. There was a slight difference between males and females in Experiment 2, with females displaying higher psi scores than males. This effect, was, however, only marginally significant, and might possibly be considered a result of multiple analyses. Some previous research on psi tasks such as the ganzfeld has suggested that male/female agent-receiver pairs tend to be the most successful (see Dalton, 1994; Dalton & Utts, 1995), and this is something which should be further investigated, as the current results are inconclusive due to the problems associated with conducting multiple analyses. It would seem that the issues concerning the dynamics between those involved in an experimental situation discussed above are further indications that interpersonal factors are important in parapsychological research. This issue deserves further study, with relationships between agent-receiver pairings being more closely scrutinised.

Finally, it is important to raise a further methodological issue that differentiates our approach from that of Stanford (1973). When investigating the response-bias effect, it is possible that the way we operationalised "less-favoured" responses could be open to question. All responses for each homophone were collated, and it was then determined how many responses there were for each interpretation. The one with fewer responses was designated the "less-favoured" response, and the author then went through all the less-favoured responses, counting how many were hits. It could be argued that established word association norms should have been used when doing this. As it stands, the sample group itself was used to establish what these "norms" were, and this may have been erroneous for one reason or another (although the consistency between Experiments 1 and 2 suggests that the biases used may be representative of the population as a whole). However, in his 1973 study, Stanford found that his subject population did not conform to the established norms he was using, perhaps suggesting that each subject population should be considered in isolation. An alternative is to attempt to determine what biases exist in each individual participant, or to manipulate this in some way. That would eliminate the possibility that the response-bias effect was due to individual idiosyncrasies in responses.

In summary, the two experiments described above have raised many important issues that are pertinent to parapsychology as a whole. The possibility of using cognitive, associative processes as a psi "vehicle" is still something that deserves more detailed study, while the role of interpersonal factors in psi experiments continues to be an important factor in our studies.

TABLE 1
HIT RATES, Z-SCORES, AND EFFECT SIZES FOR ALL APPROPRIATE
RESPONSES, AND BY AGENT

                Overall    SW sending   EP sending

Score           161/291      74/125       87/166
%                 55.3        59.2         52.4
Z                 1.82        2.05         0.62
Effect size       0.1         0.18         0.04

TABLE 2
RESPONSE BIAS DATA: OVERALL SCORES AND BY AGENT

              Overall    SW sending   EP sending

Score          43/70       21/30        22/40
%               61.4         70           55
Z               1.91        2.19         0.63
Effect size     0.23        0.4          0.14

TABLE 3
PSI PROPORTIONS AND STANDARD DEVIATION
FOR MALES AND FEMALES

                       Mean Psi Score      Standard deviation
                    (hits/hits + misses)

Males (N= 14)               0.46                  0.12
Females (N= 36)             0.51                  0.11

AUTHOR NOTE

An earlier version of this paper was presented at the 2000 Society for Psychical Research conference. The authors would like to thank the Society for Psychical Research and the Bial Foundation for providing funding at various points during this research. This paper is dedicated to Professor Robert Morris, who sadly died before publication.

NOTES

(1.) "Psi" can be defined as an interaction between the organism and its environment through means other than the recognised sensorimotor systems.

(2.) In this case, the task was nonintentional in the sense that participants were not aware of the specifics of the psi task, although some may have hypothesised that there was a psi component.

(3.) In this paper, the term "agent" is used instead of "sender." This

is because the term "sender" has specific connotations concerning what might be happening in the experimental situation. "Agent" is seen to be a more neutral term that makes fewer assumptions about processes.

(4.) These are an expansion of the norms of Paivio, Yuille, and Madigan (1968).

(5.) See Gernsbacher (1984) for a justification of using familiarity as a measure.

(6.) Except where indicated, all tests described are two-tailed in order to detect any potential "psi-missing."

(7.) This was achieved by utilising the computer's pseudo-random number generator.

(8.) Thanks to Sandie Cleland of the Psychology Department, University of Edinburgh for recording these stimuli, and to Ben Schogler for his technical assistance during recording and editing of the stimuli.

(9.) As a safeguard, a tape recorder was also present in the participant's room in order to record the responses that might be lost as a result of a computer failure. On several occasions the computer failed to record responses, and this tape recording was used. No data were lost as a result of computer failure.

(10.) Thanks to Ian Baker and Dr. Lesley Smith for agreeing to be judges.

(11.) These consisted of three questionnaires and a "false-recognition" experiment identical to Experiment 1 in Wilson (2002).

REFERENCES

BELOFF, J. (1974). The subliminal and the extrasensory. In A. Angoff & B. Shapin (Eds.), Parapsychology and the sciences. New York: Parapsychology Foundation.

BEM, D.J., & HONORTON, C. (1994). Does psi exist? Replicable evidence for an anomalous process of information transfer. Psychological Bulletin, 115, 4-18.

BURDICK, D.S. (1983). Analysing binary data in the presence of the stacking effect. Journal of Parapsychology, 47, 237-241.

BURDICK, D.S., & KELLY, E.F. (1977). Statistical methods in parapsychological research. In B.B. Wolman (Ed.) Handbook of parapsychology (pp. 81-130). New York: Van Nostrand Reinhold.

COLTHEART, M. (1981). The MRC psycholinguistic database. Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 33A (4), 497-505.

DALTON, K. (1994). A report on informal ganzfeld trials and comparison of receiver/sender sex pairing: Avoiding the file drawer. Proceedings of Presented Papers: The Parapsychological Association 37th Annual Convention, 104-113.

DALTON, K., & UTTS, J. (1995). Sex pairings, target type and geomagnetism in the PRL automated ganzfeld series. Proceedings of Presented Papers: The Parapsychological Association 38th Annual Convention, 99-109.

DALTON, K., MORRIS, R.L., DELANOY, D.L., RADIN, D.I., TAYLOR, R., & WISEMAN, R. (1996). Security measures in an automated ganzfeld system. Journal of Parapsychology, 60, 129-147.

FREEMAN, J.A. (1963). Boy-girl differences in a group precognition test. Journal of Parapsychology, 27, 175-181.

FREEMAN, J.A. (1965). Differential response of the sexes to contrasting arrangements of ESP target material. Journal of Parapsychology, 29, 251-258.

FREEMAN, J.A. (1966). Sex differences and target arrangement: High school booklet tests of precognition. Journal of Parapsychology, 30, 227-235.

FREEMAN, J.A. (1967). Sex differences, target arrangement and primary mental abilities. Journal of Parapsychology, 31, 271-279.

FREEMAN, J.A. (1968). Sex differences and primary mental ability in a group precognition test. Journal of Parapsychology, 32, 176-182.

FREEMAN, J.A. (1969). Sex differences in ESP response as shown by the Freeman Picture-Figure Test [Abstract]. In W.G. Roll, R.L. Morris, & J.D. Morris (Eds.), Research in Parapsychology 1969, pp. 49-51. New Jersey: Scarecrow Press.

GERNSBACHER, M.A. (1984). Resolving 20 years of inconsistent interactions between lexical familiarity and orthography, concreteness and polysemy. Journal of Experimental Psychology: General, 113 (2), 256-281.

GILHOOLY, K.J., & LOGIE, R.H. (1980). Meaning-dependent ratings of imagery, age of acquisition, familiarity, and concreteness for 387 ambiguous words. Behaviour Research Methods and Instrumentation, 12 (4), 428-250.

GREVILLE, T.N.E. (1944). On multiple matching with one variance deck. Annals of Mathematical Statistics, 15, 432-434.

GRIMMER, M.R., & WHITE, K.D. (1986). Psychics & ESP: The role of population stereotypes. Australian Psychologist, 21, 405-411.

HALPERN, D.F. (2000). Sex differences in cognitive abilities (3rd ed.). New Jersey: Lawrence Erlbaum Associates.

HONORTON, C., BERGER, R.E., VARVOGLIS, M.P., QUANT, M., DERR, P., SCHECHTER, E.I., & FERRARI, D.C. (1990). Psi communication in the ganzfeld: Experiments with an automated testing system and a comparison with a meta-analysis of earlier studies. Journal of Parapsychology, 54, 99-139.

IRWIN, H.J. (1979). Psi and the mind: An information processing approach. Metuchen, NJ: Scarecrow Press.

KIMURA, D. (1999). Sex and cognition. Bradford, MA: MIT Press.

NASH, C.B. (1986). Comparison of Subliminal and Extrasensory Perception. Journal of the Society for Psychical Research, 53 (805), 435-453.

PALMER, J. (1978). Extrasensory perception: Research findings. In S. Krippner (Ed.), Advances in parapsychological research 2: Extrasensory perception. New York: Plenum Press.

PAIVIO. Unpublished extension of the Paivio, Yuille & Madigan norms.

PAIVIO, A., YUILLE, J.C., & MADIGAN, S.A. (1968). Concreteness, imagery and meaningfulness values for 925 words. Journal of Experimental Psychology, 76 (1:2), 1-25.

PRATT, J.G. (1954). The variance for multiple-calling ESP data. Journal of Parapsychology, 18, 37-40.

RAO, R.K., KANTRAMANI, H., & NORWOOD, B. (1983). Sex-related differential scoring in two volitional studies. Journal of Parapsychology, 47, 7-21.

ROLL, W.G. (1966). ESP and memory. International Journal of Neuropsychiatry, 2, 505-521.

SCHMEIDLER, G.R. (1986). Subliminal perception and ESP: Order in diversity? Journal of the American Society for Psychical Research, 80, (3), 241-264.

STANFORD, R.G. (1967). Response bias and the correctness of ESP test responses. Journal of Parapsychology, 31, 280-289.

STANFORD, R.G. (1970). Extrasensory effects upon memory. Journal of the American Society for Psychical Research, 64 (2), 161-185.

STANFORD, R.G. (1973). Extrasensory effects upon associative processes in a directed free response task. Journal of the American Society for Psychical Research, 67, 147-190.

STANFORD, R.G. (1974a). An experimentally testable model for spontaneous psi events I: Extrasensory events. Journal of the American Society for Psychical Research, 68 (1), 34-57.

STANFORD, R.G. (1974b). An experimentally testable model for spontaneous psi events II: Psychokinetic events. Journal of the American Society for Psychical Research, 68 (4), 321-356.

STANFORD, R.G. (1977). Conceptual frameworks of contemporary psi research. In B.B. Wolman (Ed.) Handbook of parapsychology, pp. 823-858. London: Van Nostrand Reinhold.

STANFORD, R.G. (1982). An experimentally testable model for spontaneous extrasensory events. In I. Grattan-Guiness (Ed.), Psychical research: A guide to its history, principles and practices (pp. 195-205). UK: Aquarian Press.

STANFORD, R.G. (1990). An experimentally testable model for spontaneous psi events: A review of related evidence and concepts from parapsychology and other sciences. In S. Krippner (Ed.) Advances in parapsychological research 6 (pp. 54-161). Jefferson, NC: McFarland.

STANFORD, R.G., & SCHROETER, W. (1978). Extrasensory effects upon associative processes in a directed free-response task: An attempted replication and extension. In W.G. Roll. (Ed.), Research in parapsychology 1977 (pp. 52-64). Metuchen, NJ: Scarecrow Press.

THALBOURNE, M.A. (1982). A glossary of terms used in parapsychology. London: William Heinemann.

THOULESS, R.H., & BRIER, R.M. (1970). The stacking effect and methods of correcting for it. Journal of Parapsychology, 34, 124128.

TOGLIA, M.P., & BATTIG, W.F. (1978). Handbook of semantic word norms. Potomac, MD: Lawrence Erlbaum.

TYRELL, G.N.M. (1946). The "modus operandi" of paranormal cognition. Proceedings of the Society for Psychical Research, 48, 65120.

VAN BUSSBACH, J.G. (1953). An investigation of extrasensory perception in school children. Journal of Parapsychology, 17, 210214.

VAN BUSSBACH, J.G. (1955). A further report on an investigation of ESP in school children. Journal of Parapsychology, 19, 73-81.

VAN BUSSBACH, J.G. (1956). An investigation of ESP between teacher and pupils in American schools. Journal of Parapsychology, 20, 7181.

VAN BUSSBACH, J.G. (1959). An investigation of ESP in first and second grades of Dutch schools. Journal of Parapsychology, 23, 227-237.

VAN BUSSBACH, J.G. (1961). An investigation of ESP in first and second graders in American schools. Journal of Parapsychology, 25, 161174.

WHITE, R.A. (1977). The influence of experimenter motivation, attitudes and methods of handling subjects on psi test results. In B.B. Wolman (Ed.), Handbook of parapsychology (pp. 273-301). New York: Van Nostrand Reinhold.

WILSON, S. (2002). Psi, perception without awareness and false recognition. Journal of Parapsychology, 66 (3), 271-289.

WISEMAN, R., & SCHLITZ, M. (1997). Experimenter effects and the remote detection of staring. Journal of Parapsychology, 61, 197-207.

STUART WILSON *, ROBERT L. MORRIS **, NIKO TILIOPOULOS *, AND ERIC PRONTO

* Department of Psychology School of Social Sciences, Media and Communication Queen Margaret University College Clerwood Terrace Edinburgh EH12 8TS, UK swilson@qmuc.ac.uk

ROBERT L. MORRIS **, NIKO TILIOPOULOS *,

** Deceased Koestler Parapsychology Unit, Department of Psychology University of Edinburgh

COPYRIGHT 2004 Parapsychology Press
COPYRIGHT 2005 Gale Group