better, more precise methodologies that would help rule out nonparanormal sources of above-chance results. Eight years later Bem and Honorton (1994) published a review of ganzfield studies not included in previous reviews that did not suffer from the type of methodological flaws noted by Hyman and Honorton. Bem and Honorton concluded that these new studies did, in fact, provide “replicable evidence for an anomalous process of information transfer” (p. 4). In other words, they provided reliable evidence for the reality of ESP.
But it turned out that there was a serious problem with the Bem and Honorton (1994) review. In 1999 Milton and Wiseman published a critque of that review and an analysis of additional new ganzfeld studies. In their review Bem and Honorton had counted the results of some studies as being statistically significant when they actually were not significant. This error led Bem and Honorton to conclude that the studies they reviewed had shown, overall, that ESP was operating in the ganzfeld situation. Milton and Wiseman then reviewed thirty ganzfeld studies that had been designed to meen the rigorous methodological standards set forth in Hyman and Honorton (1986); these studies showed no effect greater than chance.
One finding that is so common in science that it is almost a general principle is that as research on new phenomena that actually exist progresses from initial discovery to detailed analysis, it becomes easier and easier to obtain the phenomena. At first, researchers aren’t sure about just which variables affect the presence, absence, or strength of the phenomena. But as research progresses, these variables become clearer and clearer as more studies are published that describe under just what specific condition that particular effect can be demonstrated. Note how different this scenario is from that seen in the more than twenty-five-year history of the ganzfeld studies and, indeed, the more than one-hundred-year history of ESP studies in general. After all this time, there is no clear way to obtain results showing any psychic phenomenon reliably. By far the most reasonable conclusion is that such effects do not now and never have existed.
Another type of ESP study is not usually considered along with the ganzfeld studies, but because there are great similarities between them, these studies will be discussed here. These are studies of ESP during dreams. The procedure is like that used in the ganzfeld study, except the subject is asleep in a laboratory. The subject’s electroencephalogram (EEG) is monitored. When she is in a dream period (indicated by the presence of rapid eye movements [REM]), the agent is signaled to begin concentrating on a target object or picture. When the subject’s REM period ends, signaling an end of dreaming, she is awakened and reports any dreams. The content of the dreams is then compared to the object or picture the agent was “sending.” The basic procedure can easily be altered to study precognition or clairvoyance (see Child 1985, for a review). Such research is time consuming and expensive, like any sleep research. For one thing, a well-equipped sleep laboratory is needed. Thus, not a great deal of parapsychological research has used this paradigm.
Initial studies at the Maimonides Medical Center in Brooklyn, New York, in the 1960s and early 1970s seemed promising. Child (1985) correctly points out that these studies have sometimes been badly misdescribed by critics. Nonetheless, serious problems remain. Akers (1984) has noted that there was a violation “of the experimental protocol” that “leaves doubts as to the rigor with which the experiment was conducted” (p. 129). Further, as Child notes, three attempts at replication have failed (Belvedere and Foulkes 1971; Foulkes et al. 1972; Globus et al. 1968). In addition, some of the Maimonides studies failed to obtain significant results, although Child reports that the series, taken as a whole, still yields statistical significance. In view of these problems, the Maimonides results cannot be taken as providing convincing evidence of psi phenomena.
RANDOM EVENTS AND REACTION TIME STUDIES
In the 1980s a series of studies by physicist Helmut Schmidt attracted the attention of parapsychologists and the field’s critics. Schmidt uses a random number generator to cause one of several lights to turn on. In his precognition studies, subjects are to press a button to predict which light will turn on. After the button is pressed, the random number generator determines the light that will turn on. In clairvoyance studies, the light that will turn on is determined before the subject’s response. Schmidt’s published reports (for a brief review see Hansel 1980; Akers 1984; Rush 1982) claim that subjects are able to show better-than-chance responses in these situations.
Schmidt’s work does face some problems, as has been pointed out by critics (Hansel 1980; Hyman 1980– 81). For one thing, the details of the random number generator change frequently in Schmidt’s work, often from one experiment to the next. Thus, one cannot gather “cumulative experience with one particular generator to fully understand its peculiarities and… properly ‘debug’ it” (Hyman 1980–81, p. 37). Further, there are problems with the control trials that Schmidt uses. These are series of trials during which the random number generator is generating random numbers, but no subjects are making attempts to predict what its output will be. This is an absolutely necessary procedure but, as Hyman (1980–81) pointed out, the control trials that Schmidt used are hundreds of times longer than the actual experimental series. Thus, the various generators Schmidt used may show temporary deviations from randomness in the short run that are obscured by the very lengthy control runs. Thus, comparing short experimental runs to very lengthy control runs would spuriously give significant results. Subjects could become aware of these deviations and adjust their predictions accordingly. This would especially be a problem in the psychokinesis (PK) studies, where subjects are to attempt to influence the counter so that one of two events occurs more than 50 percent of the time. Here the crucial comparison is between the probability of the two events in the short experimental runs and vastly longer control runs. Any short-term deviations from randomness due to the generator would be likely to be interpreted as significant PK effects. Another problem with Schmidt’s work is that the subjects are left largely unobserved and unsupervised during the experiment. Randall (1975) has said of Schmidt’s studies that they “provide us with the final proof of the reality of ESP” (p. 131). But almost the exact words have been used many times in the last one hundred years to describe the latest surefire demonstration of the existence of psi, which promptly fell to pieces upon close examination. In view of the problems with Schmidt’s work, Randall is certainly premature in his evaluation.
Schmidt is not the first investigator to have used this sort of paradigm to investigate psi. As early as 1963 Smith, Dagle, Hill, and Mott-Smith (cited in Hansel 1980), working for the U.S. Air Force, tried a similar experiment. It was a failure. Parapsychologist Charles Tart (1976) used a random number generator to study the possibility of training people to use psi. Subjects were given feedback on whether or not their responses were correct following each trial. In standard learning theory, such feedback is extremely important and enhances learning greatly (Welford 1976). Positive results were initially found, as subjects came to be able to match their responses to the numbers generated by the machine. It turned out, however, that the sequence of targets generated by the random number generator was not random. This finding renders highly problematic the contention that the experiment demonstrated psi. Tart’s response (see Akers [1984] for a brief review of this controversy) to the discovery of nonrandomness was to suggest that it was partly due to PK. Thus, a serious procedural flaw in an experiment has itself been claimed as evidence for psi, in yet another example of the use of a nonfalsifiable hypothesis.
Some years ago several of my students and I used a more sensitive method to examine claims for psi. This method uses reaction time as the major dependent variable, or measure. To date, the vast majority of studies of psi have used accuracy as their dependent measure, as can be seen from the studies described above or in any review of modern psi research (e.g., Akers 1984; Morris 1978, 1982; Palmer 1978, 1982). In cognitive psychology, which is the experimental study of human learning, memory, and higher mental processes, accuracy is rarely used as the sole dependent measure. The dependent measure of choice is reaction time. This is because reaction time is a much more sensitive indicator of cognitive processes and processing than is accuracy. In other words, phenomena that are easily shown using a reaction time measure are not revealed by an accuracy measure.
An example will make the difference between the two measures clear. In a lexical decision task, subjects see, in each of several hundred trials, a string of letters. The string either is a real word—such as
