Feeling the Future: Experimental Evidence for Anomalous Retroactive Influences on Cognition and Affect (PDF)

Journal of Personality and Social Psychology © 2011 American Psychological Association 2011, Vol. 100, No. 3, 407– 425 0...

Journal of Personality and Social Psychology © 2011 American Psychological Association 2011, Vol. 100, No. 3, 407– 425 0022-3514/11/$12.00 DOI: 10.1037/a0021524 Feeling the Future: Experimental Evidence for Anomalous Retroactive Influences on Cognition and Affect Daryl J. Bem Cornell University The term psi denotes anomalous processes of information or energy transfer that are currently unex- plained in terms of known physical or biological mechanisms. Two variants of psi are precognition (conscious cognitive awareness) and premonition (affective apprehension) of a future event that could not otherwise be anticipated through any known inferential process. Precognition and premonition are themselves special cases of a more general phenomenon: the anomalous retroactive influence of some future event on an individual’s current responses, whether those responses are conscious or noncon- scious, cognitive or affective. This article reports 9 experiments, involving more than 1,000 participants, that test for retroactive influence by “time-reversing” well-established psychological effects so that the individual’s responses are obtained before the putatively causal stimulus events occur. Data are presented for 4 time-reversed effects: precognitive approach to erotic stimuli and precognitive avoidance of negative stimuli; retroactive priming; retroactive habituation; and retroactive facilitation of recall. The mean effect size (d) in psi performance across all 9 experiments was 0.22, and all but one of the experiments yielded statistically significant results. The individual-difference variable of stimulus seek- ing, a component of extraversion, was significantly correlated with psi performance in 5 of the experiments, with participants who scored above the midpoint on a scale of stimulus seeking achieving a mean effect size of 0.43. Skepticism about psi, issues of replication, and theories of psi are also discussed. Keywords: psi, parapsychology, ESP, precognition, retrocausation The term psi denotes anomalous processes of information or Precognition and premonition are themselves special cases of a energy transfer that are currently unexplained in terms of known more general phenomenon: the anomalous retroactive influence of physical or biological mechanisms. The term is purely descriptive; some future event on an individual’s current responses, whether it neither implies that such phenomena are paranormal nor con- those responses are conscious or nonconscious, cognitive or affec- notes anything about their underlying mechanisms. Alleged psi tive. This article reports nine experiments designed to test for such phenomena include telepathy, the apparent transfer of information retroactive influence by “time-reversing” several well-established from one person to another without the mediation of any known psychological effects, so that the individual’s responses are ob- channel of sensory communication; clairvoyance (sometimes tained before the putatively causal stimulus events occur. called remote viewing), the apparent perception of objects or Psi is a controversial subject, and most academic psychologists events that do not provide a stimulus to the known senses; psy- do not believe that psi phenomena are likely to exist. A survey of chokinesis, the apparent influence of thoughts or intentions on 1,100 college professors in the United States found that psychol- physical or biological processes; and precognition (conscious cog- ogists were much more skeptical about the existence of psi than nitive awareness) or premonition (affective apprehension) of a were their colleagues in the natural sciences, the other social future event that could not otherwise be anticipated through any sciences, or the humanities (Wagner & Monnet, 1979). In fact, known inferential process. 34% of the psychologists in the sample declared psi to be impos- sible, a view expressed by only 2% of all other respondents. Although our colleagues in other disciplines would probably agree This article was published Online First January 31, 2011. with the oft-quoted dictum that “extraordinary claims require I am grateful to the students who served as head research assistants and extraordinary evidence,” we psychologists are more likely to be laboratory coordinators for their enthusiasm and dedication to this contro- familiar with the methodological and statistical requirements for versial enterprise: Ben Edelman, Rebecca Epstein, Dan Fishman, Jamison sustaining such claims and aware of previous claims that failed Hahn, Eric Hoffman, Kelly Lin, Brianne Mintern, Brittany Terner, and either to meet those requirements or to survive the test of success- Jade Wu. I am also indebted to the 30 other students who served as friendly ful replication. Several other reasons for our greater skepticism are and reliable experimenters over the course of this research program. Dean discussed by Bem and Honorton (1994, pp. 4 –5). Radin, senior scientist at the Institute of Noetic Sciences (IONS), and There are two major challenges for psi researchers, one empir- David Sherman, professor of psychology at the University of California, Santa Barbara, provided valuable guidance in the preparation of this article. ical and one theoretical. The major empirical challenge, of course, Correspondence concerning this article should be addressed to Daryl J. is to provide well-controlled demonstrations of psi that can be Bem, Department of Psychology, Uris Hall, Cornell University, Ithaca, NY replicated by independent investigators. That is the major goal in 14853. E-mail: [email protected] the research program reported in this article. Accordingly, the 407 408 BEM experiments have been designed to be as simple and transparent as noise rather than visual images as the arousing stimuli (Spottis- possible, drawing participants from the general population, requir- woode & May, 2003). A review of presentiment experiments prior ing no instrumentation beyond a desktop computer, taking less to 2006 can be found in Radin (2006, pp. 161–180). Although than thirty minutes per session, and requiring statistical analyses there has not yet been a formal meta-analysis of presentiment no more complex than a t test across sessions or participants. studies, there have been 24 studies with human participants The major theoretical challenge for psi researchers is to provide through 2009, of which 19 were in the predicted direction and an explanatory theory for the alleged phenomena that is compat- about half were statistically significant. Two studies with animals ible with physical and biological principles. Although the current were both positive, one marginally and the other substantially so absence of an explanatory theory for psi is a legitimate rationale (D. I. Radin, personal communication, December 20, 2009). for imposing the “extraordinary” requirement on the evidence, it is Most of the experiments reported in this article are also part of not, I would argue, sufficient reason for rejecting all proffered this trend toward using subliminal stimulus presentations and evidence a priori. Historically, the discovery and scientific explo- indirect or implicit response measures. Each of them modified a ration of most phenomena have preceded explanatory theories, well-established psychological effect by reversing the usual se- often by decades or even centuries. The major focus of this article quence of events, so that the individual’s responses were obtained is empirical, but I return to a brief discussion of theory at the end. before rather than after the stimulus events occurred. Table 1 As noted above, the experiments in this article are concerned provides an overview of the effects and their corresponding time- with apparent retroactive influence, a generalized form of precog- reversed experiments. nition. Experimental tests of precognition have been reported for more than half a century. Most of the early experiments used Precognitive Approach and Avoidance forced-choice designs in which participants were explicitly chal- lenged to guess which one of several potential targets would be The presentiment studies provide evidence that our physiology randomly selected at a later time. Typical targets have been ESP can anticipate unpredictable erotic or negative stimuli before they card symbols, an array of colored lightbulbs, the faces of a die, or occur. Such anticipation would be evolutionarily advantageous for visual elements in a computer display. When a participant cor- reproduction and survival if the organism could act instrumentally rectly selects the actual target-to-be, it is designated as a hit, and to approach erotic stimuli and avoid negative stimuli. The two psi performance is typically expressed as the hit rate, the percent- experiments in this section were designed to test whether individ- age of hits over trials. uals can do so. A meta-analysis of all forced-choice precognition experiments appearing in English-language journals between 1935 and 1977 Experiment 1: Precognitive Detection of Erotic Stimuli was published by Honorton and Ferrari (1989). Their analysis included 309 experiments conducted by 62 different investigators As noted above, most of the earlier experiments in precognition and involving more than 50,000 participants. Honorton and Ferrari explicitly challenged participants to guess which one of several reported a small but consistent and highly significant hit rate stimuli would be randomly selected after they recorded their guess. (mean z ⫽ 0.69, combined z ⫽ 12.14, p ⫽ 6 ⫻ 10⫺27). They also In most of these experiments, participants were also given explicit concluded that this overall result was unlikely to be significantly trial-by-trial feedback on their performance. This first experiment inflated by the selective reporting of positive results (the so-called adopts this traditional protocol, using erotic pictures as explicit file-drawer effect): There would have to be 46 unreported studies reinforcement for correct “precognitive” guesses. averaging null results for every reported study in the meta-analysis to reduce the overall significance of the database to nonsignifi- Method cance. One hundred Cornell undergraduates, 50 women and 50 men, Just as research in cognitive social psychology has increasingly were recruited for this experiment through the Psychology Depart- pursued the study of cognitive and affective processes that are not accessible to conscious awareness and control (Bargh & Ferguson, 2000), research in psi has followed the same path, moving from explicit forced-choice guessing tasks to experiments using sublim- Table 1 inal stimuli and implicit, indirect, or physiological responses. The Overview of Psychological Effects and Their Corresponding trend is exemplified by several recent “presentiment” experiments, Time-Reversed Experiments pioneered by Radin (1997), in which physiological indices of participants’ emotional arousal were monitored as participants Standard psychological effect Experiments viewed a series of pictures on a computer screen. Most of the Approach/avoidance 1. Precognitive Detection of Erotic pictures were emotionally neutral, but a highly arousing negative Stimuli or erotic image was displayed on randomly selected trials. As 2. Precognitive Avoidance of Negative expected, strong emotional arousal occurred when these images Stimuli appeared on the screen, but the remarkable finding is that the Affective priming 3. Retroactive Priming I 4. Retroactive Priming II increased arousal was observed to occur a few seconds before the Habituation 5. Retroactive Habituation I picture appeared, before the computer had even selected the pic- 6. Retroactive Habituation II ture to be displayed. The presentiment effect has also been dem- 7. Retroactive Induction of Boredom onstrated in an fMRI experiment that monitored brain activity Facilitation of recall 8. Retroactive Facilitation of Recall I 9. Retroactive Facilitation of Recall II (Bierman & Scholte, 2002) and in experiments using bursts of  FEELING THE FUTURE 409 ment’s automated online sign-up system.1 They either received participant’s arousal level to “settle down” between critical trials. one point of experimental credit in a psychology course offering This requires including many trials that do not contribute directly that option or were paid $5 for their participation. Both the to the effect being tested. recruiting announcement and the introductory explanation given to In our first retroactive experiment (Experiment 5, described participants upon entering the laboratory informed them that below), women showed psi effects to highly arousing stimuli but men did not. Because this appeared to have arisen from men’s this is an experiment that tests for ESP. It takes about 20 minutes and lower arousal to such stimuli, we introduced different erotic and is run completely by computer. First you will answer a couple of brief negative pictures for men and women in subsequent studies, in- questions. Then, on each trial of the experiment, pictures of two cluding this one, using stronger and more explicit images from curtains will appear on the screen side by side. One of them has a picture behind it; the other has a blank wall behind it. Your task is to Internet sites for the men. We also provided two additional sets of click on the curtain that you feel has the picture behind it. The curtain erotic pictures so that men could choose the option of seeing will then open, permitting you to see if you selected the correct male–male erotic images and women could choose the option of curtain. There will be 36 trials in all. seeing female–female erotic images.2 Several of the pictures contain explicit erotic images (e.g., couples From the participants’ point of view, this procedure appears to engaged in nonviolent but explicit consensual sexual acts). If you test for clairvoyance. That is, participants were told that a picture object to seeing such images, you should not participate in this was hidden behind one of the curtains, and their challenge was to experiment. guess correctly which curtain concealed the picture. In fact, how- ever, neither the picture itself nor its left/right position was deter- The participant then signed a consent form and was seated in mined until after the participant recorded his or her guess, making front of the computer. After responding to two individual- the procedure a test of detecting a future event (i.e., a test of difference items (discussed below), the participant had a 3-min precognition). relaxation period during which the screen displayed a slowly moving Hubble photograph of the starry sky while peaceful new- Results and Discussion age music played through stereo speakers. The 36 trials began immediately after the relaxation period. Across all 100 sessions, participants correctly identified the Stimuli. Most of the pictures used in this experiment were future position of the erotic pictures significantly more frequently selected from the International Affective Picture System (IAPS; than the 50% hit rate expected by chance: 53.1%, t(99) ⫽ 2.51, Lang & Greenwald, 1993), a set of 820 digitized photographs that p ⫽.01, d ⫽ 0.25.3 In contrast, their hit rate on the nonerotic have been rated on 9-point scales for valence and arousal by both pictures did not differ significantly from chance: 49.8%, t(99) ⫽ male and female raters. This is the same source of pictures used in ⫺0.15, p ⫽.56. This was true across all types of nonerotic most presentiment studies. Each session of the experiment in- pictures: neutral pictures, 49.6%; negative pictures, 51.3%; posi- cluded both erotic and nonerotic pictures randomly intermixed, tive pictures, 49.4%; and romantic but nonerotic pictures, 50.2%. and the main psi hypothesis was that participants would be able to (All t values ⬍ 1.) The difference between erotic and nonerotic identify the position of the hidden erotic picture significantly more trials was itself significant, tdiff(99) ⫽ 1.85, p ⫽.031, d ⫽ 0.19. often than chance (50%). Because erotic and nonerotic trials were randomly interspersed in The hit rate on erotic trials can also be compared with the hit the trial sequence, this significant difference also serves to rule out rates on the nonerotic trials to test whether there is something the possibility that the significant hit rate on erotic pictures was an unique about erotic content in addition to its positive valence artifact of inadequate randomization of their left/right positions. and high arousal value. For this purpose, 40 of the sessions Because there are distribution assumptions underlying t tests, comprised 12 trials using erotic pictures, 12 trials using negative the significance levels of most of the positive psi results reported pictures, and 12 trials using neutral pictures. The sequencing of the in this article were also calculated with nonparametric tests. In this pictures and their left/right positions were randomly determined by experiment, the hit rates on erotic trials were also analyzed with a the programming language’s internal random function. The re- binomial test on the overall proportion of hits across all trials and maining 60 sessions comprised 18 trials using erotic pictures and sessions, tested against a null of.5. This is analogous to analyzing 18 trials using nonerotic positive pictures with both high and low a set of coin flips without regard to who or how many are doing the arousal ratings. These included eight pictures featuring couples in flipping. It is legitimate here because the target was randomly romantic but nonerotic situations (e.g., a romantic kiss, a bride and selected on each trial and hence the trials were statistically inde- groom at their wedding). The sequencing of the pictures on these pendent, even within a single session. Across all 100 sessions, the trials was randomly determined by a randomizing algorithm de- vised by Marsaglia (1997), and their left/right target positions were 1 I set 100 as the minimum number of participants/sessions for each of the determined by an Araneus Alea I hardware-based random number experiments reported in this article because most effect sizes (d) reported in the generator. (The rationale for using different randomizing proce- psi literature range between 0.2 and 0.3. If d ⫽ 0.25 and N ⫽ 100, the power dures is discussed in detail below.) to detect an effect significant at.05 by a one-tail, one-sample t test is.80 Although it is always desirable to have as many trials as possible (Cohen, 1988). in an experiment, there are practical constraints limiting the num- 2 In describing the experiments throughout this article, I have used the ber of critical trials that can be included in this and several others plural pronouns “we” and “our” to refer collectively to myself and my experiments reported in this article. In particular, on all the exper- research team. iments using highly arousing erotic or negative stimuli, a relatively 3 Unless otherwise indicated, all significance levels reported in this large number of nonarousing trials must be included to permit the article are based on one-tailed tests and d is used as the index of effect size. 410 BEM 53.1% hit rate was also significant by a binomial test (z ⫽ 2.30, matical algorithm to generate each subsequent number from the p ⫽.011). previous number, and the sequence of numbers is random only in Individual differences. There were no significant sex differ- the sense that it satisfies (or should satisfy) certain mathematical ences in the present experiment. Over the years, however, the trait tests of randomness. It is not random in the sense of being of extraversion has been frequently reported as a correlate of psi, indeterminate because once the initial starting number (the seed) is with extraverts achieving higher psi scores than introverts. A set, all future numbers in the sequence are fully determined. meta-analysis of 60 independent experiments published between In contrast, a hardware-based or “true” RNG is based on a 1945 and 1983, involving several kinds of psi tasks, revealed a physical process, such as radioactive decay or diode noise, and the small but reliable correlation between extraversion and psi perfor- sequence of numbers is indeterminate in the quantum mechanical mance (r ⫽.09, z ⫽ 4.63, p ⫽.000004; Honorton, Ferrari, & Bem, sense. This does not in itself guarantee that the resulting sequence 1992). The correlation was observed again in a later set of telep- of numbers can pass all the mathematical tests of randomness, athy studies conducted in Honorton’s own laboratory, r ⫽.18, however; some hardware-based RNGs also fail one or more of the t(216) ⫽ 2.67, p ⫽.004 (Bem & Honorton, 1994). tests in the diehard battery (L’Ecuyer, 2001). Both Marsaglia’s The component of extraversion that underlies this correlation own PRNG algorithm and the true hardware-based Araneus Alea appears to be the extravert’s susceptibility to boredom and a I RNG used in our experiments pass all his diehard tests. tendency to seek out stimulation. Eysenck (1966) attributed the Note that a random number table is actually a PRNG, even if the positive correlation between extraversion and psi to the fact that sequence of numbers was originally generated by a true RNG. extraverts “are more susceptible to monotony... [and] respond Once the table is printed or stored electronically and an entry point more favourably to novel stimuli” (p. 59). Sensation seeking is one into the table is chosen, the resulting sequence is fully determined, of the six facets of extraversion on the Revised NEO Personality with the entry point being equivalent to the seed number of a Inventory (Costa & McCrae, 1992), and Zuckerman’s Sensation computer-based PRNG. Seeking Scale (1974), which contains a subscale of Boredom At the conceptual level, the choice of a PRNG or a hardware- Susceptibility, is significantly correlated with overall extraversion based RNG bears on the interpretation of positive findings. In the (r ⫽.47, p ⬍.01; Farley & Farley, 1967). present context, it bears on my claim that the experiments reported To assess stimulus seeking as a correlate of psi performance in in this article provide evidence for precognition or retroactive our experiments, I constructed a scale comprising the following influence. In the experiment just reported, for example, there are two statements: “I am easily bored” and “I often enjoy seeing several possible interpretations of the significant correspondence movies I’ve seen before” (reverse scored). Responses were re- between the participants’ left/right responses and the computer’s corded on 5-point scales that ranged from Very Untrue to Very left/right placements of the erotic target pictures: True and averaged into a single score ranging from 1 to 5. In the present experiment, the correlation between stimulus 1. Precognition or retroactive influence: The participant is, seeking and psi performance was.18 ( p ⫽.035). This significant in fact, accessing information yet to be determined in the correlation is reflected in the enhanced psi scores of those scoring future, implying that the direction of the causal arrow has above the midpoint on the 5-point stimulus-seeking scale: They been reversed. correctly identified the future position of the picture on 57.6% of the erotic trials, t(41) ⫽ 4.57, p ⫽.00002, d ⫽ 0.71, exact 2. Clairvoyance/remote viewing: The participant is access- binomial p ⫽.00008. The difference between their erotic and ing already determined information in real time, infor- nonerotic hit rates was itself significant, tdiff(41) ⫽ 3.23, p ⫽.001, mation that is stored in the computer. d ⫽ 0.50, with 71% of participants achieving higher hit rates on erotic trials than on nonerotic trials (exact binomial p ⫽.003). 3. Psychokinesis: The participant is actually influencing the Their psi scores on nonerotic trials did not exceed chance, 49.9%, RNG’s placements of the targets. t(41) ⫽ ⫺0.08, p ⫽.53. Finally, participants low in stimulus seeking did not score significantly above chance on either erotic or 4. Artifactual correlation: The output from the RNG is nonerotic trials, 49.9%, t(57) ⫽ ⫺0.06 and 49.9%, t(57) ⫽ ⫺0.13, inadequately randomized, containing patterns that fortu- respectively. itously match participants’ response biases. This pro- But is it precognition? The role of random number genera- duces a spurious correlation between the participant’s tors. For most psychological experiments, a random number guesses and the computer’s placements of the target table or the random function built into most programming lan- picture. guages provides an adequate tool for randomly assigning partici- pants to conditions or sequencing stimulus presentations. For both Consider, first, the clairvoyance interpretation. If an algorithm- methodological and conceptual reasons, however, psi researchers based PRNG is used for determining the successive left/right have paid much closer attention to issues of randomization. positions of the target pictures, then the computer already “knows” At the methodological level, the problem is that the random the upcoming random number before the participant makes his or functions included in most computer languages are not very good her response; in fact, once the initial seed number is generated, the in that they fail one or more of the mathematical tests used to computer implicitly knows the entire sequence of left/right posi- assess the randomness of a sequence of numbers (L’Ecuyer, 2001), tions. As a result, this information is potentially available to the such as Marsaglia’s (1995) rigorous Diehard Battery of Tests of participant through real-time clairvoyance, permitting us to reject Randomness. Such random functions are sometimes called pseudo the more extraordinary claim that the direction of the causal arrow random number generators (PRNGs) because they use a mathe- has actually been reversed. In contrast, if a true hardware-based  FEELING THE FUTURE 411 RNG is used for determining the left/right positions, the next Experiment 2: Precognitive Avoidance of Negative number in the sequence is indeterminate until it is actually gener- Stimuli ated by the quantum physical process embedded in the RNG, thereby ruling out the clairvoyance alternative. This argues for Method using a true RNG to demonstrate precognition or retroactive in- fluence. One hundred fifty Cornell undergraduates, 107 women and 43 But alas, the use of a true RNG opens up the door to the men, were recruited for this experiment through the Psychology psychokinesis interpretation: The participant might be influencing Department’s automated online sign-up system. Both the recruit- the placement of the upcoming target rather than perceiving it, a ing announcement and the opening instructions given to partici- possibility supported by a body of empirical evidence testing pants upon entering the laboratory, informed them that psychokinesis with true RNGs (Radin, 2006, pp. 154 –160). Like this is an experiment that tests for ESP (Extrasensory Perception). The the clairvoyance interpretation, the psychokinesis interpretation experiment is run entirely by computer and takes about 15 minutes.... permits us to reject the claim that the direction of the causal arrow On each trial of the experiment you will be shown a picture and its mirror has been reversed. Ironically, the psychokinesis alternative can be image side by side and asked to indicate which image you like better. The ruled out by using a PRNG, which is immune to psychokinesis computer will then flash a masked picture on the screen. The way in because the sequence of numbers is fully determined and can even which this procedure tests for ESP will be explained to you at the end of the session. be checked after the fact to confirm that its algorithm has not been perturbed. Over the course of our research program—and within Note that the participant’s task in this experiment was simply to the experiment just reported—we have obtained positive results express a preference between two closely matched pictures on using both PRNGs and a true RNG, arguably leaving precognition/ each trial; unlike in traditional precognition experiments (and reversed causality the only nonartifactual interpretation that can Experiment 1 reported above), the participant was not faced with account for all the positive results. an explicit psi challenge. This still leaves open the artifactual alternative that the output As in Experiment 1, the participant was seated in front of the from the RNG is producing inadequately randomized sequences computer and asked to respond to the two items on the stimulus containing patterns that fortuitously match participants’ response seeking scale. This was followed by the same 3-min relaxation biases. In the present experiment, this possibility was ruled out by period. Then, on each of 36 trials, the participant was shown a the twin findings that erotic targets were detected significantly low-arousal, affectively neutral picture and its mirror image side more frequently than randomly interspersed nonerotic targets and by side and asked to press one of two keys on the keyboard to that the nonerotic targets themselves were not detected signifi- indicate which neutral picture he or she liked better. Using the cantly more frequently than chance. Araneus Alea I hardware-based RNG, the computer then randomly Nevertheless, for some of the other experiments reported in this designated one of the two pictures to be the target. Note that the article, it would be useful to have more general assurance that there computer did not determine which of the two neutral pictures are not patterns in the left/right placements of the targets that might would be the target until after the participant had registered his or correlate with response biases of participants. For this purpose, her preference. Whenever the participant had indicated a prefer- Lise Wallach, professor of psychology at Duke University, sug- ence for the target-to-be, the computer flashed a positively va- gested that I run a virtual control experiment using random inputs lenced picture on the screen subliminally three times. Whenever in place of human participants (personal communication, October the participant had indicated a preference for the nontarget, the 10, 2009). In particular, if the human participant is replaced by the computer subliminally flashed a highly arousing, negatively va- same PRNG or RNG that selects the left/right target positions, this lenced picture. Pictures were again selected primarily from the maximizes the possibility that any nonrandom patterns in the IAPS set. The flashed pictures were exposed for 33 ms, followed imme- sequence of left/right target positions will be mirrored by similar diately by a masking stimulus for 167 ms. Time between flashes patterns in the left/right responses of the virtual participant (the was 500 ms. A Hubble photograph of the starry sky appeared on RNG itself), thereby producing an artifactual psi-like result. A null the screen for 3,000 ms before the onset of the next trial. A hit was result implies that no such patterns were present. defined as preferring the target-to-be, the picture that avoided the Accordingly, I ran three experiments that simulated the exper- subliminally exposed negative picture. Because participants chose iment just reported, one using the internal PRNG of the program- between two images on each trial, the psi hypothesis was that they ming language (REALbasic), one using Marsaglia’s PRNG algo- would prefer the target to the nontarget on significantly more than rithm, and one using the Araneus Alea I hardware-based RNG. 50% of the trials. Each experiment comprised 100 sessions of 36 trials per session, The pairs of neutral pictures were presented in a fixed order for in which the same PRNG or RNG provided both the left/right all sessions, and the RNG randomly determined the left/right response of a virtual participant and the subsequent left/right position of the two images. For the first 100 sessions, the flashed position of the target. These control experiments all yielded null positive and negative pictures were independently selected and results: The hit rates were 49.5%, 50.4%, and 49.5% for the sequenced randomly. For the subsequent 50 sessions, the negative internal PRNG, the Marsaglia PRNG, and the Araneus Alea I pictures were put into a fixed sequence, ranging from those that RNG, respectively (all t values ⬍ 1). The corresponding correla- had been successfully avoided most frequently during the first 100 tions (phi) between the virtual participants’ input responses and the sessions to those that had been avoided least frequently. If the RNG’s target positions were ⫺.00,.00, and ⫺.01, respectively. participant selected the target, the positive picture was flashed 412 BEM subliminally as before, but the unexposed negative picture was The fourth analysis controlled for any potential overall system- retained for the next trial; if the participant selected the nontarget, atic bias in the RNG by computing an empirical null baseline for the negative picture was flashed and the next positive and negative each participant rather than assuming it to be 50%. The output of pictures in the queue were used for the next trial. In other words, a session in this experiment can be conceptualized as a string of 36 no picture was exposed more than once, but a successfully avoided left/right decisions made by the participant compared with the negative picture was retained over trials until it was eventually corresponding string of 36 left/right target positions generated by invoked by the participant and exposed subliminally. The working the computer. The number of matches between the two strings is hypothesis behind this variation in the study was that the psi effect the number of hits obtained. An empirical baseline was computed might be stronger if the most successfully avoided negative stimuli by running each of the 150 participants through 149 virtual control were used repeatedly until they were eventually invoked. sessions by comparing his or her decision string with the computer-generated target strings from each of the other 149 Results and Discussion (nonself) sessions and calculating the associated hit rates. The mean of those hit rates becomes the empirical baseline for that The results from the last 50 sessions did not differ significantly participant. Because any systematic bias in the RNG will affect the from those obtained on the first 100 sessions, so all 150 sessions participant’s empirical baseline the same way it affects the partic- were combined for analysis. ipant’s hit rate, the bias will be subtracted out of the analysis. For Unlike in Experiment 1, which had nonerotic trials randomly descriptive purposes, the difference between the hit rate and the interspersed among the critical erotic trials, in this experiment empirical baseline is added to 50% to yield a number that can be every trial constituted a critical trial, making it necessary to con- directly compared with the more conventional hit rate where 50% firm that the successive left/right positions of the target were is the null. adequately randomized and did not contain patterns that might As Table 2 reveals, the four analyses yielded comparable re- match participants’ response biases. For this reason, I analyzed the sults, showing significant psi performance across the 150 sessions. data in four ways. The first two analyses were the familiar ones Recall, too, that the RNG used in this experiment was tested in the used in Experiment 1: a one-sample t test across participants’ hit simulation, described above in the discussion of Experiment 1, and rates, tested against a null hit rate of 50%, and a nonparametric was shown to be free of nonrandom patterns that might correlate binomial test on the proportion of hits across all trials and sessions. with participants’ responses biases. The third analysis used an alternative index of psi performance, Stimulus seeking. In the present experiment, the correlation one that corrects for unequal frequencies of left/right target posi- between stimulus seeking and psi performance was.17 ( p ⫽.02). tions within each session. The output of a session can be repre- Table 3 reveals that the subsample of high stimulus seekers sented by the 2 ⫻ 2 table shown in Figure 1. achieved an effect size more than twice as large as that of the full The conventional hit rate is defined as hits/trials, or (A ⫹ D)/(A ⫹ sample. In contrast, the hit rate of low stimulus seekers did not B ⫹ C ⫹ D). But if there is a bias in the distribution of left/right target depart significantly from chance: 50.7%–50.8%, t ⬍ 1, p ⬎.18, positions, if (A ⫹ C) ⫽ (B ⫹ D), the hit rate will not necessarily and d ⬍ 0.10 in each of the four analyses. reflect actual psi performance. For example, if the RNG places a majority of the targets on the left and the participant also has a bias Retroactive Priming favoring pictures on the left, the hit rate will be artifactually inflated. (Experimentally forcing the left/right positions of the target to be equal is not a legitimate corrective for this problem, Experiment 3: Retroactive Priming I because it destroys the statistical independence of the trials by In recent years, priming experiments have become a staple of using a sampling-without-replacement or “closed deck” proce- cognitive and cognitive social psychology (Bargh & Ferguson, dure.) 2000; Fazio, 2001; Klauer & Musch, 2003). In a typical affective An index that automatically corrects for any bias in target position priming experiment, participants are asked to judge as quickly as is the phi coefficient, the correlation between the participant’s left/ they can whether a picture is pleasant or unpleasant, and their right preferences and the RNG’s left/right target placements. Phi response time is measured. Just before the picture appears, a is computed by the formula (AD ⫺ BC)/[(A ⫹ B)(C ⫹ D)(A ⫹ positive or negative word (e.g., beautiful, ugly) is flashed briefly C)(B ⫹ D)]1/2 and ranges from ⫺1 to 1 (Siegel & Castellan, 1988). on the screen; this word is called the prime. Individuals typically Phi was computed for each participant, and a t test on phi across respond more quickly when the valences of the prime and the participants was calculated. The psi hypothesis is that phi will be picture are congruent (both are positive or both are negative) than significantly greater than 0. when they are incongruent. In our retroactive version of the pro- cedure, the prime appeared after rather than before participants made their judgments of the pictures. Target is on Target is on Because slower responding on congruent trials than on incon- the Left the Right gruent trials— called a contrast effect— has also been observed in some priming experiments (Hermans, Spruyt, De Houwer, & Participant Prefers Left Picture A (Hit) B (Miss) Eelen, 2003; Klauer, Teige-Mocigemba, & Spruyt, 2009), we also Participant Prefers Right Picture C (Miss) D (Hit) ran a standard nonretroactive priming procedure in each session to (A + C) (B + D) ensure that our protocol would produce the usual (noncontrast) priming effect (see also de Boer & Bierman, 2006). Because this Figure 1. Output of a session represented as a 2 ⫻ 2 table. turned out to be the case, the psi hypothesis was that the retroactive  FEELING THE FUTURE 413 Table 2 Four Analyses of Precognitive Avoidance of Negative Stimuli Hit rate % across participants Binomial test across trials Phi coefficient across participants Hit rate % using an empirical baseline 51.7% 2,790/5,400 ⫽ 51.7%.034 51.7% t(149) ⫽ 2.39 z ⫽ 2.44 t(149) ⫽ 2.46 t(149) ⫽ 2.37 p ⫽.009 p ⫽.007 p ⫽.008 p ⫽.010 d ⫽ 0.20 r ⫽.20 d ⫽ 0.20 d ⫽ 0.19 procedure would also produce faster responding on congruent Figure 2 displays the time sequence of events for the forward trials than on incongruent trials. priming and retroactive priming trials, respectively. In both pro- cedures, there was a 2,000-ms interval between trials during which Method a Hubble photograph of the starry sky appeared on the screen. One hundred Cornell undergraduates, 69 women and 31 men, Results and Discussion participated in a 15- to 20-min experiment. They were shown a picture on each of 64 trials and were asked to press one of two keys Several methods for analyzing response-time data from priming on the keyboard as quickly as they could to indicate whether the experiments have evolved over the years (Ratcliff, 1993). First, picture was pleasant or unpleasant. The participant’s response time trials on which a participant makes an error in judging the picture in making this judgment was the dependent variable, and the to be pleasant or unpleasant are excluded from the analysis. In the difference in mean response times between incongruent and con- present experiment, the median number of errors was three out of gruent trials is the index of a priming effect, with positive differ- 64 trials, and the data from three participants were discarded ences denoting faster responding on congruent trials. because they made errors on 16 (25%) or more of the trials, The first 32 trials constituted the retroactive priming procedure, reducing the number of participants to 97. Second, because and participants were told that a word would be flashed on the response-time data are positively skewed, each response time (RT) screen just after they made their judgment of the picture. The is usually transformed prior to analysis using either an inverse remaining 32 trials constituted the standard forward priming pro- transformation (1/RT) or a log transformation (log RT). Finally, cedure, and participants were told that “from this point on, the trials yielding very short or very long response times are consid- flashed word will appear before rather than after you have made ered to be spurious outliers and are excluded from the analysis. your response.” Prior to beginning the actual experimental proce- Ratcliff (1993) suggested using more than one cutoff criterion to dure, participants responded to the two items on the stimulus ensure “that an effect is significant over some range of nonextreme seeking scale and then had the same 3-min relaxation period cutoffs” (p. 519). Accordingly, Table 4 presents four analyses, described in the previous experiments. using both data transformations and two different cutoff criteria for The pictures were again drawn from the IAPS set and were long response times, 1,500 ms and 2,500 ms. The first criterion randomly assigned to the forward and retroactive sections of the excludes 3.1% of the trials; the second excludes 0.5% of the trials. protocol, with the restriction that an equal number of positive and As shown in the table, the standard forward priming procedure negative pictures appear in each section. The same 16 positive and produced the usual result. For example, with a 1,500-ms cutoff 16 negative prime words appeared in both sections, and a prime criterion and the inverse transformation, participants were 23.6 ms was randomly selected on each trial before the picture was pre- faster on congruent trials than on incongruent trials, t(96) ⫽ 4.91, sented (in the forward priming procedure) or after the participant p ⬍.00001, d ⫽ 0.45. The retroactive procedure also yielded the had responded to the picture (in the retroactive priming procedure). predicted psi effect: With a 1,500-ms cutoff criterion and the As a result, congruent trials and incongruent trials were randomly inverse transformation, participants were 15.0 ms faster on con- sequenced and did not necessarily occur in equal numbers. This gruent trials than on incongruent trials, t(96) ⫽ 2.55, p ⫽.006, d ⫽ made it virtually impossible for participants to anticipate the type 0.25. The results were consistent across the range defined by the of trial coming up by knowing the types of trials that had already two cutoff criteria and under both data transformations. occurred. In this experiment, randomizing was implemented by To provide an analysis that avoids distribution assumptions, Marsaglia’s PRNG algorithm. Table 4 also displays the percentage of participants who had Table 3 Four Analyses of Precognitive Avoidance of Negative Stimuli for Participants High in Stimulus Seeking Hit rate % across participants Binomial test across trials Phi coefficient across participants Hit rate % using an empirical baseline 53.5% 963/1,800 ⫽ 53.5%.079 53.6% t(49) ⫽ 3.07 z ⫽ 2.95 t(49) ⫽ 3.57 t(49) ⫽ 2.99 p ⫽.002 p ⫽.002 p ⫽.0004 p ⫽.002 d ⫽ 0.43 r ⫽.42 d ⫽ 0.50 d ⫽ 0.39 414 BEM Forward Priming Trial a picture of a menacing pit bull was paired with the positive prime of friendly and the negative prime of threatening. The computer Stimulus Fixation Prime Blank Picture Hubble Point Photograph then randomly selected either the positive or the negative prime Time (ms) 1000 150 150 Response Time 2000 before the picture was presented (in the forward priming proce- dure) or after the participant had responded to the picture (in the retroactive priming procedure). In contrast to the randomizing Retroactive Priming Trial procedure used in the previous priming experiment, this procedure provides a genuine sampling-with-replacement or “open deck” Stimulus Fixation Picture Blank Prime Blank Hubble Point Photograph procedure for determining whether a trial will be congruent or incongruent. As a result, there is no (non-psi) way for a participant Time (ms) 1000 Response 300 500 1000 2000 Time to anticipate the kind of trial coming up next. All randomizing was again implemented by Marsaglia’s PRNG algorithm. In addition, Figure 2. Timing sequences for events in the retroactive priming proce- the duration of the fixation point was increased from 1,000 ms to dure. 1,500 ms, and the time between trials (during which the Hubble photograph appeared on the screen) was decreased from 2,000 ms to 1,500 ms. positive priming scores in each condition, evaluated by an exact binomial test. For example, with a 1,500-ms outlier cutoff crite- rion, 64.9% of participants produced a positive forward priming Results and Discussion effect ( p ⫽.002), and 60.8% produced a positive retroactive priming effect ( p ⫽.021). There was no significant correlation The data from one participant were discarded because he made between stimulus seeking and either priming effect (r ⫽ ⫺.02 and errors on more than 16 (25%) of the trials, reducing the number of ⫺.05 for the forward and retroactive priming effects, respectively). participants to 99. As seen in Table 5, the results were virtually identical to those obtained in the original experiment. Once again, Experiment 4: Retroactive Priming II the standard forward priming procedure produced the usual result. For example, with a 1,500-ms cutoff criterion and the inverse (1/RT) transformation, participants were 27.4 ms faster on con- Method gruent trials than on incongruent trials, t(98) ⫽ 4.85, p ⬍.00001, The experiment just described was replicated on an additional d ⫽ 0.44. The retroactive procedure also yielded the predicted psi 100 participants (57 female and 43 male Cornell undergraduates) effect again: With a 1,500-ms cutoff criterion and the inverse with one major change and two minor timing changes. In the transformation, participants were 16.5 ms faster on congruent trials original study, the prime paired with each picture was randomly than on incongruent trials, t(98) ⫽ 2.03, p ⫽.023, d ⫽ 0.20. As in selected on each trial from the list of unused positive and negative Experiment 3, the results were consistent across the different primes. In the current replication, one fixed positive prime and analyses, and the nonparametric exact binomial analyses con- one fixed negative prime were assigned to each picture prior to the firmed both the forward and retroactive priming effects. There was experiment. These were selected to be semantically relevant to the again no correlation between stimulus seeking and either priming picture; for example, a picture of a basket of fruit was paired with effect (r ⫽ ⫺.06 and ⫺.07 for the forward and retroactive priming the positive prime of luscious and the negative prime of bitter, and effects, respectively). Table 4 Priming as a Function of Two Outlier Cutoff Criteria and Two Data Transformations Outlier cutoff criteriona ⬎1,500 ms (3.1% of trials excluded)a ⬎2,500 ms (0.5% of trials excluded)a Transformation 1/RT log RT 1/RT log RT b Forward priming (ms) 23.6 20.8 t(96) 4.91 4.47 4.71 4.00 p ⬍.00001.00001 ⬍.00001.00006 d 0.45 0.42 0.43 0.38 % with priming ⬎ 0 64.9 63.9 p (exact binomial).002.004 Retroactive priming (ms)b 15.0 15.3 t(96) 2.55 2.49 2.57 2.42 p.006.007.006.009 d 0.25 0.25 0.25 0.24 % with priming ⬎ 0 60.8 56.7 p (exact binomial).021.111 Note. RT ⫽ response time. a Also excluded are four trials (0.06%) with response times ⬍250 ms. b Incongruent trial RTs minus congruent trials RTs. Calculated prior to applying transformations. Transformations applied only for tests of statistical significance and effect sizes.  FEELING THE FUTURE 415 Table 5 Priming as a Function of Two Outlier Cutoff Criteria and Two Data Transformations Outlier cutoff criterion ⬎1,500 ms (4.7% of trials excluded)a ⬎2,500 ms (0.7% of trials excluded)a Transformation 1/RT log RT 1/RT log RT Forward priming (ms)b 27.4 31.4 t(98) 4.85 4.28 4.71 3.98 p ⬍.00001.00002 ⬍.00001.00007 d 0.44 0.40 0.43 0.37 % with priming ⬎ 0 59.6 59.6 p (exact binomial).035.035 Retroactive priming (ms)b 16.5 23.9 t(98) 2.03 2.23 2.25 2.43 p.023.014.013.008 d 0.20 0.22 0.22 0.24 % with priming ⬎ 0 58.6 61.6 p (exact binomial).054.013 Note. RT ⫽ response time. a Also excluded are seven trials (0.11%) with response times ⬍250 ms. b Incongruent trial RTs minus congruent trial RTs. Calculated prior to applying transformations. Transformations applied only for tests of statistical significance and effect sizes. Retroactive Habituation and Induction of Boredom exposure effect, with subliminal exposures being the most effec- tive. This is interpreted as showing that the effect works at an When individuals are initially exposed to a startling or emotion- unconscious level and that conscious cognitive processes actually ally arousing stimulus, they typically have a strong physiological interfere with it. As noted above, subliminal exposures were used response to it. Upon repeated exposures, the arousal diminishes, a in the Dijksterhuis–Smith habituation experiment. process known as habituation. If the stimulus is initially very The optimal number of exposures is more difficult to specify unpleasant (e.g., frightening or disgusting), the stimulus becomes because it depends upon several variables, including the complex- more neutral or less negatively arousing; if the stimulus is initially ity of the stimuli and whether different kinds of stimuli are used in very pleasant, it becomes more neutral or less positively arousing. the same session (Bornstein, 1989; Bornstein, Kale, & Cornell, This was demonstrated in an experiment in which participants who 1990). Moreover, the function relating liking to the number of had been subliminally exposed to extremely negative and ex- exposures has an inverted-U shape: As the number of exposures tremely positive words subsequently rated those words as less increases in a mere exposure experiment, liking increases, pla- extreme than matched control words to which they had not been teaus, and eventually begins to decline. To account for this, Born- exposed: Negative words were rated less negatively and positive stein proposed a two-process model in which boredom increas- words were rated less positively than the control words (Dijkster- ingly competes with habituation as the number of exposures huis & Smith, 2002). increases. Because boredom causes a stimulus to be less liked, the As Dijksterhuis and Smith noted, the procedure they used in liking curve begins to level off and then turn downward as bore- their experiment is similar to that used to demonstrate the well- dom overtakes habituation. As a result, it is not possible to specify known mere exposure effect: Across a wide range of contexts, the a priori how many exposures would be optimal in any particular more frequently humans or other animals are exposed to a partic- experiment. In this experiment, we varied the number of exposures ular stimulus, the more they come to like it. This effect has been across sessions.4 known for over a century, but it was the publication of Zajonc’s (1968) monograph, “Attitudinal Effects of Mere Exposure,” that Experiment 5: Retroactive Habituation I spurred its intensive empirical investigation. In 1989, Bornstein was able to publish a meta-analysis of 208 mere exposure exper- One hundred Cornell undergraduates, 63 women and 37 men, iments. He reported a mean effect size (r) of.26, with a combined were recruited through the Psychology Department’s automated z of 20.80 ( p ⫽ 2.2 ⫻ 10⫺96). Curiously, most mere exposure online sign-up system to serve as participants in a “20 –25 minute experiments use low-arousal, affectively neutral stimuli such as study of visual imagery that tests for ESP.” They either received nonsense words or polygons; none of the studies reviewed used one point of experimental credit in a psychology course offering stimuli that were highly arousing or strongly valenced, either that option or were paid $5 for their participation. negatively or positively. As a result, the complementary hypoth- esis that highly arousing positive stimuli might become less liked 4 This experiment was our first psi study and served as a pilot for the after repeated exposures had never been tested in a traditional mere basic procedures adopted in all the other studies reported in this article. exposure experiment. When it was conducted, I had not yet introduced the hardware-based Among the variables that affect the mere exposure effect are the random number generator or the stimulus seeking scale. Preliminary results length of each exposure and the number of exposures. The meta- were reported at the 2003 convention of the Parapsychological Convention analysis produced an unequivocal answer for the optimal length of in Vancouver, Canada (Bem, 2003); subsequent results and analyses have each exposure: The shorter the exposure, the stronger the mere revised some of the conclusions presented there. 416 BEM Upon entering the laboratory, the participant was told, he or she liked better by pressing one of two keys on the keyboard. Each pair consisted of two high-arousal negative pictures or two In this experiment, we are interested in measuring emotional reactions low-arousal, affectively neutral pictures; the left/right position of to a wide variety of visual images in a procedure that tests for ESP the two pictures was randomly determined. The computer then (Extrasensory Perception). The experiment is run completely by a randomly selected one of the two pictures to be the target and computer and takes about 20 –25 minutes. flashed it randomly on the left or the right side of the screen for 17 Each trial of the experiment involves a pair of pictures. First you will ms, followed immediately by a masking stimulus that remained on be shown the two pictures side by side and asked to indicate which the screen for 33 ms. There was a 1,000-ms blank interval between one you like better. You will then be asked to watch passively as those exposures. In this experiment, randomizing was implemented pictures are flashed rapidly on the screen. The way in which this through the programming language’s internal random function. procedure tests for ESP will be explained at the end of the session. The number of exposures varied, assuming the values of 4, 6, 8, Most of the pictures range from very pleasant to mildly unpleasant, or 10 across this experiment and its replication. Note that partic- but in order to investigate a wide range of emotional content, some of ipants were not aware that only one of the two pictures was flashed the pictures contain very unpleasant images (e.g., snakes and bodily on each trial, because the target was exposed subliminally and the injuries). instructions implied that both pictures of the pair would be flashed. The participant then signed a consent form that repeated the warning about the nature of the pictures. (The same warning also Results and Discussion appeared in the online recruiting page.) Next, the experimenter seated the participant in front of the computer and withdrew from The retroactive habituation hypothesis was supported. On trials the cubicle. with negative picture pairs, participants preferred the target sig- nificantly more frequently than the nontarget, 53.1%, t(99) ⫽ 2.23, p ⫽.014, d ⫽ 0.22. This is confirmed by a binomial test on the Method proportion of hits across all trials with negative picture pairs,.531, z ⫽ 2.09, p ⫽.018. Women achieved a significant hit rate on the This experiment adapted a binary-choice version of a mere negative pictures, 53.6%, t(62) ⫽ 2.25, p ⫽.014, d ⫽ 0.28, but exposure experiment (Kunst-Wilson & Zajonc, 1980) and, in ef- men did not, 52.4%, t(36) ⫽ 0.89, p ⫽.19, d ⫽ 0.15. This sex fect, ran it backwards. In a standard version of the experiment, the difference is not statistically significant, tdiff(62) ⫽ 0.39, p ⫽.70, following sequence of events would occur: two-tailed, but it did prompt us to introduce different pictures for 1. The participant is repeatedly exposed subliminally to a men and women in the replication (Experiment 6) reported below. picture. This picture is called the habituation target. Participants’ hit rates on the control trials with neutral picture pairs did not differ from chance expectation, 49.4%, t(99) ⫽ ⫺0.73, p ⫽ 2. The participant is then shown two pictures side by side.46, two-tailed. and asked to indicate which one he or she likes better. One of the pictures is the target; the other is a closely Experiment 6: Retroactive Habituation II matched picture that the participant has not seen be- fore. If the participant prefers the target, the trial is One hundred fifty Cornell undergraduates, 87 women and 63 scored as a hit. The hit rate expected by chance is thus men, participated in this replication and extension of Experiment 50%. 5. Most important, this replication added trials with erotic picture pairs. The retroactive habituation hypothesis for these trials was The retroactive version of this protocol simply reverses Steps 1 the opposite of that for negative trials: Participants would prefer and 2: On each trial, the participant is first shown a pair of matched the target picture—the one to be repeatedly exposed— on less than pictures on the computer screen and asked to indicate which 50% of the erotic trials. picture he or she prefers. The computer then randomly selects one There were other two other changes. First, on the basis of the of the two pictures to serve as the habituation target and displays preference data obtained in Experiment 5, we were able to better it subliminally several times. This first retroactive habituation equate the “popularity” of the two pictures within each pair (i.e., experiment comprised trials using either strongly arousing nega- the frequency with which they had been preferred in the previous tive picture pairs or neutral control picture pairs; positively arous- sessions). In a few cases, new pictures from the IAPS set were ing (i.e., erotic) picture pairs were not introduced until Experiment substituted. Second, we decided to use sets of negative and erotic 6, reported below. The retroactive habituation hypothesis was that pictures that were different for men and women. As noted above, on trials with negative picture pairs, participants would prefer the women showed a significant psi effect on the negative trials in target to the nontarget on more than 50% of the trials. Experiment 5, but men did not. Because the psi literature does not As in the experiments reported above, most of the pictures for reveal any systematic sex differences in psi ability, it seemed this study were selected from the IAPS set. The pictures in each possible that the men were simply less aroused than the women by pair were matched not only for content but for valence and arousal the negative pictures. The ratings supplied with the IAPS pictures using the ratings supplied with the IAPS set. revealed that male raters rated every one of the negative pictures in Procedure. The computer program administered the proce- the set as less negative and less arousing than did female raters. dure as outlined above. After the usual relaxation procedure, the Also, an fMRI study using IAPS pictures found that men had participant was shown two matched pictures side by side on the significantly fewer brain regions than women where activation screen on each of 48 trials and asked to indicate which of the pair correlated with concurrent ratings of their emotional experience  FEELING THE FUTURE 417 (Canli, Desmond, Zhao, & Gabrieli, 2002). So, for this replication, not being presented along with more interesting stimuli on a we supplemented the IAPS pictures for men with stronger and within-subjects basis” (p. 798). In our experiment, the control more explicit negative and erotic images obtained from Internet stimuli were arguably less interesting than either the negative or sites. We also provided two additional sets of erotic pictures, so erotic stimuli presented concurrently in the same session, so our that men could choose the option of seeing male–male erotic findings on control trials are not inconsistent with the findings images and women could choose the option of seeing female– from mere exposure experiments. This same contrast effect may female erotic images. also explain why participants in Experiment 1 were able to signif- icantly detect the position of future erotic pictures but not that of nonerotic negative, neutral, or positive pictures. Results and Discussion The control trials also yielded a significant serendipitous find- Both retroactive habituation hypothesis were supported. On ing. The hit rate on control trials was at chance for exposure trials with negative picture pairs, participants preferred the target frequencies of 4, 6, and 8. On sessions with 10 exposures, how- significantly more frequently than the nontarget, 51.8%, t(149) ⫽ ever, it fell to 46.8%, t(39) ⫽ ⫺2.12, two-tailed p ⫽.04, binomial 1.80, p ⫽.037, d ⫽ 0.15, binomial z ⫽ 1.74, p ⫽.041, thereby z ⫽ ⫺2.17, two-tailed p ⫽.03. This effect can be interpreted as the providing a successful replication of Experiment 5. On trials with retroactive induction of boredom. As with a too frequent TV erotic picture pairs, participants preferred the target significantly commercial, the many repeated exposures retroactively rendered less frequently than the nontarget, 48.2%, t(149) ⫽ ⫺1.77, p ⫽ the neutral target picture boring, or even aversive, and hence less.039, d ⫽ 0.14, binomial z ⫽ ⫺1.74, p ⫽.041. attractive than its matched nontarget. In other words, 10 exposures An overall psi score that combines the two complementary was the point at which the inverted-U-shaped function between effects can be computed by subtracting the erotic hit rate from the liking and exposure frequency began its downturn for neutral negative hit rate: This yields a difference of 3.76%, tdiff(149) ⫽ control stimuli. This is presumably the same effect as the de- 2.41, p ⫽.009, d ⫽ 0.20. Because negative and erotic trials were creased liking observed on erotic trials: Repeated exposures of an randomly interspersed in the trial sequence, this significant differ- erotic target diminish its erotic potency and render it boring ence also serves as evidence against the possibility that the signif- relative to its matched nontarget. This apparently occurs at lower icant hit rates were an artifact of inadequate randomization of exposure frequencies for erotic stimuli than it does for neutral left/right target positions. On the neutral control trials, participants control stimuli. scored at chance level: 49.3%, t(149) ⫽ ⫺0.66, p ⫽.51, two- tailed. Hit rates were not significantly correlated with the number of exposures of the target picture across the two experiments, and External Replications there were no significant sex differences on any of the measures. There have been two published replications of the retroactive Erotic stimulus seeking. Although I had not yet introduced habituation effect. Savva, Child, and Smith (2004) conducted a the stimulus seeking scale into our research program when the two conceptual replication in which all the negative pictures were retroactive habituation experiments were conducted, I did define a images of spiders and the neutral pictures were images of land- measure of erotic stimulus seeking for this replication by convert- scapes. No erotic pictures were used. The participants were 25 ing two items from Zuckerman’s Sensation Seeking Scale (1974) into true/false statements: “I enjoy watching many of the erotic spider-phobic and 25 non-spider-phobic individuals. The spider- scenes in movies” and “I prefer to date people who are physically phobic participants had a hit rate significantly above chance on the exciting rather than people who share my values.” A participant’s spider trials, 53.7%, t(24) ⫽ 1.70, p ⫽.05, d ⫽ 0.34, which was score was simply the number of items endorsed, and participants also significantly higher than the 48.2% hit rate on the neutral who endorsed both statements were defined as erotic stimulus control trials, tdiff(24) ⫽ 2.48, p ⫽.010, d ⫽ 0.50. The hit rate of seekers. These two items were administered to 100 of the 150 the non-spider-phobic participants was at chance on both spider participants in this replication prior to the relaxation period and and control trials. experimental trials. There was a highly significant positive corre- Parker and Sjödén (2010) also conducted a replication using lation between erotic stimulus seeking and psi performance (lower only negative and neutral pictures, but their participants went hit rates), r ⫽.24, p ⫽.008, with erotic stimulus seekers showing through both the retroactive habituation procedure and an adapta- a very strong retroactive habituation effect on the erotic trials, tion of Dijksterhuis and Smith’s (2002) original habituation pro- 43.1%, t(31) ⫽ ⫺3.20, p ⫽.002, d ⫽ 0.57, exact binomial p ⫽ cedure. The same stimulus pictures were used in both. In the.002. The hit rate of those who were not erotic stimulus seekers did regular habituation procedure, participants were subliminally ex- not differ from chance: 51.3%, t(67) ⫽ 0.78, p ⫽.78, d ⫽ ⫺0.09. posed six times to half of the pictures and were then asked to rate Control trials. As reported above, overall hit rates on control the valence of both the exposed and the nonexposed pictures. trials did not differ from chance in either of the two experiments. Overall, there was not a significant retroactive habituation effect This might appear inconsistent with the results of mere exposure on the negative pictures, 51.0%, t(49) ⫽ 0.51, p ⫽.31, but there studies in which neutral, nonarousing stimuli typically show in- was a highly significant correlation between habituation and ret- creased liking. Typically but not inevitably. Experiments reported roactive habituation to the negative pictures (r ⫽.34, p ⫽.008). by Bornstein et al. (1990) demonstrated that increased liking for The 34 participants who showed habituation also showed signifi- such stimuli “are only produced when subjects are not currently cant retroactive habituation, 53.9%, t(33) ⫽ 1.93, p ⫽.031, d ⫽ exposed to stimuli that are more interesting or complex. In other 0.33; the 16 participants who failed to show habituation also failed words, robust exposure effects for simple stimuli depend on their to show retroactive habituation, 44.8%, t(15) ⫽ ⫺1.32, p ⫽.90. 418 BEM Experiment 7: Retroactive Induction of Boredom active habituation experiments. Also, because 10 supraliminal exposures on each trial take up a lot of time, we had to limit the The serendipitous finding that sessions with 10 exposures pro- number of trials to 24 to avoid rendering the entire experiment duced a hit rate significantly below 50% on neutral control trials boring, not just the frequently exposed stimulus pictures. suggested that it might be possible to design an experiment spe- The random sequencing of the picture pairs, the left/right place- cifically designed to produce retroactive induction of boredom on ment of the two pictures, and the selection of the target were all nonarousing neutral stimuli as the central phenomenon. This implemented with Marsaglia’s PRNG algorithm. would be desirable for at least two reasons. First, there are large age, cultural, and individual differences in reactions to the kinds of negative and erotic pictures used in the retroactive habituation Results and Discussion experiment, making successful replication across different popu- lations less likely. Second, some psi researchers (or their institu- Across all 200 sessions, the hit rate was in the predicted direc- tional review boards) have been reluctant to conduct experiments tion but not significantly different from chance, 49.1%, t(199) ⫽ in which participants are exposed to these kinds of pictures. This ⫺1.31, p ⫽.096, d ⫽ 0.09. (I now wish I had simply continued to experiment tests directly for the retroactive induction of boredom use subliminal exposures.) Nevertheless, stimulus seeking was on neutral stimuli. again positively correlated with psi performance (lower hit rates), In a mere exposure experiment that included a measure of r ⫽.16, p ⫽.011. Participants high in stimulus seeking obtained “boredom proneness,” Bornstein et al. (1990) report that only a hit rate significantly below chance, 47.9%, t(95) ⫽ ⫺2.11, p ⫽ participants who were not prone to boredom showed a significant.019, d ⫽ 0.22, binomial z ⫽ ⫺1.94, p ⫽.026, whereas the mere exposure effect (i.e., increased liking for a frequently ex- remaining participants did not, 50.1%, t(103) ⫽ 0.17, p ⫽.43. posed stimulus). This suggests that boredom dominated habitua- As in Experiment 2 on the precognitive avoidance of negative tion for boredom prone participants. It should be apparent that our stimuli, there were no control trials randomly interspersed among two-item stimulus seeking scale (“I am easily bored” and “I often the critical trials in this experiment, making it necessary again to enjoy seeing movies I’ve seen before”; reverse scored) could confirm that the successive left/right positions of the targets were equally well be conceptualized as an index of boredom proneness. adequately randomized and did not contain patterns that might For this reason, I first introduced it into the current experiment, match participants’ response biases. Accordingly, we analyzed the with the corresponding hypothesis that those high in stimulus data on the stimulus-seeking subsample in the same four ways as seeking (high in boredom proneness) would show significantly in Experiment 2. The first two analyses were the familiar ones used decreased liking for the target. previously: a one-sample t test across participants’ hit rates, tested against a null hit rate of 50%, and a nonparametric binomial test on Method the proportion of hits across all trials and sessions for the 96 Two hundred Cornell undergraduates, 140 women and 60 men, participants high in stimulus seeking. were recruited for this experiment. From the IAPS we selected sets The third analysis used the phi coefficient, the correlation be- of matched picture pairs that ranged from mildly negative to tween the participant’s left/right preference and the PRNG’s left/ positive; no strongly negative or erotic pictures were included. right target placements, as the index of psi performance. And After the program had administered the two stimulus-seeking finally, the fourth analysis controlled for any potential systematic items and provided the 3-min relaxation period, the protocol was bias in the PRNG by computing an empirical null baseline for each essentially the same as that used in the retroactive habituation participant, as described in Experiment 2. experiments. On each of 24 trials, the participant was shown two As Table 6 reveals, all four analyses yielded comparable results, matched pictures and asked to click the mouse on the picture he or showing a significant retroactive boredom induction for partici- she preferred. The computer then randomly selected one of the two pants high in stimulus seeking. It will also be recalled from the pictures to serve as the target and flashed it on the screen 10 times. discussion of Experiment 1 that Marsaglia’s PRNG algorithm was Unlike in the retroactive habituation experiments, however, the shown in the simulation with random inputs to be free of nonran- exposures were supraliminal (750-ms duration followed by a blank dom patterns that might correlate with participants’ responses screen for 250 ms) and were enlarged to fill the entire screen. It biases. The hit rate of participants low in stimulus seeking did not was my (wrongheaded) hunch that supraliminal exposures would depart significantly from chance in any of the four analyses, with be more likely to produce boredom after 10 exposures than would hit rates ranging from 50.1% to 50.2% (all t values ⬍ 1, all ps ⬎ the subliminal exposures successfully used in our original retro-.40, and all ds ⬍ 0.02). Table 6 Four Analyses of Retroactive Induction of Boredom for Participants High in Stimulus Seeking Hit rate % across participants Binomial test across all trials Phi coefficient across participants Hit rate % using an empirical baseline 47.9% 1,105/2,304 ⫽ 47.9% ⫺.025 48.0% t(95) ⫽ ⫺2.11 z ⫽ ⫺1.94 t(95) ⫽ ⫺2.14 t(95) ⫽ ⫺2.10 p ⫽.019 p ⫽.026 p ⫽.017 p ⫽.019 d ⫽ 0.22 r ⫽.20 d ⫽ 0.22 d ⫽ 0.21  FEELING THE FUTURE 419 Retroactive Facilitation of Recall Results and Discussion In Lewis Carroll’s Through the Looking Glass, the White Queen Unlike in a traditional experiment in which all participants explains to Alice that the citizens of her country have precognitive contribute the same fixed number of trials, in the recall test each ability; or, as she puts it, “memory works both ways” in her land word the participant recalls constitutes a trial and is scored as and she herself remembers best “things that happened the week either a practice word or a control word. Accordingly, the measure after next.” When Alice says that “I’m sure mine only works one of psi in this study is a weighted differential recall (DR) score, way... I can’t remember things before they happen,” the Queen defined as the number of practice words recalled minus the number disparagingly remarks, “It’s a poor sort of memory that only works of control words recalled (P – C) multiplied by the participant’s backwards” (Carroll, 2006, p. 164). overall recall score (P ⫹ C). This gives more weight to participants who recalled more words (“contributed more trials”) and is con- Experiment 8: Retroactive Facilitation of Recall I ceptually analogous to the practice of weighting studies by their sample sizes in a meta-analysis. For descriptive purposes, the DR Inspired by the White Queen’s claim, the current experiment score is expressed as a percentage of the maximum possible DR tested the hypothesis that memory can “work both ways” by score (⫽ 576), which would be achieved if a participant recalled testing whether rehearsing a set of