Rick E. Berger, PhD Sources: off on

Psi Effects Without Real-Time Feedback

One of the most theoretically provocative questions in experimental parapsychology is whether apparent psi effects require the subject to receive immediate, trial-by-trial feedback about their performance, or whether such effects can emerge in data the subject never observes at all. Berger‘s research at the Science Unlimited Research Foundation (SURF) in the mid-to-late 1980s directly probed this question using hardware random number generators (RNGs) and paired feedback/silent data conditions, with implications for both observational theories of psi and the problem of experimenter effects.

Deeper dives — Berger:

Back to hub Forensic critique of psi datasets Automated ganzfeld methodology

Key findings

Apparent psi-direction effects emerged in RNG data samples that subjects never observed, challenging the assumption that real-time feedback is necessary for psi to occur.¹
In paired feedback/silent conditions, directional effects in the silent (never-observed) data were consistent with individual subject performance patterns across multiple experiments.¹
A preliminary review of performance across three PsiLab II computer psi games found patterns suggesting subject-related idiosyncrasies in RNG data even without feedback.²
Simulation (computer-generated baseline) data showed an anomalous peak at the mean run score, structurally similar to PEAR baseline data collected under subject observation, raising the possibility of experimenter influence on ostensible control conditions.³
A pilot study searching for individual “psychic signatures” in RNG data found post-hoc evidence of reduced run-score variance, suggesting idiosyncratic temporal patterning, though the primary hypothesis was not confirmed.⁴

Overview

A persistent assumption in psi research has been that feedback, the subject knowing in real time whether their response was correct, is a necessary ingredient for any anomalous effect to manifest. Berger’s work at SURF challenged this assumption directly. Using the PsiLab II platform he had helped develop at the Psychophysical Research Laboratories (PRL), he designed experiments in which each subject-initiated button press generated two yoked samples of RNG data: one linked to game feedback, and one “silent” sample the subject never saw. If apparent psi effects appeared only in the feedback condition, the feedback-necessity assumption would be supported. Instead, Berger found directional effects in the silent data as well, a result with significant implications for how psi experiments are designed and interpreted.¹ A key Type-II vulnerability in this literature is that silent-condition effect sizes are typically small and the experiments were not powered to detect them reliably; dismissal of null outcomes in underpowered silent conditions would be premature.

The Feedback Paradigm and Its Assumptions

Many RNG-based psi experiments in the 1980s were modeled on a feedback paradigm: subjects interacted with “black box” RNG systems via game software, receiving veridical trial-by-trial feedback about the internal state of the device. The theoretical assumption was that feedback would allow subjects to exert finer control. Berger noted that this model had a hidden implication: if psi effects require feedback, then data generated but never observed should show no effect. Observational theories (OTs) of psi, as Berger reviewed them, held that sensory feedback is necessary for psi to occur, and that effects in pre-recorded or never-observed data must therefore be attributed to retrocausal influence by a later observer. Berger’s experimental design was constructed to test whether silent-data effects could instead be attributed to real-time subject influence at the moment of data generation, which would challenge the retrocausal interpretation.¹

Theoretical Background: Observational Theories and Silent Data

Observational theories of psi, as Berger framed them, predict that psi effects in RNG data sets that were subject-gated but not subject-observed must be caused by the later observation of those data by the experimenter, a retrocausal interpretation. Berger argued that if silent-data effects could be shown to correlate with subject personality variables or display subject-specific idiosyncrasies, this would support a real-time model of psi influence rather than a retrocausal one: the effect would have occurred at time 0 (when the subject initiated the run), not at time 1 (when the experimenter later observed the data).¹

Real-Time vs. Retrocausal Models, the Falsifiability Test

Berger outlined a specific falsifiability criterion for distinguishing real-time from retrocausal psi models: if subject personality variables were found to correlate with RNG data recorded at time 0 by an experimenter and only observed later at time 1 by subjects, this would support the real-time model. Earlier work by Berger, Schechter, and Honorton (reported in Research in Parapsychology 1985) had found that subject personality and self-report measures correlated with subject-gated but never-observed RNG data, a finding Berger cited as preliminary evidence against the retrocausal interpretation.¹ The 1988 paper extended this line of inquiry across four experiments using two different PsiLab II computer psi games, with the explicit goal of accumulating evidence bearing on this theoretical distinction.¹

The Experimental Program: Paired Feedback and Silent Conditions

The core experimental design used in Berger’s feedback/silent studies involved subjects initiating runs via button presses on PsiLab II computer psi games. Each button press generated two yoked RNG data samples simultaneously: a feedback sample, contingently linked to game play so the subject received real-time performance information, and a silent sample, for which no feedback was provided and subjects were blind to its existence. This yoking meant that any difference in effect size between conditions could not be attributed to differences in subject state, motivation, or timing, the two samples were generated in the same moment by the same subject action.¹

Four Experiments, Design and Directional Results

Berger reported four experiments using two different PsiLab II computer psi games, all conducted at SURF. In each experiment, subjects were blind to the presence of the silent data condition. The 1988 paper reports that directional effects emerged in the silent data consistent with individual subject performance patterns. The 1987 conference proceedings (Research in Parapsychology 1986) reported that in the feedback-with-silent-saved (FSS+SILENT) condition, 19 of 24 series effect-size measures were in the positive direction (exact binomial p = .022, two-tailed), while the feedback-with-silent-destroyed (FSD) condition showed an opposite and consistent effect with 17 of 24 series in the negative direction (exact binomial p = .064, two-tailed).³ Simulation data showed no such directional asymmetry: 11/24 positive in FSS+SILENT simulations, 12/24 positive in FSD simulations, consistent with chance.³ A preliminary review across three PsiLab II computer psi games also found patterns suggesting subject-related performance idiosyncrasies in the absence of feedback.²

The Feedback-with-Silent-Destroyed Condition, an Unexpected Finding

One of the more theoretically striking results in Berger’s 1987 proceedings paper was the behavior of the feedback-with-silent-destroyed (FSD) condition. In this condition, silent data were collected in parallel with feedback data but then immediately destroyed, never observed by anyone. The FSD condition showed a consistent negative-direction effect (17/24 series, p = .064, two-tailed), opposite to the positive-direction effect in the feedback-with-silent-saved condition. Berger noted that the collection and then destruction of silent data in the FSD condition was originally a programming expedient to create a feedback-only condition with retained timing, it was not anticipated that the mere availability (or non-availability) of the silent data sample could influence the associated feedback data. This unexpected finding raised the possibility that even the act of generating data that is subsequently destroyed, without any observation, may carry functional significance in psi experiments.³

In Search of Psychic Signatures

A separate pilot study extended the silent-data question in a different direction: rather than asking whether effects occur in the absence of feedback, Berger asked whether individual subjects leave consistent, identifiable temporal patterns in RNG data, patterns idiosyncratic enough to serve as a kind of “psychic signature” that could be matched across independent sessions by a blind computerized procedure.⁴

Psychic Signatures Pilot, Method and Results

The pilot experiment asked two questions: (1) whether subjects interacting with a hardware RNG in a psi game produce temporal patterning in the random data, and (2) whether such patterns, if present, are unique enough to allow computerized blind matching of two independent subject-generated data sets when embedded in a matrix of 20 decoy data sets. The method used epoch averaging (signal averaging across experimental epochs) to detect idiosyncratic temporal patterns. The primary hypothesis, that a subject’s second data set (“replication”) would be identifiable as the best match to their first data set (“template”) by the highest correlation coefficient within a 20-decoy matrix, was not confirmed: the overall rank-sum analysis was not significant.⁴ However, a post-hoc finding emerged: total run-score variance was 116 against a chance expectation of 200, and at the half-run level the effect was more pronounced (observed 80, expected 200, p < 0.01, two-tailed, Monte Carlo method). Berger explicitly flagged this as a post-hoc finding, noting it suggested a psi effect in “balancing out” rank sums rather than directional scoring. This result is exploratory and requires independent replication before any inferential weight can be placed on it.⁴

Experimenter Effects and the Problem of Control Conditions

Perhaps the most methodologically consequential implication of Berger’s silent-data work is what it suggests about the integrity of control conditions in psi research. If RNG data generated without subject observation can nonetheless show directional effects correlated with subject variables, or with experimenter behavior, then the standard notion of a “control condition” in psi research may be fundamentally compromised. Berger raised this concern explicitly, noting that an experimenter with a vested interest in control data showing certain operating characteristics may inadvertently influence those data.¹

Simulation Data and the PEAR Parallel

In the 1987 proceedings paper, Berger reported an anomaly in the simulation (computer-generated baseline) data: when a histogram of simulation run scores was graphed, the mean run score of 50 protruded noticeably above the theoretical curve. The cell at the mean was independently significant (chi-square = 5.17, 1 df, p = .02). Berger noted the structural similarity of this pattern to PEAR baseline data (Nelson, Dunne, and Jahn, PEAR Technical Note 84003) collected while subjects were instructed to generate baseline data without influencing the machine, where the only deviation from the large PEAR database’s run-score histogram also occurred around the mean run score (chi-square = 5.656, 1 df, p = .017). In Berger’s simulations, the experimenter set up the number of games, pressed a button, and left the area, no subject was present. Since no real-time feedback was available in any simulation condition, all three simulation data sets (FSS, FSD, and silent) were functionally similar to the experimental silent condition. Berger concluded that a possible influence of the experimenter on simulation data must be acknowledged, and that comparison conditions intended to serve as controls may themselves constitute “experimenter psi” conditions.³

Implications for Experimental Design, No True Control?

Berger’s 1988 paper articulated a pointed methodological concern: silent-data effects suggest there may be no such thing as a “control” condition in psi research in the traditional meaning of the term. He cited Palmer and Kramer’s work (JP, 1984) in which reported psi effects arose as a function of experimenter-generated control data being deviated from chance, creating a significant difference between conditions with subject-gated experimental data. In Palmer’s study, the control condition was functionally similar to experimental conditions labeled “silent,” “hidden,” and “unseen controls” in the literature. If experimenter influence can operate on ostensible control data, then effect sizes computed as deviations from control baselines may be systematically biased, in either direction, without any subject involvement. This concern is unresolved in the literature and represents an ongoing methodological challenge for RNG-based psi research.¹

Modern Context

The methodological worry Berger surfaced — that experimenter influence on “control” data can confound subject-condition comparisons — has been independently rediscovered in the broader replication-crisis literature. Simmons, Nelson, and Simonsohn (2011) demonstrated that researcher degrees of freedom in analytic choice can manufacture significant findings from null data without any subject effect.⁵ The general principle that small expected effects require pre-specified power analyses to avoid Type-M (magnitude) and Type-II errors was formalized by Gelman and Carlin (2014), who showed that statistically significant small-effect estimates from underpowered designs systematically overstate true effect magnitudes.⁶ These mainstream methodological frameworks bear directly on the interpretive frame of Berger’s silent-data findings: the exploratory effects are best treated as hypothesis-generating, and any single-study null in the silent condition cannot be taken as definitive given typical pre-2015 sample sizes.

Skeptical Critiques and Discussion

Critique 1: Observational theories predict that silent-data effects require retrocausal explanation, and that real-time subject influence on unobserved data is theoretically incoherent

Skeptic source: Proponents of observational theories (OTs) of psi, as reviewed by Berger, held that sensory feedback is necessary for psi to occur, and that effects found in RNG data sets that were subject-gated but not subject-observed must be caused by the later observation of those data by the experimenter, a retrocausal interpretation. This position implies that Berger’s silent-data effects, if real, would themselves be evidence for retrocausation rather than real-time subject influence.¹

Response: Berger’s response was to propose a falsifiability criterion: if subject personality variables correlate with RNG data generated at time 0 and only observed at time 1, this supports real-time influence over retrocausation. Earlier work (Berger, Schechter, and Honorton, Research in Parapsychology 1985) had found personality correlates in never-observed data, and the four-experiment series extended this.¹ The directional consistency of silent-data effects with individual subject performance patterns across experiments was offered as evidence that the effect was subject-specific and real-time rather than observer-driven.¹ However, all findings in this line of research are exploratory; no preregistered adversarial test of the real-time vs. retrocausal distinction has been conducted, and the question remains open.

Analysis. The theoretical dispute between real-time and retrocausal models of psi is unresolved. Berger’s data are consistent with the real-time interpretation but do not rule out retrocausal accounts; the evidence base is exploratory and underpowered.

Critique 2: Silent-data effects may reflect experimenter influence or artifacts in ostensible control conditions rather than subject-related psi

Skeptic source: A structural critique of the silent-data paradigm is that effects attributed to subject influence on never-observed data could equally reflect experimenter influence on those same data, or on the simulation baselines used to establish chance expectation. If experimenters can inadvertently influence RNG data (a concern Berger himself raised), then the comparison between subject-run and simulation-run data may not be clean.³

Response: Berger explicitly acknowledged this concern rather than dismissing it. He noted that in his simulation data, the experimenter set up the runs, pressed a button, and left, yet the simulation data showed an anomalous peak at the mean run score (chi-square = 5.17, p = .02) structurally similar to PEAR baseline data collected under subject observation.³ He concluded that the experimenter’s possible influence on simulation data must be acknowledged, and that comparison conditions intended as controls may themselves be experimenter-psi conditions. This acknowledgment is methodologically honest but leaves the interpretation of subject-specific silent effects partially unresolved: the specific control for experimenter influence on silent data, isolating subject-related from experimenter-related variance, was not fully implemented in these studies. The artifact is mitigated by the subject-specific directional consistency of effects across sessions, but not eliminated.

Analysis. Berger’s own analysis surfaces the experimenter-influence concern as a genuine unresolved issue. The subject-specific patterning of silent effects is the primary evidence against a pure experimenter-artifact account, but the studies were not designed with sufficient controls to fully separate these sources of variance.

References

Berger, R. E. (1988). Psi effects without real-time feedback. Journal of Parapsychology, 52, 1–27. R001 [Berger 1988] ↩︎
Berger, R. E., Schechter, E. I., & Honorton, C. (1986). A preliminary review of performance across three computer psi games. Research in Parapsychology 1985, 1–3. R002 [Berger 1986] ↩︎
Berger, R. E. (1987). Psi effects without real-time feedback. Research in Parapsychology 1986, 9–12. R003 [Berger 1987] ↩︎
Berger, R. E. (1988). In search of “psychic signatures” in random data. Research in Parapsychology 1987, 81–85. R004 [Berger 1988] ↩︎
Simmons, J. P., Nelson, L. D., & Simonsohn, U. (2011). False-positive psychology: Undisclosed flexibility in data collection and analysis allows presenting anything as significant. Psychological Science, 22(11), 1359–1366. https://doi.org/10.1177/0956797611417632 R005 [Simmons 2011] ↩︎
Gelman, A., & Carlin, J. (2014). Beyond power calculations: Assessing Type S (sign) and Type M (magnitude) errors. Perspectives on Psychological Science, 9(6), 641–651. https://doi.org/10.1177/1745691614551642 R006 [Gelman 2014] ↩︎

Deeper dives — Berger:

Back to hub Forensic critique of psi datasets Automated ganzfeld methodology