The ganzfeld procedure reduces patterned sensory input by placing halved ping-pong balls over the receiver’s eyes (illuminated by diffuse red light) and delivering white noise through headphones, producing uniform, unpatterned stimulation rather than deprivation. The theoretical rationale is that reducing patterned sensory noise lowers the signal-to-noise ratio for weak psi signals. Radin’s 1994 ganzfeld study with normal and creatively talented participants (each group N = 32 sessions) found a chance hit rate (25%) in the normal population and a 41% hit rate in the creative population (exact binomial p = 0.016). Geomagnetic field fluctuations correlated inversely with psi performance in the normal group (p < 0.001, two-tailed) and positively in the creative group (p < 0.05, two-tailed), suggesting that environmental moderators interact with participant characteristics in ways that could produce heterogeneous effect sizes across studies.⁵ The proposed mechanism, that geomagnetic fluctuations modulate psi-relevant neural states, remains contested and without consensus interpretation.

The security measures documented in Dalton et al. (1996) targeted four specific artifact classes: (1) Sensory leakage, addressed by physically isolating sender and receiver in separate rooms with computer-controlled target presentation, eliminating direct cueing; (2) Experimenter cueing, mitigated by having the computer, not the experimenter, select and present targets, so no human in the room knew the target identity during the session; (3) Non-random target selection, eliminated by using a hardware random number generator for target assignment; (4) Judging contamination, partially addressed by having receivers rank all four target alternatives against their mentation report before the target was revealed. The paper noted that sender-experimenter interaction during target loading remained a partially unresolved channel, and that demand characteristics (receivers knowing they were in a psi experiment) were not controlled.¹

From August 2000 to December 2018, two online psi experiments using a five-target forced-choice protocol collected 114 million trials from an estimated 200,000 participants worldwide, an opportunity sample self-selected via web access, with no pre-screening criteria. The overall hit rate was consistent with a null effect. However, a planned secondary analysis designed to detect a predicted sequential pattern in the data yielded a small but statistically unambiguous result: with a chance expected rate of p₀ = 0.32, the combined observed p₁ = 0.320502 ± 0.000044, z = 11.28, p = 1.7 × 10⁻²⁹.² Control tests found no evidence that this deviation arose from algorithmic artifacts in the random number generator. The researcher’s preferred interpretation is that the sequential structure reflects a genuine psi-related pattern; alternative interpretations include undetected algorithmic regularities in the RNG or response-bias patterns in how participants selected targets across sequential trials. The study explicitly labels the sequential analysis as a planned secondary analysis, not the primary hypothesis, which partially addresses optional-stopping concerns but does not eliminate the possibility that the pattern reflects a previously unknown statistical artifact of the specific RNG implementation used.

The 1990 sequential analysis replication (Radin) addressed the optional-stopping artifact, the risk that an experimenter stops data collection when results look favorable, by pre-specifying stopping boundaries before data collection began. The method used sequential probability ratio tests (SPRTs) with pre-defined alpha and beta error rates. The replication confirmed that sequential methods produced results consistent with prior work while maintaining Type I error control at the pre-specified level. The extension component explored whether the method generalized across different experimental designs.⁶ This work predated the widespread adoption of preregistration in psychology and represented an early attempt to formalize stopping rules in parapsychology experiments.

The 1993 study (Radin, University of Edinburgh) used an artificial neural network to analyze RNG data from mind-matter interaction experiments and search for person-unique patterns. Using new data and new network configurations distinct from a previously reported study, eight analyses confirmed the presence of person-specific signatures in the RNG output.³ The sample consisted of participants in individual RNG sessions; specific N and session counts are not detailed in the available extract. The proposed mechanism, that individual mental states impress characteristic statistical structure on random sequences, remains without consensus interpretation. Alternative explanations include overfitting of the neural network to noise, multiple-comparisons inflation across eight analyses, and the possibility that person-specific behavioral patterns in how participants interacted with the equipment (timing, session length) produced the apparent signatures rather than any psi-related process. The multiple-comparisons artifact was partially addressed by reporting all eight analyses rather than selecting favorable ones, but no formal correction for multiple comparisons was applied.

Mossbridge and Radin (2021) extended psi-detection methodology to smartphone platforms, creating three iOS-based tasks available from 2017 to 2020 related to micro-psychokinesis and precognition. The study used a large opportunity sample drawn from app store users with no pre-screening criteria. The analysis used a “SEARCH” approach to identify demographic and personality predictors of psi performance. The researchers expected psi performance to frequently appear in the direction opposite to conscious intentions (“psi-missing”) and that gender and psi belief would predict performance.⁸ The smartphone platform addressed the demand-characteristics artifact only partially: participants knew they were in a psi experiment, but the automated delivery eliminated experimenter cueing. Sample selection bias, self-selection by individuals already interested in psi, was unresolved and represents a significant limitation for generalizing effect sizes to the broader population.

In a 2007 letter responding to Delgado-Romero and Howard’s reexamination of the ganzfeld literature, Radin noted that the Parapsychological Association adopted a policy in 1975 specifically opposing selective reporting of positive outcomes, predating the broader replication crisis discussions in psychology by decades.⁴ He further noted that the small number of active ganzfeld investigators makes a large hidden file drawer implausible, since the number of unreported null studies required to reduce the cumulative effect to chance (the “fail-safe N”) would exceed what the field’s research capacity could plausibly have produced. The strength of this argument is moderate: it addresses the implausibility of a very large file drawer but does not eliminate the possibility of a moderate file-drawer effect. The PA policy is a professional norm, not an enforcement mechanism, and compliance cannot be independently verified for all contributing laboratories.