In the standard ganzfeld protocol, a receiver is placed in a reclining position with halved ping-pong balls over the eyes and white noise delivered through headphones, producing uniform visual and auditory fields that reduce patterned sensory input without eliminating sensory experience. A sender in a separate room views a randomly selected target (typically a video clip or image). After the session, the receiver is shown four items, the actual target plus three decoys, and asked to rank or select the best match. Chance performance on this four-alternative forced-choice task is 25%. The effect size of interest is the degree to which hit rates exceed this baseline. Utts noted that the statistical power of individual studies to detect a true hit rate of, say, 33% above a 25% baseline is modest at typical sample sizes, making cumulative meta-analytic assessment essential rather than optional.¹

Utts’s 1986 paper examined the ganzfeld debate from a statistical standpoint, identifying that the dispute involved not just empirical disagreement but disagreement about appropriate statistical standards. On the proponent side, she noted that exploratory analyses of moderator variables, such as sender–receiver relationship quality or target type, were sometimes reported without adequate correction for multiple comparisons, inflating apparent significance. On the skeptic side, she observed that demands for replication under conditions of exact protocol replication were more stringent than those applied to comparable literatures in psychology and medicine, where conceptual replication across varied protocols is the norm. She also noted that the effect sizes reported in ganzfeld studies, while modest, were not smaller than effect sizes routinely accepted as meaningful in other behavioral sciences, a point she would develop more fully in later work on effect-size standards.¹

The Utts, Norris, Suess, and Johnson (2010) Bayesian analysis modeled the ganzfeld hit rate first under a simple binomial framework (X ~ Binomial(2124, p), with conjugate Beta priors) and then under a Bayesian hierarchical model that allowed for between-study heterogeneity in true effect sizes. Under the hierarchical model with a weakly informative prior, the posterior median for the per-study hit rate sat at approximately 0.33 with a 95% credible interval of roughly 0.30 to 0.36. Under the skeptic’s tight prior centered at 0.25, the data shifted the posterior median only slightly (to ≈ 0.258) — making explicit, in Utts’s framing, why skeptics “still are not convinced by the evidence, even with a p-value of 2.26 × 10⁻¹⁸”: a sufficiently tight prior on chance overwhelms even very strong likelihood evidence. The paper used this contrast as a pedagogical illustration of how identical data update different priors differently, addressing the broader concern that disagreement over psi evidence often reflects prior belief differences rather than disagreement about the data themselves.²

The Dalton and Utts (1995) analysis drew on the PRL automated ganzfeld database, which at the time represented one of the largest single-laboratory ganzfeld datasets available, spanning over a decade of trials conducted at the Princeton Research Laboratories using a standardized automated protocol. The automated system addressed sensory leakage by using computer-controlled target selection and presentation, physically isolating sender and receiver in separate rooms, eliminating direct cueing between sender and receiver, though not fully addressing all potential experimenter-mediated pathways. The paper examined three potential moderators: (1) whether same-sex versus mixed-sex sender–receiver pairs showed different hit rates; (2) whether target type (static images versus dynamic video clips) moderated performance; and (3) whether ambient geomagnetic activity on the day of testing correlated with hit rates, following earlier reports by other researchers of geomagnetic correlates in anomalous cognition tasks. The analyses were daily hit-rate and effect-value calculations using the PRL database. Because these analyses were purely exploratory and involved multiple comparisons across the three moderator variables and their interactions, the paper explicitly cautioned against treating any observed patterns as confirmed effects without independent preregistered replication.³

The 2020 Utts referee report (published via Faculty of 1000 Research as part of the open peer review process for the Stage 1 Registered Report on anomalous perception in a ganzfeld condition) addressed methodological questions including study inclusion criteria, the handling of heterogeneity across studies with different protocols, and the statistical approach to effect-size estimation. The Registered Report format is directly relevant to concerns Utts had raised across her career about the need to separate exploratory from confirmatory analyses: by reviewing the protocol before outcomes are known, referees can evaluate the methodological adequacy of the design without the results influencing their assessment. Utts’s engagement with this format as a referee represented a continuation of her long-standing position that the ganzfeld literature’s credibility depends on the quality of its statistical and methodological infrastructure, not merely on the accumulation of nominally significant results.⁴