Discovery Lab Answers

The Most Recent Ganzfeld Meta-Analyses

Coverage qualification — read this first. The ESP-Nexus library holds 57 of the 78 studies in the standard ganzfeld meta-analytic database (the Tressoldi & Storm 2024 Stage-2 Registered Report database, GZMA v13). The 16 studies not yet in the library skew toward null results, so any summary drawn from these holdings — including this one — runs somewhat more positive than the full published database. The figures below come from the published meta-analyses themselves, not just the held subset.

The two most recent meta-analyses

Tressoldi (2024) is currently the most recent large-scale synthesis. It pools 78 studies under a frequentist random-effects model (REML with Knapp-Hartung correction) and reports an overall pooled effect size of ES = 0.074 (z/√N), with a confidence interval of [0.03, 0.12] and p = .0009.

Harris (2024) takes a different approach, combining 28 original studies with 10 new ones (k = 38 total) and reporting a pooled Cohen’s h of 0.27, with z = 7.1. Because Cohen’s h and z/√N are different metrics, these two figures cannot be averaged or directly compared — the table below keeps them in separate rows.

The longitudinal pattern

Across the z/√N-metric results, earlier meta-analyses (Honorton 1990, Storm 2001, Storm 2010) reported larger pooled effects than the two most recent ones (Tressoldi 2021, Tressoldi 2024). The chart below makes this trend visible. One important interruption in that sequence: Milton and Wiseman (1999) — analyzing 30 studies published after the 1987 autoganzfeld communiqué — reported a near-null result (ES = 0.013, z = 0.70, p = .24), which is the single null result in this evidence set. When Dalton (1997) was added and the window extended to March 1999 (k = 39), the same authors’ updated figure rose to ES = 0.038, z = 2.28, p = .011 — still the smallest positive result in the series.

Pooley (2023), working with 41 studies (N = 1,624 trials, 1,496 participants), reports a hit rate of 0.32 against a mean-chance expectation of 0.25, with z = 5.81, p < .001. Hit rates are not directly comparable to standardized effect sizes, so this figure stands separately.

Unsettled signals — these must be on the table

The evidence base carries several unresolved issues:

• The null result is real. Milton and Wiseman’s 1999 analysis of 30 post-communiqué studies returned a non-significant result. Later analyses folding in additional studies recovered a positive signal, but the disagreement between that finding and the broader series has not been fully resolved in the literature — later authors including Storm (2001, 2010) and Tressoldi (2021, 2024) have argued the Milton-Wiseman corpus was too restrictively defined, while others have defended it.

• Heterogeneity. At least two pooled results in this set flag high between-study heterogeneity (I² ≥ 50%), meaning the studies are not estimating a single uniform effect. Moderator analyses — study quality, protocol type, experimenter — remain an active part of the methodological debate.

• Declining effect sizes over time. The z/√N results run from 0.29 (Honorton 1990) down to 0.074 (Tressoldi 2024). Whether this reflects publication-bias correction, protocol standardization, a genuine decline effect, or regression to a true small effect is not settled by the meta-analyses themselves. Cardeña (2018) notes that one analysis of the ganzfeld data found that when study quality increases, effect size does not decrease — the opposite of what a quality-confound account would predict.

• Metric mix. Cohen’s h values (Harris 2024, Honorton 1990) and z/√N values (Tressoldi 2024, Storm 2001/2010, Milton 1999) and Hedges’ g (Tressoldi 2021) are not interchangeable. No single pooled bottom line spans all 12 rows.

Skeptical critiques

What critics argue. Milton and Wiseman (1999) argued that the ganzfeld effect, when examined in the post-communiqué literature under stricter protocol controls, failed to replicate — returning a non-significant result that they interpreted as undermining the autoganzfeld program’s conclusions. Separately, critics have raised file-drawer concerns: that selective reporting of positive results inflates any pooled estimate, and that the apparent effect could shrink to zero under a plausible number of unreported null studies.

What the experimental data show. Subsequent meta-analyses — Storm (2001), Storm (2010), Tressoldi (2021), Tressoldi (2024) — report positive pooled effects across substantially larger corpora (up to 113 studies), with p-values well below conventional thresholds. Tressoldi (2024)’s Knapp-Hartung-corrected estimate of ES = 0.074 [0.03, 0.12] remains positive even under conservative random-effects assumptions. On the specific quality-confound claim, Cardeña (2018) summarizes analyses finding that higher-quality studies did not produce smaller effects.

Analysis. Milton and Wiseman (1999) analyzed 30 studies; Storm (2001) used 79 and arrived at a different result with the same general pool extended. The two teams reached opposite conclusions from overlapping but differently bounded corpora. Independent replication outside any single research group has been published — Storm, Tressoldi, and Harris are independent of one another — but the heterogeneity across studies means no single agreed-upon effect size has been established. File-drawer estimation methods (trim-and-fill, Egger test) have been applied in some of these meta-analyses, but the adequacy of those corrections remains debated in the methodological literature.

For deeper context on the meta-analytic methodology and the replication debate, see Meta-Analysis in Parapsychology, the Ganzfeld page, and the Replication and Meta-Analysis Debates page.