Discovery Lab Answers

What Presentiment Data Look Like

Coverage note: The figures below come from 14 studies currently held in the ESP-Nexus library. The library’s share of the published literature on presentiment has not been measured, so treat this as a summary of what the library holds rather than a settled account of the field. How much of the broader published literature is represented here is unknown.

What presentiment studies measure

Presentiment research asks whether the body registers physiological responses to emotionally significant stimuli before those stimuli are randomly selected and displayed — a potential signature of unconscious precognition. The dependent variables span several measurement systems: electrodermal activity (EDA/skin conductance), EEG, pupillary dilation, cardiac measures, and — in some later work — large-scale behavioral or text-sentiment data.

The pattern across the library’s 14 studies

All 14 result rows in the library point in the positive direction (dir=positive across the board). That said, the metrics are genuinely heterogeneous — standardized effect sizes, raw z-scores, and p-values that are not comparable to one another — so no single pooled bottom line can honestly be drawn across all rows.

Where standardized effect sizes are available:

• Mossbridge (2012), a meta-analysis of 26 studies, is the most comprehensive pooled estimate in the library. It found a fixed-effect Cohen’s d of 0.21 with a confidence interval of [0.15, 0.27], a combined z of 6.9, and a p-value reported as less than 2.7 × 10⁻¹². That is a small but consistent effect by conventional benchmarks.
• Tressoldi (2010), examining high-absorption participants, found a smaller effect (z/√N = 0.11) with a hit rate of 0.56 against a mean chance expectation of 0.50.
• Radin (2004), pooling four experiments across more than 4,500 trials, found a weighted mean pre-stimulus effect size of z/√N = 0.064 with z = 4.04 and p = 1.3 × 10⁻⁵.

Where z-scores without full standardized effect sizes are reported:

• Radin (1997) reported a combined Stouffer z of 5.0 across two early EDA experiments.
• Radin (2009) found a z of 3.17 (p = 0.0008, one-tailed) for pupillary dilation differences between emotional and calm stimuli.
• Radin (2007) found a z of 2.72 (p = 0.007) for female participants on pre-stimulus slow cortical potential differentiation.
• Radin (2011), using a sign test across 101 studies, reported that 85 of 101 were positive, with odds against chance of 1.3 × 10¹².
• Radin (2022) and Radin (2023), analyzing 13 years of Twitter sentiment data across 10 languages, found Stouffer z values of −3.825 and −3.087 respectively — both flagged as positive in the coding, reflecting pre-event sentiment shifts prior to significant negative world events.

EEG and other physiological work:

• Mossbridge (2017) applied random forest classification to EEG data (N = 40 participants) and reported p < 2.5 × 10⁻⁶ distinguishing original from scrambled pre-stimulus data.
• Mossbridge (2023) reported p < 1 × 10⁻⁶ on EEG pre-stimulus prediction of future button-press responses in a sample of 4,000 trials.
• Kittenis (2011) found pre-stimulus phase-synchronization differences at multiple intrinsic mode frequencies (IMF3–IMF5, each p < 0.001) for new versus old faces.
• Hamelin (2022) found a pre-choice GSR difference anticipating correct quiz answers (p = 0.00033) in 72 participants.
• McCraty (2010) reported positive pre-stimulus physiological differences between win and loss conditions in 13 participants, though only a direction and sample size are recorded for that row, not a test statistic.

What is not settled

Several important caveats deserve equal weight alongside the positive signal:

• Metric heterogeneity is substantial. The library holds five p-only rows, five z-only rows, and three standardized-effect-size rows. These measure different things and cannot be collapsed into a single “overall effect size” without distortion.
• Effect sizes, where available, are small. Cohen’s d = 0.21 (Mossbridge 2012) and z/√N values around 0.06–0.11 (Radin 2004, Tressoldi 2010) are modest by any standard, meaning individual studies need large samples or careful aggregation to be informative.
• No null or below-chance results appear in the library’s current holding. Whether that reflects the literature or a gap in library coverage is unknown — publication bias toward positive results is a standing concern in this area, and without measuring the library’s coverage fraction the question cannot be resolved from what is here.
• Heterogeneity across paradigms — from EDA and pupillometry to EEG to large-scale social media sentiment — makes it difficult to evaluate whether the same underlying mechanism is being measured across studies.
• Alternative explanations remain contested. Decision Augmentation Theory, experimenter effects, and conventional anticipation mechanisms have all been raised as alternative accounts; Radin’s published responses contest their sufficiency, but the debate is ongoing.

For a deeper account of the electrodermal and EEG work specifically, the ESP-Nexus page Presentiment and Electrodermal Anticipation of Future Stimuli covers Radin’s program in detail, including the meditator studies and the Decision Augmentation Theory exchange.