Discovery Lab Answers

Feeling the Future: Overall Status

One important qualification upfront: the ESP-Nexus library’s share of the published literature on precognition has not been measured, so what follows summarizes what the library holds rather than offering a settled account of the full field. The chart below plots effect size (Cohen’s d) against publication year for the studies in the evidence table that report on that metric.

Experiments

Bem’s original “Feeling the Future” paper reported nine experiments testing whether future randomly selected stimuli could retroactively influence present responses — covering precognitive recall, priming, habituation, and arousal tasks. Eight of the nine yielded results in the predicted direction. A subsequent meta-analysis by Bem, Tressoldi, Rabeyron, and Duggan synthesized 90 experiments from 33 laboratories across 14 countries, making it the primary large-scale assessment of whether the original findings replicate. A later multi-site replication study, involving a large consortium of researchers, tested Bem’s time-reversed priming paradigm specifically.

Methodology

The original experiments used within-subjects designs where participants responded to stimuli before the causal stimulus was presented, so no non-psi mechanism could allow anticipation. The meta-analysis covered both exact and modified replications, and materials were made available on request to encourage independent replication. Bem and colleagues also examined publication bias, personality correlates (particularly extraversion and stimulus-seeking), and applied both frequentist and Bayesian analyses. Critics Wagenmakers et al. (2011) and Schimmack (2012) raised concerns about exploratory analyses and low statistical power, and the meta-analysis addresses those directly.

Data

The evidence table contains mixed results across the precognition literature more broadly, including 13 null or below-chance results among the 45 result rows. For the Bem-specific findings specifically:

These are different metrics (d vs. g) and should not be averaged. The pattern shows a drop from the original experiments’ mean effect to a smaller but still statistically positive effect in independent replications. The multi-site replication of the time-reversed priming task did not reject the null hypothesis — results were not significant. The broader evidence table reflects this heterogeneity, with positive, null, and below-chance results present across studies.

Skeptical critiques

What critics argue. Wagenmakers et al. (2011) argued that Bem’s analyses were partly exploratory rather than confirmatory, and that this inflated the apparent significance of results. Schimmack (2012) argued the experiments were underpowered in ways that make false positives likely. Francis (2012) raised the possibility that the pattern of results in Bem (2011) was consistent with p-hacking or selective reporting of outcomes. Bierman (in a comment on the meta-analysis) argued that the difference between the original experiments’ effect sizes and those in the meta-analysis constituted a “failed replication” signal, posting this critique to a researcher forum before formal publication.

What the experimental data show. The 90-study meta-analysis reports a Bayes Factor of 5.1 × 10⁹ across all experiments, and when Bem’s original nine studies are excluded, the independent replications still yield a positive pooled result with a Bayes Factor of 3,853. The p-curve analysis in the meta-analysis estimated the true effect size at approximately 0.20 for the full database and 0.24 for the independent replications. Against this, the multi-site replication of the time-reversed priming paradigm did not find a significant effect, and that study involved researchers holding opposing views on the phenomenon.

Analysis. The debate centers on two distinct questions that the sources do not fully resolve: whether the original experiments’ effect sizes were inflated by analytic flexibility, and whether the effect survives in pre-registered, adequately powered independent replications. The meta-analysis and the null multi-site replication point in different directions on the second question. The personality-correlate findings (extraversion, stimulus-seeking) from the original paper have not been consistently replicated across all follow-up work, as notes the null result extended across data transformations and subgroups.

For Bem’s full research profile, see Daryl J. Bem, PhD.