Feeling the future: Experimental evidence for anomalous retroactive influences on cognition and affect.

The experimental procedures are drawn from mainstream psychology and are individually competent.

The experiments were not run under fixed protocols. Randomization methods, stimulus sets, and numbers of exposures changed between and within experiments based on interim results. In one experiment, the final 50 of 150 sessions used a different procedure for sequencing pictures, chosen based on results from the first 100. Participants were told they were in an ESP experiment, which could influence motivated responding in tasks involving conscious choices.

An unvalidated two-item "stimulus seeking" scale — constructed by the author with no published psychometric data — repeatedly rescues otherwise marginal or null results by reporting effects only in the high-scoring subgroup. None of the nine experiments were pre-registered. The procedures are competent enough for ordinary psychological research but too permissive for the extraordinary claims being made.

The individual tests — one-sample t-tests, binomial tests, phi coefficients — are appropriate, and using multiple analytic approaches (parametric and nonparametric, different outlier cutoffs, empirical baselines) for the same data addresses one specific concern about randomization artifacts.

These positives are overwhelmed by fundamental problems. The paper contains no correction for multiple comparisons despite testing dozens of hypotheses across nine experiments, multiple outcome measures, subgroup splits, and exploratory individual-difference analyses. Given the number of tests (conservatively several dozen), multiple "significant" results at the .05 level would be expected by chance alone.

One-tailed testing inflates the apparent success rate. Several headline results have p-values between .01 and .04 one-tailed, meaning they would be marginal or nonsignificant two-tailed. This clustering of p-values just below the .05 threshold is consistent with analytic flexibility — choices about test direction, subgroup, and outcome measure — nudging marginal results across the significance line.

No confidence intervals are reported. For a hit rate of 53.1% in 100 participants with 12 erotic trials each, the confidence interval would comfortably include 50%. The subgroup effect sizes — reaching 0.71 for high stimulus seekers in one experiment, based on a two-item scale with no established reliability — are implausibly large for the measurement quality of the moderator and are classic signatures of post-hoc subsetting in small samples.

The reported evidence is statistically indistinguishable from what flexible analysis of small effects in moderate samples would produce if no real effect existed.

The disparity between what the study measures versus what it claims is enormous. The measurements are tiny deviations from chance — a 53.1% hit rate where 50% is expected, a 15-millisecond reaction time difference. The claim is that causation runs backward in time. The space of mundane explanations (response biases, demand characteristics, stimulus-sequencing artifacts, researcher degrees of freedom, ordinary fluctuation) is vast, and the design cannot rule them out.

The paper's argument-by-elimination — that using both algorithmic and hardware-based random number generators rules out clairvoyance and psychokinesis, leaving precognition as the "only nonartifactual interpretation" — commits a category error. It works only if you have already accepted that some form of psychic ability produced the results and are choosing among variants. The most relevant alternative — that the results reflect methodological and statistical noise — is not in the elimination set.

The two cited external replications both required post-hoc subsetting to find any effect: one found it only in spider-phobic participants viewing spider images, the other only in participants who also showed the standard (non-reversed) version of the effect. Neither is a clean replication.

The design cannot distinguish small artifacts from genuine reversed causality.

The paper presents precognition experiments as "time-reversed" versions of well-established effects — priming, habituation, recall facilitation — implying that if you accept the standard versions, you should extend comparable credibility to the reversed versions. But the standard versions have known causal mechanisms operating in the normal direction of time. Reversing the temporal order requires an entirely different causal story, one that violates established physical principles. The framing treats this as a minor procedural detail when it is the entire substance of the claim.

The paper has a falsifiability problem. When the full sample produces a null result (as in the retroactive boredom experiment), significant effects appear in a subgroup of high "stimulus seekers." When the subgroup effect disappears (as in the final recall experiment), a stronger manipulation allegedly "restricted the range" enough to suppress the moderator's predictive power. Positive results confirm the stimulus-seeking prediction; negative results are attributed to methodological features that overwhelm it.

The paper preemptively discounts replication failures by arguing that psychologists overestimate how often significant results will replicate, citing Tversky and Kahneman. This is an accurate statistical point repurposed as a rhetorical shield. For a claim that requires overthrowing basic physics, demanding a higher evidential bar is proportional to the implausibility of the claim, not a double standard.

The paper acknowledges concerns about randomization and file-drawer effects but does not seriously engage with the possibility that its results arise from cumulative analytic flexibility — one-tailed tests, uncorrected multiple comparisons, post-hoc subgroup analyses, evolving protocols — across nine experiments conducted over several years.

None of the nine experiments were pre-registered. The paper describes extensive pilot testing and procedural evolution that makes planned analyses indistinguishable from post-hoc adjustments. Raw data are not publicly available. Analysis code is not provided. Materials are available only on request, not in a public archive. No conflict of interest statement appears, despite the author's decades-long career investment in parapsychology. No funding sources are disclosed. No ethics board approval is mentioned, despite research involving sexually explicit and emotionally disturbing imagery.

On the positive side, the file-drawer disclosure is unusually forthcoming: the author names individual-difference variables that failed, a correlate that did not survive the full dataset, and a small experiment that produced null results. Sample sizes are justified by explicit power analysis. These are rare in parapsychology and uncommon in mainstream psychology at the time.

But the disclosed exploration itself reveals the scope of analytic flexibility the design permitted — exactly the degrees of freedom that pre-registration exists to constrain.

The citation of mainstream psychology — priming, habituation, mere exposure, response-time methodology — is accurate and appropriate.

The citation of the parapsychology literature is one-sided. The paper leans heavily on a single meta-analysis of forced-choice precognition experiments (Honorton and Ferrari 1989) without mentioning skeptical methodological critiques by Hyman (1985, 1994). A meta-analysis by Milton and Wiseman (1999) that failed to find a significant ganzfeld telepathy effect — directly challenging the evidential base the paper invokes — was available a decade before submission and is not cited. Bayesian critiques of psi evidence, directly relevant to the paper's own discussion of Bayesian priors, are absent.

The two cited external replications are framed as confirmatory when one was null overall and both depended on subgroup analyses. A single failed replication is buried in a footnote and minimized as "a small pilot study" using a "different practice procedure." The physics discussion draws exclusively on voices from within the parapsychology community, and a Feynman quote about accepting well-confirmed quantum phenomena despite their counterintuitive nature is repurposed to argue for lowering evidential standards for unconfirmed claims.

The author has no financial conflicts but had spent decades building a research program in parapsychology, including an earlier high-profile publication arguing for psychic information transfer (Bem and Honorton 1994). The paper's primary outside advisor, Dean Radin, leads research at an organization whose mission depends on the reality of the phenomena being tested.

Several design choices are consistent with this career stake: one-tailed tests throughout, no pre-registration, a custom two-item moderator scale used to rescue marginal results, iteratively refined procedures, and no independent data analysis. The conclusions uniformly favor the existence of precognition, and the framing is explicitly advocatory — the author describes "simplicity and familiarity" as "tools of persuasion" and closes by urging readers to "raise their posterior probabilities."

Subsequent large-scale replication attempts by independent teams — including a registered replication (Ritchie, Wiseman, and French 2012) and a meta-analysis accounting for publication bias (Galak et al. 2012) — found no evidence for the claimed effects, consistent with the interpretation that the original results were products of researcher degrees of freedom.