Enhanced mind-matter interactions following rTMS induced frontal lobe inhibition

Freedman, M., Binns, M. A., Meltzer, J. A., Hashimi, R., & Chen, R. (2024). Enhanced mind-matter interactions following rTMS induced frontal lobe inhibition. Cortex, 172, 222–233. https://doi.org/10.1016/j.cortex.2023.10.016

AI Assessment

A study in the mainstream neuroscience journal Cortex testing a novel idea: that the brain acts as a filter suppressing psi, so temporarily inhibiting a specific left-frontal region with rTMS should let a mind-matter (psychokinesis) effect emerge. With 108 healthy participants trying to push a random-number-generator-driven arrow, the study’s pre-specified primary test was null (p = .31). A significant effect (p = .006, d = .38) appeared only after the authors applied a post-hoc trial-weighting procedure that gave more weight to early trials, on the reasoning that the rTMS effect fades over time. The design is genuinely strong (random allocation, a sham condition, and a NIST-validated generator), and the authors are unusually candid that the weighting was post hoc and the unweighted analysis did not reach significance. Whether the result is a real effect or a product of analytic flexibility is exactly the question a reader must weigh. This audit reports what the study did and found; it takes no position on whether psi exists.

Provenance

DOI. 10.1016/j.cortex.2023.10.016 · Cortex 2024, volume 172, pages 222 to 233. Open access under CC BY-NC-ND; action editor Jordan Grafman.

Study type. A randomized, sham-controlled experiment in healthy volunteers. Notably, it appears in Cortex, a mainstream cognitive-neuroscience journal, rather than a specialist parapsychology outlet.

Authors. Morris Freedman and colleagues at Baycrest Health Sciences, the Rotman Research Institute, and the University of Toronto. The work follows the group’s earlier findings in patients with frontal-lobe damage.

Funding and data. Funded by the Bial Foundation. Data are openly deposited on the Open Science Framework (osf.io/6rzy8).

Source basis. Every figure below is taken from the article’s own Methods and Results. A cataloguing note surfaced during this audit: the reference database had stored this record’s title as an author-affiliation line; the correct title and authorship, used here, are from the article itself.

What the paper reports

The authors propose that the brain may act as a psi-inhibitory filter, and specifically that the left medial middle frontal region is part of that filter. If so, temporarily inhibiting that region with repetitive transcranial magnetic stimulation (rTMS) should release a latent mind-matter interaction, measured as a participant’s apparent influence over the output of a random event generator (REG).¹ The paper frames this as a neurobiological approach to a phenomenon whose effects are usually small and hard to replicate, and reports a significant psi effect after left-frontal inhibition, in support of the a priori hypothesis.

In support of our a priori hypothesis, we found a significant psi effect following rTMS inhibition of the left medial middle frontal lobe. This significant effect was found using a post hoc weighting procedure aligned with our overarching hypothesis.

How it was run

• Design. 108 healthy participants were randomly allocated across six arms: three rTMS conditions (Left frontal, Right frontal, or Sham stimulation) crossed with trial order. Six non-completers were replaced by deterministically re-allocated participants.
• Intervention. Continuous theta-burst stimulation (a form of rTMS) targeting the frontal region via Fp1/Fp2 sites, producing a reversible virtual lesion; a sham condition provided the control.
• Task. Each participant completed 2,000 trials, 1,000 intention trials (trying to move an on-screen arrow left or right, driven by the REG output) and 1,000 control trials run after they left the room, in 500-trial Right or Left blocks.
• Primary analysis. A linear mixed-effects model on trial-level REG output, with the a priori hypothesis tested by a single cross-product term (Intention by Sham-versus-Left stimulation, among Right trials). This single-cross-product plan was pre-specified in the Bial Foundation grant but was not formally preregistered.
• The weighting procedure. Because the duration of rTMS suppression was unknown and expected to fade, the authors added a post-hoc sigmoidal weighting that gave more weight to earlier intention trials, examined at three taper midpoints (the 300th, 500th, and 700th trial). The model was run with and without this weighting.
• Generator calibration. The REG was tested against the NIST randomness suite (up to 188 tests per sequence) before, during, and after data collection to confirm it produced unbiased output.

Results, as reported

Values are reproduced from the article’s Results section. The single most important fact for a reader is the contrast between the two primary rows: the pre-specified unweighted test was null (p = .31), and statistical significance emerged only under the post-hoc weighting (p = .006). The authors present both openly.

Eleven-dimension audit

Pre-registration

Not preregistered, and this is the study’s decisive dimension. The single-cross-product analysis plan was pre-specified in the Bial Foundation grant, which is more than most, but the trial-weighting that produced significance was added post hoc. Because the pre-specified unweighted test was null and the weighted test was not, the confirmatory status of the positive result is genuinely in question. The authors argue the weighting is a standard method and that their hypothesis concerns the existence, not the duration, of an effect; a reader can accept or discount that, but the sequence (null a priori, significant post hoc) is the fact to hold onto.

Randomization

A strength. Participants were randomly allocated across the six arms, and the dependent measure is the output of a hardware random event generator. Baseline differences between arms on sex, age, and education were small, as expected under randomization (a medium difference appeared on one cognitive screening measure).

Sensory leakage

Not applicable in the ganzfeld sense; there is no sender. The analogous threat, a biased generator producing spurious structure, is directly addressed: the REG was submitted to the NIST randomness suite before, during, and after the study, and the 23 positive tests across 15 long sequences were within chance expectation for that many subtests, with no consistent deviation.

Blinding

A sham stimulation condition blinds participants to whether they received real or sham rTMS, and the experimenter left the room during trials, with the REG scoring automatically. Outcome measurement is therefore machine-automated rather than rater-judged, which limits assessor bias.

Optional stopping

The trial count (2,000 per participant) and the allocation schedule (108, with six deterministic replacements) were fixed in advance, so classic optional stopping is not the concern. The analytic flexibility here lies elsewhere, in the choice to add weighting and to examine three taper midpoints.

Outcome measure

REG output (200 bits per trial, expected sum 100) during intention versus control trials, a standard mind-matter-interaction measure. The measure itself is well defined; the issue is that the analysis of it that reached significance was the post-hoc weighted version rather than the pre-specified one.

Effect size

Small. The weighted primary effect is d = .38 and the within-Left-rTMS effect d = .35, while the unweighted primary effect is d = .11 and non-significant. These are modest effects consistent with the small-effect character the authors themselves note is typical of this field.

Multiple comparisons

This is the second major concern alongside preregistration. The headline result is one of several analyses: unweighted versus weighted, three weighting midpoints, and several related contrasts. The authors report the pattern held across all three midpoints, which is offered as robustness, but the reader should recognize that the significant result was selected from a family of analyses rather than being the single pre-specified test.

Internal replication

Some internal corroboration: the within-Left-rTMS Intention effect (t = 2.59, p = .010) supports the interaction, and the effect is reported as consistent across weighting midpoints. There is no independent replication sample within the study.

External replication

The study is itself an extension of the group’s earlier observations in patients with actual frontal-lobe damage, translated to reversible rTMS lesions in healthy volunteers. It is a novel paradigm rather than a replication of an established one, so external replication by independent groups remains to be done.

Transparency

High, and it is what most distinguishes this paper. The data are on OSF, the generator calibration is reported in full, and, critically, the authors state plainly in the abstract and results that the significant effect required a post-hoc weighting procedure and that the unweighted analysis did not reach significance (and did not converge). Nothing about the key caveat is hidden; the reader is given everything needed to judge it.

The adversarial record

• Null primary, significant post hoc. The strongest objection is straightforward and the authors supply the evidence for it themselves: the pre-specified unweighted test was null (p = .31), and the effect became significant only after a post-hoc trial-weighting was introduced. A skeptic can reasonably classify the positive result as a researcher-degrees-of-freedom outcome rather than a confirmed prediction.
• The authors’ defense. They argue the weighting is a standard way to down-weight observations that may not reflect the exposure (here, the fading rTMS effect), that their hypothesis is about the existence of an effect rather than its time course, and that the pattern held across three taper midpoints. Whether that fully answers the concern is a judgment the reader must make.
• A convergence caveat. The unweighted model did not converge, which the authors disclose while noting its focal estimate was not substantively different from the weighted one. That disclosure cuts both ways: it is honest, but it also means the cleanest comparison is technically compromised.
• What is genuinely strong. Randomized allocation, a sham control, machine-scored outcomes, a NIST-validated generator, open data, and publication in a mainstream neuroscience journal are real methodological credits, and the neurobiological framing is a novel contribution regardless of how the statistical question is resolved. The honest summary is a well-controlled study whose positive result rests on a post-hoc analytic choice that the authors did not conceal.