Anomalous ‘Retrocausal’ Effects on Performance in a Go/NoGo Task

Bierman, D. J., & Bijl, A. (2014). Anomalous ‘Retrocausal’ Effects on Performance in a Go/NoGo Task. Journal of Scientific Exploration, 28(3), 437–452.

AI Assessment

A small, statistically significant retrocausal effect that the authors themselves scrutinise for Questionable Research Practices. Participants reacted about 2% faster to a shape they would only later be trained on, consistent with a future task reaching back to influence present performance. The effect was carried entirely by “intuitive” thinkers, and the authors devote a section to how questionable research practices, and a missing formal designation of their main hypothesis as confirmatory, could have contributed.

Provenance

Source. Peer-reviewed research article: Journal of Scientific Exploration, 28(3), pp. 437–452, 2014. University of Amsterdam. Submitted 6 March 2014; accepted 1 August 2014; published 30 September 2014.

Study type. Within-subject behavioural retrocausation experiment: a reaction-time Go/NoGo task testing whether a later (future) task retroactively improves performance on an earlier one.

Funding. None stated.

Data availability. Open: the raw data (SPSS file) and the experiment program (Visual Basic source) are posted publicly by the authors.

Source basis. Figures confirmed against the published JSE article (pp. 437–452), Method, Results, and Table 1.

What the paper reports

In a first Go/NoGo task, participants responded to two of four shapes; in a second task they responded to just one of those two, randomly chosen. The shape carried into the second task is the “target”; the other is the “control.” The prediction was retrocausal: the later second task would retroactively speed responses to the target shape during the earlier first task. It did: reaction times to the target shape were about 2% faster than to the control shape (t(66) = 2.59, p = .024 two-tailed, Cohen’s d = 0.22).¹

The effect was carried entirely by participants classified as “intuitive” thinkers (t = 3.41, df = 34, p = .001 one-tailed, d = 0.40); “rational” thinkers showed nothing. Notably, the authors include a dedicated section on how Questionable Research Practices could have contributed, and report an independent reviewer’s finding that their plan never explicitly declared the main hypothesis confirmatory.

How it was run

• Across 64 trials, one of four shapes appeared at random inter-stimulus intervals (1,500 to 3,500 ms). In the first Go/NoGo task participants pressed a key to two randomly assigned “Go” shapes; in the second task they responded to only one of those two, randomly chosen (the “target”). The other first-task Go shape is the “control.”
• The prediction was that the second (future) task would retroactively act as practice on the first, so the target shape would draw faster responses than the control shape during the first task.
• Reaction times were normalized by each participant’s own baseline simple-reaction time; only correct responses were used. Shape-to-role assignment was randomized per participant to remove any intrinsic shape-recognition effect.
• After the tasks, thinking style was measured with the Human Information Processing (HIP) questionnaire; participants were split at the median into “intuitive” and “rational” groups.
• Of 69 participants, one was disregarded for an excessive error rate and one was lost to computer failure, leaving 67 analysed. Subjects came from a fixed school-environment pool, so there was no optional stopping.
• The pre-specified primary analysis was a paired-samples t-test of normalized target-vs-control reaction times on the first task; the intuitive-versus-rational comparison was a planned one-way ANOVA; HIP-subscale analyses were reported as exploratory.

Results, as reported

The primary directional prediction was confirmed at p = .024 (two-tailed), but the whole effect resides in the intuitive subgroup; the rational subgroup is flat. An exploration of the HIP subscales pointed to a factor the authors label “rigidity.”

Eleven-dimension audit

Pre-registration

The protocol was written in detail before data collection and submitted to the ethics committee and to an independent staff member tasked with checking that the final report matched the plan, which the authors describe as equivalent to a formal pre-registration; the HIP-subscale analyses were duly labelled exploratory. The crucial caveat, raised by the authors themselves: an independent KPU-registry reviewer found that the plan never explicitly stated the main retroactive hypothesis was confirmatory, and also did not pre-specify the reaction-time normalization procedure or the compound HIP thinking-style variable, leaving (in his words) too many degrees of freedom that could have been exploited post hoc. The authors state they did not use this freedom.

Randomization

Shape-to-role assignment was randomized per participant, the target shape for the second task was randomly chosen from the two first-task Go shapes, and shapes appeared in random order at randomized inter-stimulus intervals. Randomization is thorough and well-motivated (it removes intrinsic shape-recognition confounds). The authors do identify, and test by simulation, a residual confound from the non-counterbalanced random choice of target shape (an over- or under-represented shape combined with intrinsic per-shape speed differences), which they judge unlikely to account for the effect.

Sensory leakage

There is no external target to leak; the manipulation is temporal (a future task influencing earlier performance), and the comparison is within-subject (target vs control shape). Because role assignment is randomized and the second task comes after the first, the participant has no ordinary way during the first task to know which shape is the future target.

Blinding

Reaction time is an objective, automatically recorded measure with no human judge. The participant is effectively blind during the first task to which shape will become the second-task target (assignment is random and not yet revealed), and the thinking-style questionnaire was given afterward to avoid biasing performance.

Optional stopping

Explicitly controlled: participants were drawn from a fixed school-environment pool, and the authors state that no optional stopping was used. The final analysed N (67) reflects two principled exclusions (an excessive-error subject and a computer-failure data loss).

Outcome measure

The pre-specified primary measure was the normalized reaction-time difference between target and control shapes on the first task, tested with a paired t-test. It is objective; the directional prediction is reported one-tailed in the body (p = .012) and two-tailed in the abstract (p = .024).

Effect size

Small: Cohen’s d = 0.22, a roughly 2% reaction-time advantage. The intuitive subgroup’s effect is larger (t = 3.41, p = .001 one-tailed, d = 0.40, on 35 participants), but rests on a smaller sample.

Multiple comparisons

The confirmatory side is restrained: one primary paired test plus one planned individual-differences ANOVA. The HIP-subscale analyses (the “rigidity” factor) are explicitly flagged as exploratory rather than presented as confirmatory findings, which is the correct handling.

Internal replication

This is a single study with no internal repetition. The effect’s dependence on the intuitive subgroup (with the rational subgroup flat) is a within-study moderation, not a replication, and a median split on a post-task questionnaire can produce subgroup effects by chance.

External replication

The work sits in the contested retroactive-influence literature, where Bem’s “feeling the future” results (Bem, 2011)² drew sustained statistical and methodological critique (Rouder and Morey, 2011),³ and a later cumulative meta-analysis reports only a small overall precognition effect.⁴

Transparency

Exceptionally high and self-critical. The raw data and the experiment program are posted publicly, the plan was checked by an independent reviewer, exploratory analyses are labelled, and an entire section discusses how Questionable Research Practices could have contributed, including the reviewer’s note that the main hypothesis was not formally declared confirmatory. Few parapsychology papers interrogate their own result this openly.

The adversarial record

• The effect is small (d = 0.22, about a 2% reaction-time difference, p = .024 two-tailed) and rests entirely on the “intuitive” subgroup, with the “rational” subgroup showing nothing. A median split on a post-task questionnaire can manufacture such a subgroup effect by chance.
• The authors themselves raise the central caution: a dedicated Questionable Research Practices section, and an independent reviewer’s finding that the plan never explicitly designated the main hypothesis as confirmatory and did not pre-specify the reaction-time normalization procedure or the compound thinking-style variable, leaving degrees of freedom that could have been exploited post hoc (the authors say they did not use them).
• Reading against the study: a 2% reaction-time advantage in a single sample, carried by a moderator analysis, within a retrocausal paradigm whose wider literature is heavily disputed. The study’s strongest feature is not the effect but the authors’ candor about why it should be treated cautiously.