NOTE-990: Lush, Botan, Scott, Seth, Ward & Dienes — Trait phenomenological control and the rubber hand illusion

Contribution

It put a measured suggestion trait beside the rubber hand illusion in a sample large enough to see the relation. Before it, the illusion’s suggestion checks were the asynchronous condition and control statements, and two smaller studies had read non-significant correlations with suggestibility as evidence against a role for suggestion (p. 6). Here, in 353 people, trait hypnotisability predicted the ownership reports, and predicted them as much after asynchronous brushing as after synchronous brushing. It predicted the control item too. The same trait predicted vicarious pain and mirror touch in two other samples. After this paper, the usual checks cannot be cited as having controlled for suggestion.

Key insight

The rubber hand illusion’s controls compare a condition people expect to work with one they expect not to. A trait that makes people experience what they expect therefore raises the illusion score, the control score and the control item together, and the synchronous-minus-asynchronous contrast does not remove it. Whether anything is left of the illusion once expectation is matched is the question this paper sharpens and does not answer.

Assumptions

• Hypnotisability measures one context-general ability. Response to imaginative suggestion is “the top-down voluntary control of action and experience (which is experienced as involuntary) according to a stable trait ability”, and hypnosis is just one context for it (p. 2). The authors call it “phenomenological control”. They present this as consensus in hypnosis research (p. 2; PRF), which reviewer 2 disputed: what was measured is the relation between suggestibility and the illusions, which “doesn’t demonstrate control in any strong sense”.
• Responses to suggestion are genuine experiences. “If phenomenological control produced effects in embodiment experiments, the resulting reports of experiences would be genuine” (p. 6). This is taken from the hypnosis literature, not tested here.
• A shared trait means a shared mechanism. Correlations of the size between single hypnosis items and the whole scale are read as the two measures reflecting the same ability, “a parsimonious theory … which cannot currently be ruled out” (pp. 5, 7).
• SWASH subjective scores measure the trait. The subjective scale (0–5 per item, averaged) was used for its “greater reliability than the objective scale” (p. 8). In study 2 it was given by computer with a recorded induction (pp. 7–8).
• Synchronous-condition scores are the field’s measure. The headline analyses use synchronous scores, not difference scores, because the twenty most-cited studies mostly do (p. 4; SI p. 2).

Key results

• Samples (pp. 7–8). Study 1 matched existing SWASH data to 404 people screened for vicarious pain and 154 screened for mirror touch (the abstract says 156). Study 2: 385 undergraduates did SWASH screening in a class practical, and 353 completed both it and the rubber hand session. They were assigned to a synchrony-instruction group (n = 114), an asynchrony-instruction group (n = 115) or an uninstructed control group (n = 124, including nine whose instructions failed to play). Sixteen experimenters, blind to instruction, ran five set-ups at once.
• Procedure (pp. 7–8, Table 1). Right hand hidden in a box, its index finger 20 cm lateral to the rubber hand’s, both middle fingers brushed at about 1 Hz for 60 s, synchronously or with an asynchrony, in alternating order. Before induction participants rated two expectancy items (E1, E2: how strongly they expected to feel the rubber hand as theirs under synchrony and under asynchrony, −3 to +3). Drift: pointing to the felt position of the index finger on a ruler offset by a variable amount, before and after each induction. Afterwards four statements, −3 to +3: S1 touch felt where the rubber hand was touched, S2 touch caused by the brush on the rubber hand, S3 “I felt as if the rubber hand were my hand”, and S4 (control) “The rubber hand began to resemble my own (real) hand”.
• Vicarious pain (p. 3, Fig. 1). Per SWASH point, b = 1.25 more pain responses out of 16 (t(402) = 5.55, R² 0.071, rs 0.26) and b = 0.46 more intensity (R² 0.082, rs 0.27); Bayes factors 5.7 × 10⁵ and 2.6 × 10⁶.
• Mirror touch (p. 3, Fig. 2). Touch to a human: b = 0.83 responses out of 14 (t(152) = 3.07, p = 0.003, R² 0.058, rs 0.19, B = 11.98). Touch to objects: b = 0.18 out of 6 (p = 0.012, R² 0.041, rs 0.21, B = 5.23).
• The expectation manipulation failed (pp. 3–4). Mean expectancy for synchronous induction was 0.9, 0.9 and 0.7 in the synchrony, asynchrony and control groups; for asynchronous induction −0.1, 0.2 and −0.6. The instruction × induction interaction on expectancy was insensitive (F(1,255) = 2.06, p = 0.153, B = 0.72). “We are unable to directly establish a causal role for phenomenological control in the RHI” (p. 4). The authors speculate that participants knew of the illusion, that the task carries “strong expectancies inherent” in it, or that the asynchrony rationale (that 1/3 s is “the optimal delay”, SI p. 3) was not believed. Kirsch had asked whether the rationale was convincing (PRF).
• The illusion appeared by the usual check (p. 4). In the control group, synchronous drift exceeded asynchronous (0.93 against 0.38 cm, t(123) = 2.18, p = 0.031, d = 0.20, B = 4.51), and S1–S3 was higher (0.9 against −0.7, t(123) = 9.89, d = 0.89). The group’s expectations ran the same way: 0.7 for synchrony against −0.6 for asynchrony.
• Pre-registered key tests, control group (p. 4). Hypnotisability against the synchronous-minus-asynchronous difference: for drift, b = 0.64 cm per SWASH point, one-tailed p = 0.037, B = 2.91, “anecdotal”; for S1–S3, b = −0.002, B = 0.61, “no sensitive evidence for or against”.
• The analyses not pre-registered, all 353 combined by fixed-effects meta-analysis (pp. 4–5, Fig. 3; SI Table 1). Per SWASH point after synchronous brushing: S1–S3 b = 0.57 (R² 0.09, rs 0.26, B = 1.5 × 10⁶); S4 b = 0.83 (R² 0.11, rs 0.31); drift b = 0.58 cm (p = 0.008, R² 0.02, rs 0.11, B = 14.66). After asynchronous brushing: S1–S3 b = 0.70 (R² 0.09); S4 b = 0.87 (R² 0.12); drift b = 0.02 cm, B = 0.67, insensitive. So “for each 1 point increase in SWASH score … subjective report (S1–S3) increased by approximately half a point … and proprioceptive drift increased by approximately 0.5 cm” (p. 6).
• The drift slope is not uniform across groups (SI Table 1). Synchronous drift on SWASH was b = 1.06 cm [0.42, 1.70] in the control group, 0.49 [−0.31, 1.29] in the synchrony-instruction group and 0.009 [−0.72, 0.74] in the asynchrony-instruction group. The questionnaire slopes were similar across groups (0.50 to 0.65).
• Expectancy predicted report, not drift (p. 5, Fig. 3). Synchronous expectancy predicted synchronous S1–S3 (b = 0.33 per expectancy point, R² 0.06, rs 0.23, B = 1.66 × 10⁴). Its relation to synchronous drift was insensitive (b = 0.09, p = 0.551, B = 0.38).
• Combined with Walsh et al. (p. 6). Walsh et al. reported r = 0.55 between a drift difference score and hypnotisability; combining that with these data gives r = 0.22 [0.05, 0.38], B = 13.86, “even when measured in this way, proprioceptive drift is related to hypnotisability”.
• The field’s practice (SI p. 2). Of the twenty most-cited studies reporting subjective measures, none used the synchronous–asynchronous difference as the measure; most tested the difference and then reported synchronous scores. Of twenty reporting drift, seven used the difference and thirteen the synchronous magnitude. The lists include Botvinick & Cohen 1998, Rohde et al. 2011, Kalckert & Ehrsson 2012, Longo et al. 2008, Suzuki et al. 2013 and Tsakiris et al. 2011.
• What the authors conclude (pp. 5–7). “Existing approaches to control for suggestion effects in these measures are not effective” (p. 5). “Future RHI studies must ensure that expectancies and difficulty are closely matched across all control and illusion conditions” (p. 6). “To the degree to which measures of embodiment effects are driven by phenomenological control, they cannot provide a solid foundation for theories about embodiment” (p. 6). Physiological measures (skin conductance, histamine reactivity) and brain imaging are “also not evidence against suggestion effects”, since suggestion can produce them (p. 6). They predict the same relation for the full-body illusion and for reports of agency (p. 7), untested.
• What the peer review changed (PRF). The first version’s claims that the effect would “transform human behavioural science” and was “likely to confound measures across behavioural science” were removed at reviewer 2’s objection. S4 went from “is an example of demand characteristic driven visual hallucination” to “may be an example of demand characteristic driven subjective report compatible with a visual hallucination” after the reviewer wrote that “It is just ratings of agreement with a statement”.

Claims

id	claim	strength	support
C1	Trait hypnotisability predicts rubber hand ownership and location reports after synchronous brushing	strong as a correlation	n = 353, b = 0.57, rs 0.26, B ≈ 10⁶; not pre-registered, but the analysis follows the field’s usual measure
C2	It predicts the asynchronous reports and the control item as well or better	strong as a correlation	asynchronous S1–S3 b = 0.70; S4 b = 0.83 and 0.87
C3	It predicts synchronous proprioceptive drift	weak to moderate	rs 0.11, R² 0.02, B = 14.66; carried mostly by the control group (SI Table 1); asynchronous drift insensitive
C4	It predicts the synchronous-minus-asynchronous difference	not shown	pre-registered; drift anecdotal (B = 2.91), report insensitive (B = 0.61)
C5	The asynchrony and control-item checks do not control for suggestion	moderate	C2, with the control group’s expectations running in the same direction as the effect
C6	The rubber hand illusion is at least partly a phenomenological-control (implicit suggestion) effect	weak	correlational; the causal manipulation failed
C7	Hypnotisability predicts vicarious pain and mirror touch	moderate	n = 404 and 154; R² 0.04–0.08
C8	Suggestion-driven reports are genuine experiences	asserted	hypnosis literature, cited
C9	Physiological and imaging measures of the illusion are not evidence against suggestion	argued	cited hypnotic modulation of skin conductance, histamine and cortical activity; no data here
C10	Similar relations hold for the full-body illusion and reports of agency	conjecture	prediction only

Concepts

• Phenomenological control — “context-general top-down effects on perception in response to imaginative suggestion” (p. 2), measured in a hypnotic context by hypnotisability scales; voluntary, though “experienced as involuntary”.
• Imaginative suggestion — a suggestion to experience something imagined as real; hypnotic induction adds “only a small boost in response (around 10%)” (p. 2).
• Demand characteristics — after Orne, cues to the experimenter’s aims that can change behaviour; the authors stress that Orne allowed them to change “genuine” phenomenology too (p. 7).
• SWASH — the Sussex-Waterloo Scale of Hypnotisability, ten suggestions, one tactile (a mosquito on the hand), scored objectively and subjectively (pp. 6–7).
• Proprioceptive drift — the change in the pointed position of the hidden hand from before to after induction (p. 2).
• Mirror touch and vicarious pain — reports of touch or pain on one’s own body while watching touch or pain to another (p. 2).

Connections

It builds on the hypnosis literature on imaginative suggestion (Kirsch, Lynn, Terhune) and on Dienes’s cold-control theory, and on Lush 2020 in Collabra, which found that naive participants told about the procedure expected the classic pattern, the illusion under synchrony and not under asynchrony (p. 6, ref. 36). It reinterprets illusions reported without synchronous touch, from mere expectation of touch (Ferri et al.), for drawings of hands and for empty space, as suggestion effects (p. 6). It cites Rohde et al. 2011 among unimodal-stimulation results (ref. 55). It lists Botvinick & Cohen 1998, Kalckert & Ehrsson 2012 and Longo et al. 2008 among the most-cited studies whose controls it calls ineffective (SI p. 2). It disagrees with Walsh et al. 2015 and Marotta et al. 2016, which argued from non-significant correlations that drift is not suggestion (p. 6). The Nature Communications exchange that followed, Ehrsson et al. 2022 and Lush & Seth 2022, and Roseboom & Lush 2022 are being filed alongside this reading. Seth is a co-author, and his 2013 article on the illusion (LIT-1263) builds on results of the kind this paper says are not controlled for suggestion.

Bearing on the record

• THEORY-765 (the synchrony effect). This is the design that THEORY’s promote_when asks for, in form. It gives sample sizes, named tests and effect sizes for both measures, measures trait suggestibility and measures expectations. In substance it neither meets the promotion test nor refutes the THEORY.
  • Promotion is not met. Expectations were not matched: the manipulation failed, and in the control group they ran in the direction of the effect, 0.7 for synchrony against −0.6 for asynchrony. Trait suggestibility was measured, but the test of whether the synchrony effect survives it, the regression of the difference on hypnotisability, was insensitive for report (B = 0.61) and anecdotal for drift. Whether the synchrony difference remains at low hypnotisability was not reported. The data are on OSF.
  • The refutation conditions are not met either. With expectations unmatched, the first cannot be tested here. On the second, drift was larger under synchrony (0.93 against 0.38 cm, d = 0.20), so it did not appear as strongly under asynchrony.
  • It bears against the THEORY’s evidence, not its claim. The ownership reports Botvinick & Cohen relied on are shown to share variance with a suggestion trait, at both levels and in the control item. The “control items guard against blanket agreement, not against suggestion” caveat in the THEORY is borne out with data.
  • Proposal only: the promote_when could add that a measured trait must be shown not to carry the synchrony difference, for instance by the difference at low hypnotisability, since “survives” is ambiguous between those. The status should stay Proposed.
• THEORY-769 (graded common-cause belief). No parametric manipulation of reliability or correlation, so its promote_when is untouched. Two indirect bearings.
  • Ownership reports are graded, but here the gradation tracks a trait unrelated to the sensory input, so graded questionnaire ownership by itself no longer favours the common-cause account. A model-based test should include such a covariate or a matched-expectation design.
  • The paper argues that skin conductance and histamine changes can be produced by suggestion. That adds a second reason, beside the one the THEORY gives, for its exclusion of those measures from the “real hand loses ownership” prong.
  • The synchrony instruction given here was itself the common-cause rationale, and it did not raise expectations beyond the uninstructed group’s.
• THEORY-767 (interoception in ownership). The paper does not mention interoception. Its supplement lists Suzuki et al. 2013 and Tsakiris et al. 2011, the THEORY’s two named sources, among the most-cited studies using the controls it calls ineffective. Two consequences, which are the record’s inference:
  • the cardio-visual synchronous-against-asynchronous contrast is the kind of control shown here not to exclude expectation, so the first prong should also ask for matched expectations or a suggestion covariate, beside the non-bodily periodic signal it already asks for;
  • the sensitivity–susceptibility relation could be carried by a trait other than interoception, so the second prong should ask that it survive hypnotisability. This is a proposal to tighten the promote_when, not evidence against the THEORY. Seth co-authored both papers.
• THEORY-768 (minimal selfhood). Neither prong is touched: self-location and whole-body self-identification were not measured during the rubber hand illusion, and nobody without an agency history was tested. The authors predict the same relation with hypnotisability for the full-body illusion (p. 7). If it holds, the questionnaire evidence from Lenggenhager et al. 2007 that the THEORY leans on, Q3 after synchronous stroking, carries the same confound.
• THEORY-458 (ownership and agency). The paper predicts, without testing, that reports of changes in the sense of agency will track hypnotisability (p. 7). Its framing of suggestion responses as voluntary control “experienced as involuntary” is itself a claimed dissociation between being the source of an experience and feeling oneself its source, which bears on the agency half of that THEORY. It is asserted here, not shown.
• QUESTION-048 (what kinds of evidence show). A worked case. Report measures shared variance with a suggestion trait and with expectation. The behavioural measure, drift, shared much less (rs 0.11) and was not shown to follow expectancy. The authors grant that physiological and imaging measures cannot settle it either, since suggestion produces them. The paper is a direct argument that a report, however genuine the experience behind it, does not say what produced the experience.
• NOTE-372, claim C6. “Body ownership can be manipulated by multisensory conflict” assumed that the manipulation acts through the multisensory conflict. This paper shows that the reports it rests on are also predicted by a trait unrelated to the conflict. That weakens the attribution, not the manipulability.
• No instruction for machine-learning practice.

Limitations

• Correlational. The expectation manipulation failed, and the authors say they “do not directly provide evidence for anything beyond these relationships” (p. 7).
• The headline analyses were not pre-registered. The authors justify the switch to synchronous-only scores by the field’s practice, and say that the pre-registered measures “were not well motivated by the literature” (p. 4). The pre-registered tests were inconclusive.
• The effects are modest. R² is 0.06 to 0.12 for reports and 0.02 for drift; most of the variance in the illusion is not this trait.
• Brief induction, small drift. 60 s of brushing gave mean drift under 1 cm, much smaller than the 30-minute exposures of the original report.
• The asynchrony is loosely specified. It is given as “an asynchrony of approximately 0.5 Hz” (p. 8), a frequency, not a delay.
• Slips in the text. The abstract gives 156 for the mirror-touch sample where the results and methods give 154. Fig. 3’s caption puts drift in panel b and S4 in c, the text the reverse. The methods define difference scores as asynchronous minus synchronous, the results as synchronous minus asynchronous. Two printed t values do not follow from their slopes and standard errors: asynchronous S1–S3 repeats the synchronous t(347) = 5.86, and asynchronous drift gives t(347) = 1.03 for b = 0.02, SE 0.25. Two confidence intervals in Supplementary Table 1 repeat the row above. None changes an inference the authors draw.
• Sample. Undergraduates, mostly women, in a class practical; SWASH by computer.
• “Control”. Reviewer 2’s objection stands: the study measures an association between suggestibility and the illusions, and that the ability is “control” is imported from the hypnosis literature.

Open questions

• With expectations matched, by a credible manipulation or by a condition people do not know to expect, does synchronous brushing still produce ownership reports beyond asynchronous brushing? This is the test THEORY-765 needs and this study could not run.
• In the OSF data, does the synchrony difference remain among low hypnotisables? The regression intercept would say, and it is not reported.
• Is drift, which shared little variance with the trait, the measure least driven by expectation, or only the noisiest?
• Does the relation hold for full-body illusions and for agency reports, as the authors predict?
• Ehrsson et al. 2022 contest the inference; whether their reanalysis or new data change any number above is for the readings of that exchange.