Contribution
It shows, from the target study’s own data, that the size of the rubber hand illusion’s synchrony effect does not depend on hypnotisability: the least suggestible quarter of 353 participants shows the effect as clearly as the most suggestible. It also shows that the correlations between hypnotisability and illusion ratings that Lush et al. report are not specific to the illusion: they appear as strongly for the asynchronous control and for a statement describing an impossible experience. It adds no new data.
Key insight
The comment turns on one choice of measure. If the rubber hand illusion is the difference between synchronous and asynchronous brushing, hypnotisability does not predict it; if it is the rating in the synchronous condition alone, hypnotisability does, modestly. Ehrsson et al. hold that the difference is the illusion, as the field and Lush et al.’s own preregistration did, so trait suggestibility touches the general level of report and not the thing the illusion adds. Everything then hangs on whether the asynchronous condition is a valid control, which the comment assumes and the reply denies.
Assumptions
- The asynchronous condition controls for everything but timing. This is the load-bearing premise. The comment does not consider that participants might expect different experiences in the two conditions; it argues instead that if they did, “this effect should be most pronounced in highly hypnotizable individuals” (p. 3).
- A null difference across instruction groups means suggestion does not move the illusion. Figure 1d’s three groups were told which condition should work best; the comment itself says that “this manipulation failed” (p. 1), and the reply reports that the instructions did not change expectancies. The comment then uses the null as evidence that the illusion “should be resistant to suggestions” (p. 3).
- SWASH measures the capacity in question. Hypnotisability on the Sussex-Waterloo Scale is taken, with Lush et al., as the measure of trait response to suggestion.
- Questionnaire data are ordinal. Non-parametric tests throughout (Spearman, Wilcoxon, Mann–Whitney, Kruskal–Wallis), two-tailed, with default Cauchy priors for Bayes factors (Fig. 1 caption).
Key results
- The target study, as summarised (p. 1). 353 participants did synchronous and asynchronous conditions; a third were told the illusion would be stronger in the synchronous condition, a third in the asynchronous, a third told nothing. Lush et al.’s preregistered test, in the no-instruction group, found “no evidence for a relationship between the SWASH scale and the illusion ratings” and “only anecdotal evidence” for drift (p = 0.074).
- Whole-sample reanalysis (Fig. 1a, b). SWASH against the synchronous-minus-asynchronous difference in mean illusion rating (statements S1–S3): rho = −0.066, p = 0.219, 95% CI [−0.169, 0.039], BF01 = 5.685, n = 353. Against the difference in proprioceptive drift: rho = 0.047, p = 0.376, 95% CI [−0.057, 0.151], BF01 = 1.657. The first is moderate evidence for the null; the second, by the usual threshold of 3, is not evidence either way, which the text does not say.
- By item (Suppl. Fig. 1A–C). S1, touch felt on the rubber hand: rho = −0.137, p = 0.01, BF01 = 0.44, a negative relation, “directly against the authors’ (one-tailed) pre-registered hypothesis”. S2, the felt touch caused by the seen touch: rho = −0.022, BF01 = 7.88. S3, “the explicit feeling of ownership”: rho = 0.003, p = 0.962, BF01 = 7.525.
- Quartiles (Fig. 1c). In each SWASH quartile (n = 88, 89, 92, 84) synchronous ratings exceed asynchronous ones (Wilcoxon V = 2575.5 to 3198, all p < 0.001) and synchronous ratings exceed zero (all p < 0.001). The synchrony effect does not differ across quartiles (Kruskal–Wallis χ² = 2.913, df = 3, p = 0.405; Q1 against Q4, W = 3879.5, p = 0.574). “Thus, even the least suggestible participants showed clear evidence of experiencing the illusion” (p. 1). No effect size is given for any of these contrasts.
- Instructions (Fig. 1d). The synchrony effect did not differ among the groups told synchronous (n = 114), told asynchronous (n = 115) and told nothing (n = 124): pairwise Mann–Whitney p = 0.73, 0.836, 0.584. Within each, the effect was significant (p < 0.001).
- Non-specificity of the post hoc correlations (p. 2). In Lush et al.’s exploratory analysis of the synchronous condition alone, SWASH “accounted for only ~9% of the illusion ratings’ variability”; “a similar relationship was found for the asynchronous control condition (9%) and a control statement describing an ‘impossible’ experience of visual hallucinations (12%)”. Subtracting the control statement (S4) from each illusion statement leaves no positive relation (Suppl. Fig. 1D–F: S1−S4 rho = −0.189, p < 0.001, BF01 = 0.009; S2−S4 rho = −0.063, BF01 = 3.8; S3−S4 rho = −0.013, BF01 = 6.3, n = 352). The authors add that the post hoc findings may reflect “behavioral compliance or response bias, rather than ‘genuine’ phenomenological control effects”.
- The literature review (pp. 2–3). Lush et al. justified dropping the asynchronous condition by 30 highly cited RHI studies; “28 of the 30 studies used the asynchronous control to conduct critical statistical comparisons with the synchronous condition”.
- The temporal window (p. 3). Cited studies varied the asynchrony and found that “the illusion gradually diminished” and “was abolished for asynchronies >200–300 ms”, a rule that fits multisensory neurons’ integration windows and causal-inference models, and that “is not explained by phenomenological control theory because suggestions have been well controlled in studies using subtle experimental manipulations of asynchrony”. None of these studies was read here; the comment does not say how suggestion was controlled in them.
- Conclusion. The results “fit well with the view that the RHI is a perceptual illusion driven primarily by multisensory mechanisms” (p. 1); Lush et al. add to “the growing literature on individual differences in the RHI” without evidence “that fundamentally alters our understanding” (p. 3).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Hypnotisability does not predict the synchronous-minus-asynchronous difference in illusion ratings | strong as a statement about these data | n = 353, rho = −0.066, BF01 = 5.7; agrees with Lush et al.’s preregistered result and with the reply |
| C2 | Hypnotisability does not predict the synchrony effect on proprioceptive drift | weak | null, but BF01 = 1.66 is inconclusive |
| C3 | The synchrony effect on report is present and of similar size at every level of hypnotisability | moderate | Wilcoxon p < 0.001 in all quartiles, Kruskal–Wallis p = 0.405; no effect sizes |
| C4 | Explicit suggestion about which condition should work does not change the synchrony effect | weak as evidence about suggestion | null across instruction groups, but the comment calls the manipulation failed and the reply says it did not move expectancies |
| C5 | Hypnotisability’s correlation with synchronous ratings is not specific to the illusion | moderate | equal variance explained in the asynchronous condition and an impossible-experience item; no relation after subtracting that item |
| C6 | The asynchronous condition is the field’s standard control and a valid one | asserted for validity | 28 of 30 cited studies use it; validity against expectancy is not argued |
| C7 | The illusion follows a temporal window that phenomenological control does not explain | weak here | cited studies only, not read; how suggestion was controlled in them is not stated |
| C8 | The rubber hand illusion is a perceptual illusion driven primarily by multisensory mechanisms | weak as an inference from this comment | rests on C1, C3 and C6; C6 is the point in dispute |
Concepts
- Phenomenological control — Lush et al.’s term, adopted here, for the capacity to generate experiences that meet expectations, measured as hypnotisability.
- Difference measure — the illusion operationalised as the synchronous rating minus the asynchronous one, “in line with standard practice in the field and the authors’ preregistered hypothesis” (p. 1).
- Illusion statements S1–S3 and control statement S4 — S1 touch at the rubber hand, S2 felt touch caused by the seen brush, S3 ownership; S4 a visual-hallucination statement (Suppl. Fig. 1).
Connections
The comment’s frame is the multisensory account of the illusion going back to Botvinick & Cohen 1998 (LIT-1277), which it cites as ref. 2, and Tsakiris & Haggard 2005, Tsakiris 2010 and Longo et al. 2008 (a sibling filing), all cited. Its causal-inference references (Körding et al. 2007, Samad et al. 2015) are the kind of model THEORY-769 wants tested. Its target is Lush et al. 2020, filed by a sibling reader. Its reply is Lush & Seth 2022 (LIT-1267, NOTE-1003).
Who is on which side. Manos Tsakiris, last author here, is the second author of Apps & Tsakiris 2014 (LIT-1262, NOTE-991), the free-energy account the record files as THEORY-769. Anil Seth, on the other side as second author of the reply, wrote Seth 2013 (LIT-1263, NOTE-1000), the source of THEORY-767. Both 2013–14 papers are predictive accounts of ownership built on rubber-hand findings. By 2022 the two authors disagree about whether those findings show multisensory integration at all.
Bearing on the record
Read against the reply, NOTE-1003. The two NOTEs carry the same verdicts on the THEORYs; this one gives the comment’s half.
- THEORY-765 (Proposed), the synchrony effect. Its promote_when asks for a first-hand reading of a replication with sample sizes, named tests and effect sizes for questionnaire and drift, and a suggestion control: “participants’ expectations of the synchronous and asynchronous conditions are measured or matched, or … trait suggestibility is measured and the synchrony effect survives it”. This comment partly meets it. It is a large replication (n = 353) with named tests, and on the questionnaire it shows the synchrony effect surviving trait suggestibility, which is the second disjunct in the letter. It falls short in three ways: it is a reanalysis of Lush et al.’s data, so the replication is theirs and is being read here at second hand; it gives no effect sizes for the synchrony contrasts, only V statistics and p-values; and for drift it reports only the null, inconclusive relation to SWASH, not the synchrony effect on drift itself. The reply then shows that the disjunct is met while the first one fails: expectations were measured in this sample and differed between conditions, and that difference did not vary with hypnotisability, so the survival of the effect is predicted by the demand account as well. Neither refutation condition is met: no condition matched expectations, and no drift under asynchrony is reported. My proposal, set out in NOTE-1003, is to tighten the promote_when so that surviving trait suggestibility no longer suffices alone.
- THEORY-769 (Proposed), graded common-cause ownership. Not met. The comment’s strongest argument for a multisensory mechanism is the graded temporal window, which is the right kind of parametric evidence, but it is cited, not shown, and the causal-inference models it names were not fitted here. Supplementary Fig. 1C is worth noting: the ownership item’s synchrony effect has no relation to hypnotisability (BF01 = 7.5), the item the THEORY is about. Nothing here bears on the second requirement, the real hand losing ownership.
- THEORY-767 (Proposed), interoception. No direct bearing. Indirectly, by its own logic, the comment would defend synchronous against asynchronous comparisons like the cardio-visual ones the THEORY rests on; the reply is the half that bears on it.
- THEORY-768 (Proposed), minimal selfhood. No bearing. No whole-body measure is taken; the comment concerns a hand.
- THEORY-458 (Proposed), ownership against agency. No bearing; the hand is passive in every condition.
- QUESTION-048. This exchange is a worked case of the question. Both sides agree on every number and disagree on what the reports are evidence of. Each invokes response bias against the other: this comment to say that hypnotisability’s correlations may be “behavioral compliance or response bias” rather than generated experience, the reply to say that low-suggestibles’ illusion may be the same. The behavioural measure, drift, does not settle it here, because its relation to SWASH is inconclusive.
- NOTE-372, claim C6. Metzinger’s “body ownership can be manipulated by multisensory conflict” is the view this comment defends. It defends it with the asynchronous control, which is the point in dispute, so the manipulation half of C6 stands as the field’s standard reading and not as settled.
- No instruction for machine-learning practice.
Limitations
- A reanalysis of another group’s data, by authors whose own work uses the design under criticism. The comment declares no competing interests, which is accurate in the journal’s sense.
- The validity of the asynchronous control against expectancy is assumed and not argued. The comment’s test of it, that demand effects “should be most pronounced in highly hypnotizable individuals”, is not the only prediction a demand account makes, as the reply says.
- Figure 1d is used as evidence that suggestion does not move the illusion, though the comment says the instruction manipulation failed. If, as the reply reports, the instructions did not move expectation, the null cannot show that the illusion resists it.
- No effect sizes for any synchrony contrast; Bayes factors only for the SWASH relations. The drift result is called “non-significant” without noting that BF01 = 1.66 cannot support the null.
- The temporal-window argument, the one piece of evidence that a demand account would have to work to explain, rests wholly on citations.
- These analyses are not preregistered either, as the reply points out.
Open questions
- Would the synchrony effect survive with expectations matched across conditions, for example by an asynchronous condition participants are led to expect more from? Neither side reports such a test.
- Do expectations follow the graded temporal window? If participants expect the illusion to fade gradually with asynchrony, the cited studies do not separate the accounts; if they do not, the window is the comment’s best evidence.
- Why does S1, touch at the rubber hand, show a negative relation to hypnotisability in the difference measure (Suppl. Fig. 1A, D)? Neither side explains it.