nucleation

LIT-1280: Serious Problems With Interpreting Rubber Hand “Illusion” Experiments

4 min read

Properties4
StatusActive — read in full 2026-10-20 (NOTE-1001) from the accepted manuscript on PsyArXiv, with the authors' analysis notebooks for 10, 20 and 50 participants; worth reading as the clearest demonstration of what a modest correlation between hypnotisability and rubber hand report does to a field of small studies. Resampling the 351 synchronous-condition reports of Lush et al. 2020 in simulated experiments of 20, samples drawn from the upper half of the hypnotisability scale found significant agreement with the three illusion statements 99% of the time and samples from the lower half 51%; for the one ownership statement, "I felt as if the rubber hand were my hand", the figures were 74% and 4%. Read it as a reanalysis of one dataset, with no new data and no asynchronous or drift analysis. Its headline numbers depend on n = 20: at n = 50 the lower half finds the averaged illusion statements in 95% of experiments, though the ownership statement still only in 6%. The authors themselves leave open that the same people may have the illusion for other reasons.
Tagsself, psychometrics, embodied-cognition, philosophy-of-science
Cited as “paper” byNOTE-1001 — Roseboom & Lush — Serious problems with interpreting rubber hand illusion experiments
Cited as “source” byTHEORY-766 — Brushing a seen rubber hand in synchrony with the hidden real hand, against asynchronous brushing, raises reports that touch is felt at the rubber hand and, more weakly and less reliably, that the rubber hand is one's own; the proprioceptive drift toward the rubber hand that accompanies these reports is not a measure of them; the standard comparison conditions do not equalise participants' expectations, so whether the stimulation or the expectation produces the report difference is open

Graph View

Roseboom, W. & Lush, P. (2022). Serious problems with interpreting rubber hand “illusion” experiments. Collabra: Psychology 8(1), 32274. DOI 10.1525/collabra.32274. Preprint: PsyArXiv, DOI 10.31234/osf.io/uhdzs.

Key takeaways

From the reading, NOTE-1001.

  • The relationship is large enough to decide outcomes. The paper draws simulated studies of 20 from the 351 reports in Lush et al. 2020, splitting them at the median of subjective hypnotisability. Studies drawn from the upper half found the illusion (S1–S3 mean significantly above zero) 99% of the time, and studies from the lower half 51%. The authors conclude that the correlation (rs = 0.26) is not “too small to be of concern”.
  • Ownership needs the suggestible. For the one ownership statement (S3), lower-half studies found significant agreement about 4% of the time. Their p-values were roughly uniform, “the same as the probability of obtaining any other extent of 0.05”.
  • Referred touch is robust, and the measure is muddled. S1 was agreed with in 95% of lower-half studies. The authors argue that S1 is ambiguous and that only S3 concerns ownership. They also note that 12% of participants agreed with S1 while disagreeing with S2, though both are meant to measure referred touch.
  • Synchronous against asynchronous is set aside. Following Lush (2020), the authors treat the difference as confounded by expectation and analyse only the synchronous condition.
  • Three readings, one preferred. The same people may control their experience in the RHI as elsewhere (the authors’ preferred reading). They may have the RHI for distinct reasons, such as pure multisensory integration (“not explicitly ruled out”). Or they may respond as if they did for some other reason. The first reading does not exclude a multisensory contribution.
  • Implicit and animal measures. These are linked to ownership only through report, so the paper calls the inference from drift, neuroimaging or a “rubber tail” in mice to subjective ownership a reverse inference with “no direct evidence”.

Standing in the record

Filed on 2026-10-20 at the owner’s request for experimental work on the rubber hand illusion that bears on the record’s four Proposed THEORYs about it. Read in full from the accepted manuscript and the authors’ notebooks: Active.

It is not new data. It re-expresses the correlation in Lush et al. 2020 (filed alongside) as outcomes for a field of small studies, and it answers the commentary by Ehrsson and colleagues, the exchange filed alongside as Ehrsson et al. 2022 and Lush & Seth 2022. The reading in the NOTE separates two things. The part that holds even at n = 50 is that agreement with the ownership statement itself depends heavily on trait phenomenological control. The part that depends on sample size is the claim that the averaged illusion statements “require” such participants.

What it does to the four THEORYs, set out in the NOTE:

  • THEORY-765. It does not meet the promote_when: it has no asynchronous or drift analysis, and it does not test whether the synchrony effect survives trait suggestibility. It bears against the ownership clause more than the referred-touch clause.
  • THEORY-767. It suggests that the interoceptive-sensitivity result should be re-tested with trait phenomenological control as a covariate.
  • THEORY-769 and THEORY-768. It qualifies their evidence without testing them.

It also bears on QUESTION-048, on reverse inference from implicit and animal measures.

Filed alongside, in the 2026-10-20 batch: Lush 2020, demand characteristics (LIT-1276), on which this paper’s set-aside of the asynchronous condition rests; Lush et al. 2020; Ehrsson et al. 2022 and Lush & Seth 2022; Rohde, Di Luca & Ernst 2011; Abdulkarim & Ehrsson 2015; Longo et al. 2008; Kalckert & Ehrsson 2012. It carries no instruction for machine-learning practice.

One of the sources of THEORY-766 (filed 2026-10-20), which restates the rubber hand synchrony effect as a difference in reports, separates drift from those reports, and leaves open whether expectation produces the difference.

Backlinks