NOTE-992: Abdulkarim & Ehrsson — No causal link between hand position sense and limb ownership

Contribution

Earlier work, Rohde et al. (2011) above all, showed that drift can occur without ownership, but left open that drift might still be needed for ownership. This study manipulates felt hand position directly during the illusion, by moving the real hand without the participant noticing, and finds that ownership does not follow. It is the converse dissociation: a change in position sense with no change in ownership.

Key insight

If the felt position of the real hand coming closer to the rubber hand were part of what makes the rubber hand feel owned, then moving the real hand closer should strengthen the illusion and moving it away should weaken it. Neither happened. Ownership rides on the timing of seen and felt touch; where the hidden hand is felt to be, within a range of about 8 to 24 cm, does not enter.

Assumptions

• Mechanical displacement mimics drift. The hand rested on Plexiglas sheets driven by a motor at 0.9 mm/s, “causing an 8-cm displacement to take approximately 90 s”, aimed at no more than 0.3 deg/s at the elbow, below the cited threshold for sensing passive movement (p. 709). The authors grant that it differs from illusory drift, since moving the limb “likely produced increased proprioceptive inflow” (p. 717), and argue that the end result, a changed sense of position without felt movement, is the same.
• Participants did not notice. Statement S4 (“my real hand were drifting toward or away”) was denied, and in Experiment 2 a post-experiment questionnaire asked about movement (pp. 709–710, 715–716).
• The questionnaire measures the illusion. Nine statements adapted from Botvinick and Cohen, −3 to +3: S1–S3 for the illusion, S4 for felt displacement, S5–S9 as “control statements for suggestibility and task compliance” (Table 1, p. 709). The outcome is an “illusion index”, the mean of S1–S3 minus the mean of S4–S9. It was given “after the first unique repetition of each condition” (p. 710), so once per condition although each condition ran three times.
• Pointing measures felt position. Eyes closed, the left index finger slid along the table’s edge until judged to be above the right index finger, from a random start (p. 710). Drift is post minus pre, “No corrections were thus made for the veridical hand position” (p. 710).
• Asynchrony is a 500 ms lag. The rubber hand was touched first and the real hand about 500 ms later, at one stroke per second (p. 709).

Key results

• Set-up (pp. 708–709, Fig. 1). Right rubber hand on a small table; real right hand hidden 12.5 cm below on the displacement apparatus, always lateral to the rubber hand. Earplugs and headphones masked the motor. 90 s of brushing per trial; each condition three times. Experiment 1: n = 18 (9 women, mean age 29); Experiment 2: n = 18 (8 women, mean age 28). Sample size “based on previous studies” (p. 711).
• Experiment 1: toward against away (2 × 2 within subjects; pp. 711–713, Figs. 2–4). The hand started 24 cm or 8 cm lateral to the rubber hand and ended at 16 cm. Illusion index: main effect of synchrony F(1,17) = 12.424, p = .003, η² = .125; no effect of direction F(1,17) = 1.321, p = .266; no interaction (p = .840). Synchronous against asynchronous: toward d = 0.577, away d = 0.620. From Fig. 3A, the synchronous indices are about 1.85 toward and 1.45 away, not significantly different. Drift: main effect of direction F(1,17) = 112.211, p < .001, η² = .682; no effect of synchrony (p = .165). From Fig. 3B, about +6 cm toward and −5 cm away, for an 8 cm movement. Shift and illusion did not correlate significantly (toward r = .417, p = .085; away r = .304, p = .220).
• Experiment 2: away against static (2 × 2; pp. 713–716, Figs. 5–8). The hand started 8 cm lateral to the rubber hand and either stayed or moved to 16 cm. Illusion index: synchrony F(1,17) = 25.826, p < .001, η² = .379; displacement F(1,17) = 0.638, p = .443; no interaction (p = .884). Synchronous against asynchronous: static d = 1.759, away d = 1.232. From Fig. 6A, about 2.0 static and 2.1 away under synchrony. Drift: displacement F(1,17) = 251.245, p < .001, η² = .602; synchrony F(1,17) = 11.314, η² = .067. Static synchronous against asynchronous “t(17) = 1.984, p = .003 (two-tailed), d = 0.435”; away, “t(1, 17) = 1.943, p = .057”. From Fig. 6B, about +1.3 cm and −0.3 cm static, −4.5 cm and −5.5 cm away. In the static conditions illusion and shift correlated, “r = .562, p = .015” (Fig. 8); in the away conditions they did not (r = .123, p = .628).
• Awareness of the movement (pp. 715–716, Fig. 9). “11 participants experienced no movement”; seven reported some, but “none of them described a movement that could be attributed to the mechanical displacement”. The authors read those seven reports as “most likely … confabulation, task compliance, or imagination” (p. 717).
• Conclusion (pp. 716–717). Manipulating felt position “does not influence the subjective illusion, which implies that proprioceptive drift either happens as a consequence of the illusion or constitutes an independent, parallel process”. Against Rohde et al.’s open possibility, “our results suggest that proprioceptive drift is not a prerequisite for the illusion”. They are careful about the other direction: “our results merely demonstrate that it is possible to induce an illusion without it being accompanied by proprioceptive drift”, and drift in other contexts “does not mean that the subjective rubber hand illusion cannot cause proprioceptive drift” (p. 717).
• Models (pp. 717–718). The result is “difficult to reconcile” with Botvinick and Cohen’s three-way model and with Tsakiris’s comparator model, whose posture comparison should have been affected by moving the hand. It fits Makin et al.’s peripersonal-space model, since the rubber hand stayed within reach of the hand, and “appear[s] to be consistent with” Samad et al.’s (2015) Bayesian causal-inference model, in which the illusion fades only near 30 cm. They “did not attempt to replicate the ‘vision-only’ condition”, so “we cannot determine whether the asynchronous condition eliminates proprioceptive drift or the synchronous condition triggers it” (p. 718).
• On measurement (p. 718). “One should not rely on the proprioceptive drift measure alone”, but drift “can be used as a valuable indirect measure … if it is used with caution in well-controlled rubber hand illusion experiments when the real hand is immobile (or when only one finger is moved; see Kalckert & Ehrsson, 2012)”. Skin temperature changes “appear to be quite difficult to reproduce”.

Claims

id	claim	strength	support
C1	Slow, unnoticed displacement of the real hand changes felt hand position	strong	drift main effects η² .68 and .60; pointing follows the hand by about 5–6 cm of 8
C2	Changing felt hand position toward or away from the rubber hand, within 8–24 cm, does not change the ownership illusion	moderate	no main effect of direction (p = .266) or displacement (p = .443), n = 18 each; a null without an equivalence test
C3	Drift is not a prerequisite or cause of the illusion	moderate as an inference from C2	limited to the distances tested and to mechanically induced, not illusory, drift
C4	With the hand still, synchronous brushing gives more drift than asynchronous, and the shift correlates with the illusion	weak to moderate	synchrony main effect on drift; the static t-test’s reported p does not match its t; r = .562, n = 18
C5	Drift is a process distinct from the core processes of ownership	moderate, read with Rohde et al. 2011	the authors’ conclusion “in conjunction with” Rohde et al.
C6	Drift remains a valid indirect measure when the hand is immobile and other measures are used	weak	argued from correlations like C4; the study does not test it
C7	Participants were unaware of the displacement	moderate	S4 denied; 11 of 18 reported no movement and 7 reported movements unlike the real one, Exp 2 only

Concepts

• Proprioceptive drift and shift — drift is post minus pre pointing in one condition; shift is synchronous drift minus asynchronous drift, “a synchrony-specific measure” (p. 710).
• Illusion index — mean of the illusion statements minus mean of the control statements (p. 709).
• Peripersonal (perihand) space — the region around the hand within which visuotactile integration refers touch to a seen hand; the rubber hand here was “never farther than 24 cm” from the real one (p. 718).

Connections

The study is built as a reply to Rohde, Di Luca and Ernst 2011 (LIT-1281, NOTE-1009): they showed drift without ownership and left open that drift could be a prerequisite of ownership; this tests that and rejects it. It cites Holle et al. (2011), drift without ownership for rotated hands, and Holmes and colleagues on visual capture. It reads its result as consistent with Makin, Holmes and Ehrsson’s peripersonal-space model and Samad et al.’s Bayesian causal inference model, and against Botvinick and Cohen’s and Tsakiris’s. Kalckert and Ehrsson 2012, filed in parallel by another reader, is cited for a moving-finger variant in which drift is still usable.

Bearing on the record

• NOTE-1006 and LIT-1277 (Botvinick & Cohen 1998). That note’s open question, whether drift measures ownership, is answered in part: felt position can be changed by several centimetres without any change in ownership. The authors themselves say the result is hard to reconcile with Botvinick and Cohen’s three-way model, in which distortion of position sense is how the illusion reconciles vision and touch.
• THEORY-765 (Proposed). It replicates the synchrony effect on the ownership items, twice, with named tests and effect sizes (d 0.58–1.76), so it partly meets the first half of promote_when. It does not meet the suggestion control: S5–S9 are labelled controls “for suggestibility and task compliance”, which is the kind of control the THEORY’s promote_when says is not enough, and expectation was not measured or matched. Its static condition reproduces drift toward the rubber hand under synchrony, but the key t-test is internally inconsistent (see Limitations), and without a vision-only condition it cannot tell, as the authors say, whether synchrony causes drift or asynchrony removes it. So it does not refute the drift clause either; that refutation comes from Rohde et al. One more thing it offers, mine to point out: 7 of 18 participants affirmed in a forced choice that their hand had moved and then described movements unlike the one the apparatus made. The authors call this “confabulation, task compliance, or imagination”. It is a small, direct sign that forced-choice reports in this set-up are open to compliance, which is the worry behind the THEORY’s suggestion control.
• THEORY-769 (Proposed). Two points, both modest. First, as with Rohde et al., ownership and drift came apart, so a promotion test that scores a model on “ownership and drift” is scoring two outputs, not one. Second, a common-cause model in which the spatial discrepancy between felt and seen hand lowers the probability of a common cause predicts some loss of ownership when the felt hand is moved 5–6 cm farther away. None was found. The authors answer this themselves: in Samad et al.’s model the spatial term matters only near 30 cm, beyond the range tested. So the result constrains the shape of such a model (flat spatial dependence under about 24 cm) rather than counting against the graded account. Gradedness and the real hand’s loss of ownership are not tested.
• THEORY-767 (Proposed). As for Rohde et al.: drift in its promote_when is not a safe measure of ownership. This paper’s own advice, drift only with the hand immobile and alongside other measures, would at least be a minimum for the cardio-visual studies.
• THEORY-768 (Proposed). No bearing on whole-body self-identification or self-location, which were not measured.
• THEORY-458 (Proposed). No bearing: the hand was moved passively and unnoticed, and agency was not at issue. The paper notes in passing that drift remains usable “when only one finger is moved”, which concerns the moving rubber hand of Kalckert and Ehrsson 2012, not this design.
• QUESTION-048. The design is an intervention on the putative mechanism, not another correlation: change the behavioural variable and see whether the report follows. That is a stronger test of what a correlate shows than any correlation, and here the answer is that it does not follow. The same study then reports a correlation between the two in its static conditions (r = .562) and uses it to defend drift as a measure, which shows how a correlate can be valid as a sign and invalid as a mechanism.
• No instruction for machine-learning practice.

Limitations

• A null from 18 per experiment. No power analysis, equivalence test or Bayes factor; “no causal link” in the title is stronger than “no significant effect”. In Experiment 1 the synchronous index was numerically higher toward (about 1.85) than away (about 1.45), read off Fig. 3A, the direction a causal link would predict.
• Within the illusion’s known range. The real hand was 8–24 cm from the rubber hand throughout. A null within the range where distance is already known to matter little is a weaker test than one crossing it.
• Mechanical, not illusory, drift. Moving the arm adds afferent signals the classic illusion lacks, as the authors grant.
• Statistics that do not match. The static synchronous–asynchronous drift test is given as “t(17) = 1.984, p = .003”; a t of 1.984 on 17 degrees of freedom gives a two-tailed p of about .06 (my computation). Likewise F(1,17) = 11.314 is reported as p = .001 and F(1,17) = 0.982 as p = .326, where I get about .004 and .34; “t(1, 17)” is malformed. The synchrony main effect on drift in Experiment 2 survives (p about .004), but the paper’s one test of the classic drift effect, the validation of its set-up, is uncertain as printed.
• A hint of a synchrony effect in the displaced hand. In Experiment 2 the away-synchronous drift (about −4.5 cm) was about 1 cm less negative than away-asynchronous (about −5.5 cm), p = .057, much as in the static case. The discussion treats this as no difference.
• Questionnaire once per condition, after the first of three repetitions, and the awareness check in Experiment 2 only.
• No vision-only condition, so it cannot separate its result from Rohde et al.’s account of what drives drift.
• No suggestion control beyond the control statements.

Open questions

• Does displacing the real hand beyond the illusion’s spatial limit, past about 30 cm, reduce ownership? That would separate “position sense never matters” from “it matters only at larger distances”.
• With a vision-only condition added, does the static synchronous drift exceed vision-only drift, contrary to Rohde et al.?
• Does ownership cause any part of drift? The authors leave this open, and the 1 cm synchronous residue in the displaced conditions is a place to look.
• How many of the reports in a rubber hand questionnaire are compliance? The seven unfounded movement reports suggest the question is not idle.