Contribution
The paper builds a rubber hand illusion out of movement instead of touch, with a physical model hand, and uses it to manipulate the sense of ownership and the sense of agency separately in one paradigm. Each can be present without the other: passive movement gives ownership without agency, and an anatomically impossible hand gives agency without ownership. Before it, studies of moving hands had always had agency together with ownership, so “it was not possible to test for double dissociation” (p. 2).
Key insight
What makes a moving hand feel like mine and what makes me feel I am moving it depend on different conditions. Ownership needs the seen hand to be where and how one’s own hand could be, and to move in time with it; it does not need me to be the one moving it. Agency needs me to be the one moving it, and the movement to come in time; it does not need the hand to be mine. Both need synchrony. Yet when both are present they are not simply added: agency over an owned hand is rated stronger, and only then do the two ratings go together.
Assumptions
- The questionnaire measures the two senses. Sixteen statements on a −3 to +3 scale: four for ownership, four for agency, and four control statements for each, which “served as controls for task compliancy, suggestibility, and expectancy effects” (p. 4, Table 1). Items were adapted from Botvinick & Cohen (1998) and Longo et al. (2008). Scores are the mean of each group of four, and a mean ≥ +1 counts as affirmed.
- Drift measures ownership, and only with the questionnaire. One pointing movement before and one after each induction, with the left index finger to the felt height of the right, eyes closed; three trials per condition. “We do not rely on experimental measures in isolation”; the illusion requires both significant drift and affirmed ratings, citing Rohde et al. (2011) (p. 5). There is no objective measure of agency.
- Passive means without efference. In passive trials the experimenter moved the rod, with the link to the participant’s finger kept, and checked by feel for “the lack of active or reactive forces” (p. 4). Both fingers then move together, so the participant gets kinaesthetic and visual feedback with no motor command.
- The asynchronous and passive movements imitate the participant’s. The experimenter copied amplitude, frequency and double taps from earlier trials (p. 4). In the asynchronous condition the rod was detached and the experimenter moved the model finger by hand with “approximately 500 ms delay”, so that the fingers “appeared to be moving in opposite phases”. The participants did not learn the mechanism by debriefing (p. 4).
- Between-group experiments. Each experiment is a separate sample; conditions within an experiment are within-subject. The comparison of asynchrony with passive movement crosses experiments.
Key results
Setup (pp. 3–4, Fig. 1). The participant’s gloved right hand lay in a box 30 cm in front; a gloved, life-sized wooden model hand lay on the box, a blanket covering the gap to the shoulder. A rod through the box joined the two index fingers. Participants tapped at about 1 Hz with occasional double taps, for 3 minutes per condition in Experiments 1 and 3.
Sample. 104 naïve participants: 20 (Exp. 1), 20 (Exp. 2), 32 (Exp. 3), 32 (Exp. 4); no one took part twice (p. 3). Exp. 2 lost three questionnaires, so its correlations have n = 17 (p. 6).
Experiment 1: synchrony, questionnaire (p. 5, Fig. 2).
- Synchronous: ownership 1.63 (SD 1.27), agency 2.44 (SD 0.56); “75% of the participants” affirmed ownership.
- Asynchronous: ownership −0.99 (SD 1.5), agency 0.66 (SD 1.19).
- Wilcoxon: ownership against its control, synchronous, Z = 3.792, p < 0.001; ownership synchronous against asynchronous, Z = 3.304, p = 0.001; agency against its control, Z = 3.825, p < 0.001; agency synchronous against asynchronous, Z = 3.767, p < 0.001 (pp. 5–6).
- On the rendered Figure 2, agency under asynchrony (0.66) is still marked above its control (about −0.8, ”**”). Agency was reduced, not abolished, by asynchrony.
Experiment 2: synchrony, drift (p. 6, Figs. 3–4).
- Drift 1.73 cm (SD 2.05) synchronous against −0.35 cm (SD 1.04) asynchronous; Wilcoxon Z = 3.062, p = 0.002.
- Drift correlated with ownership ratings in the synchronous condition, “Spearman r = 0.557, n = 17, p = 0.020”, and r = 0.532 (n = 16) without an outlier; pooled over conditions, r = 0.676 (n = 34). Not with agency ratings (synchronous r = 0.274, p = 0.287). “Thus, the proprioceptive drift is related to the experience of ownership but is not related to agency” (p. 6).
Experiment 3: active/passive × congruent/incongruent, questionnaire (pp. 7–9, Fig. 5). All four conditions synchronous.
| condition | ownership | agency |
|---|---|---|
| active congruent | 2.12 (SD 1.03) | 2.52 (SD 0.55) |
| active incongruent | −0.62 (SD 1.62) | 2.2 (SD 0.78) |
| passive congruent | 1.57 (SD 1.3) | −0.61 (SD 1.5) |
| passive incongruent | −0.91 (SD 1.6) | −1.5 (SD 1.64) |
- Each affirmed score exceeded its control (Z between 4.920 and 4.940, p < 0.001) (p. 7).
- Ownership differed across conditions (Friedman χ²(3, n = 32) = 65.287, p < 0.001); congruent above incongruent for active (Z = 4.863) and passive (Z = 4.656), both p < 0.001.
- Active above passive ownership in the congruent position, Z = 2.765, p < 0.01, but not in the incongruent position, Z = 1.961, p = 0.050, against a Bonferroni threshold of 0.0125. The interaction test was not significant (t = 0.744, df = 31, p = 0.462) (p. 8).
- Agency differed across conditions (Friedman χ² = 75.247, p < 0.001); active above passive in both positions (Z = 4.863 and 4.772, p < 0.001). Agency was also higher in congruent than incongruent positions, for active (Z = 2.631) and passive (Z = 3.207, p = 0.001), “even between the passive conditions … in which participants did not affirm experiencing agency per se”. The interaction for agency approached significance (t = −1.950, df = 31, p = 0.060) (pp. 8–9).
- Responders: 87.5% and 78.1% for ownership (active and passive congruent), 96.9% and 93.8% for agency (active congruent and incongruent); McNemar p = 0.375 and 1.00. The authors conclude that the mean differences come from “the strength of these experiences among responders” (p. 9).
- Agency control statements rose “in the passive congruent and passive incongruent conditions” without reaching +1 (p. 9). On the rendered figure they sit near +0.3 and +0.8, above the agency scores in both passive conditions.
- Ownership and agency correlated only in the active congruent condition, “r = 0.503, n = 32, p = 0.003”; elsewhere r = 0.19 to 0.22, n.s. (pp. 9–10, Fig. 6).
Experiment 4: the same design, drift (p. 9, Fig. 7). Drift +0.77 cm active congruent, +0.6 cm passive congruent, −0.35 cm active incongruent, −0.004 cm passive incongruent. Repeated-measures ANOVA: position F(1,31) = 7.878, p < 0.01; mode F(1,31) = 0.144, p = 0.707; no interaction. In the incongruent trials “the position of the wooden index finger was kept constant” (p. 4), so the moving finger was seen at the same place in all four conditions.
Asynchrony against passive movement (p. 9). A post hoc unpaired t test across experiments: agency under passive congruent movement (−0.61) was lower than under asynchrony (0.66), t = 3.105, df = 50, p = 0.003, “demonstrating that passive movement eliminated agency more completely than visuo-somatic asynchrony”.
Interpretation (pp. 9–12).
- Ownership: “meaningful correlations among all available sensory channels are critical for body ownership rather than signals from a specific sensory modality” (p. 10). The illusion took “approximately 10–30 s” to develop, by anecdote.
- Agency: “as long as the intention to move is present, there exists an inherent tendency to attribute authorship of actions to the self, which is then cancelled by the detection of mismatches” (p. 11). “The sense of agency is best explained by a model in which both motor intentions and comparator mechanisms are essential contributing factors.”
- Body agency: three observations (stronger agency when congruent, the near-significant interaction, the correlation only in the owned condition) “provide indirect evidence that body agency is distinct from agency over external objects” (p. 11). Participants said informally they moved the model “exactly like when I move my real hand”.
- Ownership feeds agency: “even in the absence of an engaged agency mechanism, the presence of ownership still drives a residual sense of agency” (pp. 11–12).
- Agency and ownership: “Our data do not provide conclusive evidence that agency enhances ownership”; the active–passive difference in ratings may come from “general differences in task demands” (p. 12).
- Conclusion: “the first direct demonstration of the double dissociation of ownership and agency in a single rubber hand illusion paradigm” (p. 12).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Synchronous movement of a model finger with one’s own hidden finger induces ownership of the model hand and drift toward it; asynchrony removes both | moderate | Exp. 1 (n = 20) questionnaire, Z = 3.304; Exp. 2 (n = 20) drift 1.73 against −0.35 cm, Z = 3.062 |
| C2 | Passive movement abolishes agency but leaves ownership and drift | moderate | Exp. 3 (n = 32) ownership 1.57, agency −0.61; Exp. 4 (n = 32) no effect of mode on drift |
| C3 | An anatomically incongruent model hand abolishes ownership and drift but leaves agency | moderate | Exp. 3 agency 2.2, ownership −0.62; Exp. 4 position effect F = 7.878 |
| C4 | Ownership and agency are therefore distinct processes (a double dissociation) | moderate as a behavioural dissociation on report; weak as a claim about distinct processes | C2 and C3; agency is measured by questionnaire only |
| C5 | Synchrony is needed for both ownership and agency | moderate for ownership; weaker for agency | Exp. 1; agency under asynchrony was reduced to 0.66 but still above its control |
| C6 | Proprioceptive drift tracks ownership, not agency | moderate | Exp. 2 correlations (n = 17); Exp. 4 drift follows position, not mode |
| C7 | Agency over an owned hand is stronger and more tied to ownership than agency over an external object (“body agency”) | weak | the congruent–incongruent agency difference, a p = 0.06 interaction, one correlation; the authors call it indirect |
| C8 | Motor intention gives a default tendency to self-attribute that feedback mismatch cancels | weak | one post hoc, cross-experiment comparison of explicit ratings (t = 3.105, df = 50) |
| C9 | Agency does not conclusively enhance ownership | moderate as a null | drift equal for active and passive; the questionnaire difference has no interaction and may be task demand |
Concepts
- Sense of ownership — the experience of the body as one’s own, measured by four statements such as “I felt as if the rubber hand was my hand”.
- Sense of agency — “the sense of authorship over bodily movements” (p. 2), measured by four statements such as “I felt as if I was causing the movement I saw”.
- Body agency and external agency — agency over one’s own limb, constrained by the body’s geometry, against agency over an object or a tool, which rests on “learned arbitrary mappings between an action and a goal” (pp. 2, 11). The distinction is the authors’ and is offered as preliminary.
- Moving rubber hand illusion — ownership induced by correlated visual and somatic (proprioceptive, kinaesthetic, tactile) signals of movement, without brushing.
- Responder — a participant whose score is ”> 1”; a non-responder scores ”≤ 0” (p. 9).
Connections
The paper takes the rubber hand illusion from Botvinick & Cohen (1998) and the anatomical-incongruence manipulation from Ehrsson et al. (2004) and Tsakiris & Haggard (2005). Its account of agency is Frith et al.’s comparator, qualified by intention-based accounts (Haggard 2005; Wegner; Sato). It answers Tsakiris et al. (2006), who compared active and passive moving hands on drift alone and found them equal, but had no condition with agency and no ownership. It draws on Farrer et al. (2008) for the finding that delayed feedback is still judged self-generated or “biased” rather than “other”. Rohde, Di Luca & Ernst (2011), filed by a sibling reader, is cited for the risk of relying on drift alone, and Longo et al. (2008), also filed by a sibling, for the questionnaire and for a “feeling of control” in the static illusion. None of these was read here.
Bearing on the record
- THEORY-458 (Proposed). This is the direct experimental test of the
dissociation half of Gallagher’s split, and it is the one most affected.
- Its refutation clause includes “if reports of ownership and agency could not be dissociated outside involuntary movement”. The active incongruent condition dissociates them outside involuntary movement, in the direction involuntary movement cannot give: agency without ownership (C3). So this route to refutation is closed for reports of a model hand. The passive condition is involuntary movement and so repeats the known direction.
- It fits the mechanism half without testing it. Ownership followed the sensory conditions (position and synchrony) and not the motor command; agency followed the motor command. That is the pattern THEORY-458 states: ownership from sensory feedback, agency from what “precedes action”. But there is no forward-model or efference-copy measure, and the authors’ own account adds intention to the comparator (C8).
- It qualifies the split. The two are not independent: congruent position raised agency even in passive movement, and agency and ownership correlated only in the owned, active condition (C7). THEORY-458 says the two are “distinct and come apart”; this says they come apart and interact.
- It does not touch thought insertion, the patient studies or the immunity principle, which are what THEORY-458’s promote_when asks for. So it does not meet the promote_when. I would propose (proposal only) that THEORY-458’s refutation clause cite this result as having addressed the dissociation outside involuntary movement for the hand, so that what remains open is the comparator and thought.
- THEORY-765 (Proposed). It partly meets the promote_when
and does not meet its core.
- Met: a first-hand reading of a replication, in a visuomotor rather than visuo-tactile form, with sample sizes, named tests (Wilcoxon, Friedman, ANOVA, Spearman) and means and SDs for both questionnaire and drift. No standardised effect sizes are given.
- Not met: the suggestion control. The control statements are the same kind of guard the THEORY already says is insufficient (“control items guard against blanket agreement, not against suggestion specific to the touch and ownership items”). Expectations were not measured or matched, and trait suggestibility was not measured. Each participant saw all conditions of an experiment, and the predicted pattern (a rotated hand is not mine; a finger moved for me is not moved by me) is the obvious one. A double dissociation that matches what anyone would expect does not, on its own, rule out expectation. That is my reading; the paper does not discuss demand characteristics. The sibling filings of Lush 2020 and Lush et al. 2020 are the place this is argued.
- On the drift refutation condition, it bears against the alternative. Drift did not appear under asynchrony (−0.35 cm), and in Experiment 4 the moving finger was seen at the same place in all conditions yet drift appeared only with the anatomically placed hand. So drift here does not track visual exposure. It also tracked ownership ratings and not agency ratings (C6). That is a stronger behavioural tie than Botvinick & Cohen’s, though still on ratings collected after a separate, shorter induction.
- Its scope is brushing; this result is movement. I would not change the THEORY’s status. Proposal only: the promote_when could say whether a visuomotor replication counts, since it changes the stimulus the THEORY names.
- THEORY-769 (Proposed). Does not meet the promote_when,
and is consistent with the account.
- Nothing is varied parametrically and no model is fitted.
- The responder analysis is suggestive of graded ownership: similar proportions of responders for active and passive congruent movement, with different means among them (p. 9). But the responder rule leaves scores between 0 and 1 unclassified, and the comparison is post hoc.
- The anatomical-congruence effect is the prior the account invokes (a hand where no hand could be is not owned), measured here as an effect, not apart from it, which is the circularity the THEORY’s own text names.
- The real hand losing ownership, the second commitment, is not tested. The control statement “as if my right hand had disappeared” bears on it, but it was averaged into the ownership control score, which stayed negative in every condition; the item is not reported alone.
- Drift did not differ between active and passive movement, which fits the account’s treatment of motor signals as one more prediction with no special standing for ownership; the authors draw no such inference.
- THEORY-767 (Proposed). No bearing. The paper uses no interoceptive signal. It does show visuomotor correlation inducing ownership on its own, which is the signal Seth reports overshadowing the cardiac cue; it says nothing about the cardiac cue itself.
- THEORY-768 (Proposed). No bearing on promotion.
- First count: self-location and whole-body self-identification were not measured. The study is about part ownership only.
- Second count: the passive condition shows ownership without agency in adults whose agency is “intact but unused”, which the promote_when names as evidence that cannot settle the agency thesis.
- Read at the level of the hand, the results fit “causally enabling, not constitutive”: ownership arose without agency, and active movement raised ownership ratings only weakly and not drift (C9). But the THEORY’s thesis is about minimal selfhood, which by its own terms the rubber hand does not reach. The finding that ownership raises agency (C7) runs the other way from the thesis, and concerns a hand.
- THEORY-322 (Proposed). Weak support for a feedback-independent, intention-based contribution to agency: with intention present and feedback mismatched, agency was 0.66; with feedback matched and no intention, −0.61 (C8). That fits the claim that agency draws on signals from action selection, “not only” on comparison. But it is a post hoc, cross-experiment comparison of explicit judgements, which THEORY-322’s promote_when sets aside, and it does not vary outcome probability. It does not meet the promote_when.
- LIT-1009, NOTE-827 (Haggard). NOTE-827’s C1, that involuntary movement gives ownership without agency, has here an experimental instance with a model hand (C2). Haggard’s C8, that the comparator explains non-agency but not the positive experience, is close to these authors’ reading that intention supplies a default that mismatch cancels; they reach it from data, he from argument.
- NOTE-372, C6 (Metzinger, LIT-450). Supports the manipulation half again, with visuomotor rather than visuo-tactile correlation. Nothing on “simplest form of self-awareness”.
- QUESTION-048. A worked case of report and behaviour diverging. Active movement raised ownership ratings (Z = 2.765) without raising drift, and the authors decline to trust the ratings alone, calling the combination of measures their “golden standard” (p. 12). The agency half of the dissociation rests on report alone, with no implicit measure such as intentional binding.
- LIT-453, NOTE-377 (Gallagher). The paper cites Gallagher (2000) as one source of the distinction it tests.
- No instruction for machine-learning practice.
Limitations
- Report only for agency. The double dissociation is in questionnaire scores. Drift is a measure of ownership; nothing objective measures agency.
- No expectation control. The control statements guard against agreeing with everything, not against expecting the predicted pattern. Conditions were within-subject, so participants could compare them. The experimenter, who moved the rod in the passive and asynchronous conditions, knew the hypotheses.
- Asynchrony is not matched. The asynchronous movements were made by hand by the experimenter, imitating the participant, at “approximately” 500 ms; the delay was not measured or held fixed.
- No standardised effect sizes. Means and SDs are given; Z and F values are given; effect sizes are not.
- A reporting slip. The active congruent against active incongruent agency comparison is given as “Z = 2.631, p < 0.001” (p. 8). A two-tailed Z of 2.631 gives p ≈ 0.009, and Figure 5 marks the comparison ”**”, p < 0.01. It is still significant at the corrected threshold.
- Agency controls in passive conditions. The agency control scores rose in both passive conditions, and on the rendered figure exceed the agency scores there. Statements such as “the rubber hand was controlling my movements” may be partly true when an external force moves one’s finger; that reading is mine.
- Experiment 4’s questionnaire is not reported. The short questionnaire was given at the end of Experiments 2 and 4 (p. 5), but only Experiment 2’s correlations appear.
- Cross-experiment comparison. The asynchrony against passive result (C8) compares different groups post hoc.
- Small drift. In Experiment 4 drift in the owned conditions was under 1 cm, from one pointing before and one after each trial.
- “Body agency” is preliminary. The authors say themselves that it rests on “indirect evidence” and that “an important goal” is to test it objectively.
Open questions
- Does the dissociation survive with expectations measured or matched, for example by asking participants beforehand what each condition will do to them?
- Is there an implicit agency measure, such as intentional binding, that shows the same pattern, so that the agency half does not rest on report?
- Is “body agency” a distinct process, or does ownership only raise ratings of agency, as it raised them in passive conditions where no agency was felt?
- Does the real hand lose ownership as the model gains it? The disappearing-hand item could be analysed alone.
- Does a moving rubber hand reach whole-body self-location or self-identification, which THEORY-768 says the static one does not?