Does somatic feedback guide decisions?

Opened with the Bechara & Damasio (2005) ingest — the wiki’s first primary source for the somatic-marker-hypothesis, which until now was present only through other people’s readings.

The lopsidedness is fixed. This page carried a standing warning that the proponents’ case was here first-hand and the critics’ only through Friedman’s (2010) one-line mention, and told readers not to mistake the asymmetry for support. The Dunn, Dalgleish & Lawrence (2006) ingest supplies the critics’ case in full, plus a detailed secondhand account of Maia & McClelland (still not in raw/). Both sides are now read.

What the critics actually say is more interesting than ‘no.’ Dunn et al. return a split verdict — the neural substrate survives, the psychological mechanism does not — and decline to call it a falsification. The useful consequence for this page is that the ‘three questions wearing one name’ framing below is not the wiki’s editorial convenience. It is roughly the structure the critical literature itself found, arriving independently and twenty years earlier.

Three questions wearing one name

The debate is usually conducted as though it were one dispute. It is at least three, and they can come apart:

1. Is the anticipatory somatic signal doing causal work, or is it an epiphenomenon of knowledge? This is the Maia & McClelland challenge. Bechara et al.’s (1997) design probes knowledge every ten cards with open questions; the objection is that this underestimates what subjects know, and that with finer probing, explicit knowledge arrives early enough to explain the behavioural shift without any unconscious bias. If that is right, the anticipatory SCR is a correlate of emerging knowledge rather than its precursor, and the framework’s headline claim dissolves.

Note what would not rescue the hypothesis: the VM patient data. Those show that damage to a region abolishes both the signal and good performance — consistent with the somatic marker account, and equally consistent with the VM cortex doing something else entirely that both the SCR and the choices depend on. The pre-hunch timing is what makes the causal claim, and the pre-hunch timing is what is contested.

The challenge can now be stated with its design, via Dunn et al. (2006). Focused questions after each block of 20 trials, n = 20: advantageous performance was nearly always accompanied by reportable knowledge sufficient to guide behaviour. The finding inside the finding is the one that bites — subjects reported accurate knowledge more reliably than they chose advantageously. Knowing the right thing and doing the wrong thing is what the Iowa programme treats as the signature of a missing marker in VM patients. If healthy controls do it too, it stops being diagnostic. And the obvious rebuttal was pre-empted: a second group of 20 given Bechara’s original open questions behaved identically, so the fine probing detected knowledge rather than creating it.

Damasio anticipated this objection in 1996 and answered it badly. Damasio (1996) names the alternative precisely — that normal subjects reason early on that certain decks are bad and, on the basis of that cognition, generate a somatic response the SCR indexes — eight years before the critique was published. He rejects it on two grounds:

1. His perspective on the evolutionary biology and adaptive value of an automated somatic marker device. This is not evidence about the task; it is a prior about what the mechanism ought to be for.
2. “Recent studies in our laboratory” showing normals produce anticipatory SCRs to bad decks long before they can testify to any notion of the decks’ nature — cited to nothing, a year before Bechara et al. (1997) supplied the four-period knowledge probe that the 2005 review rests the claim on.

So the state of play is worth stating exactly: the objection is not one the framework overlooked, and the framework’s answer at the time was an evolutionary argument plus an uncited in-house observation. What later made the answer respectable was Bechara et al. (1997) — which is precisely the study Maia & McClelland attack. The pre-hunch claim was believed before it was measured.

And the eventual answer gave the claim away. Bechara et al. (2005, TICS) reply that implicitness was never the central feature — ‘emotion-related signals assist cognitive processes even when they are non-conscious.’ That survives the penetrability result, but Dunn et al. note the cost: it retreats from Tranel et al. (1999) and ‘makes the SMH hard to distinguish from other accounts of decision-making.’ A theory that emotion influences decisions, sometimes non-consciously, is not one many people dispute. The interesting claim was the strong one.

One caveat the wiki should keep. Bechara et al. also argue the SCRs appear early enough to precede even the knowledge Maia & McClelland detect; Maia & McClelland (2005) reply that insight data from different labs with very different probes may not be comparable at all. Nobody has run the timing comparison within a single study. So question 1 is damaged, not closed.

2. Does the signal have to be felt? The framework’s answer is no — and this is where it parts company with the rest of this wiki. The striatal level is explicitly covert (“knowledge without awareness”); the anticipatory SCRs of the pre-hunch period bias choice “without any awareness of why the choice was made.” So the somatic marker hypothesis is a theory about somatic signalling, and interoception enters only as an assumption.

This matters more than it sounds. SCR is an efferent measure: it records what the body did, not what the subject perceived. Nothing in the Iowa programme establishes that anyone feels the anticipatory response. So the hypothesis could be entirely true and tell us nothing about interoception — or interoception could turn out to be the moderator that decides when it holds. Which of those is the case is question 3.

2b. Does the signal have to be bodily? Added with the Damasio (1996) ingest, which forced this page to split question 2 in half. The wiki had these merged and they are different retreats:

• Need not be felt — the version argued with above. This is still a bodily theory: real peripheral events, unperceived, biasing choice. SCR is an appropriate measure for it, and the framework’s evidence base fits its claims.
• Need not be peripheral — Damasio’s 1996 position. The as-if loop is the mode most frequently in operation, the periphery is bypassed, and only the neural structures representing body states are modified. This is not a bodily theory. It is a theory about somatosensory cortex, in which the body is a developmental scaffold subsequently dispensed with.

Asked point-blank by Everitt whether manipulating the periphery would change gambling behaviour, Damasio said it should have relatively little impact. That is the founder of this framework, in 1996, telling an experimentalist not to bother with the body.

And it was still the position in 2000. The Bechara, Damasio & Damasio (2000) ingest supplies a midpoint reading, and the sentence is verbatim continuous with 1996: “From both evolutionary and ontogenetic perspectives, the ‘body loop’ is the original mechanism but has been superseded by the ‘as-if body loop’ and is possibly used less frequently than it.”

That matters for how this page reads the 2005 indexing claim. The wiki had two datapoints and could describe the change as a trajectory of unknown shape. With three, the shape is legible: the as-if loop is the official default across nine years and two papers, and the body loop’s job — ambiguity engages it — arrives only in the third. So “the somatic marker hypothesis holds that bodily signals guide decisions” is not the framework’s constant commitment being progressively specified. It is a claim the framework’s own primary statements largely declined to make until 2005, and the one thing that ever made it falsifiable arrived with it.

Two problems follow, and the second is worse.

First, the framework’s central measure is instrumented on the wrong loop. SCR records a peripheral event; if the as-if loop usually runs, the absence of an SCR says nothing about whether a marker occurred, and its presence indexes the exceptional case. Neither Damasio nor Bechara notes this.

Second — and this is why the 1996 position is not merely a historical curiosity — it means the framework’s most-cited claim is better supported by the later paper than the earlier one. The 2005 indexing gives the body a job back. Read forward, that looks less like a standing view finally specified and more like a reversal, unflagged in either text. So the wiki should stop treating “the somatic marker hypothesis says bodily signals guide decisions” as the framework’s constant commitment. It is the 2005 commitment. In 1996 the same author held that they mostly do not.

3. Does perceiving bodily signals help? The question this wiki actually cares about, and the one the Iowa programme never asked. It is also where the answer is least likely to be a clean yes: is-more-interoceptive-awareness-better already records several sources in which more bodily contact accompanies worse outcomes.

The critics named this gap in 2006, and the wiki should stop presenting it as its own observation. Dunn et al. close their paper by proposing that “an individual differences approach could be adopted, for example looking at how variance in interoceptive ability relates to IGT performance” (citing Katkin et al. 2001). That is question 3, identified as the missing programme. Four pages in this wiki independently observe that the Iowa programme measures somatic signalling and never asks whether anyone perceives the signal; the observation was in print before any of them.

And it has now been answered — by the same author, and the answer is “neither”

The Dunn et al. (2010) ingest closes this item. Four years after naming the missing study, Dunn ran it: heartbeat counting paired with a decision task, in 92 participants, with anticipatory EDA and HR recorded before each choice.

The result is not yes and not no.

interoceptive accuracy ↔ decision quality	r = .08, p = .46
interoceptive accuracy ↔ how informative one’s own body was	r = .07, p = .56
bodily differentiation ↔ decision quality	r = .41, p < .001
bodily differentiation × interoceptive accuracy → decision quality	ΔR² = .08, p < .01

Perceiving the body accurately has no relationship whatever to deciding well. What it does is set how strongly the body’s signal transmits: better perceivers decided better when their anticipatory signals favoured the profitable decks, and worse when those signals favoured the unprofitable ones. Interoception is a gain term, not a benefit — it amplifies signal and noise indifferently, because nothing in the perceiving distinguishes them.

So question 3’s fifteen-year-old framing (“does perceiving help?”) is malformed in the way the answer reveals. Perceiving does not help or hurt. It determines how much whatever your body is saying gets to matter — and the body’s accuracy is a separate variable that interoception neither improves nor tracks (r = .07).

Why this bears on question 1 as well. The moderation is an argument for causal work that no main-effect design could make. The standing critic’s position is that bodily responses arrive late and are consequences rather than causes of cognitive-affective activity (Moors 2009). If that is right, how well someone perceives their body should be irrelevant to how tightly bodily change couples to choice — you cannot perceive an epiphenomenon back into causal relevance. Interoceptive accuracy moves the coupling. That is the best structural argument for somatic influence anywhere in this wiki, and it comes from the framework’s principal critic.

Two things stop it settling anything. The alternative Dunn et al. never consider: interoceptive accuracy is confounded with cardiodynamics (stroke volume predicts heartbeat-counting performance — heartbeat-detection-task), so “good perceiver” partly means “loud heart,” and a loud heart is also a bigger cardiac signal to couple to report. The interaction might be a signal-strength effect wearing a perceptual label. And the paper reports two unreplicated interactions at n = 58 and n = 92, with no simple-slope tests, in 2010 — see the limitations on dunn-2010-listening-to-your-heart.

Also: it is not the IGT. Study 2 uses the intuitive-reasoning-task, a task Dunn built to repair the IGT’s flaws. Better science; weaker bridge. This does not tell you what interoception does to Iowa data, and nobody has run the version that would.

The proponents’ own datum, and a correction

The sharpest datum from the other side. Shiv, Loewenstein, Bechara, Damasio & Damasio found that in an investment task where normal individuals drop out because of anxiety after a streak of losses, the poor somatic reaction of neurological patients let them keep investing and outperform normals.

Correction, from the Dunn et al. (2010) ingest: this is not unpublished. This page and is-more-interoceptive-awareness-better both recorded it as “unpublished observations” reported in Bechara & Damasio (2005), and instructed the wiki to upgrade or drop the claim when Shiv et al. became available. It was already available — Investment behavior and the negative side of emotion, Psychological Science, 16, 435–439 (2005) — cited in full and endorsed by Dunn et al. (2010), who read it as agreeing with their own result. The wiki was carrying a published Psych Science paper as an unsourced in-house observation for three ingests, because the review it first met the finding in described it that way. It is still not in raw/, so it still cannot be read first-hand, and the n and statistics remain unknown here. But it is a citable result, not a rumour.

Read carefully, that is not a small caveat. The framework’s central patient group — defined by a somatic deficit, characterized across fifteen years as impaired — outperforms healthy controls under a specifiable condition. Bechara & Damasio absorb it as consistent (the somatic state was unrelated to the optimal strategy, hence disruptive; §2.4’s integral/incidental distinction). That absorption is available, but notice what it costs: whether a given somatic state counts as “integral” or “unrelated” is determined after seeing whether it helped. The distinction has no independent criterion in this paper.

Everitt’s experiment has been run, and the wiki had this wrong

The single largest correction the Dunn et al. (2006) ingest forces on this page.

This page and body-loop-and-as-if-body-loop both recorded B. J. Everitt asking Damasio, in the 1996 Royal Society discussion, whether experimentally manipulating peripheral somatic change would alter gambling-task behaviour — and both said the experiment was “thirty years old and still unrun.” It was run. By 2005 it had been attempted four times, including by the two routes Damasio himself named in his reply:

manipulation	SMH prediction	result
peripheral neuropathy (Bechara et al. 1998, n = 20)	impaired	mildly impaired — but an unrefereed conference abstract
pure autonomic failure (Heims et al. 2004, n = 6)	impaired	outperformed controls
spinal cord section at C6 (North & O’Carroll 2001)	impaired	no deficit
vagus nerve stimulation (Martin et al. 2004)	improved	no overall group difference

The PAF result is the one to hold. Pure autonomic failure denervates autonomic neurons peripherally — the body loop is cut. The patients did not merely fail to show a deficit, they were significantly superior to controls, which is why the null cannot be dismissed as low power (it is not a null). And Critchley et al.’s (2003) VBM finding that long-standing PAF also thins grey matter in ACC and insula closes the obvious escape: these patients have a compromised body loop and a degraded substrate for the as-if loop, and still decide well.

Which Damasio does this vindicate? The 1996 one. He predicted the periphery could be manipulated with relatively little behavioural impact, and on the gambling task that is roughly what happened. The 2005 indexing claim — ambiguity engages the body loop — predicts the opposite, and the gambling task is its paradigm case of ambiguity. So the framework’s two primary statements make opposite predictions here, the experiment has been done, and the answer favours the statement in which the body barely matters.

Two honest qualifications, both from Dunn et al. themselves. The samples are small — Heims’ n = 6 is the same order as the six VM patients whose sample size the wiki criticizes on iowa-gambling-task, and consistency requires saying so. And long-standing cord damage or PAF may have permitted adaptive reliance on the as-if loop, so a deficit might yet appear if patients were tested immediately post-injury. That second qualification is itself the escape hatch below, which is Dunn et al.’s point.

The framework’s own escape hatches, listed

Recorded not as an accusation but because a reader of this wiki should know where the theory is unfalsifiable-as-stated and where it is not.

This analysis is not the wiki’s own, and the record should say so. The table below was assembled from reading Bechara & Damasio and Damasio (1996) directly. Dunn et al. (2006) reached the same structural conclusion twenty years earlier, from the empirical side, with more of the primary literature in hand:

Nearly all of the negative findings can be explained away through some aspects of peripheral feedback remaining intact in the patient groups studied (for example, the somatic system in PAF, endocrine feedback following spinal injury) or through recourse to the ‘as-if’ loop. It seems virtually impossible to test the theory in a scenario where all peripheral feedback routes are disturbed and the ‘as-if’ loop cannot be utilised, therefore making it difficult to disprove.

Independent arrival at the same complaint is worth more than either statement alone — and it means the wiki should present this as corroborated, not as a reading of its own.

move	what it absorbs	is it constrained?
the as-if body loop (body-loop-and-as-if-body-loop)	any decision showing no peripheral somatic change	Yes, in principle — the certainty→as-if / ambiguity→body mapping forbids appealing to as-if under ambiguity. But that mapping is one preliminary, statistics-free comparison across two dissimilar tasks. If it fails, this is a free parameter.
integral vs. unrelated emotion (§2.4)	any case where emotion hurt decisions	No. No independent criterion is offered; the classification follows the result.
background states (background-somatic-states)	any case where a triggered signal failed to bias	Partly — the signal-to-noise model makes a real prediction (framing effects modulated by streaks), which the paper does not test.
posterior/anterior VM gradients	any lesion result	No, as currently specified. Four gradients on one axis, never dissociated, each invoked where it fits. See ventromedial-prefrontal-cortex.
the as-if loop as default (Damasio 1996)	the periphery, entirely	Not an escape hatch — an advertised mechanism. In 1996 the as-if loop is not invoked when the body is quiet; it is what the body loop was superseded by. The 2005 indexing is the only thing that ever constrained it, and it postdates this position by nine years.

The as-if row is the important one, and it is why body-loop-and-as-if-body-loop flags the loop-indexing claim as the most useful open empirical question the hypothesis poses to interoception research. With the indexing, the framework is falsifiable. Without it, “somatic marker” is satisfiable with no body involved — and a bodily theory of decision that does not need the body has given the game away.

The fourth row changes the character of that complaint. The wiki had been treating the as-if loop as an unfalsifiability risk — something a defender could reach for. The 1996 text shows it was, at one point, the stated default. So the objection is not “this theory could be defended without a body if pressed”; it is “this theory was, for at least one of its two primary statements, a theory without much body in it, and said so when asked.” That is a stronger claim and the wiki can source it.

What would settle it

Concretely, and with the wiki’s own queue in mind. Revised after the Dunn et al. (2006) ingest, which closed two of these and downgraded a third.

• Everitt’s experiment, which is thirty years old and still unrun. Done — see the section above. Four attempts by 2005; the aggregate favours Damasio’s 1996 prediction of little peripheral impact over the 2005 indexing claim. What remains open is the acute version: every patient group tested had long-standing damage and may have adapted onto the as-if loop. Dunn et al.’s proposed successors are the studies to want — the peripherally-acting beta-blocker nadolol in healthy volunteers, patients with facial paralysis (Keillor et al. 2002), and peripheral blockade administered within PAF patients to close the residual somatic routes that let Heims et al.’s result be explained away.
• Read Maia & McClelland (2004). Now available in detail secondhand, via Dunn et al.’s report — design, n, control condition, and the reply exchange. Still worth reading in the original, and still not in raw/, so quote Dunn’s account rather than the source. Downgraded from “the single most load-bearing absent source on this page.”
• The within-study timing comparison nobody has run. Bechara et al. say SCRs precede even finely-probed knowledge; Maia & McClelland (2005) say cross-laboratory insight data are not comparable. Both are right that the existing evidence cannot settle it. One study, one probe, SCR and knowledge measured together, would.
• Ingest Dunn et al. (2010). Done — see question 3 above. Question 3 asked directly, by the author who named the design as the missing programme in 2006. The answer: interoception does not relate to decision quality (r = .08); it moderates (ΔR² = .08). See dunn-2010-listening-to-your-heart, intuitive-reasoning-task.
• Ingest Werner et al. (2009) Done — see werner-2009-cardiac-perception-decision-making and the Werner/Schandry field position above. The prospective conflict is real but does not resolve in the proponents’ favour, and the reconciliation the queue note guessed (“Werner’s task supplied bodies that were mostly right — Dunn’s moderation restated”) is unsupportable from this paper: Werner never measured the per-choice body, so it cannot instantiate the moderation. What it is instead is a main-effect, extreme-groups study — the decision-quality twin of Pollatos’s arousal main effect, on Schandry’s own instrument — so the Werner–Dunn disagreement is the Pollatos–Dunn disagreement again: selected sample finds the main effect, unselected sample does not, and sampling inflation and mixture purification both predict that pattern with no fact about deciding settled. The IGT-with-marker-differentiation study — Werner’s design plus Dunn’s per-choice bodily term — still has not been run, so the moderation has never been measured on the paradigm that generated this literature.
• A replication of the Dunn moderation, in anyone else’s laboratory. The single most valuable absent study on this page. Two interactions at n = 58 and n = 92, from the lab that built both the hypothesis and the task, 2010, never repeated in this wiki’s sources. The claim is load-bearing enough — it is the only evidence here that interoception does anything to decision-making at all — that it should not rest on one paper.
• Simple-slope tests, or the data to run them. Dunn et al. plot regression lines at ±1 SD and never test them against zero, so it is unknown whether poor perceivers are uncoupled from their bodies or inversely coupled. Both figures show the worse-interoception line sloping negative. Those are different theories; the existing report cannot distinguish them. See dunn-2010-listening-to-your-heart.
• The IGT version of the Dunn study. Dunn tested interoception against the intuitive-reasoning-task, not the Iowa task, so the moderation has never been measured on the paradigm that generated the entire literature this page is about.
• A test of the loop indexing. Whether ambiguity really engages the body loop and certainty the as-if loop, on tasks matched for everything except epistemic character. No source in this wiki does this, and the PAF result makes it more urgent rather than less: if the body loop is what ambiguity recruits, patients without one should have failed the most ambiguous task in the literature, and they did not.
• An IGT that restores impenetrability. Dunn et al.’s suggestion, and the constructive form of question 1: a variant using an artificial-grammar structure, with the reversal confound removed, deck position counterbalanced, and decks classified by running rather than eventual net outcome. If the pre-hunch effect survives that, it means something.
• An insula-lesion study that measures the body. Added with the Jones et al. (2010) ingest and the field position above. Three lesion studies target the insula and produce a decision deficit; none records anticipatory SCR, heart rate, or any bodily term, and none measures interoception. So the wiki now has causal evidence that damaging interoceptive cortex degrades risky choice, and no evidence about why — the deficit is equally consistent with losing the somatic signal, losing the perception of it, and losing an expected-value computation that has nothing to do with the body. Clark’s design (focal insula vs. focal vmPFC vs. dorsolateral control) plus Dunn’s per-choice bodily-differentiation term would be the study.
• A discriminating test of integral vs. unrelated emotion — i.e. any criterion for the distinction that can be applied before the outcome is known.
• Any multi-channel autonomic measure. The framework asserts positive and negative somatic states are physiologically distinguishable and cites Cacioppo et al. (2000) for it; SCR alone could never show it. Dunn et al. make the same demand independently — facial EMG and ECG alongside SCR — and add a subtler reason: bodily responses to environmental demand cannot be modelled on a single arousal dimension at all (Lacey 1967), so an isolated SCR change may be one limb of a response complex (Graham 1979) rather than a scalar to be read up or down. See autonomic-specificity-of-emotion.

Why this is not the same as the wiki’s other debates

• autonomic-specificity-of-emotion asks whether emotions have distinct bodily signatures. This page asks whether bodily signatures guide choice — a hypothesis that could be true even if every emotion had the same signature, since a single valence dimension would suffice to bias.
• feedforward-vs-predictive-interoception asks which way the signals run. The somatic marker hypothesis has no position on it: the as-if loop activates stored patterns “directly,” and the source never says whether that is prediction, memory read-out, or efference copy.
• is-more-interoceptive-awareness-better asks whether bodily contact is good for wellbeing. This page asks whether it is good for deciding. The Shiv et al. observation sits on both, and is the reason the two pages cross-reference.
• where-are-feelings-constituted asks where feelings happen. This page is largely about signals that are never felt at all — which is precisely why the somatic marker hypothesis has less to say to this wiki than its citation count suggests.