Skin conductance response (SCR)

The operational definition of “somatic state activation” in the Iowa programme, and the measure on which the somatic-marker-hypothesis rests almost entirely. Also, in a different guise, the physiological half of self-report-physiology-congruence — the wiki’s two electrodermal literatures, arriving from opposite directions.

The anticipatory SCR

The reason this measure matters here. In the iowa-gambling-task, the interval between two card selections is divided into two windows (Bechara & Damasio 2005, Fig. 4):

• Reward/Punishment (R/P) — 5 s immediately after a card is turned. Responses here are reactions to winning or losing.
• Anticipatory — the period before the next selection, while the subject is pondering which deck to choose.

The anticipatory window is the innovation. A response in it cannot be a reaction to an outcome; it precedes the choice. Normal subjects, once experienced with the task, generate anticipatory SCRs that are larger before the risky decks — and they do so during the pre-hunch period, before any conscious knowledge of which decks are bad. This is the observation that licenses calling the somatic signal a bias rather than a consequence.

Three reasons that licence is doubtful

Added with the Dunn et al. (2006) ingest, and they attack the window itself rather than the valence limitation this page already recorded. The distinction matters: the criticisms below apply to the results the framework actually rests on, not to its superstructure.

1. The response may track variance, not value. Tomb et al. (2002) rebuilt the task so the advantageous decks carried the larger rewards and punishments. Subjects still learned the good decks — and now showed larger anticipatory SCRs before the good decks. In the original design the bad decks are also the high-variance ones, so “SCR marks badness” and “SCR tracks magnitude/variance” were confounded from the start. Damasio et al. (2002) replied that markers record positive futures too, and that the effect may be a non-conscious danger signal overridden by conscious assessment — a reply which, note, requires the SCR to mean one thing when it fits and another when it does not.
2. The response is not necessary. Crone et al. (2004), n = 96: only the best-performing third showed anticipatory SCR (and heart-rate) differentiation. The moderate performers learned the task with no psychophysiological differentiation at all.
3. The window may not be anticipatory. Amiez et al. (2003), in rhesus monkeys on an equivalent task, found SCRs tracking anticipation of reward after the decision rather than before it. Given SCR’s slow, smeared time course (above), “prior to selection” and “prior to feedback” are hard to separate at all. Dunn et al.’s proposed fix is the right one: acquire at higher temporal resolution and decompose tonic from phasic components (Lim et al. 1997), or use a faster-responding channel.

Point 3 is the one that should worry a reader of this page, because it is not a limitation of SCR in general — it is the possibility that the specific window this measure was celebrated for isolating does not isolate what it claims. Combined with the iowa-gambling-task’s attentional-sweep problem (the response is labelled by the deck eventually chosen, though subjects scan all four), the anticipatory SCR may be indexing a post-decisional expectancy attached to a choice already made.

Point 1 has since been answered, in a task built for the purpose

The variance interpretation is the one item above that a later study addressed head-on, and the wiki should record the outcome rather than leaving the objection standing at full strength.

Tomb et al.’s objection has force because in the IGT the disadvantageous decks are also the high-variance decks — “SCR marks badness” and “SCR tracks magnitude” are confounded by construction. Dunn et al. (2010) built the intuitive-reasoning-task with magnitude crossed against profitability in a 2 × 2, and recorded anticipatory EDA before each choice in 92 participants. Bodily responses tracked profitability and were entirely unrelated to magnitude — no main effect of magnitude, no magnitude × profitability interaction, Fs < 1. The composite differentiation index then correlated with decision quality at r = .41, p < .001.

So the electrodermal response can mark goodness rather than variance, when the two are pulled apart. Note carefully what this does and does not do:

• It does not decontaminate IGT data. In the Iowa task the confound is exactly as Tomb et al. described, and remains so in every result derived from it. This is a new task answering the question the old task could not.
• It is one study, and the EDA main effect carrying half the composite is p = .05, reported as “a strong trend toward a significantly greater anticipatory EDA maximum response” — which at the paper’s own two-tailed alpha it is not. The composite is carried by the cardiac half (p < .01).

Still, it is the best evidence in this wiki that the anticipatory electrodermal response indexes something about an option’s long-run value, and it comes from the laboratory that catalogued the reasons to doubt it.

The startle/deep-breath control, and why it matters

The Iowa programme’s other SCR experiment, recorded only in Damasio (1996) (from Damasio et al. 1990, 1991; Tranel 1994; Tranel et al. 1995). Three groups — normal controls, patients with non-frontal lesions, patients with ventromedial lesions — viewed emotionally neutral images (landscapes, abstract patterns) and emotionally charged ones (social catastrophe, body mutilation).

The design’s good part is the control condition. Autonomic responsivity was independently established in all three groups using startling stimuli (loud noises) and deep breath, both of which reliably elicit SCRs. All three groups responded normally. Only then does the experimental result mean anything: the ventromedial group alone failed to react to the emotionally charged images.

This is worth recording as a methodological template, not just a finding. The single most obvious confound in any between-group SCR comparison is that one group’s electrodermal system is simply less responsive — the non-responder minority and the enormous between-individual variance noted below make it a live risk in every such study. Establishing normal SCRs to a non-emotional elicitor first converts “these patients don’t respond” into “these patients don’t respond to this.” The deficit is in emotional evocation, not autonomic capacity.

No n’s are reported for this experiment in the 1996 paper, and the neutral-image condition’s floor performance means the effect is carried entirely by the charged images.

The SCR depends on what you ask the subject to do

The sharpest methodological caution in this material, from Bechara, Damasio & Damasio (2000) reporting Damasio et al. (1990), and it qualifies the startle/deep-breath template above rather than replacing it.

VM patients failed to generate SCRs to emotionally charged pictures when they viewed them passively. The same patients generated normal-magnitude SCRs to the same pictures when asked to view and describe their content.

So the deficit established by the 1996 experiment is not “these patients do not respond to emotional images.” It is “these patients do not respond to emotional images under one instruction.” The startle/deep-breath control establishes that the electrodermal system works; it does not establish that a null under passive viewing is a deficit in emotional evocation rather than in spontaneously engaging with the stimulus. Those are different claims and the framework needs the first.

The same source adds a related gradation: during the gambling task VM patients did generate SCRs when they lost a large sum of money, but never as large as controls’. Together the two observations suggest a weakened and instruction-sensitive emotional responsivity rather than an absent one — which is also what the paper’s emotional-imagery study found from the other direction (anger reliably re-experienced, happiness and sadness mostly not).

Two consequences for reading this literature:

1. Between-group SCR nulls in this programme are task-set dependent and should not be reported as capacity claims. Very few of the studies specify what subjects were instructed to do with the stimulus.
2. It gives the framework a live alternative it never addresses: if the VM patient responds normally when directed to engage, the anticipatory failure in the gambling task might be a failure to engage spontaneously with an imagined outcome rather than a failure to generate a somatic state at all. Nobody has run the gambling task with an instruction to describe each deck’s prospects aloud before choosing — which would be the discriminating version.

What it can and cannot say

The measure’s value and its ceiling come from the same fact: it is one sympathetic channel.

It can say: a somatic state was activated, at roughly this magnitude, at this moment, without the subject reporting it. That is enough to establish the somatic marker hypothesis’s core claim — that a bodily signal precedes and predicts advantageous choice, and that its absence accompanies bad choice.

It cannot say: which state, of what valence. And this bites harder than it first appears, because the somatic marker framework’s later apparatus needs valence discrimination. The background-somatic-states signal-to-noise model is built entirely on congruous vs. incongruous states — undefined unless positive and negative somatic states are physiologically distinct. Bechara & Damasio assert the distinction (§4.1.1) and cite Cacioppo et al. (2000) for it. They do not, and with SCR alone could not, show it in their own data.

So the framework’s empirical core is SCR-supported and its theoretical superstructure is not. Recorded on autonomic-specificity-of-emotion, where this is the same gap Friedman (2010) identifies from the Jamesian side.

The inference it does not license

Worth stating plainly, because it is the trap this wiki is positioned to fall into. SCR is an efferent measure. It records what the body did. It does not record what the subject perceived, and nothing in the Iowa programme establishes that the anticipatory SCR is felt, or that people who feel bodily signals more accurately generate more useful ones.

The somatic marker hypothesis is often read here and elsewhere as an interoception result. On this evidence it is not one — it is a result about somatic signalling, with the perceiving left as an assumption. Closing that gap is what motivated pairing decision tasks with the heartbeat-detection-task, and it is the live question on does-somatic-feedback-guide-decisions.

A user of the measure concedes the point, in a footnote. Dunn et al. (2010), note 3: “We conceptualize EDA as reflecting somatic marker activity, consistent with previous work on the Iowa gambling task… The source of feedback from EDA to the brain is currently unclear, however, and requires further specification.”

That is the whole problem in two sentences, stated by someone building a study on the measure anyway. Eccrine sweat-gland activity is an efferent sympathetic event; no one has identified an afferent route by which it returns to the brain to be felt. So an index treated throughout this literature as the bodily signal that biases choice has no established path back to the chooser.

The awkwardness is sharpest in Dunn et al.’s own design, and they do not remark on it: the moderator is cardiac (heartbeat counting), while half the moderated signal is electrodermal. The paper’s claim is that people who perceive their bodies well are more influenced by their bodily responses — averaged across a channel they can plausibly feel and a channel that, by their own footnote, may not be perceptible at all. That the interaction survives when EDA and HR are analysed separately (reported, ps < .05) is the right check and partly rescues it. But the composite remains a sum of a signal and a rumour.

And there is a second inference the measure does not license, which the 1996 source makes urgent. SCR records a peripheral event. But Damasio (1996) holds that the as-if body loop — in which the periphery is bypassed entirely and only somatosensory representations change — is the mode most frequently in operation. If that is right, then an electrodermal measure is instrumented on the arm of the loop the theory says is usually not running, and the absence of an SCR is uninformative about whether a somatic marker occurred. The framework’s central measure and its central mechanism are not matched. Neither Damasio nor Bechara notes this, and it is a sharper problem than the valence limitation above, because it applies to the results the framework actually rests on rather than to its superstructure.

The wiki’s other electrodermal literature

Oldroyd et al. (2019) use electrodermal activity for a different purpose: not as an index of bodily state, but as the objective term in a correspondence — within-person congruence between electrodermal change and moment-to-moment self-report, as a proxy for interoceptive coherence usable in children. See self-report-physiology-congruence.

The contrast is instructive. Bechara & Damasio use SCR to show a bodily signal exists and biases behaviour without being perceived. Oldroyd et al. use the same signal to ask whether it is perceived. Same electrodes, opposite question — and only the second is about interoception in this wiki’s sense.