Event-related potentials

The wiki’s first electrocortical method, arriving with Pollatos, Kirsch & Schandry (2005). Its place in this collection is defined by contrast: every other brain measure here (fMRI, PET, lesion, morphometry) trades time for space. ERP trades space for time.

The components that matter in this literature

• N100 (~100-150 ms) — early sensory/perceptual response. In Pollatos et al. interoceptive accuracy had no effect here, which is the paper’s cleanest inference: whatever interoception does to emotional processing, it does late.
• P300 (~290-500 ms) — a large positive deflection, maximal at posterior/parietal sites for visual stimuli. Read as an index of attention, processing capacity, motivational relevance, or task difficulty — the ambiguity is the problem. Larger for emotional than neutral pictures.
• Late positive slow wave (~400 ms onward) — sustained positivity, sensitive to emotional content, interpreted as continued perceptual processing and allocation of resources to motivationally relevant input (Kok 1997; Keil et al. 2002).

Why the N100 null is the most defensible thing ERP contributed here

Worth isolating, because it runs against the usual reading of null results.

Pollatos et al. make several anatomical claims (insula, ACC, medial prefrontal, somatosensory cortices) that their method cannot support — they are imported from Damasio et al. (2000) and PET/fMRI work, and the paper concedes that P300 generator studies disagree with each other before reasoning from them anyway. Those claims are decoration.

The N100 null is not. It requires no source model at all: if interoceptive accuracy modulated the earliest cortical response, an ERP would show it, and it did not. That licenses “interoception enters late” — a claim about timing, which is the one thing this method is authoritative about. On Damasio’s staging that points away from emotion trigger sites and toward first- and second-order structures.

The general lesson for reading this wiki’s ERP content: trust the latency, discount the anatomy.

The between-groups amplitude problem

A caution Pollatos et al. do not raise and which bears on their central ERP finding.

Their P300 effect is a comparison between people — good vs poor heartbeat perceivers — not between conditions within a person. Raw ERP amplitude varies between individuals for reasons with nothing to do with psychology: skull thickness, cortical geometry, electrode impedance, and tonic arousal all scale the recorded voltage. A within-subject condition contrast cancels these; a between-groups contrast does not.

This is exactly why the emotion-specificity of an effect carries the interpretive weight. Their P300 difference appeared across all 60 slides including neutral household objects, which is what a generic amplitude difference between two groups of people looks like. Their slow-wave difference appeared for affective slides and not neutral ones, which is not — a nuisance factor scaling everything cannot produce a condition-selective effect. The slow wave is the finding; the P300 main effect is a group difference of undetermined origin.

What an ERP component is, on the predictive-coding reading (Friston 2005)

Everything above treats components as non-specific indices — the P300 “indexes attention, capacity, relevance and difficulty at once,” and the honest response is to discount the anatomy and trust the latency. Friston’s “A theory of cortical responses” (2005) makes a stronger, riskier claim: an evoked response is prediction error being explained away. Decompose the response into error units and representational units; higher levels take time to settle on the cause, so the error signal at each level waxes and is then suppressed as the explanation arrives — a damped transient, with late components reflecting inference about higher-order, more global causes.

The mismatch negativity (MMN) is the sharp case. On this reading it is not the output of dedicated change-detection neurons; it is the failure to suppress prediction error to a deviant that violates a learned regularity, and it shrinks over repeated standards (“roving” paradigm) because that is the M-step — learning-related plasticity in backward and lateral connections — happening in experimental time. The pharmacology fits: ketamine (NMDA antagonism, and NMDA carries the slow plasticity the M-step needs) cuts MMN amplitude ~20%, which Friston ties to a disconnection account of schizophrenia.

This says what the component is — at the price of committing to the predictive-coding architecture to say it. That price is exactly the one this page keeps flagging. The MMN-as-prediction-error claim is a theoretical interpretation of scalp timing plus a source model (Friston’s own demonstration reaches it via dynamic causal modelling, not from the scalp waveform), so it inherits both cautions: trust the latency, discount the anatomy, and read the mechanistic story as a hypothesis the DCM makes plausible rather than as something the ERP shows on its own. See predictive-coding.

The cardiac ERP: the heartbeat-evoked potential

Everything above locks the average to an external stimulus. The heartbeat-evoked potential (HEP) is the same machinery locked to an internal one — the R-wave of the participant’s own ECG — and it is the wiki’s one electrocortical measure of interoception, arriving first-hand with García-Cordero et al. (2017). It inherits the two cautions this page insists on (trust the latency, discount the anatomy; distrust between-groups amplitude), and adds one with no counterpart in visual ERP: the cardiac field artefact, the heartbeat’s own electrical field projected onto the scalp at exactly the latency the HEP occupies, which must be removed before the signal can be read. See its page.

Relation to the rest of the wiki’s toolkit

ERP measures the central response to an emotional stimulus. It says nothing about the peripheral response — and Pollatos et al. recorded ECG without ever analysing it against the pictures, so their study has cortex and self-report and no body in between. Compare skin-conductance-response and the cardiac measures in Dunn et al. (2010), which have the body and no cortex. No study in this wiki yet has both.