Stimulus-Preceding Negativity (SPN) Overview
- Stimulus-Preceding Negativity is a slow cortical ERP marked by a negative shift preceding anticipated stimuli and reflecting cognitive uncertainty.
- SPN is employed to study anticipatory processing using paradigms like mixed-reality gaze tasks that manipulate feedback and intention.
- SPN amplitude offers actionable insights into neural precision weighting and resource allocation in predictive coding models.
Stimulus-Preceding Negativity (SPN) is a slow cortical event-related potential (ERP) characterized by a negative shift in electrical activity over centro-parietal and parieto-occipital regions, emerging in the interval preceding the delivery of an anticipated, task-relevant stimulus. SPN is not contingent on overt motor preparation but reflects anticipatory processing under uncertainty, constituting a distinct neurophysiological marker for cognitive states related to expectancy and prediction. In contemporary neurocognitive models, SPN has been interpreted as an index of anticipatory uncertainty or “precision weighting” within predictive coding frameworks, with its amplitude modulated by the predictability and relevance of imminent stimuli (Chiossi et al., 26 Jan 2026).
1. Theoretical Foundations and Functional Significance
SPN was originally defined as a slow, negative-going ERP component occurring immediately before an expected stimulus, independent of motor contingencies (Damen & Brunia, 1994; Brunia et al., 2012; van Boxtel & Brunia, 1994; Chwilla & Brunia, 1991). Unlike the Readiness Potential or Lateralized Readiness Potential, which are tied to overt movement planning, SPN is reliably elicited during mere anticipation of sensory input or feedback.
Contemporary accounts grounded in predictive coding conceptualize SPN as a reflection of anticipatory uncertainty: SPN is enhanced when prediction precision is low, requiring up-regulation of attention and cognitive resources (Catena & Perales, 2012; Tanovic & Joormann, 2019; Feldman & Friston, 2010; Clark, 2013). More negative SPN amplitudes correspond to increased uncertainty about incoming events, operationalizing the “precision weighting” mechanism in neural resource allocation (Chiossi et al., 26 Jan 2026).
2. Experimental Paradigms and Methods
Recent research has employed ecologically valid mixed-reality (MR) interaction paradigms to interrogate SPN’s functional profile. In a key study, participants (N = 28) engaged in gaze-based MR tasks utilizing a Varjo XR-4 headset (200 Hz per-eye, 51 pixels/°, 120 × 105° field of view). The experimental design was a 2 × 2 factorial crossing of Intention (Select vs. Observe) and Feedback (With vs. Without), with three scenarios: App Launcher (icon grid), Document Editor, and Video Player.
Trial structure involved an instruction (2000 ms), central fixation (1000 ms + jitter), target presentation (dwell 750 ms to trigger Select/Observe), UI action for Select (3000 ms), feedback in “With” blocks (500 ms), and a 1000 ms ISI. EEG was recorded with 64-channel LiveAmp (Brainproducts), referenced at FCz, ground at FPz, sampled at 500 Hz. Eye-tracking data synchronized gaze events at millisecond resolution (Chiossi et al., 26 Jan 2026).
Preprocessing included RANSAC bad-channel detection, 50 Hz notch and 1–15 Hz bandpass filtering, common average referencing, ICA-based artifact rejection (ICLabel), epoching (–1000…0 ms), baseline correction (–1000…–750 ms), and autoreject-based noise removal. ERP quantification leveraged posterior ROI electrodes: O1, Oz, O2, Iz, PO7, PO3, POz, PO4, PO8, P5, P6, P7, P8, P9, P10.
3. SPN Quantification and Statistical Analysis
SPN amplitude was operationalized as the mean voltage during –750…0 ms pre-stimulus (relative to UI display onset), baseline-corrected against the –1000…–750 ms window:
Statistical modeling employed a linear mixed-effects model (lme4 package, REML) predicting SPN amplitude from main effects and interaction of Intention (Select=0, Observe=1) and Feedback (With=0, Without=1), with random intercepts for participant, channel, and trial. The conditional R² was .33, with marginal R² = .0013.
Key parameter estimates (in µV):
| Condition | Mean SPN (µV) |
|---|---|
| Select + With Feedback (baseline) | –8.17 |
| Observe + With Feedback | –7.15 |
| Select + No Feedback | –7.73 |
| Observe + No Feedback | –6.71 |
Effects: Passive observation (Observe) yields more negative (larger) SPN than intentional selection, and absence of feedback further increases negativity, especially in Observe trials. The Intention×Feedback interaction was significant (β₃ = –0.96, p<.001), indicating that uncertainty is maximized with observation and no feedback (Chiossi et al., 26 Jan 2026).
4. Empirical Findings: Amplitude, Topography, and Context Sensitivity
Grand-average waveforms and scalp topographies confirm a robust, slow negativity centered over parieto-occipital regions in the –750…0 ms window across task contexts. The most negative SPN amplitudes were obtained under Observe–No Feedback, indicating maximal anticipatory uncertainty in the absence of both intention to act and external feedback.
The observed interaction effect replicated across all MR scenes (App Launcher, Editor, Video), as no three-way interaction with scene was found (p=0.440). Presence of feedback had a pronounced effect only during passive observation. Feedback had minimal impact in Select trials, suggesting that intentional action already reduces uncertainty and attenuates SPN (Chiossi et al., 26 Jan 2026).
A plausible implication is that SPN indexes a state of greater cognitive resource allocation when action or outcome predictability is diminished.
5. Decoding Intention from SPN: Deep-Learning Approaches
Single-trial, baseline-corrected SPN intervals (–750…0 ms) served as input (64 channels, 375 samples at 500 Hz) to several deep neural architectures: EEGNetv4, ShallowFBCSPNet, Deep4Net, EEGResNet, and EEGInceptionERP. Classifiers were trained in both person-dependent (stratified by participant) and person-independent paradigms.
Person-dependent mean accuracies ranged from 73.2% (Deep4Net) to 78.4% (EEGInceptionERP), with individual participant peaks up to 97%. Person-independent performance peaked at .692 (Deep4Net; 95% CI [.673, .712]). Significant pairwise differences were found in favor of EEGInceptionERP vs. EEGNetv4 (t(27)=–3.18, p=.003) and Deep4Net (p=.002); EEGResNet significantly exceeded Deep4Net (p=.020).
| Architecture | Person-Dependent Accuracy (%) | Person-Independent Accuracy |
|---|---|---|
| EEGInceptionERP | 78.4 ± 9.0 | .631 [.610, .651] |
| EEGResNet | 75.8 ± 8.0 | — |
| ShallowFBCSPNet | 74.2 ± 8.0 | .642 [.621, .662] |
| EEGNetv4 | 73.6 ± 6.7 | .662 [.642, .681] |
| Deep4Net | 73.2 ± 7.8 | .692 [.673, .712] |
| EEGConformer | — | .506 [.487, .525] |
Model sizes ranged from 0.44 MB (EEGNetv4) to 4.87 MB (Deep4Net); inference times ranged from 0.67 ms (EEGNetv4) to 3.07 ms (Deep4Net). SPN-based decision latency (750 ms window + <5 ms processing + inference) is compatible with typical MR dwell durations.
Saliency mapping via LIME revealed that classifier-driven decision-making depends on centro-parietal/parieto-occipital ROI features within the –750…0 ms window, corroborating the neurophysiological specificity of the learned representations (Chiossi et al., 26 Jan 2026).
6. Applications: Intention-Aware Interaction and Adaptive Systems
Leveraging SPN for intention detection offers several avenues for adaptive human-computer interaction:
- Dynamic Dwell-Time Adjustment: Real-time SPN amplitude can index user uncertainty, permitting adaptive lengthening or shortening of dwell requirements in gaze-based UI selection.
- Uncertainty-Contingent Feedback: Confirmation modalities (none, subtle highlight, explicit gesture) can be graduated according to SPN-derived certainty.
- Personalized Interaction Profiles: Calibration of each user’s baseline SPN and feedback sensitivity supports individualized optimization of dwell and feedback parameters.
SPN emerges as a robust, implicit, non-motor marker of anticipatory uncertainty. Its amplitude is modulated jointly by internal intention and external feedback, providing a neurophysiological basis for intention-aware MR interactions that mitigate spurious activations (the “Midas Touch” issue) (Chiossi et al., 26 Jan 2026).