AI agent skill · by @NeuroAIHub

Erp Analysis

Domain-validated pipeline and parameter guidance for event-related potential analysis, from preprocessing through statistical testing

Archived 2026-10-23 · Source updated 2026-10-02 · License: AGPL-3.0. Check the original repository for the latest version.

SKILL.md instructions

# ERP Data Analysis

## Purpose

This skill encodes expert methodological knowledge for analyzing event-related potentials (ERPs) from EEG data. It provides domain-specific parameter recommendations, processing order guidance, component identification criteria, and statistical analysis strategies that a general-purpose programmer or data scientist would not know without specialized training.

## When to Use This Skill

- Designing an ERP preprocessing pipeline for a new study
- Choosing filter settings, reference schemes, or artifact rejection criteria
- Identifying which ERP component to measure and how to define its time window and ROI
- Selecting appropriate amplitude measures (mean, peak, area) for a given component
- Choosing between traditional ANOVA-based analysis and mass univariate approaches
- Reviewing or troubleshooting an existing ERP analysis pipeline

## Research Planning Protocol

Before executing the domain-specific steps below, you MUST:

1. **State the research question** — What specific question is this ERP analysis addressing?
2. **Justify the method choice** — Why ERP (not fMRI, behavior-only, etc.)? What alternatives were considered?
3. **Declare expected outcomes** — Which component(s) do you expect to differ, in what direction?
4. **Note assumptions and limitations** — What does this method assume? Where could it mislead?
5. **Present the plan to the user and WAIT for confirmation** before proceeding.

For detailed methodology guidance, see the `research-literacy` skill.


## ⚠️ Verification Notice

This skill was generated by AI from academic literature. All parameters, thresholds, and citations require independent verification before use in research. If you find errors, please [open an issue](https://github.com/HaoxuanLiTHUAI/awesome_cognitive_and_neuroscience_skills/issues).

## Preprocessing Pipeline Overview

The standard ERP preprocessing pipeline involves 7 ordered steps. Processing order matters and can influence results (the "multiverse" problem; see Lonedo et al., 2020).

### Recommended Default Order

1. **Import and inspect raw data** -- Check for gross artifacts, disconnected electrodes
2. **Filter** -- Apply bandpass and optional notch filter
3. **Re-reference** -- Choose and apply reference scheme
4. **Bad channel identification and interpolation** -- Identify and interpolate noisy channels
5. **ICA-based artifact correction** -- Remove eye blinks, saccades, cardiac artifacts
6. **Epoching and baseline correction** -- Segment continuous data around events
7. **Epoch rejection** -- Remove remaining bad epochs by amplitude threshold

> **Multiverse note**: Steps 2-5 interact. Filtering before ICA improves decomposition quality (Winkler et al., 2015). Some researchers re-reference after ICA. Document your choices and consider running key alternatives to assess robustness.

For detailed parameters at each step, see `references/preprocessing-pipeline.md`.

### Key Parameter Defaults

| Step | Parameter | Default | Source |
|------|-----------|---------|--------|
| Filter | High-pass | **0.1 Hz** | Luck, 2014, Ch. 5; Tanner et al., 2015 |
| Filter | Low-pass | **30 Hz** | Luck, 2014, Ch. 5 |
| Filter | Filter type | FIR, zero-phase | Widmann et al., 2015 |
| Re-reference | Scheme | Average reference | Luck, 2014, Ch. 5; Keil et al., 2014 |
| Artifact rejection | Threshold | **+/-100 uV** | Luck, 2014 |
| Bad channels | Max proportion | **< 10%** of total channels | Keil et al., 2014 |
| Epoching | Window | **-200 ms to 800 ms** | Luck, 2014, Ch. 5 |
| Baseline | Window | **-200 ms to 0 ms** | Luck, 2014, Ch. 5 |

## ERP Component Identification

To measure an ERP component correctly, you need three pieces of information:

1. **Latency range** -- The expected time window for the component
2. **Scalp distribution** -- Where on the scalp the component is maximal
3. **Functional context** -- What experimental manipulation elicits the component

### Component Selection Decision Logic

1. **Identify your cognitive process of interest** (e.g., semantic processing, error monitoring)
2. **Look up the canonical component** in `references/erp-components.md`
3. **Verify the component matches your paradigm** -- The same voltage deflection in a different paradigm may reflect a different component (Luck, 2014, Ch. 2)
4. **Use the recommended ROI and time window** as your starting point, then verify against your own grand average waveform
5. **For disputed components**, read the Interpretation Notes in `references/erp-components.md` before committing to a theoretical interpretation

### Component Domains at a Glance

| Domain | Key Components | Reference File Section |
|--------|---------------|----------------------|
| Visual perception | P1, N1, N170, N2pc | `erp-components.md` Section 1 |
| Language | N400, P600/LPC, ELAN, LAN | `erp-components.md` Section 2 |
| Executive control | ERN/Ne, Pe, N2, CNV | `erp-components.md` Section 3 |
| Memory | FN400, parietal old/new, Dm | `erp-components.md` Section 4 |

## Statistical Analysis Strategy

### Amplitude Measurement

Choose your measurement approach based on the component:

| Measure | Best For | Avoid When | Source |
|---------|----------|------------|--------|
| **Mean amplitude** | Broad components (N400, P300, LPC) | Component is sharp and brief | Luck, 2014, Ch. 9 |
| **Peak amplitude** | Sharp, well-defined peaks (P1, N1) | Component has no clear peak or has multiple peaks | Luck, 2014, Ch. 9 |
| **Peak latency** | Measuring processing speed | Component lacks a clear peak | Luck, 2014, Ch. 9 |
| **50% fractional area latency** | Latency with unequal amplitudes across conditions | Rarely inappropriate; preferred over peak latency | Luck, 2014, Ch. 9; Kiesel et al., 2008 |
| **Signed/unsigned area** | Components spanning positive and negative voltages | Simple, unipolar components | Luck, 2014, Ch. 9 |

### Time Window and ROI Selection

1. **A priori selection** (preferred): Choose time window and electrodes based on prior literature before looking at your data (Luck & Gaspelin, 2017)
2. **Collapsed localizer**: Average across all conditions to identify the window/ROI, then test differences between conditions within that window (Luck & Gaspelin, 2017)
3. **Data-driven**: Use mass univariate approach to avoid arbitrary window selection (see below)

> **Critical**: Never select a time window or ROI by looking at the difference between conditions. This inflates Type I error (Luck & Gaspelin, 2017).

### Choosing a Statistical Framework

```
Is your hypothesis about a specific, well-characterized component?
 |
 +-- YES --> Do you have a priori time window and ROI?
 | |
 | +-- YES --> Traditional ANOVA on mean/peak amplitude
 | |
 | +-- NO --> Use collapsed localizer, then ANOVA
 |
 +-- NO --> Is your effect potentially distributed across time/space?
 |
 +-- YES --> Cluster-based permutation test (Maris & Oostenveld, 2007)
 |
 +-- NO --> Mass univariate with FDR correction (Groppe et al., 2011)
```

For detailed statistical method descriptions, see `references/statistical-approaches.md`.

## Common Pitfalls

1. **Double-dipping**: Selecting time windows or electrodes based on the effect of interest, then testing that same effect in the selected window (Kriegeskorte et al., 2009)
2. **Excessive filtering**: High-pass cutoffs above **0.1 Hz** can distort slow components like the P300, N400, and LPC (Tanner et al., 2015; Widmann & Schroger, 2012)
3. **Confounding component overlap**: Apparent differences in one component may be driven by overlap from an adjacent component; consider difference waves and component-specific analyses (Luck, 2014, Ch. 2)
4. **Ignoring trial count imbalance**: Unequal trial counts across conditions produce differential noise levels, biasing peak amplitude and latency measures (Luck, 2014, Ch. 9)
5. **Cluster-based tests for latency**: Cluster permutation tests control family-wise error but cannot localize effects to specific time points or channels (Maris & Oostenveld, 2007)
6. **Reporting only p-values**: Always report effect sizes (partial eta-squared for ANOVA, Cohen's d for t-tests) alongside p-values (Keil et al., 2014)

## Minimum Reporting Checklist

Based on Keil et al. (2014) and Luck (2014):

- [ ] Number of accepted trials per condition (minimum **30** per condition recommended; Boudewyn et al., 2018)
- [ ] Filter settings (type, cutoff frequencies, roll-off)
- [ ] Reference scheme
- [ ] Artifact rejection method and criteria (ICA, threshold, proportion rejected)
- [ ] Number and identity of interpolated channels
- [ ] Epoch window and baseline correction window
- [ ] Component time window and ROI electrodes (with justification)
- [ ] Amplitude measure used (mean, peak, area)
- [ ] Statistical test, correction method, and effect sizes

## References

- Boudewyn, M. A., Luck, S. J., Farrens, J. L., & Kappenman, E. S. (2018). How many trials does it take to get a significant ERP effect? *Psychophysiology*, 55(6), e13049.
- Groppe, D. M., Urbach, T. P., & Kutas, M. (2011). Mass univariate analysis of event-related brain potentials/fields I. *Psychophysiology*, 48(12), 1711-1725.
- Keil, A., et al. (2014). Committee report: Publication guidelines and recommendations for studies using EEG and MEG. *Psychophysiology*, 51(1), 1-21.
- Kiesel, A., Miller, J., Jolicoeur, P., & Brisson, B. (2008). Measurement of ERP latency differences. *Psychophysiology*, 45(4), 517-523.
- Kriegeskorte, N., Simmons, W. K., Bellgowan, P. S., & Baker, C. I. (2009). Circular analysis in systems neuroscience. *Nature Neuroscience*, 12(5), 535-540.
- Lonedo, A., et al. (2020). The multiverse of ERP analysis pipelines. *NeuroImage*, 209, 116465.
- Luck, S. J. (2014). *An Introduction to the Event-Related Potential Technique* (2nd ed.). MIT Press.
- Luck, S. J., & Gaspelin, N. (2017). How to get statistically significant effects in any ERP experiment (and why you shouldn't). *Psychophysiology*, 54(1), 146-157.
- Maris, E., & Oostenveld, R. (2007). Nonparametric statistical testing of EEG- and MEG-data. *Journal of Neuroscience Methods*, 164(1), 177-190.
- Tanner, D., Morgan-Short, K., & Luck, S. J. (2015). How inappropriate high-pass filters can produce artifactual effects. *Psychophysiology*, 52(8), 997-1009.
- Widmann, A., Schroger, E., & Maess, B. (2015). Digital filter design for electrophysiological data. *Journal of Neuroscience Methods*, 250, 34-46.
- Winkler, I., Debener, S., Muller, K. R., & Tangermann, M. (2015). On the influence of high-pass filtering on ICA-based artifact reduction in EEG-ERP. *Proceedings of the IEEE EMBC*, 4101-4105.

See `references/` for detailed parameter tables, component database, and statistical method descriptions.