メニュー
Expert guidance for designing EEG paradigms optimized to isolate specific ERP components, with domain-validated timing, trial count, and control condition parameters
デフォルトでは、最初にソースを確認する Prompt が選択されています。直接コマンドに切り替えるか、ローカルコピーをダウンロードすることもできます。
インストールを決める前に、SKILL.md と SkillsMP に表示されている付属ファイルをお読みください。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
直接コマンドでは確認用 Prompt が省略されます。実行前にソースを確認してください。
npx skills add https://github.com/NeuroAIHub/BrainPilot --skill eeg-paradigm-designerコマンドは1行のまま表示されます。コピー前に横へスクロールして全体を確認してください。
ローカルで確認しますか？SkillsMP が現在取得できるファイルをダウンロードできます。
SOC 職業分類に基づく
SKILL.md を表示中
| name | eeg-paradigm-designer |
| description | Expert guidance for designing EEG paradigms optimized to isolate specific ERP components, with domain-validated timing, trial count, and control condition parameters |
| domain | cognitive-neuroscience |
| version | 1.0.0 |
| authors | ["AI-assisted (Claude)"] |
| papers | ["Luck, 2014","Kappenman & Luck, 2010","Kappenman et al., 2021","Boudewyn et al., 2018","Keil et al., 2014","Jensen & MacDonald, 2023"] |
| dependencies | {"required":["research-literacy"],"recommended":["erp-analysis","cognitive-paradigm-design"]} |
| review_status | ai-generated |
This skill encodes expert knowledge for designing EEG experimental paradigms that reliably isolate specific event-related potential (ERP) components. Designing an EEG paradigm differs fundamentally from designing a behavioral experiment: timing constraints are stricter, stimulus properties must be controlled to avoid confounding sensory ERPs with cognitive ERPs, trial counts must be higher to achieve adequate signal-to-noise ratios, and the choice of control condition directly determines which neural process can be isolated via subtraction. A general-purpose programmer or experimental psychologist without EEG training would get many of these decisions wrong.
For ERP preprocessing and analysis after data collection, see the erp-analysis skill. For general experimental paradigm selection (behavioral focus), see the cognitive-paradigm-design skill.
Before executing the domain-specific steps below, you MUST:
For detailed methodology guidance, see the research-literacy skill.
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.
Map the research question to a specific ERP component. The component determines everything else: paradigm type, timing, electrode montage, trial count, and analysis strategy.
Use the Component-Paradigm Quick Reference below or consult references/component-paradigm-map.md for full details.
| Component | Canonical Paradigm | Key Manipulation | Latency (ms) | Max Site |
|---|---|---|---|---|
| P1 | Spatial attention (Posner) | Attended vs. unattended location | 80--130 | O1/O2 |
| N1 | Spatial attention / discrimination | Attended vs. unattended stimulus | 150--200 | PO7/PO8 |
| N170 | Face perception | Faces vs. non-face objects | 140--200 | P7/P8 |
| MMN | Passive oddball | Deviant vs. standard (no response) | 100--250 | Fz/FCz |
| N2pc | Lateralized visual search | Contralateral vs. ipsilateral to target | 200--300 | PO7/PO8 |
| P3a | Novelty oddball | Novel/unexpected stimuli | 250--350 | Fz/Cz |
| P3b | Target oddball | Rare targets vs. frequent standards | 300--600 | Pz |
| N400 | Semantic violation / priming | Incongruent vs. congruent words | 300--500 | Cz/CPz |
| P600 | Syntactic violation | Ungrammatical vs. grammatical | 500--800 | Pz/CPz |
| ERN | Speeded response (flanker, Go/NoGo) | Error vs. correct (response-locked) | 0--100 post-resp | FCz |
| LRP | Choice-RT with lateralized responses | Contralateral vs. ipsilateral motor cortex | Sustained pre-resp | C3/C4 |
| CNV | S1-S2 foreperiod | Warning signal before imperative stimulus | Sustained | Cz/FCz |
| SSVEP | Frequency tagging / flicker | Periodic visual stimulation at fixed Hz | Steady-state | Oz |
Once the target component is identified, select the appropriate paradigm class and configure its parameters. See references/component-paradigm-map.md for detailed paradigm specifications per component, and references/timing-parameters.md for timing configurations.
EEG paradigms have stricter timing requirements than behavioral experiments for three reasons a non-specialist would not anticipate:
ERP overlap: When stimuli arrive too quickly, the ERP to one stimulus overlaps with the ERP to the next, making components unresolvable. The minimum ISI must be long enough for the slowest ERP component of interest to resolve -- typically >= 1000 ms for fast components (P1, N1) and >= 1500--2000 ms for slow components (N400, P300, P600) (Luck, 2014, Ch. 6).
Alpha-band contamination: Rhythmic stimulation near 10 Hz (ISI ~ 100 ms) entrains alpha oscillations, producing steady-state responses that obscure transient ERPs. Avoid ISIs that create stimulus rates in the 8--13 Hz range unless studying SSVEPs (Luck, 2014, Ch. 6).
Habituation and refractoriness: Sensory ERPs (P1, N1) are attenuated by repetition. Short ISIs (< 500 ms) produce refractory-period suppression of early components, reducing sensitivity to experimental manipulations. For paradigms targeting P1/N1, use ISIs of >= 1000 ms or jitter ISIs widely (Luck, 2005; Coles & Rugg, 1995).
Always jitter the ISI to prevent anticipatory CNV buildup from contaminating the pre-stimulus baseline and to support regression-based overlap correction (e.g., LIMO, unfold). Recommended jitter: +/- 200--500 ms uniform or exponential distribution around the mean ISI (Luck, 2014, Ch. 6; Woldorff, 1993).
Trial counts for EEG must be substantially higher than for behavioral studies because the ERP signal is extracted from noisy single-trial EEG by averaging, and the signal-to-noise ratio improves with the square root of the number of trials.
There is no universal minimum trial count. The required number depends on the interaction of effect magnitude, number of participants, and component-specific noise levels (Boudewyn et al., 2018; Jensen & MacDonald, 2023). The table below provides component-specific starting recommendations for typical effect sizes in well-designed paradigms:
| Component | Minimum Trials/Condition | Recommended Trials/Condition | Rationale |
|---|---|---|---|
| P3b (oddball) | 30 | 50--80 | Large effect; SNR good at Pz (Luck, 2014, Ch. 9; Kappenman et al., 2021) |
| N400 (semantic) | 30 | 40--60 | Large effect for strong violations; more for graded manipulations (Boudewyn et al., 2018) |
| N170 (faces) | 40 | 60--80 | Moderate effect; requires adequate face and control exemplars (Rossion & Jacques, 2008) |
| N2pc (search) | 100 | 150--200 | Small lateralized difference; many trials needed (Luck, 2014, Ch. 3; Kappenman et al., 2021) |
| MMN (oddball) | 150 (deviants) | 200--300 (deviants) | Small amplitude; passive paradigm adds noise (Naatanen et al., 2007; Duncan et al., 2009) |
| ERN (errors) | 6 | 10--15 | Large amplitude but depends on error rate (Olvet & Hajcak, 2009; Boudewyn et al., 2018) |
| LRP (lateralized) | 40 | 80--100 | Small lateralized difference; high trial-to-trial variability (Boudewyn et al., 2018) |
| P600 (syntactic) | 30 | 40--60 | Large effect for clear violations (Osterhout & Holcomb, 1992) |
| CNV (foreperiod) | 30 | 40--60 | Moderate amplitude; slow wave requires low-frequency filtering (Brunia et al., 2012) |
| SSVEP (flicker) | 10--20 blocks | 30+ blocks of 10--20 s | Frequency-domain; SNR depends on block duration (Norcia et al., 2015) |
Critical note: These are minimum retained trials after artifact rejection. Plan for 20--30% attrition from artifacts. If you need 40 clean trials, design for at least 50--55 trials per condition (Luck, 2014, Ch. 6).
ERP components are best isolated using difference waveforms that subtract overlapping activity common to two conditions, leaving only the neural process of interest (Luck, 2014, Ch. 2; Kappenman et al., 2021).
Design principle: For every target component, explicitly define the subtraction that will isolate it.
| Component | Subtraction | What It Removes |
|---|---|---|
| N400 | Incongruent minus Congruent | Sensory ERP, P1/N1, baseline activity |
| P3b | Target minus Standard | Sensory response to frequent stimuli |
| MMN | Deviant minus Standard | Obligatory auditory response |
| N2pc | Contralateral minus Ipsilateral | Bilateral sensory activity, P1/N1 |
| ERN | Error minus Correct (response-locked) | Motor preparation, baseline activity |
| LRP | (C3-C4 left hand) averaged with (C4-C3 right hand) | Non-lateralized activity |
| N170 | Faces minus Control objects | Low-level visual ERPs |
Warning: The subtraction is only valid if the two conditions are matched on all low-level stimulus properties (luminance, spatial frequency, size, contrast, position) and differ only on the cognitive dimension of interest. Failure to match stimuli is the most common source of confounded ERP results (Luck, 2014, Ch. 2; Kappenman & Luck, 2010).
The required electrode density depends on the spatial precision needed:
| Montage | Channels | Best For | Not Sufficient For |
|---|---|---|---|
| Low-density | 32 | P3b, N400, ERN, MMN (midline components) | N2pc, LRP, source localization |
| Medium-density | 64 | N2pc, LRP, N170, most ERP research | High-resolution source localization |
| High-density | 128--256 | Source localization, CSD analysis, spatial mapping | Overkill for standard ERP analysis on midline components |
Decision rules (Luck, 2014, Ch. 4; Keil et al., 2014):
Before finalizing the paradigm, check for these non-obvious EEG-specific design flaws:
Overlapping ERPs from adjacent events: If ISI < the duration of the slowest component, ERPs overlap. For P3b (300--600 ms), this means ISIs under ~1200 ms create overlap. For P600 (500--800+ ms), ISIs under ~1500 ms are problematic. Use the ADJAR procedure or linear modeling (e.g., unfold toolbox) if fast ISIs are required (Woldorff, 1993; Ehinger & Dimigen, 2019).
Stimulus confounds masquerading as cognitive ERPs: Differences in luminance, contrast, spatial frequency, size, or retinal position between conditions produce P1/N1 differences that are sensory, not cognitive. Always equate low-level stimulus properties or use difference waveforms that cancel them (Luck, 2014, Ch. 2).
Inadequate baselines: If pre-stimulus activity differs between conditions (e.g., from a preceding cue or from CNV buildup during fixed foreperiods), standard baseline correction (-200 to 0 ms) will distort post-stimulus ERP measurements. Use jittered ISIs and verify baseline equivalence (Luck, 2014, Ch. 6; Alday, 2019).
Motor confounds with cognitive ERPs: If conditions differ in response requirements (e.g., one condition has button press, the other does not), motor-related ERPs (LRP, readiness potential) contaminate the cognitive ERP. Use conditions with identical motor responses or analyze only stimulus-locked, pre-response windows (Luck, 2014, Ch. 6).
Probability confounds in oddball paradigms: In P3b oddball designs, the rare target differs from the frequent standard in both probability and task relevance. To disentangle these, include a rare non-target condition (three-stimulus oddball) or use an equiprobable control (Luck, 2014, Ch. 3; Polich, 2007).
Physical-deviance confound in MMN: The standard and deviant stimuli differ in physical features, which can produce differential N1 responses independent of memory-trace mismatch. Use a "many-standards" or "flip-flop" control design where the same physical stimulus serves as both standard and deviant across blocks (Naatanen et al., 2007; Jacobsen & Schroger, 2001).
Lateralized eye movements confounding N2pc: Saccades toward the target produce HEOG artifacts that mimic the contralateral negativity of the N2pc. Enforce fixation, reject trials with HEOG deviations > +/- 16 uV (corresponding to ~1 degree eye movement), or use residual HEOG correction (Luck, 2014, Ch. 3; Woodman & Luck, 2003).
Insufficient error trials for ERN: Error rate depends on task difficulty. If the task is too easy (< 5% errors), you will not accumulate enough error trials. Titrate difficulty to achieve 10--25% error rate using adaptive procedures or speed-emphasis instructions (Gehring et al., 1993; Olvet & Hajcak, 2009).
Confounding component overlap in language ERPs: In sentence paradigms, an apparent N400 reduction may be driven by an overlapping P600 in the same condition, and vice versa. Report and interpret both components; consider component-overlap modeling (Luck, 2014, Ch. 2; Brouwer et al., 2017).
High-pass filter artifacts for slow components: If you plan to study CNV, P3b, N400, or P600, ensure the recording system and preprocessing pipeline allow high-pass cutoffs of <= 0.1 Hz. Cutoffs at 0.5 Hz or above create artificial distortions of broad components (Tanner et al., 2015; see erp-analysis skill).
When adapting a behavioral paradigm for EEG, apply these modifications:
| Feature | Behavioral Design | EEG Adaptation | Reason |
|---|---|---|---|
| ISI | 500--1500 ms | 1200--2500 ms | Avoid ERP overlap (Luck, 2014, Ch. 6) |
| ISI variability | Fixed or blocked | Jittered +/- 200--500 ms | Prevent CNV, enable overlap correction |
| Trial count | 40--80/condition | 50--200+/condition (component-dependent) | SNR from averaging |
| Response hand | Any | Counterbalanced across blocks | LRP contamination |
| Rest breaks | Every 50--100 trials | Every 30--60 trials (1--2 min breaks) | Reduce muscle artifact, blink accumulation |
| Block length | 5--10 min | 3--5 min | Alpha drift, impedance changes |
| Stimulus duration | Until response | Fixed 100--300 ms (for transient ERPs) | Standardize sensory input |
| Practice | 10--20 trials | 20--40 trials with artifact feedback | Reduce blinks, movements in early blocks |
See references/ for detailed component-paradigm mapping and timing parameter tables.