Pitfall gallery
What it looks like when a technique is misapplied — and the check that catches it.
By category: acquisition · cohort · filtering · preprocessing · erp · time-frequency · connectivity · statistics — by level: L0 · L1 · L2 · L3 · L4 · L5 · L6 · L7
acquisition
Dropped samples and clock drift
Events drift relative to the data across the recording.
Electrode bridging
Two neighbouring channels are near-identical; interpolation "works".
EMG reported as gamma
Broadband high-frequency power over temporal/frontal sites tied to jaw or neck.
Eye movements contaminate lateralized effects
N2pc / lateralized gamma tracks HEOG.
Hardware notch or band-limit inside the analysis band
Spectral hole at 50/60 Hz or a roll-off at 40–50 Hz mistaken for a physiological dip; double-notching in software.
Uncorrected stimulus–trigger latency
Components shifted by a constant; latencies differ between labs.
Usable bandwidth far below the Nyquist frequency
"Gamma" analysed at 128 Hz output from a headset whose hardware rolls off near 43 Hz; a 500 Hz recording with a ~30 Hz hardware low-pass.
cohort
Age, sex or medication state differ between groups
Patients older and sex-skewed relative to controls; patients recorded on medication.
Site, amplifier or cap confounded with group or time
Two amplifiers at two sites with different sampling rates; two cap systems over 20 years; 50 vs 60 Hz mains by country.
filtering
Downsampling without anti-alias filtering
High-frequency content folds into the band of interest.
Filtering across discontinuities
Ringing at file joins, breaks, or buffer edges.
Filtering short epochs
Edge artifacts inside the analysis window.
High-pass cutoffs that distort slow ERPs
Artifactual early components; attenuated late components.
Over-wide notch or strong IIR notch
Gamma hole and ringing around 50/60 Hz.
Zero-padding mistaken for resolution
Smooth spectrum interpreted as finer resolution.
preprocessing
Comparing amplitudes across references
Component at Pz differs by half between studies.
ICA after interpolation without rank adjustment
Duplicated or degenerate components; ICA fails to converge.
Rejection that differs by condition
One condition loses 30% of trials; effect appears or disappears.
Removing brain components that carry the effect
Effect shrinks after "cleaning"; components with alpha or posterior topography removed.
erp
time-frequency
Band power changes that are slope changes
"More beta" with no beta peak; all bands shift together.
Narrowband filtering creates oscillations
Any noise looks rhythmic after a 2 Hz-wide filter.
Non-sinusoidal waveforms produce spurious coupling
PAC between a rhythm and its harmonic.
Time-frequency edge artifacts in the baseline
Low-frequency power ramps at epoch edges; baseline inflated.
connectivity
statistics
Cluster p-values read as latency/location claims
"The effect began at 312 ms" from a cluster test.
Information leakage across CV folds
Cross-validated accuracy collapses on a true held-out set.
Mass univariate tests without correction
Significant blobs everywhere.
Windows chosen from the tested data
Effect only in the window that looked biggest.
Level 0
- Dropped samples and clock drift (acquisition)
- Electrode bridging (acquisition)
- EMG reported as gamma (acquisition)
- Eye movements contaminate lateralized effects (acquisition)
- Overclaiming source depth or precision (connectivity)
- Uncorrected stimulus–trigger latency (acquisition)
- Usable bandwidth far below the Nyquist frequency (acquisition)
Level 1
- Band power changes that are slope changes (time-frequency)
- Downsampling without anti-alias filtering (filtering)
- Filtering across discontinuities (filtering)
- Filtering short epochs (filtering)
- Hardware notch or band-limit inside the analysis band (acquisition)
- High-pass cutoffs that distort slow ERPs (filtering)
- Over-wide notch or strong IIR notch (filtering)
- Usable bandwidth far below the Nyquist frequency (acquisition)
- Zero-padding mistaken for resolution (filtering)
Level 2
- Comparing amplitudes across references (preprocessing)
- Electrode bridging (acquisition)
- Filtering short epochs (filtering)
- High-pass cutoffs that distort slow ERPs (filtering)
- ICA after interpolation without rank adjustment (preprocessing)
- Rejection that differs by condition (preprocessing)
- Removing brain components that carry the effect (preprocessing)
- Uncorrected stimulus–trigger latency (acquisition)
Level 3
- Baseline contains condition differences (erp)
- Cluster p-values read as latency/location claims (statistics)
- Comparing amplitudes across references (preprocessing)
- Eye movements contaminate lateralized effects (acquisition)
- Mass univariate tests without correction (statistics)
- Peak amplitude inflated by noise (erp)
- Windows chosen from the tested data (statistics)
Level 4
- Band power changes that are slope changes (time-frequency)
- Cluster p-values read as latency/location claims (statistics)
- Hardware notch or band-limit inside the analysis band (acquisition)
- Mass univariate tests without correction (statistics)
- Narrowband filtering creates oscillations (time-frequency)
- Non-sinusoidal waveforms produce spurious coupling (time-frequency)
- Time-frequency edge artifacts in the baseline (time-frequency)
Level 5
- Comparing amplitudes across references (preprocessing)
- Information leakage across CV folds (statistics)
- Overclaiming source depth or precision (connectivity)
- Zero-lag connectivity from one source (connectivity)
Level 6
- Age, sex or medication state differ between groups (cohort)
- Cluster p-values read as latency/location claims (statistics)
- Information leakage across CV folds (statistics)
- Mass univariate tests without correction (statistics)
- Site, amplifier or cap confounded with group or time (cohort)
- Windows chosen from the tested data (statistics)
Level 7
- Age, sex or medication state differ between groups (cohort)
- Band power changes that are slope changes (time-frequency)
- EMG reported as gamma (acquisition)
- Filtering across discontinuities (filtering)
- Information leakage across CV folds (statistics)
- Site, amplifier or cap confounded with group or time (cohort)
- Usable bandwidth far below the Nyquist frequency (acquisition)