Skip to content

Case Study 1 · ERP correlates of neuropsychological performance

A hypothesis-driven look at whether two classic ERP components — N400 and P300 — relate to neuropsychological performance in post-stroke patients, using four scalar features and linear regression.

See the code

4 ERP features Linear regression Synthetic data

The two hypothesised associations in Case Study 1, with regression lines

Research question

Are the amplitude and latency of specific ERP components associated with neuropsychological performance and demographic characteristics in post-stroke patients?

This is a hypothesis-driven question: rather than exploring the full spectral or spatiotemporal structure of the EEG signal (see Case Study 2), it targets two specific, well-characterised ERP components:

  • N400 — a negative deflection ~400 ms post-stimulus, indexing semantic processing and language comprehension (Kutas & Federmeier, 2011)
  • P300 — a positive deflection ~300 ms post-target, indexing attentional resource allocation (Polich, 2007)

For each component we consider amplitude (mean µV in a predefined window) and latency (ms of peak amplitude) — four participant-level neural predictors in total.

Data

The code uses a synthetic (mock) dataset of 40 simulated post-stroke patients, so it can be run end to end without access to real clinical data:

  • dataset_NPSY.csv — demographics (age, sex) and neuropsychological scores (CRIq, APACS, Attentive Matrices, MoCA)
  • dataset_ERP.csv — the four ERP features (N400/P300 amplitude and latency), linked by subject_id

Pipeline

1. Data Wrangling

  • Load & merge
  • Range & missing-data checks
  • Normality & outlier detection

2. Analysis

  • Descriptive statistics
  • Bivariate correlations
  • Three regression models (APACS; Attentive Matrices, kitchen-sink vs. theory-guided)
  • Post-hoc diagnostics (Q-Q, Cook's distance, VIF)

3. Export

  • Figures (PNG)
  • Tables (CSV + Excel)
  • Ready to reuse in a report or a slide

A taste of the results

  • Correlation heatmap

    Pearson correlations between all ERP and neuropsychological variables, annotated with significance stars.

  • Key associations

    The two hypothesised associations (P300 amplitude ↔ Attentive Matrices; N400 latency ↔ APACS), with regression line and 95% CI band.

Key takeaway

More predictors do not make a better model — and how you decide which ones to drop matters as much as the decision itself. Choosing predictors a priori, from theory, avoids the circular reasoning of dropping whatever happens to be non-significant in a bigger model. See the full reasoning in the code, Section 2.3.

See the code

Strengths and limitations

Strengths — reduces high-dimensional EEG to four interpretable scalar features; uses regression models that are familiar to clinical researchers; N400 and P300 rest on decades of validation in neurological populations.

Limitations — the predefined time windows and electrode sets may miss effects that are delayed, spatially atypical, or otherwise deviate from the normative pattern the features were designed to capture — a particular risk in a clinical population such as stroke patients. When the question is broader or more exploratory, see Case Study 2.