Flanker Interference Reanalysis

This repository documents a reproducible reanalysis of a small-sample Eriksen flanker dataset. The project reorganizes a raw spreadsheet export into block-level and subject-level datasets, reproduces the original repeated-measures ANOVA, and extends the workflow with a block-level mixed-effects model and descriptive error checks.

The emphasis is methodological rather than claim-oriented: the repository focuses on data restructuring, repeated-measures inference, hierarchical thinking, transparent limitations, and reproducible reporting.

Research Question

How do arrow load and flanker consistency affect the congruency effect in an Eriksen flanker task?

The reanalysis examines a 2×2 within-subject design with two experimental factors:

• Arrow load: low (5 arrows) vs high (7 arrows)
• Flanker consistency: consistent vs inconsistent

The outcome of interest is the congruency effect in milliseconds, defined as the difference between incongruent and congruent reaction times.

Data Structure

The repository preserves both the original spreadsheet export and the reconstructed analysis-ready datasets.

Raw data

• data/raw/flanker_block_export.csv
  Original spreadsheet-style export retained for provenance.

• data/raw/flanker_subject_summary_original.csv
  Original subject-condition summary file used as a validation target.

Processed data

• data/processed/flanker_block_level.csv
  Reconstructed block-level dataset with 240 rows:

  • 12 participants
  • 5 repeated observations per condition
  • 4 within-subject conditions

• data/processed/flanker_subject_summary.csv
  Reconstructed subject-level summary with 48 rows:

  • 12 participants
  • 4 within-subject conditions

The processed summary matches the original summary exactly.

Analysis Workflow

The workflow proceeds in four steps:

1. Parse the original export into a tidy block-level dataset.
2. Reconstruct subject-level summaries and validate them against the original summary file.
3. Reproduce the repeated-measures ANOVA reported in the original analysis.
4. Extend the analysis with a block-level mixed-effects model and descriptive error checks.

Main Findings

1. Validation of reconstructed summaries

The reconstructed subject-level summary reproduces the original summary exactly:

• 48 out of 48 rows matched
• Maximum absolute difference: 0 ms

2. Repeated-measures ANOVA

The subject-level repeated-measures ANOVA produced the following results:

• Arrow load: F(1, 11) = 4.80, p = 0.051, partial eta squared = 0.304
• Flanker consistency: F(1, 11) = 67.56, p < 0.001, partial eta squared = 0.860
• Interaction: F(1, 11) = 2.82, p = 0.121, partial eta squared = 0.204

The strongest and most stable signal is the main effect of flanker consistency.

3. Condition means

Mean congruency effects by condition:

• Low load / Inconsistent: 33.48 ms
• Low load / Consistent: 66.02 ms
• High load / Inconsistent: 35.02 ms
• High load / Consistent: 82.50 ms

These condition means indicate substantially larger congruency effects under consistent flankers.

4. Within-load contrasts

Pairwise contrasts comparing consistent vs inconsistent flankers within each load condition:

• Low load: +32.54 ms, t(11) = 4.19, Holm-adjusted p = 0.0015
• High load: +47.48 ms, t(11) = 9.19, Holm-adjusted p < 0.001

5. Block-level mixed-effects model

Using the reconstructed 240-row block-level dataset, the mixed-effects model showed:

• A strong consistency effect under low load
• Little evidence for a main effect of arrow load on its own
• No clear evidence for the interaction

The block-level reanalysis therefore supports the same broad interpretation as the subject-level ANOVA: flanker consistency is the dominant effect in this dataset.

6. Error patterns

Error counts are included as a descriptive secondary outcome. Two participants have missing low-load error entries in the original export, so error analyses should be interpreted cautiously.

The error summaries suggest that accuracy should be reported alongside reaction-time effects rather than ignored.

Figures and Tables

The repository includes:

• condition mean plots with 95% confidence intervals
• participant-level trajectory plots
• block-level distribution plots
• descriptive error-pattern plots
• ANOVA, contrast, mixed-model, and validation tables

See results/figures/ and results/tables/ for outputs.

Repository Structure

flanker-interference-reanalysis/
├── data/
│   ├── raw/
│   │   ├── flanker_block_export.csv
│   │   └── flanker_subject_summary_original.csv
│   └── processed/
│       ├── flanker_block_level.csv
│       └── flanker_subject_summary.csv
├── docs/
│   ├── attribution.md
│   ├── data_dictionary.md
│   ├── limitations.md
│   └── study_design.md
├── results/
│   ├── figures/
│   ├── tables/
│   └── analysis_summary.md
├── scripts/
│   ├── 01_parse_raw_export.py
│   ├── 02_validate_reconstructed_summary.py
│   ├── 03_reproduce_rm_anova.py
│   ├── 04_block_level_mixed_model.py
│   ├── 05_error_analysis.py
│   ├── 06_make_figures.py
│   └── utils.py
├── README.md
└── requirements.txt