CWT fMRI Project - Confidence Weighting Task

📊 Quick Access to Results

🎯 View GLMM Analysis Report

Interactive report with model coefficients, fit statistics, and embedded figures. Features dark mode toggle for comfortable viewing.

---

Overview

This repository contains the analysis pipeline for a Continuous Wavelet Transform (CWT) fMRI study investigating the confidence weighting task. The project examines how participants learn to predict emotional faces based on predictive cues in a reversal learning paradigm, with a focus on confidence, accuracy, and response time measures.

📋 Experimental Design: See docs/experimental_design.md for detailed task description and stimulus structure, including visual diagram of the experimental design.

Project Structure

CWT_fmri/
├── data/
│   ├── raw/                    # Original data files
│   │   ├── CWT_vmp1_master.csv (3.7MB, 53,592 trials)
│   │   ├── CWT_VMP1_master_table.mat
│   │   └── CWT_VMP1_master_table_extended.mat
│   └── processed/              # Processed data files
├── code/
│   ├── preprocessing/          # Data preparation scripts
│   │   └── 01_import_and_clean_data.R
│   ├── analysis/              # Statistical analysis scripts
│   │   ├── 01_basic_analysis_and_plots.R
│   │   ├── 02_run_glmm_main.R (current main analysis script)
│   │   ├── 02_validate_glmm.R (validation script)
│   │   ├── 03_run_anova_analyses.R
│   │   ├── 05_reversal_learning_plots.R
│   │   ├── theme_nature_neuroscience.R
│   │   └── backup/            # Archived previous versions
│   │       ├── 02_run_glmm_models_original.R
│   │       └── 02_run_glmm_models_debug.R
│   └── modeling/              # Computational modeling scripts
│       └── 04_run_computational_models.R
├── results/
│   ├── figures/               # Generated plots
│   │   ├── glmm_models/       # GLMM model plots
│   │   ├── learning_curves/   # Learning analysis plots
│   │   ├── confidence_analysis/ # Confidence-specific plots
│   │   ├── basic_analysis/    # Basic analysis plots
│   │   └── predictive_checks/ # Model validation plots
│   ├── tables/                # Output tables
│   └── models/                # Model results (.rds files)
├── docs/
│   ├── experimental_design.md # Task description and stimulus structure
│   ├── experimental_design.png
│   ├── glmm_report_simple.Rmd # Reproducible GLMM analysis report
│   ├── glmm_report.html # Generated HTML report with dark mode
│   ├── render_report.R # Script to regenerate the report
│   ├── dark-mode.css # Dark mode styling
│   ├── dark-mode-header.html # Dark mode toggle functionality
│   └── README.md # Documentation for the docs directory
└── README.md

Experimental Design

Task Overview

• Participants: 202 subjects
• Trials: 250 trials per subject (53,592 total trials)
• Task: Predict emotional faces (Happy/Angry) based on visual cues
• Paradigm: Reversal learning with changing cue:stimulus associations

Cue:Stimulus Associations

• Predictive blocks (0.8 probability): Cue strongly predicts face emotion
• Non-predictive blocks (0.5 probability): Cue provides no predictive information
• Antipredictive blocks (0.2 probability): Cue predicts opposite face emotion

Key Manipulations

• Stimulus Noise: High noise (ambiguous) vs low noise (clear) faces
• Cue Validity: Valid vs invalid vs non-predictive trials
• Probability Reversals: Learning blocks with changing contingencies

Analysis Pipeline

The analysis is designed as a series of simple, standalone scripts that can be run in sequence:

1. Data Import and Cleaning

source("code/preprocessing/01_import_and_clean_data.R")

• Loads raw CSV data (53,592 trials)
• Cleans and recodes variables
• Filters out error trials (RawConfidence == 888)
• Creates the main data frame df with 48,199 trials

2. Basic Analysis and Plots

source("code/analysis/01_basic_analysis_and_plots.R")

• Creates descriptive statistics
• Generates basic plots (accuracy, RT, confidence distributions)
• Saves figures to results/figures/basic_analysis/

3. GLMM Analysis (Main Analysis)

source("code/analysis/02_run_glmm_main.R")

• Accuracy Model: High noise trials only, predicts correct/incorrect responses
• Choice Model: High noise trials only, predicts Happy vs Angry choices
• Response Time Model: All trials, predicts response times
• Confidence Model: All trials, predicts confidence ratings (0-1 scale)

4. ANOVA Analysis

source("code/analysis/03_run_anova_analyses.R")

• Runs traditional ANOVA analyses
• Provides additional statistical tests

5. Reversal Learning Analysis

source("code/analysis/05_reversal_learning_plots.R")

• Analyzes learning dynamics around reversals
• Generates learning curve plots

6. Computational Modeling

source("code/modeling/04_run_computational_models.R")

• Fits reinforcement learning models using hBayesDM
• Uses ug_delta model for uncertainty-guided learning
• Saves model results to results/models/

GLMM Analysis Report

A comprehensive, reproducible report of the GLMM analysis is available in the docs/ directory:

Interactive HTML Report

• File: docs/glmm_report.html
• Features:
  • Dynamic tables with model coefficients and fit statistics
  • Embedded figures from existing analysis
  • Dark mode toggle for comfortable viewing
  • Professional styling with interactive elements
  • Self-contained (no external dependencies)

Reproducible Source

• File: docs/glmm_report_simple.Rmd
• Features:
  • Loads pre-estimated models from results/models/
  • Extracts coefficients and fit statistics programmatically
  • Embeds existing figures from results/figures/glmm_models/
  • Updates automatically when analysis changes

Regeneration

To update the report after analysis changes:

# From project root
cd docs
Rscript render_report.R

View Online: The report is hosted on GitHub Pages and can be viewed directly in your browser.

Repository Metadata

• Topics: fMRI, neuroscience, psychology, statistics, GLMM, mixed-models, confidence, emotion-recognition, reversal-learning, R, reproducible-research
• Description: CWT fMRI Study: Confidence Weighting Task Analysis with interactive GLMM reports
• Website: https://embodied-computation-group.github.io/CWT_fmri/glmm_report.html

Key Variables

Experimental Design

• SubNo: Subject identifier (factor)
• TrialNo: Trial number within session (1-250)
• TrialValidity: Whether cue correctly predicts face emotion (0=Invalid, 1=Valid)
• StimNoise: Noise level in stimulus (0=low, 1=high)
• TrialsSinceRev: Number of trials since last reversal

Stimulus Variables

• CueImg: Cue image type (0/1)
• FaceEmot: Actual face emotion (0=Angry, 1=Happy)
• TrialValidity2: Extended validity (Valid/Invalid/non-predictive)

Response Variables

• Accuracy: Binary accuracy (0=miss, 1=hit)
• ResponseRT: Response time in seconds
• RawConfidence: Confidence rating (0-100, converted to 0-1)
• FaceResponse: Participant's choice (0=Angry, 1=Happy)

Processed Variables

• TrialsSinceRev_scaled: Z-scored within subjects
• TrialValidity2_numeric: Recoded as 1/0/-1
• FaceResponse_numeric: Numeric version for modeling (0=Angry, 1=Happy)

Quick Start Guide

For New Users

1. Clone the repository and open in RStudio
2. Install dependencies (see Dependencies section below)
3. Run the main analysis pipeline:

   source("code/preprocessing/01_import_and_clean_data.R")
   source("code/analysis/01_basic_analysis_and_plots.R")
   source("code/analysis/02_run_glmm_main.R")

For Researchers Taking Over

1. Review experimental design: Read docs/experimental_design.md
2. Check current results: Browse results/figures/ and results/models/
3. Run validation: Execute code/analysis/02_validate_glmm.R
4. Modify analyses: Edit scripts in code/analysis/ (backup versions in backup/)

Dependencies

Required R Packages

# Core analysis
library(tidyverse)
library(lme4)
library(lmerTest)
library(glmmTMB)

# Specialized models
library(ordinal)
library(DHARMa)

# Visualization and reporting
library(sjPlot)
library(sjmisc)
library(ggplot2)

# Computational modeling
library(hBayesDM)

System Requirements

• R (version 4.0+ recommended)
• RStudio (for interactive development)
• MATLAB (optional, for additional .mat file processing)

Output Files

Generated Plots

• results/figures/glmm_models/: GLMM model predictions
• results/figures/learning_curves/: Learning dynamics
• results/figures/confidence_analysis/: Confidence-specific analyses
• results/figures/basic_analysis/: Descriptive statistics
• results/figures/rl_models/: Reinforcement learning model plots
• results/figures/predictive_checks/: Model validation

Model Results

• results/models/: RDS files containing fitted models
• results/models/glmm_model_summaries.txt: Complete model summaries

Tables

• results/tables/: CSV and HTML output tables

Reports

• docs/glmm_report.html: Interactive HTML report with dark mode
• docs/glmm_report_simple.Rmd: Reproducible R Markdown source
• docs/render_report.R: Script to regenerate the report

Notes

• Script Safety: Original scripts are preserved in code/analysis/backup/
• White Backgrounds: All plots use white backgrounds for publication
• Modular Design: Each script is standalone and can be run independently
• Progress Tracking: Scripts include progress messages for debugging
• Documentation: See README.ai.md for AI assistant context

Citation

If you use this code in your research, please cite the original study and include a reference to this repository.