Behavioral Data Analysis of Memory Fidelity (Mnemonic Similarity Task)

A data analysis project studying how event boundaries influence recognition memory and pattern separation using behavioral experiment logs from the Mnemonic Similarity Task (MST).

Overview

This project analyzes behavioral data from a replication and extension of Morse et al. (2023), using the Mnemonic Similarity Task (MST) to study how event segmentation affects memory.

The core goals were to:

• clean and structure raw experimental logs
• engineer trial-level behavioral variables
• compute memory-related indices
• run statistical analyses to test hypotheses about encoding and recognition

Dataset

The dataset consists of behavioral logs collected from 160 participants across three experimental conditions:

• Both (n = 51)
• Item_Only (n = 56)
• Task_Only (n = 53)

Each participant produced:

• one task/encoding CSV
• one test-phase CSV

The experiment included:

• 40 events of 7 items each during encoding
• 150 test images per participant:
  • 60 targets
  • 60 lures
  • 30 foils

Problem Statement

The project investigates whether event boundaries help or hurt memory.

In particular, it tests whether:

• post-boundary items are slower to encode
• post-boundary items have poorer recognition
• event boundaries affect lure discrimination / pattern separation
• lure similarity bins show a graded difficulty effect

Data Processing Pipeline

The preprocessing workflow included:

• parsing raw participant CSV logs
• classifying stimuli as target / lure / foil
• deriving event position labels: pre-boundary, mid, post-boundary
• computing reaction times from multiple response fields
• assigning lure similarity bins
• validating dataset integrity through manipulation checks and summary plots

Key Metrics

Two main behavioral metrics were analyzed:

• Recognition Memory Index (REC)
  Measures target recognition corrected by foil false alarms

• Lure Discrimination Index (LDI)
  Measures the ability to distinguish similar lures from previously seen items

Methods

Statistical analyses included:

• Generalized Linear Mixed Models (GLMMs)
• Repeated-measures ANOVA
• Holm-Bonferroni corrected comparisons
• Bayesian paired t-tests

Main Findings

• Post-boundary items were slower to encode
• Recognition memory (REC) dropped for post-boundary items
• Lure discrimination (LDI) did not significantly differ across event positions
• LDI increased monotonically from similarity Bin 1 to Bin 5
• Boundary effects differed across experimental conditions

Repository Structure

• brsm_data/ — raw and processed data files
• analysis scripts / notebooks — preprocessing, modeling, and visualization
• README.md — project overview

Tech Stack

• R
• lme4
• emmeans
• ggplot2

Why this project matters

This repository demonstrates an end-to-end behavioral data analysis workflow:

1. cleaning messy experimental logs
2. engineering trial-level variables
3. validating dataset quality
4. applying mixed-effects statistical models
5. interpreting human memory behavior through data