Quantitative Analysis Pipeline: Emotion Encoded

This page hosts the quantitative data pipeline and statistical modeling notebooks for Emotion Encoded, an independent research initiative investigating cognitive-emotional barriers, metacognition, and trust calibration in artificial intelligence across medical, legal, and technical sectors.

To analyze structured survey responses and expert interview metrics, the analysis was constructed using Python within a Jupyter Lab environment, leveraging core data science and visualization libraries (pandas, matplotlib, and seaborn) to process datasets and execute logistic regression models.

Technical Stack & Libraries

  • Environment Jupyter Lab (Interactive Python Notebooks)
  • Data Manipulation & Cleaning pandas, numpy
  • Statistical Modeling statsmodels / scikit-learn (Logistic Regression)
  • Data Visualization matplotlib, seaborn

Analysis Pipeline & Methodology

1. Data Ingestion & Preprocessing (pandas)

` Raw quantitative survey data from normal, ordinary people in SKN are ingested, cleaned, and structured into dataframes.

  • Handling missing values and standardizing categorical response scales regarding trust calibration, automation bias, and perceived algorithmic risk.
  • Constructing binary and ordinal target variables to measure clinical/legal resistance versus adoption intent.

2. Exploratory Data Analysis & Visualization (matplotlib & seaborn)

Before fitting predictive models, exploratory data analysis is performed to uncover underlying distributions and behavioral clustering:

  • Distribution Plots: Visualizing frequency distributions of algorithm aversion and status quo bias scores across different professional cohorts.
  • Correlation Heatmaps: Identifying multicollinearity among independent variables (e.g., workload fatigue, perceived technical transparency, and reliance on automated suggestions).

3. Logistic Regression Modeling

To evaluate the predictors of clinical and professional AI adoption, binary logistic regression models are fitted using the survey metrics:

  • Model Specification: Estimating the log-odds of a professional rejecting or adopting an AI tool as a function of cognitive load, data skepticism, and explainability requirements.
  • Coefficient Interpretation: Extracting odds ratios (OR) to quantify how specific psychological barriers (such as the accountability gap or fear of epidemiological mismatch) impact decision thresholds.

Qualitative Analysis & Thematic Coding Pipeline: Emotion Encoded

This page hosts the qualitative data organization and thematic analysis framework for Emotion Encoded, an independent research initiative investigating cognitive-emotional barriers, metacognition, and trust calibration in artificial intelligence across medical, legal, and technical sectors.

To process and synthesize qualitative insights from over 45 expert interviews with healthcare specialists, attorneys, and behavioral scientists, this workflow leverages Python within a Jupyter Lab environment, utilizing data manipulation libraries to parse, categorize, and visualize qualitative thematic distributions.

Technical Stack & Workflow

  • Environment Jupyter Lab (Interactive Python Notebooks)
  • Data Processing & Parsing pandas, re (Regular Expressions) for structuring interview transcripts and coded excerpts.
  • Data Visualization matplotlib, seaborn for mapping code frequencies, co-occurrence matrices, and thematic distributions across professional cohorts.

Qualitative Methodology & Pipeline

1. Transcript Parsing & Data Ingestion (pandas)

Raw qualitative notes and interview transcripts are cleaned, standardized, and ingested into structured DataFrames.

  • Metadata Tagging: Categorizing excerpts by participant profession (e.g., physicians, surgeons, clinical psychologists, legal experts), regional context (Caribbean SIDS), and institutional background.
  • Corpus Management: Structuring transcripts to enable programmatic text searching and sentiment tagging across large interview sets.

2. Thematic Coding Analysis (Taguette & Python Integration)

Following qualitative coding workflows (incorporating Taguette export data), the notebook processes applied thematic codes to evaluate core psychological and cognitive barriers:

  • Algorithm Aversion & Automation Bias: Tracking qualitative markers where clinicians or professionals describe resistance to flawed machine logic or over-reliance on automated outputs.
  • Status Quo Bias & Workflow Inertia: Quantifying the recurrence of statements favoring traditional manual methods over digital integration.
  • The Accountability Gap & Explainability (XAI): Analyzing qualitative quotes surrounding professional autonomy, lack of transparent model reasoning, and the "black box" dilemma.
  • Regional Data Skepticism: Parsing concerns regarding "Epidemiological Mismatch" and the applicability of imported Global North AI models within Caribbean healthcare systems.

3. Visualizing Thematic Frequencies (matplotlib & seaborn)

  • Code Frequency Distributions: Generating bar plots and categorical charts illustrating the prevalence of specific behavioral themes across different medical and legal specialties.
  • Thematic Co-Occurrence: Constructing heatmaps to examine how psychological barriers (such as algorithm aversion and status quo bias) intersect with structural concerns like data scarcity and explainability requirements.