Master's Thesis Research Project — An empirical study investigating whether and how gender shapes the way people prompt AI coding assistants, and whether these differences translate into measurable differences in generated code quality.

This repository contains the full analytical pipeline for a study in which participants solved real-world programming tasks using LLMs (ChatGPT, Claude). Their conversations were collected via an online survey, stored in a structured database, and analysed across three dimensions:

1. Prompting Style — Do cisgender men and women differ in how they write prompts to LLMs (length, formality, politeness markers, sentence structure, request framing)?
2. Code Quality — Does the gender of the prompter correlate with the quality of the LLM-generated code?
3. Gender Predictability — Can a machine learning model reliably predict a user's gender from their prompts alone?

Online Survey (LimeSurvey)
        │
        ▼
Playwright Scraper ──► Raw chat HTML from ChatGPT / Claude share links
        │
        ▼
Importer ──► SQLite (giicg.db)
        │
        ├── Language Detection  (xlm-roberta-base-language-detection)
        ├── Translation DE/IT → EN  (HuggingFace Helsinki-NLP)
        ├── Spelling Correction  (oliverguhr/spelling-correction-english-base)
        └── Contraction Expansion
        │
        ▼
Prompt Parser (GPT-4o via LangChain)
    Segments each user message into:
      conversational | code | other
        │
        ▼
Analysis Notebooks
    ├── Linguistic analyses (spaCy, statsmodels, scipy, pingouin)
    ├── Code quality  (Pylint, Radon)
    └── Gender prediction  (RoBERTa fine-tune, LIME)

• Group comparison: Welch's t-test, Mann-Whitney U, Fisher's exact test
• Effect sizes: Cohen's d, odds ratio
• Multiple testing correction: Bonferroni
• Normality checks: Shapiro-Wilk
• Explainability: LIME (Local Interpretable Model-agnostic Explanations)

pip install -r requirements.txt

python -m spacy download en_core_web_sm

Create a .env file in the project root:

OPENAI_API_KEY=your_openai_key_here
ANTHROPIC_API_KEY=your_anthropic_key_here

jupyter notebook

This project is part of a Master's thesis. All code is provided for academic transparency. The survey dataset is not redistributed due to participant privacy.